A novel methodology is proposed to robustly map oil seed rape (OSR) flowering phenology from time series generated from the Copernicus Sentinel-1 (S1) and Sentinel-2 (S2) sensors. The time series are averaged at parcel level, initially for a set of 229 reference parcels for which multiple phenological observations on OSR flowering have been collected from April 21 to May 19, 2018. The set of OSR parcels is extended to a regional sample of 32,355 OSR parcels derived from a regional S2 classification. The study area comprises the northern Brandenburg and Mecklenburg-Vorpommern (N) and the southern Bavaria (S) regions in Germany. A method was developed to automatically compute peak flowering at parcel level from the S2 time signature of the Normalized Difference Yellow Index (NDYI) and from the local minimum in S1 VV polarized backscattering coefficients. Peak flowering was determined at a temporal accuracy of 1 to 4 days. A systematic flowering delay of 1 day was observed in the S1 detection compared to S2. Peak flowering differed by 12 days between the N and S. Considerable local variation was observed in the N-S parcel-level flowering gradient. Additional in-situ phenology observations at 70 Deutscher Wetterdienst (DWD) stations confirm the spatial and temporal consistency between S1 and S2 signatures and flowering phenology across both regions. Conditions during flowering strongly determine OSR yield, therefore, the capacity to continuously characterize spatially the timing of key flowering dates across large areas is key. To illustrate this, expected flowering dates were simulated assuming a single OSR variety with a 425 growing degree days (GDD) requirement to reach flowering. This GDD requirement was calculated based on parcel-level peak flowering dates and temperatures accumulated from 25-km gridded meteorological data. The correlation between simulated and S2 observed peak flowering dates still equaled 0.84 and 0.54 for the N and S respectively. These Sentinel-based parcel-level flowering parameters can be combined with weather data to support in-season predictions of OSR yield, area, and production. Our approach identified the unique temporal signatures of S1 and S2 associated with OSR flowering and can now be applied to monitor OSR phenology for parcels across the globe.

The natural world has multiple, sometimes conflicting, sometimes synergistic, values to society when viewed through the lens of the Sustainable Development Goals (SDGs), Spatial mapping of nature's contributions to the SDGs has the potential to support the implementation of SDG strategies through sustainable land management and conservation of ecosystem services. Such mapping requires a range of spatial data. This paper examines the use of remote sensing and spatial ecosystem service modelling to examine nature's contribution to targets under SDG 6, also highlighting synergies with other key SDGs and trade-offs with agriculture. We use a wide range of remotely sensed and globally available datasets (for land cover, climate, soil, population, agriculture) alongside the existing and widely used spatial ecosystem services assessment tool, Co$tingNature. With these we identify priority areas for sustainable management to realise targets under SDG 6 (water) at the country scale for Madagascar and at the basin scale for the Volta basin, though the application developed can be applied to any country or major basin in the world. Within this SDG 6 priority areas footprint, we assess the synergies and trade-offs provided by this land for SDG 15 (biodiversity) and SDG 13 (climate action) as well as SDG 2 (zero hunger). Results highlight the co-benefits of sustainably managing nature's contribution to SDG 6, such as the protection of forest cover (for SDG target 15.2), carbon storage as a contribution to the Paris climate agreement and nationally determined contributions (SDG 13) and biodiversity (for SDG target 15.5) but also trade-offs with the zero hunger goal (for SDG 2). Such analyses allow for better understanding of land management requirements for realising multiple SDGs through protection and restoration of green infrastructure. We provide a freely available tool, within the Co$tingNature platform, based on a variety of remotely sensed products, that can be used by SDG practitioners to carry out similar analyses and inform decision-making at national or sub-national levels globally.

Accurate remote assessment of phytoplankton chlorophyll-a (Chla) concentration in turbid case-2 waters is a challenge, owing largely to terrestrial substances (such as minerals and humus) that are optically significant but do not co-vary with phytoplankton. Here, we propose an improved Quasi-Analytical Algorithm (QAA) (denoted as TC2) for retrieving Chla concentrations from remote sensing reflectance (Rrs(λ)) which can be applied to Sentinel-3 Ocean and Land Colour Instrument (OLCI) images in turbid case-2 waters. TC2 has two main extensions when compared with QAA. First, TC2 makes an additional assumption to separate the total non-water absorption at 665 nm (anw(665)) into phytoplankton absorption (aph(665)) and yellow matter (aym(665)), which is the sum of colored dissolved matter (CDOM) and detritus. Second, for selecting the position of the near-infrared (NIR) band which is used to estimate the signal of total backscattering coefficient (bb(λ0)) at QAA reference band (λ0), we take into account the assumption that the absorption of pure water should be dominant at this band, as well as the impact of the signal-to-noise ratio (SNR) in the NIR band on the Chla concentration estimating model. When applied to in situ Rrs(λ) and OLCI match-up Rrs(λ) data in this study, TC2 provided more accurate Chla estimation than previous Cha concentration retrieval algorithms for turbid case-2 waters. TC2 has the potential for use as a simple and effective algorithm for monitoring Chla concentrations in the turbid case-2 waters at a global scale from space.

Structure-from-motion (SfM) and multi-view stereo (MVS) algorithms coupled with the use of unmanned aerial vehicles (UAVs) have become a popular tool in the geosciences for modelling complex landscapes on-demand allowing for high-resolution topographic change-detection studies to be conducted at minimal cost. To identify the effects of UAV image orientation on the accuracy of SfM-MVS 3D surface models, we tested four different UAV image acquisition schemes that incorporated both nadir and oblique imagery of an agricultural field. The coupling of nadir and oblique imaging angles led to the highest surface model accuracy in the absence of ground control points (GCPs; vertical RMSE: 0.047 m, horizontal RMSE: 0.019 m), while with a normative distribution of GCPs the nadir-only image sets had similar accuracy metrics (vertical RMSE 0.028 m, horizontal RMSE 0.017 m) to surface models generated with nadir and oblique imaging angles (vertical RMSE 0.028 m, horizontal RMSE 0.013 m). Homologous keypoint matching between nadir and oblique imagery was poor when the survey conditions were bright and the surface texture of the field was homogeneous, leading to broad-scale vertical noise in the generated surface models. Results indicate that a nadir-only image set accompanied with a dense deployment of GCPs is the most ideal for SfM-MVS agricultural 3D surface reconstructions. The diachronic analysis of surface models generated from nadir-only image sets were able to detect surface-change >0.040 m in depth (i.e., rill and gully erosion, depositional zones) and were comparable to results obtained from a terrestrial laser scanner. Stable GCPs should be used where possible to ensure precise co-registration between subsequent UAV surveys.

The value of combining Landsat time series and airborne laser scanning (ALS) data to produce regional maps of forest structure has been well documented. However, studies are often performed over single study areas or forest types, preventing a robust assessment of the approaches that produce the most accurate estimates. Here, we use Landsat time series data to estimate forest attributes across six Canadian study sites, which vary by forest type, productivity, management regime, and disturbance history, with the goal of investigating which spectral indices and time series lengths yield the most accurate estimates of forest attributes across a range of conditions. We use estimates of stand height, basal area, and stem volume derived from ALS data as calibration and validation data, and develop random forest models to estimate forest structure with Landsat time series data and topographic variables at each site. Landsat time series predictors, which were derived from annual gap-free image composites, included the median, interquartile range, and Theil Sen slope of vegetation indices through time. To investigate the optimal time series length for predictor variables, time series length was varied from 1 to 33 years. Across all six sites, increasing the time series length led to improved estimation accuracy, however the optimal time series length was not consistent across sites. Specifically, model accuracies plateaued at a time series length of ~15 years for two sites (R2 = 0.67–0.74), while the accuracies continued to increase until the maximum time series length was reached (24–29 years) for the remaining four sites (R2 = 0.45–0.70). Spectral indices that relied on shortwave infrared bands (Tasseled Cap Wetness and Normalized Burn Ratio) were frequently the most important spectral indices. Adding Landsat-derived disturbance variables (time since last disturbance, type of disturbance) did not meaningfully improve model results; however, this finding was largely due to the fact that most recently disturbed stands did not have predictions of forest attributes from ALS, so disturbed sites were poorly represented in the models. As model accuracies varied regionally and no optimal time series length was found, we provide an approach that can be utilized to determine the optimal time series length on a case by case basis, allowing users to extrapolate estimates of forest attributes both spatially and temporally using multispectral time series data.

Temperate grasslands are globally widespread, play an important role as carbon storage, and harbor unique biodiversity. Livestock grazing is the most widespread land use in temperate grasslands, and understanding the impact of grazing on grassland ecosystems is therefore important. However, monitoring grazing pressure and how it changes is hampered by a lack of adequate tools. The Eurasian steppe belt, extending from Eastern Europe to China has experienced marked changes in grazing pressure. Most notably, livestock numbers in the steppes of Kazakhstan and Russia declined by up to 80% after the breakdown of the Soviet Union in 1991, yet how this impacted spatial patterns of grazing pressure is unclear. To address this research gap, we used all available Landsat data from 1985 to 2017 together with extensive ground reference data on grazing pressure to evaluate a broad range of spectral-temporal metrics regarding their ability to capture grazing pressure. While Tasseled Cap-based disturbance indices performed best, combining all spectral-temporal metrics in a binary random forest classification yielded a grazing class membership probability that strongly outperformed all individual metrics. This new index of grazing pressure correlated well with a range of field-based grazing indicators (e.g., number of dung piles, herbaceous biomass) and yielded highly plausible spatial patterns of grazing pressure. We used this index to reconstruct annual changes in grazing pressure across our 360,000 km2 study region, and used LandTrendr time series segmentation to identify trends in grazing pressure. Aggregated grazing pressure followed closely known trends in total livestock numbers over the time period we studied. The spatial footprint of heavy grazing was very large before 1991, but decreased by 73 (±2) % until 2017. This now leaves large areas virtually ungrazed, even in close vicinity to settlements and agricultural areas, and despite a recent recovery of livestock numbers. Our analyses uncovered previously unknown hot-spots of heavy grazing during Soviet times (e.g., around watering points). Our findings suggest potential for a further revival of the livestock sector as well as for the restoration of steppe ecosystems. More broadly, our study highlights how the Landsat archive, in combination with field data on grazing, can be used to map grazing pressure reliably across large areas and over long time spans.

Insights into the dynamics of coastlines and tidal flats at fine spatial and temporal resolutions are essential for sustainable development. Previous studies were generally conducted at relatively coarse temporal intervals, which hardly captured detailed coastal dynamics, especially in rapidly developing islands. In this study, we developed a new method to map the monthly changes in coastlines and tidal flats in the Zhoushan Archipelago during 1985–2017 using the full time series of Landsat images based on the Google Earth Engine (GEE) platform. First, we built the full time series of the Modified Normalized Difference Water Index (MNDWI). Second, we derived temporal segments of MNDWI using a binary segmentation algorithm. Third, we classified the corresponding coastal cover types (i.e., water, tidal flats, and land) for each temporal segment based on the features of MNDWI and regional tidal heights. Finally, we identified the change information including conversion types, and turning years and months. Results indicate that the proposed method can well identify turning years with an overall accuracy of 90% and map coastal cover types with overall accuracies of 89–94% in 1985 and 87–92% in 2017. Significant coastline expansions and declines in tidal flats were found in the study area. The areas of water and tidal flats decreased by 6% and 10% during 1985–2017, respectively, while the land area increased by 18%. The land reclamation was accelerated in the recent decade, and mainly occurred on the medium-large islands, their surrounding small islands, and the islands close to the mainland. The proposed framework based on the GEE platform is transferable to investigate coastal dynamics in other areas. The derived information of changes in coastlines and tidal flats is of great use for sustainable management and ecological studies in coastal areas.

