Monitoring temperate forest degradation on Google Earth Engine using Landsat time series analysis
Introduction
Definitions of forest degradation vary in the literature (Simula, 2009). For example, Food and Agriculture Organization (FAO)’s definition of forest degradation emphasizes the reduction of the capability of forests to provide goods and services, including both anthropogenic and natural disturbances (FAO, 2012). Intergovernmental Panel on Climate Change (IPCC)’s definition, however, focuses on carbon loss and constrains forest degradation to human activities (IPCC, 2003). Definitions of forest degradation in other international or national organizations can vary. Based on the existing definitions of forest degradation and the capability of remote sensing, forest degradation in this study is defined as forests that experienced loss of tree cover due to anthropogenic or natural causes but remain as forest. Forest is defined as places with tree cover greater than 10%, excluding orchards, and deforestation as conversion from forest to non-forest. Forest disturbance includes forest degradation and deforestation.
The issue of forest degradation is gaining attention in the scientific community, since carbon emissions from forest degradation are potentially substantial. FAO (2015) and Federici et al. (2015) estimated that global emissions from forest degradation have increased recently and contributed about 25% of the net CO2 emissions from forests. Pearson et al. (2017) found that emissions from forest degradation were higher than those from deforestation in 28 of 74 countries. Baccini et al. (2017) found that carbon loss from forest degradation, defined as reductions in carbon density within standing forests, accounted for 69% of the overall loss from forest degradation and deforestation in the tropics.
The estimated extent and rate of forest degradation have traditionally been associated with large uncertainties with the result that the impact of forest degradation on terrestrial carbon dynamics is poorly understood (FAO, 2015; Federici et al., 2015; Pearson et al., 2017). However, the situation is changing. Recent advancements, including the open data policy (Woodcock et al., 2008), global consolidation of Landsat satellite data (Wulder et al., 2016), cloud computing platforms (Gorelick et al., 2017), the development of time series-based approaches (Kennedy et al., 2010; Verbesselt et al., 2010; Kennedy et al., 2014; Bullock et al., 2020a) and the combined use of sampling techniques and remote sensing (Olofsson et al., 2013; Olofsson et al., 2014), have increased our capability of monitoring and estimating the extent of forest degradation (Mitchell et al., 2017; Bullock et al., 2020b; Quintano et al., 2013). Recent studies using these new advancements show that the extent and rate of forest degradation have been previously underestimated. For example, Bullock et al. (2020b) showed that when considering forest degradation, the area of forest disturbance in the Amazon Ecoregion since 1995 was 60% higher than previously estimated. Matricardi et al. (2020) also found that the extent and rate of forest degradation surpassed that of deforestation in Brazilian Amazon from 1992 to 2014.
Despite an increased focus, forest degradation remains understudied in many parts of the world. In the 2020 Global Forest Resource Assessment by FAO (2020a), only 58 countries reported on forest degradation, representing only 38% of global forest area. In Europe, only 7 countries reported on forest degradation. Furthermore, most studies of forest degradation from the last decade are in the tropics (e.g., Rappaport et al., 2018; Souza et al., 2013; Miettinen et al., 2014; Bullock et al., 2020b; Matricardi et al., 2020). This is partly because forest degradation in the tropics is a more severe issue than in temperate regions. While forest degradation in the tropics is important due to the associated impacts on carbon dynamics, biodiversity, and local climate (Herold et al., 2011), it is likely that newly discovered patterns and dynamics of forest degradation are not confined to the tropics.
Temperate forests, roughly defined as forests located in the temperate zone between 35- and 50-degree latitudes both north and south, represent a wider range of biomes than tropical forests (Strahler, 2011) and account for 16% of the global forest area (FAO, 2020b). In the 2020 Global Forest Resource Assessment (FAO, 2020b), only 15% of the total forest area of countries in temperate regions reported on forest degradation, as compared to 72% in the tropics. One reason why fewer countries in temperate regions reported forest degradation is because forest degradation is not a major problem in these countries. Another reason is that national forest inventories in some temperate countries, like the country of Georgia, are either out of date or did not include forest degradation. Compared to the tropics, degraded temperate forests recover more slowly due to climate constraints. Also, regardless of biome, forest degradation is always associated with environmental consequences, such as carbon emissions, soil erosion, water quality decline, and loss of wildlife habitat.
