Sentinel-1 based soil freeze/thaw estimation in boreal forest environments

https://doi.org/10.1016/j.rse.2020.112267Get rights and content

Highlights

  • Soil freezing and thawing in boreal forests can be detected with C-band SAR

  • Backscatter of ground and canopy are retrieved using a zeroth order forward model

  • Stem volume is the preferred forest parameter, but also canopy cover is feasible

  • Sentinel-1 enables continuous near real-time monitoring of soil freeze/thaw state

Abstract

A method for the retrieval of soil freeze/thaw (F/T) state in the boreal forest region using SAR is presented in this paper. The method utilizes Sentinel-1 data and is thus suitable for continuous near real-time monitoring. The main challenge with the C-band VV-polarization signal is the sensitivity to vegetation and especially to forest canopies. A relatively simple zeroth-order model is used for the retrieval of the ground and the canopy backscatter contributions in 1 km cell size. These backscatter components are then used to identify the F/T state of the soil by comparing them to corresponding reference values representing frozen and thawed conditions. The classification algorithm is based on threshold values applied on the Euclidian distances between the retrieved backscatter and the reference values. The method is tested for three test areas across Finland, having different forest properties: Sodankylä, Nurmes and Tampere, located in northern, central and southern Finland, respectively. We first evaluated whether the use of canopy cover (CC) or stem volume (SV) as the parameter describing the forest conditions provide better model accuracy. We then assessed the Sentinel-1 based soil F/T estimates by comparing them to automatic in situ observations and the SMOS (Soil Moisture and Ocean Salinity) based soil F/T product. The model performance was generally better when SV was used as the forest parameter. Nevertheless, for both CC and SV, the RMSE between the modeled and the observed backscatter was considerably lower than the seasonal variation of the backscatter. In Sodankylä and Nurmes, the Sentinel-1 based F/T estimates were well in line with the in situ observations and the SMOS F/T product. The Sentinel-1 retrievals measuring the top soil layer were fast to react to air temperature changes between negative and positive Celsius degrees, showing similarity of 94–99% with the air temperature measurements. In Tampere the method showed weaker results; the similarity with the air temperature observations was 64%. Overall, a correct vertical freezing pattern of the soil was demonstrated in this study, with Sentinel-1 sensitive to the top soil layer, in situ sensors measuring at 5 cm depth, and SMOS reaching to 5–15 cm soil depth. Additional assessment should be conducted in southern Finland.

Introduction

Sentinel-1 (S1) satellites provide frequent and regular observations of microwave backscatter at a relatively high spatial resolution. These data are available almost on a daily basis from the boreal forest regions of northern Europe. The C-band observations enable the detection of soil conditions for the top soil layer up to 5 cm depth, such as soil freezing and thawing (Bruckler et al., 1988; Ulaby et al., 1982). Thanks to the short revisit times, these estimates can be used for continuous near real-time monitoring of the soil freeze/thaw (F/T) state, useful for e.g. the estimation of terrain trafficability for forest machinery or other heavy equipment, and estimation of soil hydrology. A major limitation when using C-band observations for soil applications, compared to the lower frequency L- or P-bands, is the decreased penetration of the signal through forest canopies (Lang et al., 2008; Townsend, 2002; Ulaby et al., 1986; Wang et al., 1995). Nevertheless, forest backscatter models can be used to extract the signal components scattered from the ground surface and the forest canopy (Liang et al., 2005; Lin and Sarabandi, 1999; Liu et al., 2010; Sun and Ranson, 1995), thus allowing a compensation of the effect of the forest canopies on the observed signal (Cohen et al., 2015; Koskinen et al., 2010; Luojus et al., 2009; Luojus et al., 2007; Pulliainen et al., 2001; Pulliainen et al., 1999; Pulliainen et al., 1994).

Active and passive microwave observations are ideal for detecting soil moisture conditions, because higher soil water content increases the dielectric constant of the soil. Soil freezing decreases the amount of liquid water, thus lowers the dielectric constant of the soil. For active instruments, this causes higher penetration of the incoming radiation into the soil, leading to a decreased returning signal, namely lower backscatter (Ulaby et al., 1982). As to passive microwave radiometry, soil water content influences the microwave emission, thus altering the measured brightness temperature (Njoku and Entekhabi, 1996; Ulaby et al., 1986).