In the face of a sparse global river gauging station network in decline, new approaches are needed to reconstruct and monitor river discharge from satellite observations. Where in-situ river discharge measurements are not available, it may be possible to use discharge estimates from a hydrological model, provided the model simulations are of sufficient quality, to construct satellite-based discharge gauging. We tested this approach by developing model- and gauge-based satellite gauging reaches (SGRs) using 0.05° MODIS optical remote sensing at ~10,000 gauged and ~370,000 ungauged river reaches globally. Model-based SGRs are aimed to infer temporal flow patterns and reflect unusually high or low river discharge behavior (i.e. flood or drought conditions), if not necessarily absolute discharge volumes. The model-based SGRs achieved a discharge prediction skill that was often similar to gauge-based SGRs, and sometimes better than the model itself. Our results showed promising opportunities to develop model-based SGRs in sparsely gauged basins in South America, Africa, and Asia. We selected river reaches, with mean widths ranging from 67 to 3105 m, representing both poor and successful SGRs in different environments for case studies to analyze conditions for successful SGR development. River size and morphology were the main factors determining the performance of SGRs. Wide channels with strong temporal variations, broad floodplains and multiple braided or anastomosing channels provided the best conditions for SGRs. The probability of constructing a successful SGR could be predicted from high-resolution inundation summary data available globally, and can thus be predicted anywhere. Ongoing increases in the spatial and temporal resolution of remote sensing will further increase the number of river reaches for which satellite-based discharge gauging will become possible.

Satellite radar altimetry has been widely used in the monitoring of water levels of lakes, rivers and wetlands in the past decades. The conventional pulse-limited radar altimeters have a relatively large ground footprint, which limits their capability to retrieve surface elevation information over small and medium-sized water bodies. A new generation of satellite radar altimeter system, a dual-frequency SAR radar altimeter (SRAL) onboard the Copernicus Sentinel-3 satellite, has produced densely sampled elevation measurements with a smaller footprint for the Earth's surfaces since June 2016, owing to the Delay-Doppler processing technique. Four standard SRAL SAR altimetry waveform retracking algorithms (known as retrackers) have been designed to retrieve elevation measurements for different types of surfaces: Ice-Sheet retracker for polar ice sheets, SAMOSA-3 retracker for open ocean and coastal zones, OCOG retracker for sea-ice margins, and Sea-Ice retracker for sea ice. In this research, we evaluated the performances of the Sentinel-3 SRAL SAR altimetry retrackers over lakes, particularly over seasonally ice-covered lakes in one hydrological cycle. For 15 lakes and reservoirs with different sizes and at varying latitudes in the northern hemisphere, we compared the lake water levels estimated by each of standard SRAL SAR retrackers against in-situ water level measurements for different seasons (a full hydrologic cycle) during 2016–2017. Our evaluation shows that Sea-Ice retracker was unable to provide continuous estimates of lake water levels, as a result of the high rate of missing data. Although the precision and relative accuracy of lake water level estimates from these three standard SRAL SAR retrackers are similar, the SAMOSA-3 retracker has the least bias in comparison with ground-based gauge measurements. When the lakes in the mid- and high-latitude regions were covered by ice in the winter season, these three standard SAR retrackers generated erroneous lake water level measurements, significantly lower than the true lake water levels recorded by in-situ gauge stations. The measurement errors of these three standard retrackers increase with the growth of the lake ice thickness. To address the negative effect of the seasonal ice cover, we developed a new bimodal correction algorithm. We demonstrate that our bimodal correction algorithm can retrieve the ice thickness and reliably estimate water levels for the ice-covered lakes in winter, hence enabling the generation of temporally consistent lake water level measurements throughout all seasons for lake hydrological analysis.

Several decade-long satellite retrievals of solar-induced chlorophyll fluorescence (SIF) have become available during the past few years, but understanding the long-term dynamics of SIF and elucidating its co-variation with historical gross primary production (GPP) remains a challenge. Part of the challenge is due to the lack of direct comparability among these SIF products as they are derived from various satellite platforms with different retrieval methods, instruments characteristics, overpass time, and viewing-illumination geometries. This study presents a framework that circumvents these discrepancies and allows the harmonization of SIF products from multiple instruments to achieve long-term coverage. We demonstrate this framework by fusing SIF retrievals from SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) and Global Ozone Monitoring Experiment 2 (GOME-2) onboard MetOp-A developed at German Research Center for Geosciences (GFZ). We first downscale both original SIF datasets from their native resolutions to 0.05° (SIF¯GOME2_005 and SIF¯SCIA_005 respectively) using machine learning (ML) algorithms imposed with regionalization constraints to account for the varying relationships between predictors and SIF in space and time. We then apply the cumulative distribution function (CDF) matching technique to correct the offset between SIF¯GOME2_005 and SIF¯SCIA_005 inherited from the original instrumental discrepancies to generate a harmonized SIF time series from 2002 to present (SIF¯005). Finally, we quantify the uncertainty of SIF¯005. SIF¯005 is validated with 1) the original retrievals to ensure the spatial and temporal variabilities are preserved, 2) airborne SIF derived from the Chlorophyll Fluorescence Imaging Spectrometer (CFIS, R2 = 0.73), and 3) ground-based SIF measurements at a subalpine coniferous forest (R2 = 0.91). The SIFyield derived from SIF¯005 has high seasonal consistency with the ground measurements (R2 = 0.93), suggesting that the harmonized product SIF¯005 carries physiological information beyond the absorbed photosynthetically active radiation. Additionally, SIF¯005 has a good capability for large-scale stress monitoring as demonstrated with several major historical drought and heatwave events. The framework developed in this study sets the stage for future development of even more advanced SIF products from all SIF-capable satellite platforms once issues related to inter-sensor calibration are resolved and SIF physiology is better understood.

Coral reefs are the foundation of productive ecosystems in the global, tropical oceans and are under threat from a variety of local to global scale stressors. Satellite imagery provides a tool to identify and understand the processes that control coral reef degradation, however due to the dynamic nature of seawater constituents, current spaceborne multispectral sensors cannot reliably discriminate between the many coral reef benthic classes necessary to detect change. Hyperspectral imagers may provide sufficient spectral resolution to estimate water column properties and differentiate benthic classes, however, the effects of depth, seawater constituents, and classification algorithm on the accuracy of benthic classifications have not been systematically assessed. Here, we simulate the ability of a spaceborne hyperspectral imager to accurately map fractional cover of coral reef benthic classes under a variety of conditions. Benthic reflectance is simulated by combining pure reflectance spectra of coral, algae, and sand and projecting these mixed spectra through a fully crossed set of water columns. We then use a semi-analytical optimization procedure to estimate the water column properties and multiple endmember spectral mixture analysis to estimate the fractional cover of the benthic classes using many independent endmember spectra. We compare our estimated benthic class fractions to the original, actual fractions used to produce the mixed coral reef spectra to quantify several measures of error. We found that multiple endmember spectral mixture analysis decreases fractional retrieval error, which is also reduced when the first derivative of the mixed and endmember spectra is used prior to unmixing. The estimation of fractional benthic class cover is most accurate for depths ≤3 m for most water conditions. Depths ≥5 m should be classified only if chlorophyll and sediment concentration are <0.1 mg m−3 and <0.1 g m−3, respectively. Our results indicate that the fractional cover of coral and algae should be at least 25% for accurate benthic class estimates (mean relative error < 50%), however there will be many ways to leverage the repeat measurements of a hyperspectral satellite sensor, such as a stable depth retrievals and benthic cover estimates, to produce more accurate and useful fractional cover data. We show how this simulation analysis can be used to generate maps of predicted benthic cover fractional retrieval uncertainty across a coral reef system using aerial hyperspectral imagery acquired over Hawaii, USA, although reef-specific, within pixel variations in depth and benthic class complexity should be considered.

Recently, there have been significant efforts in the integration of in-situ and satellite observations for effective monitoring of coastal areas (e.g., the Copernicus program of the European Space Agency). In this study, a 15-year diurnal variation of Water Constituent Concentrations (WCCs) was retrieved from multi-sensor satellite images and in-situ hyperspectral measurements using Radiative Transfer (RT) modeling in the Dutch Wadden Sea. The existing RT model 2SeaColor was inverted against time series of in-situ hyperspectral measurements of water leaving reflectances (Rrs [sr−1]) for the simultaneous retrieval of WCCs (i.e., Chlorophyll-a (Chla), Suspended Particulate Matter (SPM), Dissolved Organic Matter (CDOM)) on a daily basis between 2003 and 2018 at the NIOZ jetty station (the NJS) located in the Dutch part of the Wadden Sea. At the same time, the existing coupled atmosphere-hydro-optical RT model MOD2SEA was used for the simultaneous retrieval of WCCs from time series of multi-sensor satellite images of the MEdium Resolution Imaging Spectrometer (MERIS) onboard ENVISAT, Multispectral Instrument (MSI) onboard Sentinel-2 and Ocean and Land Colour Instrument (OLCI) onboard Sentinel-3 between 2003 and 2018 over the Dutch Wadden Sea. At the NJS, a direct comparison (Taylor diagram and statistical analysis) showed strong agreement between in-situ and satellite-derived WCC values (Chla: R2 ≥ 0.70, RMSE ≤7.5 [mg m−3]; SPM: R2 ≥ 0.72, RMSE ≤5.5 [g m−3]; CDOM absorption at 440 nm: R2 ≥ 0.67, RMSE ≤1.7 [m−1]). Next, the plausibility of the spatial variation of retrieved WCCs over the study area was evaluated by generating maps of Chla [mg m−3], SPM [g m−3], and CDOM absorption at 440 nm [m−1] from MERIS and OLCI images using the MOD2SEA model. The integration of the spatio-temporal WCC data obtained from in-situ measurements and satellite images in this study finds applications for the detection of anomaly events and serves as a warning for management actions in the complex coastal waters of the Wadden Sea.

Recent advances in L-band passive microwave remote sensing provide an unprecedented opportunity to monitor soil moisture at ~40 km spatial resolution around the globe. Nevertheless, retrieval of the accurate high spatial resolution soil moisture maps that are required to satisfy hydro-meteorological and agricultural applications remains a challenge. Currently, a variety of downscaling, otherwise known as disaggregation techniques have been proposed as the solution to disaggregate the coarse passive microwave soil moisture into high-to-medium resolutions. These techniques take advantage of the strengths of both the passive microwave observations of soil moisture having low spatial resolution and the spatially detailed information on land surface features that either influence or represent soil moisture variability. However, such techniques have typically been developed and tested individually under differing weather and climate conditions, meaning that there is no clear guidance on which technique performs the best. Consequently, this paper presents a quantitative assessment of the existing radar-, optical-, radiometer-, and oversampling-based downscaling techniques using a singular extensive data set collected specifically for that purpose, being the Soil Moisture Active Passive Experiment (SMAPEx)-4 and -5 airborne field campaigns, and the OzNet in situ stations, to determine the relative strengths and weaknesses of their performances. The oversampling-based soil moisture product best captured the temporal and spatial variability of the reference soil moisture overall, though the radar-based products had a better temporal agreement with airborne soil moisture during the short SMAPEx-4 period. Moreover, the difference between temporal analysis of products against in situ and airborne soil moisture reference data sets pointed to the fact that relying on in situ measurements alone is not appropriate for validation of spatially enhanced soil moisture maps.

