SMAP products for prediction of surface soil moisture by ELM network model and agricultural drought index

In recent decades, the earth's surface data have been collected more efficiently using remote sensing, which needs drought indexes update. In this study, soil moisture (SM) data were collected from the surface layer of a high humidity climate in northern Iran using Soil Moisture Active Passive (SMAP) and field measurement. After analyzing the data, we found that the average RMSE between the field and SMAP measurement was 0.054 m(3)/m(3). Considering the same agricultural land use and the strong correlation of 0.92 between them, the validated SMAP data were used to propose an agricultural drought index. After data validation, the extreme learning machine (ELM) model was put to the test using sigmoid, triangular, sine, and hard-limit activation functions. Of all the activation functions tested, the model with the sigmoid activation function yielded the lowest amount of error and was therefore chosen. Five years of continuous daily SM as a target, five-year daily normalized difference vegetation index, land surface temperature, and precipitation were inputs to predict one-year daily SM time series in the humid climate. From 2021 to 2022, daily surface SM was predicted with the average RMSE = 0.03 m(3)/m(3) compared to the SMAP data. Finally, a new regional agricultural drought index based on 4 years of SMAP and 1-year prediction of SMAP from 2022 to 2023 was proposed. Further investigation is needed to conclude that the application of the presented index is reliable in other climates.