Publications

Khiavi, AN (2025). Application of remote sensing indicators and deep learning algorithms to model the spatial distribution of soil moisture. EARTH SCIENCE INFORMATICS, 18(2), 404.

Abstract
This study aimed to spatially model Soil Moisture (SM) in the Cheshmeh-Kileh Watershed, Iran, using a comprehensive approach that integrated geo-environmental criteria, Remote Sensing (RS) indices, and advanced Deep Learning Algorithms (DLAs), specifically Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Recurrent Neural Networks (RNN). The research utilized a rich dataset in two categories of geo-environmental variables and RS indicators related to drought derived from MODIS products, Sentinel-2, and Landsat 8 via Google Earth Engine (GEE). Python was employed to implement the DLAs for SM modeling, allowing for the classification of spatial variations into very low, low, moderate, high, and very high categories. Among the DLAs, the RNN model was identified as the most accurate, exhibiting an R-2 of 0.950, Mean Absolute Error (MAE) of 0.014, Mean Square Error (MSE) of 0.012, and an Area Under the Curve (AUC) of 0.98. The analysis revealed a notable gradient in SM levels, which decreased from the watershed's mid-reaches to the upstream areas, in conjunction with reduced RS indices (GVMI, PDI, SWCI), indicating diminished moisture retention capabilities. Furthermore, the study identified a strong correlation (0.91) between the outputs of the CNN and RNN models, while Alpha Diversity Indices (ADI) such as Simpson and Shannon exhibited consistent variability across the modeling approaches. The distribution of SM levels across sub-watersheds revealed that 46.67% fell into the very low category, 13.33% as low, 20% as moderate, 6.67% as high, and 13.33% as very high. Collectively, these findings underscore a critical and significant deficit of SM in the upstream areas of the watershed, with implications for water resource management and land use planning in the region.

DOI:
10.1007/s12145-025-01842-9

ISSN:
1865-0481