A Machine Learning-Based High-Resolution Soil Moisture Mapping and Spatial–Temporal Analysis: The mlhrsm Package

Abstract

Soil moisture is a key environmental variable. There is a lack of software to facilitate non-specialists in estimating and analyzing soil moisture at the field scale. This study presents a new open-sourced R package mlhrsm, which can be used to generate Machine Learning-based high-resolution (30 to 500 m, daily to monthly) soil moisture maps and uncertainty estimates at selected sites across the contiguous USA at 0–5 cm and 0–1 m. The model is based on the quantile random forest algorithm, integrating in situ soil sensors, satellite-derived land surface parameters (vegetation, terrain, and soil), and satellite-based models of surface and rootzone soil moisture. It also provides functions for spatial and temporal analysis of the produced soil moisture maps. A case study is provided to demonstrate the functionality to generate 30 m daily to weekly soil moisture maps across a 70-ha crop field, followed by a spatial–temporal analysis.

Keywords

• remote sensing
• quantile random forest
• visualization
• leaflet
• spatial–temporal analysis
• water resources management