In the face of the growing challenges brought about by human activities, effective planning and decision-making in biodiversity and ecosystem conservation, restoration, and sustainable development are urgently needed. Ecological models can play a key role in supporting this need and helping to safeguard the natural assets that underpin human wellbeing and support life on land and below water (United Nations Sustainable Development Goals; SDG 15 & 14). The urgency and complexity of safeguarding forest (SDG 15.2) and mountain ecosystems (SDG 15.4), for example, and halting decline in biodiversity (SDG 15.5) in the Anthropocene requires a re-envisioning of how ecological models can best support the comprehensive assessments of biodiversity and its change that are required for successful action. A key opportunity to advance ecological modeling for both predictive and explanatory purposes arises through a collaboration between ecologists and the Earth observation community, and a close integration of remote sensing and species distribution models. Remote sensing products have the capacity to provide continuous spatiotemporal information about key factors driving the distribution of organisms, therefore improving both the use and accuracy of these models for management and planning. Here we first survey the literature on remote sensing data products available to ecological modelers interested in improving predictions of species range dynamics under global change. We specifically explore the key biophysical processes underlying the distribution of species in the Anthropocene including climate variability, changes in land cover, and disturbances. We then discuss potential synergies between the ecological modeling and remote sensing communities, and highlight opportunities to close the data and conceptual gaps that currently impede a more effective application of remote sensing for the monitoring and modeling of ecological systems. Specific attention is given to how potential collaborations between the two communities could lead to new opportunities to report on progress towards global agendas - such as the Agenda 2030 for sustainable development of the United Nations or the Post-2020 Global Biodiversity Framework of the Convention for Biological Diversity, and help guide conservation and management strategies towards sustainability.

The kappa coefficient is not an index of accuracy, indeed it is not an index of overall agreement but one of agreement beyond chance. Chance agreement is, however, irrelevant in an accuracy assessment and is anyway inappropriately modelled in the calculation of a kappa coefficient for typical remote sensing applications. The magnitude of a kappa coefficient is also difficult to interpret. Values that span the full range of widely used interpretation scales, indicating a level of agreement that equates to that estimated to arise from chance alone all the way through to almost perfect agreement, can be obtained from classifications that satisfy demanding accuracy targets (e.g. for a classification with overall accuracy of 95% the range of possible values of the kappa coefficient is −0.026 to 0.900). Comparisons of kappa coefficients are particularly challenging if the classes vary in their abundance (i.e. prevalence) as the magnitude of a kappa coefficient reflects not only agreement in labelling but also properties of the populations under study. It is shown that all of the arguments put forward for the use of the kappa coefficient in accuracy assessment are flawed and/or irrelevant as they apply equally to other, sometimes easier to calculate, measures of accuracy. Calls for the kappa coefficient to be abandoned from accuracy assessments should finally be heeded and researchers are encouraged to provide a set of simple measures and associated outputs such as estimates of per-class accuracy and the confusion matrix when assessing and comparing classification accuracy.

Earth remote sensing using reflected GNSS signals is currently an emerging trend especially in ocean surface wind measurements. Unlike the existing scatterometer missions, GNSS-R uses L-Band navigation signals that can penetrate through clouds and rain. Rain may have a negligible impact on the transmitted signal in terms of path attenuation at this wavelength. However, there are other effects due to rain, such as changes in surface roughness and rain induced local winds, which can significantly alter the measurements. Currently, there is no observation-based characterization of all possible impacts of rain on radar forward scatter, which is the nature of operation of GNSS-R missions. In this study, we propose a 3-fold rain model which accounts for attenuation, surface effects of rain and rain induced local winds. We utilize the large dataset of measurements made by the CYGNSS mission to separate these different effects of rain. The attenuation model suggests that a total of at least 96% transmissivity exists at L-Band up to a rain rate of 30 mm/h. A perturbation model is used to characterize the other rain effects. It suggests that rain is accompanied by an overall reduction in the scattering cross-section of the ocean surface and, most importantly, this effect is observed only up to surface wind speeds of 15 m/s, beyond which the gravity capillary waves dominate the scattering in the quasi-specular direction. Observations also suggest that, at very low wind speeds, the lower bound on wavenumber of the portion of the surface roughness spectrum that influences the measurements deviates from the geometric optics approximation normally used. This work binds together several rain-related phenomena and enhances our overall understanding of rain effects on GNSS-R measurements.

Monitoring irrigation is essential for an efficient management of water resources in arid and semi-arid regions. We propose to estimate the timing and the amount of irrigation throughout the agricultural season using optical and thermal Landsat-7/8 data. The approach is implemented in four steps: i) partitioning the Landsat land surface temperature (LST) to derive the crop water stress coefficient (Ks), ii) estimating the daily root zone soil moisture (RZSM) from the integration of Landsat-derived Ks into a crop water balance model, iii) retrieving irrigation at the Landsat pixel scale and iv) aggregating pixel-scale irrigation estimates at the crop field scale. The new irrigation retrieval method is tested over three agricultural areas during four seasons and is evaluated over five winter wheat fields under different irrigation techniques (drip, flood and no-irrigation). The model is very accurate for the seasonal accumulated amounts (R ~ 0.95 and RMSE ~ 44 mm). However, lower agreements with observed irrigations are obtained at the daily scale. To assess the performance of the irrigation retrieval method over a range of time periods, the daily predicted and observed irrigations are cumulated from 1 to 90 days. Generally, acceptable errors (R = 0.52 and RMSE = 27 mm) are obtained for irrigations cumulated over 15 days and the performance gradually improves by increasing the accumulation period, depicting a strong link to the frequency of Landsat overpasses (16 days or 8 days by combining Landsat-7 and -8). Despite the uncertainties in retrieved irrigations at daily to weekly scales, the daily RZSM and evapotranspiration simulated from the retrieved daily irrigations are estimated accurately and are very close to those estimated from actual irrigations. This research demonstrates the utility of high spatial resolution optical and thermal data for estimating irrigation and consequently for better closing the water budget over agricultural areas. We also show that significant improvements can be expected at daily to weekly time scales by reducing the revisit time of high-spatial resolution thermal data, as included in the TRISHNA future mission requirements.

Heterogeneous crop identification has been the subject of much concern, since smallholder farms less than 1 ha are the main agricultural form in many areas, especially China. Remote sensing with high spectral and spatial resolutions via aerial platforms such as unmanned aerial vehicles (UAV) provides a potential alternative technique for the monitoring of heterogeneous crops in smallholder agriculture. Although this new type of remote sensing data with high spectral and spatial resolutions provides the possibility of fine classification, it also brings some challenges, such as bands contaminated with severe noise, the nonuniform distribution of the discriminative spectral information, and the spectral variability of crops. In this study, we attempted to resolve these problems by developing a robust spectral-spatial agricultural crop mapping method based on conditional random fields (SCRF), which learns the sensitive spectral information of the crops by a spectrally weighted kernel, and uses the spatial interaction of pixels to improve the classification performance. Data from a manned aircraft platform and a UAV platform were chosen to validate the effectiveness of the proposed algorithm. The experimental results showed that the proposed algorithm can effectively use the relative utility of each spectral band to detect the bands contaminated with severe noise, and it uses the spectrally weighted kernel to consider the sensitive spectral information of the crops. The algorithm with only a spectrally weighted kernel showed an improvement of more than 4% over the classical support vector machine and random forest methods. Moreover, the spatial information was proved to be of crucial importance for crop classification, and both the object-oriented method and the proposed SCRF method can improve the classification performance in terms of both visualization and the quantitative metrics by considering the spatial information. Compared with the object-oriented method, SCRF can deliver a better classification performance, with an accuracy improvement of more than 2%.

Efficient water use assessment and irrigation management is critical for the sustainability of irrigated agriculture, especially under changing climate conditions. Due to the impracticality of maintaining ground instrumentation over wide geographic areas, remote sensing and numerical model-based fine-scale mapping of soil water conditions have been applied for water resource applications at a range of spatial scales. Here, we present a prototype framework for integrating high-resolution thermal infrared (TIR) and synthetic aperture radar (SAR) remote sensing data into a soil-vegetation-atmosphere-transfer (SVAT) model with the aim of providing improved estimates of surface- and root-zone soil moisture that can support optimized irrigation management strategies. Specifically, remotely-sensed estimates of water stress (from TIR) and surface soil moisture retrievals (from SAR) are assimilated into a 30-m resolution SVAT model over a vineyard site in the Central Valley of California, U.S. The efficacy of our data assimilation algorithm is investigated via both the synthetic and real data experiments. Results demonstrate that a particle filtering approach is superior to an ensemble Kalman filter for handling the nonlinear relationship between model states and observations. In addition, biophysical conditions such as leaf area index are shown to impact the relationship between observations and states and must therefore be represented accurately in the assimilation model. Overall, both surface and root-zone soil moisture predicted via the SVAT model are enhanced through the assimilation of thermal and radar-based retrievals, suggesting the potential for improving irrigation management at the agricultural sub-field-scale using a data assimilation strategy.

Remote sensing assessments of land use and land cover change (LULCC) are critical to improve understanding of socio-economic, institutional and ecological processes that lead to and stem from land use change. This is particularly crucial in the emerging frontiers of Southern Africa, where there is a paucity of LULCC studies relative to the humid tropics. This study focuses on Gurué District (5606 km2) of Zambezia province of Mozambique, one of many countries in the region that has experienced a recent growth in foreign investments in agriculture through large-scale land acquisitions, often resulting in land use conversions and modifications. Previous LULCC assessments covering Mozambique have focused on dynamics between natural and anthropogenic land categories, with limited efforts to distinguish the different land use agents associated with these changes, and relating this with social, economic and technological processes. In this study we built a new LULC assessment methodology that leverages the power of open remote sensing data and tools to integrate categorical and continuous training and validation data obtained from field surveys and Collect Earth software within Google Earth Engine. We then examined the suitability of five pixel-based compositing techniques for generating cloud-free Landsat images that can support analysis of land use dynamics in persistently cloudy, mosaic landscapes with more limited Landsat archives. Drawing upon the spectral and textural features of Landsat data in pixel-based composites, we classified land use over three time periods, 2006, 2012 and 2016, and characterized land use change, focusing on changes between small-scale cropland, large-scale mechanized cropland, and other land uses. This method can be upscaled and applied in many parts of Africa with similar historic image availability challenges, and similar economic contexts with great disparities between small-scale unmechanized cropland and very large-scale mechanized cropland, to explore land consolidation dynamics and agent-specific pathways of land use change.

We present the first flight data using a Computational Reconfigurable Imaging Spectrometer (CRISP) system. CRISP (Sullenberger et al., 2017) is a novel hyperspectral thermal imaging spectrometer that uses computational imaging to enable high sensitivity measurements (via spectral multiplexing) from smaller, noisier, and less-expensive components (e.g., uncooled microbolometers) making it useful on small space and air platforms with strict size, weight, and power requirements. In contrast to other multiplexing hyperspectral solutions (e.g Michelson interferometer), it does not require moving parts, allowing for a robust system without aggressive engineering solutions. We discuss flight system design and calibration. Spectra from ground targets and gaseous species are compared to performance expectations. We successfully demonstrate the ability to extract airborne longwave infrared (8–12 μm) imagery and spectra from an uncooled camera-based CRISP system.

