This document discusses using machine learning techniques to forecast agricultural drought by incorporating high-resolution soil moisture data. It aims to 1) forecast soil water deficit index (SWDI) up to one week using support vector machines (SVM) improved with dual ensemble Kalman filters, and 2) evaluate satellite-derived soil moisture against in situ observations to assess its use in drought indices. The results show dual EnKF greatly improves SVM predictions of SWDI at different soil layers and SMAP satellite soil moisture captures the dynamics of root-zone soil moisture compared to in situ observations.