A regional model for soil moisture concentration from remote sensing data and its application for agricultural drought hazard mapping

Abstract

Soil moisture constitutes one essential variable in agriculture drought monitoring. However, because spatial and temporal soil moisture datasets from in situ observations are not accessible for all locations, remote sensing constitutes an indispensable approach in the assessment of surface soil moisture on a regional scale. In this study, a method to estimate regional-scale distribution of soil moisture (0–30 cm) from remote sensing observations is presented and applied to produce a drought hazard map, taking as case study area the arid region of Jiroft plain, Iran. For this study area, we dispose of remote sensing data available within the broad time span from 2007 to 2022, including satellite Vegetation Index and Land Surface Temperature, as well as observed soil moisture at a regional scale with a spatial resolution of 1 km². Based on the spatial distribution of soil moisture appraise from these datasets, we calculate the relative exit of soil moisture associated with eight severe droughts in the Jiroft plain and the associated inventory map of agricultural drought. Machine learning models, including improved regression trees, multivariate discriminant analysis and support vector machine, are then applied to predict agricultural drought hazards. Using these different models, a model for agricultural drought hazard (ADH) is produced from ten independent variables characterizing environmental factors in the area. We find that plant available water capacity constitutes, together with soil moisture, the most important factor in ADH modeling. Furthermore, our results further indicate that, over the machine learning methods considered in our study, the support vector machine leads to the highest model accuracy in agricultural drought mapping (AUC = 0.95). We show how the ADH estimated with our model can be applied for predicting drought occurrence throughout Jiroft plain in future years. The results of our study provide quantitative information for drought risk assessment and management in Jiroft plain, and deliver insights that will help in the future development of agricultural drought hazard mapping in other arid regions of our planet – especially in areas with limited hydro-meteorological data.