The seasonal snow-cover is one of the most rapidly varying natural surface features on Earth. It strongly modulates the terrestrial water, energy, and carbon balance. Fractional snow-covered area (fSCA) is an essential snow variable that can be retrieved from multispectral satellite imagery. In this study, we evaluate fSCA retrievals from multiple sensors that are currently in polar orbit: the operational land imager (OLI) on-board Landsat 8, the multispectral instrument (MSI) on-board the Sentinel-2 satellites, and the moderate resolution imaging spectroradiometer (MODIS) on-board Terra and Aqua. We consider several retrieval algorithms that fall into three classes: thresholding of the normalized difference snow index (NDSI), regression on the NDSI, and spectral unmixing. We conduct the evaluation at a high-Arctic site in Svalbard, Norway, by comparing satellite retrieved fSCA to coincident high-resolution snow-cover maps obtained from a terrestrial automatic camera system. For the lower resolution MODIS retrievals, the regression-based retrievals outperformed the unmixing-based retrievals for all metrics but the bias. For the higher resolution sensors (OLI and MSI), retrievals based on NDSI thresholding overestimated the fSCA due to the mixed pixel problem whereas spectral unmixing retrievals provided the most reliable estimates across the board. We therefore encourage the operationalization of spectral unmixing retrievals of fSCA from both OLI and MSI.

Small islands face environmental issues directly or indirectly related to land cover/land use changes (LCLUC), such as natural hazards, climate change, loss of biodiversity and proliferation of invasive alien species, some of which are caused by direct human exploitation. A Land Cover/Land Use Change (LCLUC) detection approach based on PCA and vegetation indices derived from low cost high-resolution RapidEye multispectral satellite data and available vegetation maps was developed to assess vegetated/forested aboveground carbon stocks and their changes in Madeira Island, Portugal, for the period between December 2009 and August 2011 due to catastrophic events occurred in 2010. During this period, the identified LCLUC revealed a relevant decrease of vegetated areas (especially those dominated by native/endemic communities) substituted by increases of non-vegetated and human-managed vegetated/forested areas. In particular, there was a decrease of 2% of vegetated areas, 30% of which were represented by native/endemic vegetation. The largest and most accurate LCLUC detected were used to estimate changes in aboveground biomass carbon (AGC) stocks. In 2010 more than 25,000 Mg of AGC stocks may have been released. Both relevant LCLUC and AGC stocks depletion in such short period of time may have been strongly enhanced by two catastrophic events that affected Madeira in February (flashflood and landslides) and August 2010 (wildfires). This straightforward and cost-effective methodological approach may be successfully applied in remote territories such as islands or mountainous areas, where the logistic and economic costs associated to periodic and standard airborne remote sensing campaigns for mapping, assessing and monitoring aboveground biomass carbon stocks are generally unaffordable for most decision-makers and land managers in these territories.

The accurate representation of spatio-temporal patterns of precipitation is an essential input for numerous environmental applications. However, the estimation of precipitation patterns derived solely from rain gauges is subject to large uncertainties. We present the Random Forest based MErging Procedure (RF-MEP), which combines information from ground-based measurements, state-of-the-art precipitation products, and topography-related features to improve the representation of the spatio-temporal distribution of precipitation, especially in data-scarce regions. RF-MEP is applied over Chile for 2000—2016, using daily measurements from 258 rain gauges for model training and 111 stations for validation. Two merged datasets were computed: RF-MEP3P (based on PERSIANN-CDR, ERA-Interim, and CHIRPSv2) and RF-MEP5P (which additionally includes CMORPHv1 and TRMM 3B42v7). The performances of the two merged products and those used in their computation were compared against MSWEPv2.2, which is a state-of-the-art global merged product. A validation using ground-based measurements was applied at different temporal scales using both continuous and categorical indices of performance. RF-MEP3P and RF-MEP5P outperformed all the precipitation datasets used in their computation, the products derived using other merging techniques, and generally outperformed MSWEPv2.2. The merged P products showed improvements in the linear correlation, bias, and variability of precipitation at different temporal scales, as well as in the probability of detection, the false alarm ratio, the frequency bias, and the critical success index for different precipitation intensities. RF-MEP performed well even when the training dataset was reduced to 10% of the available rain gauges. Our results suggest that RF-MEP could be successfully applied to any other region and to correct other climatological variables, assuming that ground-based data are available. An R package to implement RF-MEP is freely available online at https://github.com/hzambran/RFmerge.

Leaf area index (LAI) is a key variable for characterizing crop growth conditions and estimating crop productivity. Despite continuing efforts to develop LAI estimation algorithms, LAI datasets still need improvement at spatial and temporal resolutions to meet the requirements of agricultural applications. Advancements in data fusion technique and the emergence of new satellite data provide opportunities for LAI data at higher resolutions in both space and time. In this study, we derived new LAI estimations by leveraging novel satellite remote sensing datasets, STAIR fusion (MODIS-Landsat fusion) and Planet Labs' CubeSat data (through a reprocessed pipeline) for a typical agricultural landscape in the U.S. Corn Belt. The STAIR fused data and our reprocessed CubeSat data have both fine spatial resolutions (30 m and 3.125 m, respectively) and high frequencies (daily for both). To reliably estimate LAI from these advanced satellite datasets, we used two methods: inversion of a radiative transfer model (RTM), and empirical relationship with vegetation index (VI) calibrated from field measured LAI. Compared to the ground-truth LAI collected at 36 sites across the study region, reliable approximations were achieved by both LAI estimations based on PROSAIL RTM (STAIR: R2 = 0.69 and root mean squared error (RMSE) = 1.12 (m2 m−2), CubeSat: R2 = 0.76 and RMSE = 1.09 (m2 m−2)), and LAI estimations based on Green Wide Dynamic Range Vegetation Index (GrWDRVI) (STAIR: R2 = 0.75, RMSE = 1.10 (m2 m−2), CubeSat: R2 = 0.76, RMSE = 1.08 (m2 m−2), where validation ground-truth is independent from calibration data). Newly estimated high-resolution LAI data were aggregated at 500 m resolution and compared with MODIS and VIIRS LAI products, revealing substantial uncertainties and biases in these two products. We also demonstrated phenology stage estimation at fine spatial resolutions based on our high-frequency LAI data. The proposed LAI estimation methods at both high spatial resolution and temporal frequency can be applied to the entire U.S. Corn Belt and provide significant advancement to crop monitoring and precision agriculture.

Sentinel-1 mission with its wide spatial coverage (250 km), short revisit time (6 days), and rapid data dissemination opened new perspectives for large-scale interferometric synthetic aperture radar (InSAR) analysis. However, the spatiotemporal changes in troposphere limits the accuracy of InSAR measurements for operational deformation monitoring at a wide scale. Due to the coarse node spacing of the tropospheric models, like ERA-Interim and other external data like Global Navigation Satellite System (GNSS), the interpolation techniques are not able to well replicate the localized and turbulent tropospheric effects. In this study, we propose a new technique based on machine learning (ML) Gaussian processes (GP) regression approach using the combination of small-baseline interferograms and GNSS derived zenith total delay (ZTD) values to mitigate phase delay caused by troposphere in interferometric observations. By applying the ML technique over 12 Sentinel-1 images acquired between May–October 2016 along a track over Norway, the root mean square error (RMSE) reduces on average by 83% compared to 50% reduction obtained by using ERA-Interim model.

The attention towards possible link of earthquakes (EQs) and ionosphere in the form of seismo ionosphere anomalies (SIAs) has increased exponentially by utilizing new data and more accurate observations. The integrated atmosphere and ionosphere monitoring satellites has played a decisive role in this development and provided detection and analysis of anomalies attributed to seismic processes. In this paper, we study EQ anomalies in ionosphere from IGS permanent Global Navigation Satellite Systems (GNSS) based Total Electron Content (TEC) and foF2 parameter (highest frequency reflected from the main ionospheric F2 layer on a vertical propagation path), retrieved from stations operating within the seismogenic zone in Japan for EQs of magnitude Mw > 6.0. Furthermore, spatial composite maps of geopotential height, air temperature, and Outgoing Long-wave Radiation (OLR) from National Oceanic and Atmospheric Administration/National Center for Environmental Prediction (NOAA/NCEP) are analyzed to support the hypothesis of diffusion of SIAs through the atmosphere over the epicenter during the seismic preparation zone. We find evidences of TEC and foF2 anomalies in the analysis of nearby IGS permanent stations within seismogenic zone on main shock day, when geomagnetic activities remain quiet. In addition, atmospheric composite indices manifest anomalies attributed to the EQ on the same day as TEC and foF2 perturbations. Similarly, differential ionosphere and atmosphere values indicate that EQ abnormalities are significant on main shock day during UT = 10–12. Our results show that atmospheric and ionospheric measurements may play a role for the analysis and prediction of EQs.

Antarctic pack-ice seals, a group of four species of true seals (Phocidae), play a pivotal role in the Southern Ocean foodweb as wide-ranging predators of Antarctic krill (Euphausia superba). Due to their circumpolar distribution and the remoteness and vastness of their habitat, little is known about their population sizes. Estimating pack-ice seal population sizes and trends is key to understanding how the Southern Ocean ecosystem will react to threats such as climate change driven sea ice loss and krill fishing. We present a functional pack-ice seal detection pipeline using Worldview-3 imagery and a Convolutional Neural Network that counts and locates seal centroids. We propose a new CNN architecture that detects objects by combining semantic segmentation heatmaps with binary classification and counting by regression. Our pipeline locates over 30% of seals, when compared to consensus counts from human experts, and reduces the time required for seal detection by 95% (assuming just a single GPU). While larger training sets and continued algorithm development will no doubt improve classification accuracy, our pipeline, which can be easily adapted for other large-bodied animals visible in sub-meter satellite imagery, demonstrates the potential for machine learning to vastly expand our capacity for regular pack-ice seal surveys and, in doing so, will contribute to ongoing international efforts to monitor pack-ice seals.

Cropping intensity has undergone dramatic changes worldwide due to the effects of climate changes and human management activities. Cropping intensity is an important factor contributing to crop production and food security at local, regional and national scales, and is a critical input data variable for many global climate, land surface, and crop models. To generate annual cropping intensity maps at large scales, Moderate Resolution Imaging Spectroradiometer (MODIS) images at 500-m or 250-m spatial resolution have problems with mixed land cover types within a pixel (mixed pixel), and Landsat images at 30-m spatial resolution suffer from low temporal resolution (16-day). To overcome these limitations, we developed a straightforward and efficient pixel- and phenology-based algorithm to generate annual cropping intensity maps over large spatial domains at high spatial resolution by integrating Landsat-8 and Sentinel-2 time series image data for 2016–2018 using the Google Earth Engine (GEE) platform. In this pilot study, we report annual cropping intensity maps for 2017 at 30-m spatial resolution over seven study areas selected according to agro-climatic zones in China. Based on field-scale sample data, the annual cropping intensity maps for the study areas had overall accuracy rates of 89–99%, with Kappa coefficients of 0.76–0.91. The overall accuracy of the annual cropping intensity maps was 93%, with a Kappa coefficient of 0.84. These cropping intensity maps can also be used to enable identification of various crop types from phenological information extracted from the growth cycle of each crop. These algorithms can be readily applied to other regions in China to generate annual cropping intensity maps and quantify inter-annual cropping intensity variations at the national scale with a greatly improved accuracy.

