Ensemble surface soil moisture estimates at farm-scale combining satellite-based optical-thermal-microwave remote sensing observations

Abstract

Remote sensing presents great potential for soil moisture mapping, although a lack of high-resolution, all-season suitable data and model-based complications hinders continuous mapping at the farm scale. The current study attempts to access optical (Sentinel-2) and microwave (Sentinel-1) based Water Cloud Model (WCM), optical and/or thermal-based Thermal-Optical TRApezoid Model (TOTRAM), and Optical TRApezoid Model (OPTRAM) utilizing Landsat-8 data, and an ensemble of these three models synergistically using microwave, optical and thermal remote sensing data for fourteen dates spanning two post-monsoon periods over a semi-arid irrigated agricultural farm. The ensemble model developed with Gradient Boosting Machine (GBM) offered a superior SSM estimate for the entire crop growth season. The ensemble model had the lowest RMSE (0.053 m3 m−3), followed by OPTRAM (SWIR2) (0.06 m3 m−3), while TOTRAM recorded the highest RMSE (0.092 m3 m−3), even though the model performances varied throughout crop growth cycles. For SSM mapping in the early phases of crop growth, WCM or TOTRAM were found superior, while during the latter growth stages, the performance of OPTRAM was better. In contrast to TOTRAM, which needs local calibration as the land surface temperature is sensitive to atmospheric conditions, the study emphasizes the shortcomings of WCM in terms of calibration requirements for changing vegetation structure utilizing in situ data. OPTRAM is suitable for generating SSM maps in semi-arid environments with a predominance of clear skies throughout the post-monsoon season. It is straightforward, low data, and resource intensive with universal surface reflectance-soil moisture association. The ensemble technique can be recommended for farm-level irrigation mapping over a post-monsoon period.