SAR instruments are particularly useful in soil F/T retrievals of local to regional scales, due to their high spatial resolution. However, there are some limitations concerning the signal properties and imaging geometry which should be considered. The main factors influencing the applicability of SAR in soil F/T detection are signal frequency and polarization, and in some extent, also imaging angle. All these factors are related to canopy penetration. Higher canopy penetration, which is of course an advantage in soil related applications, can be achieved with lower frequency co-polarized signals (Evans et al., 1986; Kuga et al., 1990; Wu and Sader, 1987), and typically with steeper incidence angles (Lang et al., 2008). Lower frequencies such as L- and P-bands are less sensitive to smaller sized particles such as canopy branches and leaves/needles, or snow particles. The effect of volume scatter in dry fresh snow with small particle size can be considered small even at C- and X-bands. However, matured snow with larger particle size is expected to contribute to the measured backscatter, at least at X-band (Tan et al., 2015). A recent study has also found a relation between C-band's co- and cross-polarized backscatter ratio and snow depth over deep snow cover in mountains, owing to increases in the detected backscatter from snow at cross polarization (Lievens et al., 2019). Wet snow prevents the signal from reaching the soil layer for all typical SAR frequencies (Luojus et al., 2007; Matzler, 1998; Nagler and Rott, 2000), which implies, that SAR based soil F/T detection is more feasible during the freezing period and less during the spring snowmelt season. Co-polarized signals penetrate through forest canopy better than cross-polarization, and more specifically, HH-polarization penetrates better than VV-polarization, due to the vertical orientation of the tree trunks (Bourgeau-Chavez et al., 2001; Lavender and Lavender, 2015; Pierdicca et al., 2013; Townsend, 2002; Wang et al., 1995). Steeper incidence angles are usually preferable in soil applications, because the canopy volume along the signal propagation path decreases in more vertical directions. The signal frequency is also related to the soil penetration depth, such that lower frequencies penetrate deeper into the soil. C-band signal's soil penetration have been shown to vary between 1 and 5 cm, in wet and dry soil respectively, whereas the penetration depth of L-band signals is up to 15 cm (Bruckler et al., 1988; Rautiainen et al., 2012; Ulaby et al., 1982).

Previous studies have investigated the use of active microwave sensors in general, and SAR in particular, for the estimation of soil F/T state. Studies utilizing C-band SAR observations for soil F/T state retrieval started with ERS-1 (Rignot and Way, 1994), and continued with Envisat (Park et al., 2011). Soil freezing has an impact also on the dynamics of the polarimetric entropy and mean scattering alpha angle, enabling the detection of soil F/T state with fully polarimetric C-band data (Jagdhuber et al., 2014). Recent studies show that S1 data has the potential to enable continuous monitoring of soil F/T state in agricultural areas when combined with a semi-empirical dielectric constant model (Baghdadi et al., 2018; Fayad et al., 2020), in boreal forests when using a semi-empirical forest backscatter model (Cohen et al., 2019) and in arctic environments due to large backscatter dynamic ranges caused by soil freezing and thawing (Rodionova, 2019). A method for the retrieval of soil F/T designed for the SMAP (Soil Moisture Active Passive) L-band active sensor was introduced by Xu et al. (2016), who tested the method on Aquarius L-band scatterometer observations. The algorithm is based on a seasonal thresholding approach, where the soil state is estimated by comparing the observed backscatter with reference values representing freeze and thaw conditions. A soil surface state product using the C-band ASCAT (Advanced Scatterometer) instrument provided in 25 km spatial resolution has also been developed (Naeimi et al., 2012).

Several methods utilizing passive microwave instruments for detecting the F/T state of the ground surface can also be found in literature. Methods based on relatively high frequency K-band (Kim et al., 2011) and Ka-band (Zhao et al., 2011) instruments have been introduced, however, these methods acknowledge the interference of factors such as forest canopy and the presence of snow cover to the retrieved F/T estimates. Following the launch of SMOS (Soil Moisture and Ocean Salinity) and SMAP radiometers operating at L-band, more efficient methods for the detection of soil F/T state have been developed (Derksen et al., 2017; Rautiainen et al., 2016; Rautiainen et al., 2014), less affected by the presence of vegetation. For SMOS and SMAP, the F/T state of the soil is estimated from the brightness temperature measurements having a spatial resolution of a few tens of kilometers and a daily global coverage. The observations are further resampled to the Equal-Area Scalable Earth (EASE) grid, providing daily estimates of the soil F/T state globally in 36 km and 9 km pixel size for SMAP, and for the Northern Hemisphere in 25 km pixel size for SMOS.