Geostationary Ocean Color Imager (GOCI) sensor onboard the COMS (Communication, Ocean and Meteorological Satellite) launched in 2010 was primarily designed to provide high-frequency observations in and around the Korean Peninsula to ensure the thorough monitoring of ocean properties. Owing to its pixel resolution of 500 m and large set of spectral solar channels, GOCI can also be considered for applications related to the characterization of vegetation and the retrieval of aerosol properties over land. However, to apply it for the full characterization of land, it is mandatory to properly remove clouds from the images. Such a procedure has limitations when there is a lack of thermal bands, as is the case with GOCI. However, GOCI data are impacted by shadows and radiation scattering effects during the daily course of the sun. Although this yields strong directional effects, the bidirectional reflectance distribution function (BRDF) can be determined to a high level of accuracy. This information is used as a reference to detect clouds over land because surface BRDF varies slowly with time compared to that of clouds. The proposed algorithm relies on knowledge of the BRDF field derived from the application of a semi-empirical model that simulates the minimum difference between top and bottom of atmosphere reflectance values as the baseline of clear atmosphere. This step also serves to estimate background surface reflectance underneath clouds. Accuracy assessment of the new GOCI cloud mask product is appraised through a comparison with high-resolution vertical profiles of lidar data from the polar orbiting Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO). The results for the Probability Of Detection (POD) of all cloud types was found to be 0.831 for GOCI; this is comparable to that of MODIS (0.772). For the case of only thin cirrus, GOCI POD value was assessed to be 0.849, similar to that of MODIS, underlining the improved efficiency of determining thin cloud pixels.

The cloud-top height (CTH) product derived from passive satellite instrument measurements is often used to make climate data records (CDR). CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations) provides CTH parameters with high accuracy, but with limited temporal-spatial resolution. Recently, the Advanced Himawari Imager (AHI) onboard Japanese Himawari-8/-9, provides high temporal (every 10 min) and high spatial (2 km at nadir) resolution measurements with 16 spectral bands. This paper reports on a study to derive the CTH from combined AHI and CALIPSO using advanced machine learning (ML) algorithms with better accuracy than that from the traditional physical (TRA) algorithms. We find significant CTH improvements (1.54–2.72 km for mean absolute error, MAE) from four different machine learning algorithms (original MAE from TRA method is about 3.24 km based on CALIPSO data validation), particularly in high and optically thin clouds. In addition, we also develop a joint algorithm to combine optimal machine learning and traditional physical (TRA) algorithms of CTH to further reduce MAE to 1.53 km and enhance the layered accuracy (CTH < 18 km). While the ML-based algorithm improves CTH retrieval over the TRA algorithm, the lower or higher clouds still exhibit relatively large uncertainty. Combining both methods provides the better CTH than either alone. The combined approach could be used to process data from advanced geostationary imagers for climate and weather applications.

A combined snow modelling approach integrating remote sensing data, in-situ data, and an improved hydrological model is presented. Complementary information sources are evaluated in terms of its value for constraining the model parameters and to overcome limitations of individual data such as inadequate scale representation. The study site consists of the Upper Fagge river basin in the Austrian Alps featuring the Weisssee Snow Research Site. The available remote sensing datasets include Terra MODIS based medium resolution and Landsat-7/8 and Sentinel-2A based high resolution fractional snow covered area maps. Recently, Sentinel-1 based wet snow covered area maps have become increasingly available. To the knowledge of the authors the first evaluation of their value for snow-hydrological modelling is presented. Besides conventional small footprint station data, in-situ time-series of snow water equivalent (SWE) of a Cosmic-Ray Neutron Sensor (CRNS) having a footprint of several hectares is additionally used. For including these data the model now provides respective outputs such as fractional snow cover, wet/dry snow surface and SWE areal means equivalent to the CRNS sensor footprint. By means of 40,000 model runs the high complementary value of representative SWE data and remote sensing information was assessed with most promising results achieved by combining high resolution fractional snow covered area maps with CRNS-SWE data. Regarding mean SWE or mean snow covered area in the catchment the ensemble spreads are reduced by two thirds compared to the results of a benchmark simulation based only on runoff for model calibration. Wet snow covered area maps have a high potential for simulating SWE at Weisssee Snow Research Site but introduce additional uncertainties for runoff simulations likely caused by the uncertain detection of the snow covered area from Sentinel-1 backscatter. The approach has high potential for water resources management in gauged and ungauged mountain basin and gives guidance for efficient data assimilation schemes.

Remote sensing-derived cropland products have depicted the location and extent of agricultural lands with an ever increasing accuracy. However, limited attention has been devoted to distinguishing between actively cropped fields and fallowed fields within agricultural lands, and in particular so in grass fallow systems of semi-arid areas. In the Sahel, one of the largest dryland regions worldwide, crop-fallow rotation practices are widely used for soil fertility regeneration. Yet, little is known about the extent of fallow fields since fallow is not explicitly differentiated within the cropland class in any existing remote sensing-based land use/cover maps, regardless of the spatial scale. With a 10 m spatial resolution and a 5-day revisit frequency, Sentinel-2 satellite imagery made it possible to disentangle agricultural land into cropped and fallow fields, facilitated by Google Earth Engine (GEE) for big data handling. Here we produce the first Sahelian fallow field map at a 10 m resolution for the baseline year 2017, accomplished by designing a remote sensing driven protocol for generating reference data for mapping over large areas. Based on the 2015 Copernicus Dynamic Land Cover map at 100 m resolution, the extent of fallow fields in the cropland class is estimated to be 63% (403,617 km2) for the Sahel in 2017. Similar results are obtained for five contemporary cropland products, with fallow fields occupying 57–62% of the cropland area. Yet, it is noted that the total estimated area coverage depends on the quality of the different cropland products. The share of cropped fields within the Copernicus cropland area is found to be higher in the arid regions (200–300 mm rainfall) as compared to the semi-arid regions (300–600 mm rainfall). The woody cover fraction within cropped and fallow fields is found to have a reversed pattern between arid (higher woody cover in cropped fields) and semi-arid (higher woody cover in fallow fields) regions. The method developed, using cloud-based Earth Observation (EO) data and computation on the GEE platform, is expected to be reproducible for mapping the extent of fallow fields across global croplands. Future applications based on multi-year time series is expected to improve our understanding of crop-fallow rotation dynamics in grass fallow systems being key in teasing apart how cropland intensification and expansion affect environmental variables, such as soil fertility, crop yields and local livelihoods in low-income regions such as the Sahel. The mapping result can be visualized via a web viewer (https://buwuyou.users.earthengine.app/view/fallowinsahel).

Optical properties of clouds and heavy aerosol retrieved from satellite measurements are the most important elements for calculating surface solar radiation (SSR). The Himawari-8/Advanced Himawari Imager (AHI) satellite measurements receive high spatial, temporal and spectral signals, which provides an opportunity to estimate cloud, aerosol and SSR accurately. In this study, we developed the AHI official cloud property product (version 1.0) for JAXA P-Tree system. A look-up table (LUT) method was used to calculate high-temporal (10 min) and high-spatial (5 km) SSR from AHI cloud properties. First, the LUT of the SSR estimation was optimized through a radiative transfer model to account for solar zenith angle, cloud optical thickness (COT), effective particle radius (CER), aerosol optical thickness and surface albedo. Following this, COT and CER were retrieved from the AHI data, with ice cloud parameters being retrieved from an extended Voronoi ice crystal scattering database and water cloud parameters being retrieved from the Mie–Lorenz scattering model. The retrieved COT and CER for water clouds were compared well with MODIS collection 6 cloud property products, with correlation coefficients of 0.77 and 0.82, respectively. The COT of ice cloud also shows good consistency, with a correlation coefficient of 0.85. Finally, the SSR was calculated based on the SSR LUT and the retrieved cloud optical parameters. The estimated SSR was validated at 122 radiation stations from several observing networks covering the disk region of Himawari-8. The root-mean-square error (RMSE) at CMA (China Meteorological Administration) stations was 101.86 Wm−2 for hourly SSR and 31.42 Wm−2 for daily SSR; RMSE at non-CMA stations was 119.07 Wm−2 for instantaneous SSR, 81.10 Wm−2 for hourly SSR and 26.58 Wm−2 for daily SSR. Compared with the SSR estimated from conventional geostationary satellites, the accuracy of the SSR obtained in this study was significantly improved.

Drone applications are becoming increasingly common in the arena of forest management and forest inventories. In particular, the use of photogrammetrically derived drone-based image point clouds (DIPC) in individual tree detection has become popular. Use of an area-based approach (ABA) in small areas has also been considered. However, in-situ field measurements of sample plots substantially increase the cost of small area forest inventories. Therefore, we examined whether small-scale forest management inventories could be carried out without local field measurements. We used nationwide and regional ABA models for stem volumes fitted with airborne laser scanning (ALS) data to predict stem volumes using corresponding metrics calculated from DIPC data. The stem volumes were predicted at the cell level (15 × 15 m) and aggregated to test plots (30 × 30 m). Height metrics for the dominant tree layer from the DIPC data showed strong correlations with similar metrics computed from the ALS data. The ALS-based models applied with DIPC metrics performed well, especially if the ABA model was fitted in the same geographical area (regional model) and the inventory units were disaggregated to coniferous and deciduous dominated stands using auxiliary information from Multi-source National Forest Inventory data (root mean square error at 30 × 30 m level was 13.1%). The corresponding root mean square error associated with the nationwide ABA model was 20.0% with an overestimation (mean difference 9.6%).

This paper presents the validation results of GOES-16 Satellite's Advanced Baseline Imager (ABI) obtained from a reflectance-based field campaign undertaken at the Salar de Uyuni in Bolivia in June 2017. In situ ground measurements are used to characterize the surface reflectance and the atmosphere in order to constrain the radiative transfer code and predict at-sensor reflectance (also referred to as top-of-atmosphere (TOA) reflectance) to compare with concurrent GOES-16 ABI measurements. The five-day field campaign provided repeated TOA reflectance estimates, allowing assessment not only of the calibration accuracy of the ABI reflective channels 1, 2, 3, 5 and 6, but also of its stability over the duration of the campaign. The results show that the accuracy of the ABI reflective channels calibration is within specification for channels 1, 3, 5, and 6 - average biases within 2%; for channel 2 the bias is 5%. The estimated uncertainty on the derived biases is 2–2.4%. Some calibration stability issues were present in the ABI calibration at the time of the campaign: (i) a jump on the order of 2% in channels 1 and 6, coincident with an ABI solar calibration event, reflects an instability of the ABI gains in these channels, and (ii) short-term variability in channels 1 and 2 is due to striping (ABI detector-to-detector calibration differences). Continued validation and subsequent reprocessing of ABI reflectance imagery would allow Earth scientists to fully benefit from the high spatial and spectral fidelity of the GOES-16 ABI diurnal measurements at the continental scale.

Suspended particulate matter (SPM) consists both of an organic (OSPM) and an inorganic fraction (ISPM) and the latter can be used as an indicator for coastal influence in the Baltic Sea. The concentration of SPM can be derived from particle scatter if the specific scattering properties of the respective water body are known. In this paper we show that likewise, ISPM can be derived reliably from remotely sensed particle scatter. An empirical algorithm between particle scatter (AC9 data) and ISPM concentration (measured gravimetrically) was derived from in-water measurements. This regional algorithm was then applied to the iop_bpart level 2 product (i.e. the particle scatter measured at 443 nm) derived from OLCI data on Sentinel-A (S3A) using the C2RCC neural network and validated against an independent data set. The standard error of the derived OLCI match-up data was 10%, and was thus within the goal of the mission requirements of Sentinel-3. The generated S3 composite images from spring and autumn 2018 show that in the Baltic Sea most of the ISPM falls out rather close to the shore, whereas only a very small proportion of ISPM is carried further off-shore. This is also supported by in situ ISPM transects measured in the coastal zone. The ISPM images clearly highlight the areas that are most strongly influenced by terrestrial matter. Differences between the NE Baltic and the SE Baltic proper can be explained by the difference in hydrology and coastal influence as well as bathymetry and wind-wave stirring. The method is of interest for coastal zone management and for assessing the effect of seasonal changes in terrestrial run-off and wind-driven resuspension of sediments. It can also be used to evaluate the effect of climate change which has led to an increase of extreme storm and flooding events that are usually accompanied by increased erosion and run-off from land. Last but not least, turbidity caused by particles influences the light conditions in inner coastal areas and bays, which has a profound effect on pelagic productivity, the maximum growth of macroalgae as well as fish behaviour.

