Monitoring Peatland Water Table Depth with Optical and Radar Satellite Imagery

Peatland water table depth (WTD) and wetness have widely been monitored with optical and synthetic aperture radar (SAR) remote sensing but there is a lack of studies that have used multi-sensor data, i.e., combination of optical and SAR data. We assessed how well WTD can be monitored with remote sensing data, whether multi-sensor approach boosts explanatory capacity and whether there are differences in regression performance between data and peatland types. Our data consisted of continuous multiannual WTD data from altogether 50 restored and undrained Finnish peatlands, and optical (Landsat 5–8, Sentinel-2) and Sentinel-1 C-band SAR data processed in Google Earth Engine. We calculated random forest regressions with dependent variable being WTD and independent variables consisting of 21 optical and 10 SAR metrics. The average regression performance was moderate in multi-sensor models (R2 43.1%, nRMSE 19.8%), almost as high in optical models (R2 42.4%, nRMSE 19.9%) but considerably lower in C-band SAR models (R2 21.8%, nRMSE 23.4%) trained separately for each site. When the models included data from several sites but were trained separately for six habitat type and management option combinations, the average R2 was 40.6% for the multi-sensor models, 36.6% for optical models and 33.7% for C-band SAR models. There was considerable site-specific variation in the model performance (R2 −3.3–88.8% in the multi-sensor models ran separately for each site) and whether multi-sensor, optical or C-band SAR model performed best. The average regression performance was higher for undrained than for restored peatlands, and higher for open and sparsely treed than for densely treed peatlands. The most important variables included SWIR-based optical metrics and VV SAR backscatter. Our results suggest that optical data works usually better than does C-band SAR data in peatland WTD monitoring and multi-sensor approach increases explanatory capacity moderately little.