The optimal microwave sensor for monitoring the soil F/T state in a local to a regional scale in the boreal forest region would be, based on the criteria mentioned above, a relatively low frequency SAR, such as L-band, and preferably HH-polarization. This would allow better canopy penetration and deeper soil penetration compared to higher frequency C- or X-bands, and higher spatial resolution compared to passive microwave sensors. Fully polarimetric C-band SAR could also be suitable (Jagdhuber et al., 2014). Nonetheless, the availability of L-band or fully polarimetric SAR data is currently not sufficient for continuous monitoring purposes, which leaves the C-band single polarization as the only valid option for near real-time monitoring of soil F/T state in a higher spatial resolution than currently provided by the methods based on passive microwave observations. The S1 satellites do not provide continuous HH-polarization data from land areas, and therefore VV-pol was used here.

The goal of this work was to develop and evaluate a method for the retrieval of soil F/T state in the boreal forest region, suitable for near real-time monitoring. A relatively simple forest backscatter model was first utilized to extract the contributions of the ground and the canopy to the total observed backscatter. We assessed whether the use of canopy cover (CC) or stem volume (SV) as the forest parameter provide better model performance, namely a better similarity between the modeled and the observed backscatter in a range of various forest properties. The retrieved ground and canopy backscatter values were then used to classify the observations to frozen, thawed and uncertain soil conditions according to their Euclidean distances from corresponding reference freeze and thaw values in a 2-d space. The method was applied for three test areas in Finland, each one representing a different type of boreal forest environment. The S1 soil F/T retrieval results were evaluated by comparing them with in situ measurements and with the SMOS based soil F/T product (Rautiainen et al., 2019; Rautiainen et al., 2016). The in situ measurements included air temperature observations and soil F/T indices derived from liquid water content (LWC) and temperature measurements at 5 cm soil depth.

Section snippets

Test areas and data

Three test areas were used to assess the developed soil F/T retrieval method; Sodankylä, Nurmes and Tampere, located in northern, central and southern Finland, respectively. Although all these areas are in the boreal forest zone, differences regarding forest properties exist, mainly in terms of tree density, height and species. These may have influence on the performance of the model applied in the retrieval algorithm. Near Sodankylä, forests contain predominantly needle-leaved pine trees,

Methods

This section first explains how the soil F/T index was calculated from the in situ measurements in Sodankylä and Tampere test areas (section 3.1). Then, the model used for the extraction of the ground and the canopy backscatter components is introduced in section 3.2, and the methodology used for the classification of the retrieved values to frozen, thawed and uncertain soil conditions is presented in section 3.3.

Results

Simulation of the SAR backscatter with respect to varying forest properties was performed with the zeroth order backscatter model once, when forest CC was used as the parameter describing the forest properties (FP in Eqs. (6), (7)) and again, when SV was used as the forest parameter. A comparison of the model results when changing the forest parameter used is presented in section 4.1 of this chapter. The feasibility of the presented method for the detection of the F/T state in forested terrain

Discussion

A method for the retrieval of soil F/T state in the boreal forest region using S1 VV-pol SAR data, suitable for near real-time monitoring, is demonstrated and assessed in this paper. The developed method is based on the approach introduced by Cohen et al. (2019), who demonstrated that the retrieved ground and canopy backscatter values representing freeze conditions can be separated from values representing thaw. Cohen et al. (2019) used CC as the forest parameter in the forward model and

Conclusions

A method for the estimation of the F/T state of the soil in the boreal forest region with Sentinel-1 (S1) data is presented in this paper. The soil F/T is estimated by comparing the retrieved ground and the canopy-ground backscatter difference values to the corresponding reference values representing freeze and thaw conditions. The ground and the canopy backscatter values are retrieved by applying a zeroth order model on the C-band data, simulating the backscatter as a function of a parameter

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.

