Downscaling and validating SMAP soil moisture using a machine learning algorithm over the Awash River basin, Ethiopia

Area of Science:

• Environmental Science
• Remote Sensing
• Hydrology

Background:

• Global soil moisture data from Soil Moisture Active Passive (SMAP) satellites offer valuable insights but suffer from coarse spatial resolution (36km x 36km).
• This limitation restricts their application in basin-scale hydrological studies and water resource management.
• Surface roughness significantly influences microwave remote sensing signals, necessitating its evaluation for accurate soil moisture retrieval.

Purpose of the Study:

• To evaluate the capability of Synthetic Aperture Radar (SAR) for retrieving surface roughness variables in the Awash River basin.
• To assess the performance of a Random Forest (RF) regression model for downscaling SMAP satellite soil moisture data over the Awash River basin.
• To validate the downscaled soil moisture data using in-situ measurements within the river basin.

Main Methods:

• A Random Forest (RF) based downscaling approach was employed to enhance the spatial resolution of SMAP soil moisture data from 36km x 36km to 1km x 1km.
• SAR data was utilized to retrieve surface roughness variables, contributing to the downscaling process.
• In-situ soil moisture measurements from the Middle and Upper Awash sub-basins were used for validation.

Main Results:

• Fine spatial resolution (1km) soil moisture data for the Awash River basin was successfully generated.
• The downscaled soil moisture product exhibited a strong spatial correlation with the original SMAP data, providing richer soil moisture information.
• Validation revealed a Pearson correlation of 0.69 between downscaled and in-situ soil moisture, outperforming the original SMAP data's correlation of 0.53.

Conclusions:

• The Random Forest regression model effectively downscales SMAP satellite soil moisture data to a finer resolution, enhancing its utility for basin-scale applications.
• The downscaled soil moisture data demonstrates improved accuracy and provides more detailed spatial information compared to the original coarse-resolution product.
• Future research should explore the temporal aspects of SMAP satellite soil moisture product downscaling in the study region.