Comparison of Different Methods of Soil Moisture Mapping in Peatlands Using Polarimetric Synthetic Aperture Radar (SAR) Data

English Abstract:

Peatland surface soil moisture is important in many biophysical models but access to peatlands is often limited or difficult. Remote sensing provides a synoptic view of inaccessible areas and Synthetic Aperture Radar (SAR) is generally thought to be the most promising for soil moisture retrieval due to its sensitivity to the dielectric constant. Accurate estimation of soil moisture (volumetric water content, VWC) from remotely sensed data has been a long standing challenge and an operational solution that allows retrieval of VWC from a variety of landscapes and vegetation conditions has yet to be developed. Several different modelling approaches have been used for relating in-situ measurements SAR, however the high degree of variability in both soil and vegetation parameters is widely cited as the fundamental barrier to the development of reliable predictive models. Throughout the summer of 2014 in-situ VWC measurements were acquired at repeat monitoring stations throughout a peatland in Eastern Ontario. Measurements were acquired using two different VWC measurement devices (Hydrosense and Diviner EnviroScan) during the same day as RADARSAT-2 Fine Quad acquisitions. Here, we present results of several different modeling techniques (multiple linear regression (MLR), temporal difference analysis, and the Water Cloud Model) for estimating VWC in peatlands using polarimetrc SAR data (Fine Quad), as well as LiDAR DEM, DSM and canopy derivatives and optical data. Results show differences between the two measurement devices. With the Hydroscan (n = 32), SAR intensity and polarimetric parameters were not sufficient for predicting VWC in peatlands (MLR R² < 0.4). The addition of elevation improved the model (R² > 0.75), however it alone accounted for most of the variability in the model. Several polarimetric parameters (e.g. pedestal height, cross-pol and co-pol ratio) were found to be weakly correlated with Diviner VWC (n = 8), but their addition to SAR intensity in models increased predictability (R² from 0.2 to > 0.6). Using temporal differences between both SAR intensity and Diviner VWC, MLR resulted in R² > 0.6. The addition of LiDAR elevation and vegetation parameters improved the model (R² > 0.8). Results of the Water Cloud Model were highly variable depending on date and specific model parameters employed.