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests:

References (68)

  • S.E. Park et al.

    Monitoring freeze/thaw cycles using ENVISAT ASAR global mode

    Remote Sens. Environ.

    (2011)
  • N. Pierdicca et al.

    Observing floods from space: experience gained from COSMO-SkyMed observations

    Acta Astronaut.

    (2013)
  • J. Pulliainen et al.

    Semi-empirical modeling of the scene reflectance of snow-covered boreal forest: validation with airborne spectrometer and LIDAR observations

    Remote Sens. Environ.

    (2014)
  • K. Rautiainen et al.

    Detection of soil freezing from L-band passive microwave observations

    Remote Sens. Environ.

    (2014)
  • K. Rautiainen et al.

    SMOS prototype algorithm for detecting autumn soil freezing

    Remote Sens. Environ.

    (2016)
  • E. Rignot et al.

    Monitoring freeze-thaw cycles along north-south Alaskan transects using ERS-1 SAR

    Remote Sens. Environ.

    (1994)
  • A. Roy et al.

    L-band response to freeze/thaw in a boreal forest stand from ground- and tower-based radiometer observations

    Remote Sens. Environ.

    (2020)
  • Y. Wang et al.

    Understanding the radar backscattering from flooded and nonflooded Amazonian forests: results from canopy backscatter modeling

    Remote Sens. Environ.

    (1995)
  • N. Baghdadi et al.

    Detection of frozen soil using Sentinel-1 SAR data

    Remote Sens.

    (2018)
  • L.L. Bourgeau-Chavez et al.

    Analysis of space-borne SAR data for wetland mapping in Virginia riparian ecosystems

    Int. J. Remote Sens.

    (2001)
  • J. Cohen et al.

    The effect of boreal Forest canopy in satellite snow mapping-a multisensor analysis

    IEEE Trans. Geosci. Remote Sens.

    (2015)
  • J. Cohen et al.

    A modeling-based approach for soil frost detection in the northern boreal forest region with C-band SAR

    IEEE Trans. Geosci. Remote Sens.

    (2019)
  • O. Conrad et al.

    System for automated geoscientific analyses (SAGA) v. 2.1.4. Geosci

    Model Dev.

    (2015)
  • D.R. DeWalle et al.

    Principles of snow hydrology, Principles of Snow Hydrology

    (2008)
  • F. Domine et al.

    Seasonal evolution of the effective thermal conductivity of the snow and the soil in high Arctic herb tundra at Bylot Island

    Canada. Cryosph.

    (2016)
  • D.L. Evans et al.

    Multipolarization radar images for geologic mapping and vegetation discrimination

    IEEE Trans. Geosci. Remote Sens.

    (1986)
  • I. Fayad et al.

    Near real-time freeze detection over agricultural plots using Sentinel-1 data

    Remote Sens.

    (2020)
  • J. Ikonen et al.

    The Sodankylä in situ soil moisture observation network: an example application of ESA CCI soil moisture product evaluation. Geosci

    Instrumentation, Methods Data Syst.

    (2016)
  • J. Ikonen et al.

    Spatially distributed evaluation of ESA CCI soil moisture products in a northern boreal Forest environment

    Geosciences

    (2018)
  • T. Jagdhuber et al.

    Identification of soil freezing and thawing states using SAR polarimetry at C-band

    Remote Sens.

    (2014)
  • M. Kärkkäinen

    Puutiede

    (1985)
  • Y. Kim et al.

    Developing a global data record of daily landscape freeze/thaw status using satellite passive microwave remote sensing

    IEEE Trans. Geosci. Remote Sens.

    (2011)
  • J.T. Koskinen et al.

    Seasonal comparison of HUTSCAT ranging scatterometer and ERS-1 SAR microwave signatures of boreal forest zone

    IEEE Trans. Geosci. Remote Sens.

    (1999)
  • J.T. Koskinen et al.

    Monitoring of snow-cover properties during the spring melting period in forested areas

    IEEE Trans. Geosci. Remote Sens.

    (2010)
  • Cited by (12)

    • Retrieving freeze-thaw states using deep learning with remote sensing data in permafrost landscapes

      2024, International Journal of Applied Earth Observation and Geoinformation
    View all citing articles on Scopus
    View full text