A physically based metamodel is proposed to describe the dependency of canopy reflectance on the wavelength, leaf and soil optical properties. The four-stream solution is first applied to describe the interaction between the soil background and the vegetation layers. This leads to the calibration of four terms for a given canopy structure, observation configuration and leaf properties. This number can be reduced to two terms by using a linear approximation which shows a slight degradation when the multiple scattering contribution is significant. The dependency of each of the two or four terms on wavelength and leaf properties is described using the leaf total absorption coefficient. Our approach requires only 12 (linear approximation) to 24 (four-stream solution) simulations of a reference model to describe the full canopy reflectance dependency on wavelength, leaf and soil properties. The approach was evaluated against reference canopy reflectance simulations using the ray tracing LuxCoreRender model. LuxCoreRender was first compared against reference radiative transfer models. The reference dataset corresponds to a range of detailed 3D maize canopies showing variation of leaf and background properties and one heterogeneous scene including vegetation elements of different shapes that is classically used for model inter-comparison exercises under different view and sun directions in a set of wavebands. Results demonstrate that our approach provides an accurate description of the dependency of canopy reflectance on wavelength, leaf and soil properties with RMSE = 0.0017 for the four-stream solution and RMSE = 0.0022 for the linear approximation. The proposed approach appears therefore computationally effective and well suited to generate a large number of canopy reflectance simulations with detailed 3D radiative transfer models that can be used to retrieve vegetation characteristics from remote sensing observations.

Satellite observations are used to detect surface waters but uncertainties such as instrument noise or retrieval errors can introduce noise or missing-data in the resulting water maps, especially for datasets at the global scale. In this study, spatial filters based on several a priori information are proposed to reduce noise and perform spatial interpolation to fill missing-data in satellite-based surface water maps such as wetlands, rivers, lakes. Four main sources of a priori of information are considered: (1) historical information at the pixel level, (2) neighbouring information constraints based on a historical record, (3) constraints based on topography, and (4) hydrological constraints based on a floodability index. Experiments are conducted over synthetic but realistic data, as well as over real Sentinel 1 (SAR) and 2 (visible) water map retrievals. Mis-classification quantitative results over these three types of data show that simple determinist spatial filters allow reducing noise and filling missing-data. The four sources of a priori information can be exploited and combined to improve observed water maps. This opens some ways to develop post-processing tools for improving surface water maps at high spatial resolution from missions such as SWOT (Surface Water and Ocean Topography) to be launched in 2020.

Many coastal waters include large areas of Optically Shallow Waters (OSWs) where the sea-bottom affects above-water observations of remote sensing reflectance (Rrs [sr−1]). If not treated, the effect of bottom reflectance will interfere with the correct retrieval of Water Constituent Concentrations (WCCs) from hyperspectral and multispectral remote sensing observations. To study this phenomenon in more detail, the existing semi-infinite 2SeaColour Radiative Transfer (RT) model was modified into a finite water layer model, bounded by a diffusely reflecting surface at the sea-bottom. From simulations with the new model, called Water - Sea Bottom (WSB) model, it was observed that a ratio of spectral bands in the Near-Infrared, bands 750 nm and 900 nm, is nearly insensitive to the WCCs and increases with the shallowness of the water, and therefore can be used as a robust index to detect OSWs. The newly established Near-Infrared Bottom Effect Index (NIBEI) was applied to a series of satellite observations over the Wadden Sea during high and low tidal phases. Images from the MEdium Resolution Imaging Spectrometer (MERIS) and the Ocean and Land Colour Instrument (OLCI) were processed to retrieve WCCs of the study area. The results indicate that the sea-bottom effect in OSWs affects the accuracy of atmospheric correction and retrievals. On the other hand, applying the NIBEI to flag OSWs improves the reliability and consistency of WCCs maps. The application of proposed NIBEI on satellite images requires only Top Of Atmosphere (TOA) radiances at 750 nm and 900 nm and does not depend on atmospheric correction and ancillary local input data (e.g., bathymetry map, bottom type, empirical coefficients, in-situ measurements). As a result, the proposed NIBEI can readily be applied to detect OSWs on various ocean colour remote sensors in various shallow coastal regions.

An accurate estimation of biomass burning emissions is partially limited by the lack of knowledge of fire burning phase (smoldering vs. flaming). In recent years, several fire detection products have been developed to provide information of fire radiative power (FRP), location, size, and temperature of fire pixels, but no information regarding fire burning phase is retrieved. The Day-Night band (DNB) aboard Visible Infrared Imaging Radiometer Suite (VIIRS) is sensitive to visible light from flaming fires in nighttime scenes. In contrast, VIIRS 4 μm moderate resolution band #13 (M13), though capable of detecting fires at all phases, has no direct sensitivity for discerning fire phase. However, the hybrid usage of VIIRS DNB and M-bands data is hampered by their different scanning technology and spatial resolution. In this study, we present a novel method to rapidly and accurately resample DNB pixel radiances to the footprint of M-band pixels, accounting for onboard detector aggregation schemes and bowtie effect removals. The visible energy fraction (VEF) is subsequently introduced as an indicator of fire burning phase. VEF is calculated as the ratio of visible light power (VLP) to FRP for each fire pixel retrieved from the VIIRS 750 m active fire product. A global distribution of VEF values is quantitatively obtained, showing smaller VEF values in regions with mostly smoldering wildfires, such as peatland fires in Indonesia, larger VEF values in regions with flaming wildfires over grasslands and savannas in the sub-Sahelian region, and the largest VEF values associated with gas flaring in the Middle East. Mean VEF for different land cover types or regions is highly correlated with modified combustion efficiency (MCE). These results, together with a case study of the 2018 California Camp Fire, show that the VEF has the potential to be an indicator of fire combustion phase for each fire pixel, appropriate for estimating emission factors at the satellite pixel level.

From the EO-1 Hyperion imaging spectrometer to the newly launched Chinese satellite hyperspectral imagers, stripe noise is a ubiquitous phenomenon that seriously degrades the data quality and usability. Although previous efforts have achieved inspiring results, hyperspectral image (HSI) destriping remains a challenging task, as the stripe degradations are sometimes more complicated than the predefined assumptions, i.e., the preselected reference, filter, or handcrafted priors. With the rapid advances in deep learning technologies, convolutional neural networks (CNNs) provide a new potential to learn essential priors in an automatic manner. However, the training of CNNs is highly reliant on a large high-quality standard dataset, which is difficult to acquire for hyperspectral spaceborne sensors. In this paper, an innovative approach termed the satellite-ground integrated destriping network (SGIDN) is proposed for HSIs. Rather than using self-training, a satellite-ground integrated strategy is proposed, for the first time, to mitigate the data dependency, so that a large set of striped-clean pairs is generated from the ground-based HSIs. Considering the varied stripes among different bands, a unique CNN architecture design, including the combination of 3D convolution and 2D convolution, residual learning, and supplementary gradient channels, is integrated to capture the intrinsic spectral-spatial features in the HSIs and the unidirectional property of stripe noise. Compared with the traditional methods, SGIDN can be flexibly extended to specific HSI destriping tasks, e.g., coexisting horizontal and vertical stripes, and generalizes well to different hyperspectral satellite sensors. Given the same study area (Shanghai, China), three HSIs acquired by the EO-1 Hyperion imaging spectrometer, the Chinese HJ-1A HSI sensor, and the wide-range hyperspectral imager onboard the Chinese SPARK spectral micro-nano satellite are adopted to assess the proposed SGIDN model. Both simulated and real-data experiments confirm that SGIDN can consistently outperform the benchmark methods, with a higher degree of efficiency. Moreover, the land-cover mapping results further demonstrate the necessity of destriping and the suitability of the destriped results for use in further applications.

Methods to retrieve urban surface temperature (Ts) from remote sensing observations with sub-building scale resolution are developed using the Discrete Anisotropic Radiative Transfer (DART, Gastellu-Etchegorry et al., 2012) model. Corrections account for the emission and absorption of radiation by air between the surface and instrument (atmospheric correction), and for the reflected longwave infrared (LWIR) radiation from non-black-body surfaces (“emissivity” correction) within a single modelling framework. The atmospheric correction a) can use horizontally and vertically variable distributions of atmosphere properties at high resolution (<5 m); b) is applied here with vertically extrapolated weather observations and MODTRAN atmosphere profiles; and c) is a solution to ray tracing and cross section (e.g. absorption) conflicts (e.g. cross section needs the path length but it is typically unavailable during ray tracing). The emissivity correction resolves the reflection of LWIR radiation as a series of scattering events at high spatial (<1 m) and angular (ΔΩ ≈ 0.02 sr) resolution using a heterogeneous distribution of radiation leaving the urban surfaces. The method is applied to a novel network of seven ground-based cameras measuring LWIR radiation across a dense urban area (extent: 420 m × 420 m) where a detailed 3-dimensional representation of the surface and vegetation geometry is used. Our unique observation set allows the method to be tested over a range of realistic conditions as there are variations in: path lengths, view angles, brightness temperatures, atmospheric conditions and observed surface geometry. For pixels with 250 (±10) m path length the median (5th and 95th percentile) atmospheric correction magnitude is up to 4.5 (3.1 and 8.1) K at 10:10 on a mainly clear-sky day. The detailed surface geometry resolves camera pixel path lengths accurately, even with complex features such as sloped roofs. The atmospheric correction method evaluation, with simultaneous “near” (~15 m) and “far” (~155 m) observations, has a mean absolute error of 0.39 K. Using broadband approximations, the emissivity correction has clear diurnal variability, particularly when a cool and shaded surface (e.g. north facing) is irradiated by warmer (up to 17.0 K) surfaces (e.g. south facing). Varying the material emissivity with bulk values common for dark building materials (ε = 0.89 → 0.97) alters the corrected roof (south facing) surface temperatures by ~3 (1.5) K, and the corrected cooler north facing surfaces by less than 0.1 K. Corrected observations, assuming a homogeneous radiation distribution from surfaces (analogous to a sky view factor correction), differ from a heterogeneous distribution by up to 0.25 K. Our proposed correction provides more accurate Ts observations with improved uncertainty estimates. Potential applications include ground-truthing airborne or space-borne surface temperatures and evaluation of urban energy balance models.

Choosing appropriate scales for remotely sensed image classification is extremely important yet still an open question in relation to deep convolutional neural networks (CNN), due to the impact of spatial scale (i.e., input patch size) on the recognition of ground objects. Currently, the optimal scale selection processes are extremely cumbersome and time-consuming requiring repetitive experiments involving trial-and-error procedures, which significantly reduce the practical utility of the corresponding classification methods. This issue is crucial when trying to classify large-scale land use (LU) and land cover (LC) jointly (Zhang et al., 2019). In this paper, a simple and parsimonious Scale Sequence Joint Deep Learning (SS-JDL) method is proposed for joint LU and LC classification, in which a sequence of scales is embedded in the iterative process of fitting the joint distribution implicit in the joint deep learning (JDL) method, thus, replacing the previous paradigm of scale selection. The sequence of scales, derived autonomously and used to define the CNN input patch sizes, provides consecutive information transmission from small-scale features to large-scale representations, and from simple LC states to complex LU characterisations. The effectiveness of the novel SS-JDL method was tested on aerial digital photography of three complex and heterogeneous landscapes, two in Southern England (Bournemouth and Southampton) and one in North West England (Manchester). Benchmark comparisons were provided in the form of a range of LU and LC methods, including the state-of-the-art joint deep learning (JDL) method. The experimental results demonstrated that the SS-JDL consistently outperformed all of the state-of-the-art baselines in terms of both LU and LC classification accuracies, as well as computational efficiency. The proposed SS-JDL method, therefore, represents a fast and effective implementation of the state-of-the-art JDL method. By creating a single, unifying joint distribution framework for classifying higher order feature representations, including LU, the SS-JDL method has the potential to transform the classification paradigm in remote sensing, and in machine learning more generally.

