High-Resolution Mapping of Peatland Water Table Depth Using Innovative Multi-Level Downscaling of SMAP Data

The Water Table Depth (WTD) in peatlands plays a crucial role in habitats, agriculture, and CO2 emissions. WTD observations often face limitations in terms of record length and spatial distribution, which can impact modeling results. Soil Moisture Active Passive (SMAP) data, with its sub-daily temporal resolution, provides a valuable resource for WTD monitoring in peatlands. However, SMAP data with an 11-km spatial resolution is large-scale and requires downscaling to achieve finer resolution for detailed analysis. In this study, an innovative downscaling technique was used to convert the 11-km SMAP-WTD data into 10-m resolution. Employing a multi-level machine learning downscaling approach, the SMAP-WTD data is first downscaled from 11-km to 1-km, and subsequently from 1-km to 10-m using input data at corresponding scales. Elevation, land use, precipitation, and NDVI were used as independent variables, and the Classification and Regression Trees (CART) algorithm was applied for downscaling SMAP-WTD. The model's performance was evaluated using R, RMSE, MBE, and MAE indices, while the TRE index was employed to assess the importance of the model inputs.