Due to the variety of the biomes and forest definitions, characterizing change in temperate forest extent and condition is challenging. The difficulty is highlighted by the different estimated trends in temperate forest area between the Forest Resource Assessments (FAO, 2015) which suggests an increase based on self-reported national inventories and remote sensing-based studies (FAO and JRC, 2012; Hansen et al., 2013) that have found that the area of temperate forest is declining in many parts of the world. Recent research even suggests that forest harvest and biomass loss have accelerated considerably across Europe in the last few years (Ceccherini et al., 2020). Compared with tropical forests, temperate forests exhibit different phenology and reflectance characteristics, and it remains an open question if approaches developed for monitoring tropical forest degradation will also work in temperate forests. One example of the difficulty associated with mapping forest degradation in temperate forests is presented in Olofsson et al. (2010), who attempted to map forest changes in the country of Georgia after the collapse of the Soviet Union. Most of the forest disturbance in Georgia was degradation rather than complete forest cover loss, which traditional approaches to remote sensing-based change detection failed to map. The failure was evident by large map errors, and a margin of error of almost 100% of the area estimate of forest disturbance (Olofsson et al., 2010). The same approach applied in neighboring countries, with less forest degradation and more complete forest cover loss, yielded substantially more accurate results (Kuemmerle et al., 2011; Olofsson et al., 2011). Although there are a number of studies on land cover and land use change in Eastern Europe (Lewińska et al., 2020; Baumann et al., 2012; Potapov et al., 2015), we are unaware of efforts to monitor forest degradation in the Caucasus region (Georgia, Armenia, and Azerbaijan) using remote sensing since Olofsson et al. (2010), and the country of Georgia has no reliable estimate of the area affected by forest degradation.
There are several dimensions of complexity to monitoring forest degradation. First, area estimates of forest degradation in the literature vary partly due to different definitions of forest degradation and forestland (Simula, 2009; FAO, 2012; IPCC, 2003). The spatial and temporal scale associated with different definitions may also lead to inconsistent results. Second, forest degradation may not result in complete land cover change but sub-pixel canopy loss that triggers only subtle changes in spectral reflectance. In Georgia, most forest disturbances do not result in complete land cover change, and detecting such subtle changes in satellite imagery is inherently difficult. Third, forest degradation can be either an abrupt event or a gradual process, or both. For example, illegal logging may initially be an abrupt degradation followed by gradual degradation as extraction of individual tree continues. Monitoring combinations of abrupt and gradual degradation processes are difficult and understudied (Vogelmann et al., 2016). Fourth, separating forest degradation from natural intra-annual variability is challenging, particularly in deciduous forests.
The objectives of our study are: (1) to develop an approach to monitor forest degradation in temperate regions and (2) to map forest degradation and estimate the area of forest degradation in Georgia from 1987 to 2019. Here, we developed an approach, Continuous Change Detection and Classification - Spectral Mixture Analysis (CCDC-SMA), which combines recent advancements in remote sensing for monitoring degradation in temperate forests, including spectral mixture analysis (SMA) (Souza et al., 2005; Souza et al., 2013), Continuous Change Detection and Classification (CCDC) (Zhu and Woodcock, 2014b) and COntinuous DEgradation Detection (CODED) (Bullock et al., 2020a) algorithms. CCDC-SMA can monitor both abrupt and gradual forest degradation and runs on Google Earth Engine (GEE), which is easy to adjust and apply to other regions.
Section snippets
Study area
Our study area (Fig. 1) is the country of Georgia. Georgia is rich in forests, with more than 3 million hectares, or about 40% of the country's total area (Forestry Department at the Georgian Ministry of Environment and Agriculture, 2006; FAO, 2019). Georgia's forests are rich in biodiversity since Georgia has a wide variety of climates and topography despite being a relatively small country. The Caucasus Ecoregion, where Georgia is located at, is among the top 25 richest and most endangered
Overview
We used all available Landsat Collection 1 surface reflectance data for our analysis and Google Earth Engine as the data processing platform. Fourteen Landsat scenes are required to cover Georgia (Fig. 1), and the study period was defined as 1987–2019 to include the end of the Soviet era and the subsequent civil war in 1991–1992. We tested harmonization (Roy et al., 2016a) and bidirectional reflectance distribution function (BRDF) - correction (Roy et al., 2016b) of Landsat time series data.
Sensitivity of SMA-based indices
Based on the visual interpretation of time series of the 113 locations that we explained in Section 3.4, we found that SMA-based indices are more effective in observing changes compared to the original Landsat bands (red, NIR and SWIR1) and traditional vegetation indices (NDVI and EVI). Table 5 shows the proportion of sites in each forest type and indices. For example, the cell for row “NDFI” and column “DE” has a value of 0.93, which means that in 93% of the deciduous forest sites a decrease
Evaluation with field conditions
To compare the time series and our map with field conditions, we visited several sites that were mapped as forest degradation around Tbilisi and in the Borjomi region in Georgia in the summer of 2019. In the Borjomi region, forest degradation was prevalent as evident by the presence of stumps, woody debris, branches, or dead trees, in most of the field sites mapped as forest degradation. We compared the time series and the forest conditions in the field. Fig. 21 shows the time series of NDFI
Conclusions
In this study, we developed an approach, CCDC-SMA, to monitor abrupt and gradual degradation in temperate forests. CCDC-SMA runs on Google Earth Engine without the need to download data and can be easily adjusted and applied to other regions. We found that fraction of GV and shade performed differently in response to degradation of different types of forest (deciduous and coniferous), whereas NDFI and fraction of soil performed more consistently. Based on our analysis, using different metrics
Description of author's responsibilities
Shijuan Chen: Conceptualization, Data curation, Formal analysis, Methodology, Software, Validation, Visualization, Writing - original draft, Writing - review & editing. Curtis E. Woodcock: Conceptualization, Supervision, Resources, Writing - review & editing. Eric L. Bullock: Software, Writing - review & editing. Paulo Arévalo: Software, Writing - review & editing. Paata Torchinava: Resources, Writing - review & editing. Siqi Peng: Validation, Writing - review & editing. Pontus Olofsson:
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This research was funded by support from the NASA Land-Cover and Land-Use Change Program (grant number: 80NSSC18K0315) to Boston University (PI: Pontus Olofsson) and USGS Landsat Science Team Program for Better Use of the Landsat Temporal Domain: Monitoring Land Cover Type, Condition and Change (grant number: G12PC00070) (PI: Curtis Woodcock). Finally, the authors are grateful to the editors and three anonymous reviewers for their insightful and constructive comments, which greatly helped to
References (62)
- et al.