Optimal estimation (OE) is a core method in quantitative Earth observation. The optimality of OE depends on the errors in the prior, measurements and forward model being zero mean and having well-known error covariance. Often these assumptions are not met. We show how to use matches of satellite observations to in situ reference measurements to estimate parameters for use in OE that bring the retrieval framework closer to the theoretical optimality. This is done by retrieving bias correction and error covariance parameters. Bias correction parameters for some components of the retrieved state and for the satellite radiances are anchored by the in situ reference measurements, and are obtained by a modification of Kalman filtering. Error covariance matrices for the prior state and for the observation-simulation difference are iteratively obtained by applying equations for diagnosing internal retrieval consistency. The theory is applied to the case of OE of sea surface temperature from a sensor on a geostationary platform. Relative to an initial OE implementation, all measures of retrieval performance are improved when the optimised OE is tested on independent data: mean difference from validation data is reduced from −0.08 K to −0.01 K, and the standard deviation from 0.47 to 0.45 K; retrieval sensitivity to sea surface temperature increases from 71% to 76%; and a 20% underestimation of retrieval uncertainty is corrected. Perhaps more significant than the quantitative improvements are the coherent new insights into the forward model simulations and prior assumptions that are also obtained. These include estimates of prior bias in the absence of in situ information, an important consideration when in situ information is not globally distributed. Biases and lack of information about error covariances arise in remote sensing very often. While illustrated here for a particular case, the principles and methods we present for constraining that lack of knowledge systematically using ground truth will be widely applicable in remote sensing.

Forest aboveground biomass is a key variable in remote sensing based forest monitoring. Active sensor systems, such as lidar, can generate detailed canopy height products. Relationships between canopy height and biomass are commonly established via regression analysis using information from ground-truth plots. In this way, many site-specific height-biomass relationships have been proposed in the literature and applied for mapping in regional contexts. However, such relationships are only valid within the specific forest type for which they were calibrated. A generalized relationship would facilitate biomass estimation across forest types and regions. In this study, a combination of lidar-derived and ancillary structural descriptors is proposed as an approach for generalization between forest types. Each descriptor is supposed to quantify a different aspect of forest structure, i.e., mean canopy height, maximum canopy height, maximum stand density, vertical heterogeneity and wood density. Airborne discrete return lidar data covering 194 ha of forest inventory plots from five different sites including temperate and tropical forests from Africa, Europe, North, Central and South America was used. Biomass predictions using the best general model (nRMSE = 12.4%, R2 = 0.74) were found to be almost as accurate as predictions using five site-specific models (nRMSE = 11.6%, R2 = 0.78). The results further allow interpretation about the importance of the employed structure descriptors in the biomass estimation and the mechanisms behind the relationships. Understanding the relationship between canopy structure and aboveground biomass and being able to generalize it across forest types are important steps towards consistent large scale biomass mapping and monitoring using airborne and potentially also spaceborne platforms.

The three dimensional distribution of water vapour around mountainous terrain can be highly variable. This variability can in turn affect local meteorological processes and geodetic techniques to measure ground surface motion. We demonstrate this general problem with the specific issues of a small tropical island, Montserrat. Over a period of 17 days in December 2014 we made observations using InSAR and GPS techniques, together with concurrent atmospheric models using the WRF code. Comparative studies of water vapour distribution and its effect on refractivity were made at high spatial resolution (300 m) over short distances (~10 km). Our results show that model simulations of the observed differences in water vapour distribution using WRF is insufficiently accurate. We suggest that better use could be made of the knowledge and observations of local water vapour conditions at different scales, specifically the Inter Tropical Convergence Zone (ITCZ), the trade wind fields and the mountain flow (~30 m) perhaps using eddy simulation. The annual perturbations of the ITCZ show that the range of humidity is approximately the same expressed as the differential phase of InSAR imaging (~100 mm). Trade wind direction and speed are particularly important at high wind speeds driving vigorous asymmetrical convection over the island's mountains. We also show that the slant angles of radar can follow distinct separate paths through the water vapour field. Our study is novel in demonstrating how synoptic-scale features and climate can advise the modelling of mesoscale systems and sub-seasonal InSAR imaging on tropical islands.

This paper applies the Fully-Focussed SAR (FF-SAR) algorithm to CryoSat-2 full-bit-rate data to measure water levels of lakes and canals in the Netherlands, and validates these measurements by comparing them to heights measured by gauges. Over Lake IJssel, a medium-sized lake, the FF-SAR height is biased about 6 cm below the gauge height, and a similar bias is found at six sites where CryoSat-2 crosses rivers and canals. The precision of the FF-SAR measurements depends on the extent of multi-looking (incoherent averaging along-track) applied. Over Lake IJssel the precision varies from 4 to 11 cm, decreasing as multi-looking increases. The precision of FF-SAR with 100 m of multi-looking is equivalent to that of the standard delay/Doppler processing, which has an along-track resolution of about 300 m. The width and orientation of rivers and canals limits the maximum available multi-looking. After removing the 6 cm bias, FF-SAR heights of rivers and canals have an accuracy between 2 cm and several decimeters, primarily depending on the presence of other water bodies lying within the cross-track measurement footprint, as these contaminate the waveform. We demonstrate that FF-SAR processing is able to resolve and measure small ditches only a few meters in width. The visibility of these signals depends on the angle at which CryoSat-2 crosses the ditch and on whether or not the ditch remains straight within CryoSat-2’s field of view. In the best-case scenario, straight ditches at nearly 90° to the CryoSat-2 ground track, the ditch signal has high enough signal-to-noise to allow sub-decimeter accuracy of FF-SAR height measurement.

The Antarctic Ice Sheet holds ~91% of the global ice mass making it the biggest potential contributor to global sea-level-rise. In order to evaluate the ice sheet's present and future response to global climate change, detailed information about spatial and temporal ice mass dynamics is required. While ice sheet surface mass balance can be estimated from regional climate models, ice discharge is often determined from spaceborne measurements of ice thickness and ice motion at selected flux gates. The retrieval of ice motion is not only critical for analysing the rate of ice transport into the ocean but also for investigating the spatial and temporal pattern or the driving forcings of Antarctic ice flow dynamics. This study provides a comprehensive review of state-of-the-art spaceborne ice velocity measurements in Antarctica. Based on 201 scientific papers, published over the last three decades, our analysis revealed a dominant interest in local and regional scale investigations. Yet, the launch of the ERS-1/-2 and RADARSAT-1 SAR satellites in the 1990s dramatically increased data coverage over the Antarctic continent and enabled the first circum-antarctic velocity product to be published in the late 2000s. In recent years, the improved imaging capabilities of Landsat 8 led to a shift towards using mainly optical satellite data and enabled great advances in circum-antarctic velocity mapping. Data from the currently operating Sentinel-1/-2 missions was applied in fewer studies but will play a crucial role in future ice motion studies e.g. as part of ESA CCI projects. In addition, we found a growing interest in high temporal resolution velocity change analyses. Based on the reviewed studies, we moreover summarized the highly dynamic and locally variable flow pattern over the Antarctic continent being in accordance with varying local and regional driving mechanisms. Despite the great advances in spaceborne ice velocity mapping in Antarctica, we report some major drawbacks and discuss future challenges and requirements. To start with, our study uncovered strong variations regarding the spatial availability of investigations. While we found an aggregation of studies over selected glaciers, we report a major lack of studies in all three Antarctic sectors. Additionally, many velocity datasets were created with strongly heterogeneous output parameters and were provided at coarse temporal resolution or at point-based measurement locations only. Future studies should consequently aim at creating more uniform ice motion products at defined spatial and temporal coverage and resolution standards and at analysing particularly the least studied glaciers.

Satellite-based active fire data provide indispensable information for monitoring global fire activity and understanding its impacts on climate and air quality. Yet the limited spatiotemporal sampling capacities of current satellites result in considerable uncertainties in fire observation and emissions estimation. The mitigation of these uncertainties mainly relies on new remote-sensing technology. The Advanced Baseline Imager (ABI) onboard the Geostationary Operational Environmental Satellite-R (GOES-R) Series observes fires across North and South Americas at an unprecedentedly spatiotemporal resolution of nominal 2 km every 5–15 min. This study evaluated the GOES-16 (the first GOES-R satellite) ABI active fire product using active fire data derived from the 30-m Landsat-8 and the 375-m and 750-m Visible Infrared Imaging Radiometer Suite (VIIRS), and ground-based burning data across the southeastern Conterminous United States (CONUS) during the 2018 peak fire season. Specifically, we characterized the overall fire detection performance of the ABI active fire detections, estimated omission and commission errors, and evaluated ABI fire radiative power (FRP). The results showed that the ABI fire detection probability and its omission and commission errors were highly related to fire size and temporal period. ABI detection probability was higher than 95% for the fire pixel that contained over 114 Landsat-8 (30 m) fire detections or 11 VIIRS (375 m) detections. During a period of ±8 h, ABI detected 19% and 29% more fires observed by Landsat-8 and 375-m VIIRS, respectively. Correspondingly, the omission error could reduce by up to 33%. Further, ABI was able to detect 6–22% and 31–42% more ground-recorded fires than VIIRS in Georgia and Florida States, respectively, but ABI still missed many very small fires because ABI was hard to detect fires smaller than ~34.5 MW. Additionally, compared with 750-m VIIRS FRP, ABI FRP was ~30–50% larger in individual fire events but was overall similar at a regional scale.

Complex dune-field landscape patterns (i.e., dune morphologies and their spatial arrangements) are recognized as indicators of dune self-organization and are required for investigating dune-field landscapes from remotely sensed images. However, dune-field landscape pattern maps are rarely available due to the insufficient attention paid to representing dune-field landscape mosaics and the labor-intensive manual interpretation processes in prior studies. Moreover, automatic or semi-automated dune-field landscape pattern mapping methods are still absent, resulting from a lack of appropriate image units to represent the heterogeneous dune-field landscape patches (DLPs). To address above issues, this study makes the first attempt to apply geoscenes, a new kind of complex units proposed recently, in dune-field landscapes and proposes a multiscale segmentation approach. The proposed approach is able to represent DLPs at multiple scales on images by integrating multisource features (i.e., class-related spatial-pattern features and the line-spatial-pattern features) in a unified segmentation framework. Visual and quantitative assessments of multiscale segmentation results demonstrate that our approach can delineate DLPs effectively. Comparisons among our approach, the multiresolution segmentation in eCognition software and an urban geoscene segmentation approach also reveal the superiority of our approach. This study is hence of great significance for dune-field landscape pattern mapping and offers a potential to be coupled into other landscape investigations.

Preharvest crop yield prediction is critical for grain policy making and food security. Early estimation of yield at field or plot scale also contributes to high-throughput plant phenotyping and precision agriculture. New developments in Unmanned Aerial Vehicle (UAV) platforms and sensor technology facilitate cost-effective data collection through simultaneous multi-sensor/multimodal data collection at very high spatial and spectral resolutions. The objective of this study is to evaluate the power of UAV-based multimodal data fusion using RGB, multispectral and thermal sensors to estimate soybean (Glycine max) grain yield within the framework of Deep Neural Network (DNN). RGB, multispectral, and thermal images were collected using a low-cost multi-sensory UAV from a test site in Columbia, Missouri, USA. Multimodal information, such as canopy spectral, structure, thermal and texture features, was extracted and combined to predict crop grain yield using Partial Least Squares Regression (PLSR), Random Forest Regression (RFR), Support Vector Regression (SVR), input-level feature fusion based DNN (DNN-F1) and intermediate-level feature fusion based DNN (DNN-F2). The results can be summarized in three messages: (1) multimodal data fusion improves the yield prediction accuracy and is more adaptable to spatial variations; (2) DNN-based models improve yield prediction model accuracy: the highest accuracy was obtained by DNN-F2 with an R2 of 0.720 and a relative root mean square error (RMSE%) of 15.9%; (3) DNN-based models were less prone to saturation effects, and exhibited more adaptive performance in predicting grain yields across the Dwight, Pana and AG3432 soybean genotypes in our study. Furthermore, DNN-based models demonstrated consistent performance over space with less spatial dependency and variations. This study indicates that multimodal data fusion using low-cost UAV within a DNN framework can provide a relatively accurate and robust estimation of crop yield, and deliver valuable insight for high-throughput phenotyping and crop field management with high spatial precision.

Sustainable Development Goal (SDG) indicator 15.1.1 proposes to quantify “Forest area as a proportion of total land area” in order to achieve SDG target 15.1. While area under forest cover can provide useful information regarding discrete changes in forest cover, it does not provide any insight on subtle changes within the broad vegetation class, e.g. forest degradation. Continental or national-level studies, mostly utilizing coarse-scale satellite data, are likely to fail in capturing these changes due to the fine spatial and long temporal characteristics of forest degradation. Yet, these long-term changes affect forest structure, composition and function, thus ultimately limiting successful implementation of SDG targets. Using a multi-scale, satellite-based monitoring approach, our goal is to provide an easy-to-implement reporting framework for South Asian forest ecosystems. We systematically analyze freely available remote sensing assets on Google Earth Engine for monitoring degradation and evaluate the potential of multiple satellite data with different spatial resolutions for reporting forest degradation. Taking a broad-brush approach in step 1, we calculate vegetation trends in six south Asian countries (Bangladesh, Bhutan, India, Nepal, Pakistan, and Sri Lanka) using the Moderate Resolution Imaging Spectroradiometer (MODIS) Normalized Difference Vegetation Index (NDVI) during 2000–2016. We also calculate rainfall trends in these countries using the Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) rainfall data, and further calculate Rain-Use Efficiency (RUE) that shows vegetation trends in the context of rainfall variability. In step 2, we focus on two protected area test cases from India and Sri Lanka for evaluating the potential of finer-resolution satellite data compared to MODIS, i.e. Landsat 8, and Sentinel-2 data, for capturing forest degradation signals, which will ultimately contribute towards SDG indicators 15.1.1 and 15.1.2. We find that most countries show a fluctuating trend in vegetation condition over the years, along with localized greening and browning. The Random Forest (RF) classifier utilized in step 2 was able to generate accurate maps (87% and 91% overall accuracy for Indian and Sri Lankan test cases, respectively) of non-intact forest within the protected areas. We find that almost one-third of the Indian test case is degraded forest, even though it shows overall greening as per the broad-brush approach. This finding corroborates our argument that utilizing higher-resolution satellite data (e.g. 10-m) than those normally used for national-level studies will be crucial for reporting SDG indicator 15.2.1: “progress towards sustainable forest management”.

Evapotranspiration (ET) from the land surface is an important component of the terrestrial hydrological cycle. Since the advent of Earth observation by satellites, various models have been developed to use thermal and shortwave remote sensing data for ET estimation. In this review, we provide a brief account of the key milestones in the history of remote sensing ET model development in two categories: temperature-based and conductance-based models. Temperature-based ET models utilize land surface temperature (LST) observed through thermal remote sensing to calculate the sensible heat flux from which ET is estimated as a residual of the surface energy balance or to estimate the evaporative fraction from which ET is derived from the available energy. Models of various complexities have been developed to estimate ET from surfaces of different vegetation fractions. One-source models combine soil and vegetation into a composite surface for ET estimation, while two-source models estimate ET of soil and vegetation components separately. Image contexture-based triangular and trapezoid models are simple and effective temperature-based ET models based on spatial and/or temporal variation patterns of LST. Several effective temporal scaling schemes are available for extending instantaneous temperature-based ET estimation to daily or longer time periods. Conductance-based ET models usually use the Penman-Monteith (P-M) equation to estimate ET with shortwave remote sensing data. A key put to these models is canopy conductance to water vapor, which depends on canopy structure and leaf stomatal conductance. Shortwave remote sensing data are used to determine canopy structural parameters, and stomatal conductance can be estimated in different ways. Based on the principle of the coupling between carbon and water cycles, stomatal conductance can be reliably derived from the plant photosynthesis rate. Three types of photosynthesis models are available for deriving stomatal or canopy conductance: (1) big-leaf models for the total canopy conductance, (2) two-big-leaf models for canopy conductances for sunlit and shaded leaf groups, and (3) two-leaf models for stomatal conductances for the average sunlit and shaded leaves separately. Correspondingly, there are also big-leaf, two-big-leaf and two-leaf ET models based on these conductances. The main difference among them is the level of aggregation of conductances before the P-M equation is used for ET estimation, with big-leaf models having the highest aggregation. Since the relationship between ET and conductance is nonlinear, this aggregation causes negative bias errors, with the big-leaf models having the largest bias. It is apparent from the existing literature that two-leaf conductance-based ET models have the least bias in comparison with flux measurements. Based on this review, we make the following recommendations for future work: (1) improving key remote sensing products needed for ET mapping purposes, including soil moisture, foliage clumping index, and leaf carboxylation rate, (2) combining temperature-based and conductance-based models for regional ET estimation, (3) refining methodologies for tight coupling between carbon and water cycles, (4) fully utilizing vegetation structural and biochemical parameters that can now be reliably retrieved from shortwave remote sensing, and (5) to improve regional and global ET monitoring capacity.

The original data produced by the Shuttle Radar Topography Mission (SRTM) tend to have an abundance of voids in mountainous areas where the elevation measurements are missing. In this paper, deep learning models are investigated for restoring SRTM data. To this end, we explore generative adversarial nets, which represent one state-of-the-art family of deep learning models. A conditional generative adversarial network (CGAN) is introduced as the baseline method for filling voids in incomplete SRTM data. The problem regarding shadow violation that possibly arises from the CGAN restored data is investigated. To address this deficiency, shadow geometric constraints based on shadow maps of satellite images are devised. In addition, a shadow constrained conditional generative adversarial network (SCGAN), which incorporates the shadow geometric constraints into the CGAN, is developed. Training the SCGAN model requires both the remote sensing observations (i.e., the original incomplete SRTM data and satellite images) and the ground truth data (i.e., the complete SRTM data, which are manually refined from the incomplete SRTM data with the reference of in-situ measurements). The integration of the multi-source training data enables the SCGAN model to be characterized by comprehensive information including both mountain shape variation and mountain shadow geometry. Experimental results validate the superiority of the SCGAN over the comparison methods, i.e., the interpolation, the convolutional neural network (CNN) and the baseline CGAN, in SRTM data restoration.

The use of time series analysis with moderate resolution satellite imagery is increasingly common, particularly since the advent of freely available Landsat data. Dense time series analysis is providing new information on the timing of landscape changes, as well as improving the quality and accuracy of information being derived from remote sensing. Perhaps most importantly, time series analysis is expanding the kinds of land surface change that can be monitored using remote sensing. In particular, more subtle changes in ecosystem health and condition and related to land use dynamics are being monitored. The result is a paradigm shift away from change detection, typically using two points in time, to monitoring, or an attempt to track change continuously in time. This trend holds many benefits, including the promise of near real-time monitoring. Anticipated future trends include more use of multiple sensors in monitoring activities, increased focus on the temporal accuracy of results, applications over larger areas and operational usage of time series analysis.

Synthetic Aperture Radar (SAR) enables 24-hour, all-weather flood monitoring. However, accurate detection of inundated areas can be hindered by the extremely complicated electromagnetic interaction phenomena between microwave pulses, and horizontal and vertical targets. This manuscript focuses on the problem of inundation mapping in areas with emerging vegetation, where spatial and seasonal heterogeneity makes the systematic distinction between dry and flooded backscatter response even more difficult. In this context, image interpretation algorithms have mostly used detailed field data and reference image(s) to implement electromagnetic models or change detection techniques. However, field data are rare, and despite the increasing availability of SAR acquisitions, adequate reference image(s) might not be readily available, especially for fine resolution acquisitions. To by-pass this problem, this study presents an algorithm for automatic flood mapping in areas with emerging vegetation when only single SAR acquisitions and common ancillary data are available. First, probability binning is used for statistical analysis of the backscatter response of wet and dry vegetation for different land cover types. This analysis is then complemented with information on land use, morphology and context within a fuzzy logic approach. The algorithm was applied to three fine resolution images (one ALOS-PALSAR and two COSMO-SkyMed) acquired during the January 2011 flood in the Condamine-Balonne catchment (Australia). Flood extent layers derived from optical images were used as validation data, demonstrating that the proposed algorithm had an overall accuracy higher than 80% for all case studies. Notwithstanding the difficulty to fully discriminate between dry and flooded vegetation backscatter heterogeneity using a single SAR image, this paper provides an automatic, data parsimonious algorithm for the detection of floods under vegetation.

The quantification of vegetation height for the circumpolar Arctic tundra biome is of interest for a wide range of applications, including biomass and habitat studies as well as permafrost modelling in the context of climate change. To date, only indices from multispectral data have been used in these environments to address biomass and vegetation changes over time. The retrieval of vegetation height itself has not been attempted so far over larger areas. Synthetic Aperture Radar (SAR) holds promise for canopy modeling over large extents, but the high variability of near-surface soil moisture during the snow-free season is a major challenge for application of SAR in tundra for such a purpose. We hypothesized that tundra vegetation height can be derived from multispectral indices as well as from C-band SAR data acquired in winter (close to zero liquid water content). To test our hypothesis, we used C-band SAR data from Sentinel-1 and multi-spectral data from Sentinel-2. Results show that vegetation height can be derived with an RMSE of 44 cm from Normalized Difference Vegetation Index (NDVI) and 54 cm from Tasseled Cap Wetness index (TC). Retrieval from C-band SAR shows similar performance, but C-VV is more suitable than C-HH to derive vegetation height (RMSEs of 48 and 56 cm respectively). An exponential relationship with in situ height was evident for all tested parameters (NDVI, TC, C-VV and C-HH) suggesting that the C-band SAR and multi-spectral approaches possess similar capabilities including tundra biomass retrieval. Errors might occur in specific settings as a result of high surface roughness, high photosynthetic activity in wetlands or high snow density. We therefore introduce a method for combined use of Sentinel-1 and Sentinel-2 to address the ambiguities related to Arctic wetlands and barren rockfields. Snow-related deviations occur within tundra fire scars in permafrost areas in the case of C-VV use. The impact decreases with age of the fire scar, following permafrost and vegetation recovery. The evaluation of masked C-VV retrievals across different regions, tundra types and sources (in situ and circumpolar vegetation community classification from satellite data) suggests pan-Arctic applicability to map current conditions for heights up to 160 cm. The presented methodology will allow for new applications and provide advanced insight into changing environmental conditions in the Arctic.