Using the Landsat record to detect forest-cover changes during and after the collapse of the Soviet Union in the temperate zone of European Russia
Remote Sens. Environ.
(2012) - et al.
Land-cover change in the Caucasus Mountains since 1987 based on the topographic correction of multi-temporal Landsat composites
Remote Sens. Environ.
(2020) - et al.
Monitoring tropical forest degradation using spectral unmixing and Landsat time series analysis
Remote Sens. Environ.
(2020) - et al.
New estimates of CO2 forest emissions and removals: 1990-2015
For. Ecol. Manag.
(2015) - et al.
Google Earth Engine: planetary-scale geospatial analysis for everyone
Remote Sens. Environ.
(2017) - et al.
Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1. LandTrendr - temporal segmentation algorithms
Remote Sens. Environ.
(2010) - et al.
Remote sensing of forest degradation in Southeast Asia-Aiming for a regional view through 5-30 m satellite data
Glob. Ecol. Conserv.
(2014) - et al.
Making better use of accuracy data in land change studies: estimating accuracy and area and quantifying uncertainty using stratified estimation
Remote Sens. Environ.
(2013) - et al.
Good practices for estimating area and assessing accuracy of land change
Remote Sens. Environ.
(2014) - et al.
Mitigating the effects of omission errors on area and area change estimates
Remote Sens. Environ.
(2020)
Eastern Europe’s forest cover dynamics from 1985 to 2012 quantified from the full Landsat archive
Remote Sens. Environ.
Multiple Endmember Spectral Mixture Analysis (MESMA) to map burn severity levels from Landsat images in Mediterranean countries
Remote Sens. Environ.
Characterization of Landsat-7 to Landsat-8 reflective wavelength and normalized difference vegetation index continuity
Remote Sens. Environ.
A general method to normalize Landsat reflectance data to nadir BRDF adjusted reflectance
Remote Sens. Environ.
Combining spectral and spatial information to map canopy damage from selective logging and forest fires
Remote Sens. Environ.
Detecting trend and seasonal changes in satellite image time series
Remote Sens. Environ.
Perspectives on monitoring gradual change across the continuity of Landsat sensors using time-series data
Remote Sens. Environ.
The global Landsat archive: status, consolidation, and direction
Remote Sens. Environ.
Object-based cloud and cloud shadow detection in Landsat imagery
Remote Sens. Environ.
Automated cloud, cloud shadow, and snow detection in multitemporal Landsat data: An algorithm designed specifically for monitoring land cover change
Remote Sens. Environ.
Continuous change detection and classification of land cover using all available Landsat data
Remote Sens. Environ.
Continuous monitoring of forest disturbance using all available Landsat imagery
Remote Sens. Environ.
Generating synthetic Landsat images based on all available Landsat data: Predicting Landsat surface reflectance at any given time
Remote Sens. Environ.
A suite of tools for continuous land change monitoring in Google Earth Engine
Front. Clim.
Tropical forests are a net carbon source based on aboveground measurements of gain and loss
Science
Satellite-based estimates reveal widespread forest degradation in the Amazon
Glob. Chang. Biol.
Georgia’s forestry sector: current problems and ways forward
World For. Congr.
Statistical Inference
Abrupt increase in harvested forest area over Europe after 2015
Nature
Sampling Techniques
National forest concept for Georgia
J. Exp. Psychol. Gen.
Cited by (63)
Estimating fractional vegetation cover from multispectral unmixing modeled with local endmember variability and spatial contextual information
2024, ISPRS Journal of Photogrammetry and Remote SensingInterannual changes of urban wetlands in China's major cities from 1985 to 2022
2024, ISPRS Journal of Photogrammetry and Remote SensingReview of drivers of forest degradation and deforestation in Southeast Asia
2024, Remote Sensing Applications: Society and EnvironmentExploring natural and social drivers of forest degradation in post-Soviet Georgia
2024, Global Environmental ChangeNot-so-random forests: Comparing voting and decision tree ensembles for characterizing partial harvest events
2023, International Journal of Applied Earth Observation and Geoinformation