Soil Moisture Mapping in Indian Tropical Island Ecosystem using C-band SAR Data and Neural Network Models

Soil moisture is one of the critical variables in identifying the stress or health of plants and its accurate information in agricultural land use is of greater importance under the threat of climate change. This study focuses on mapping the spatio-temporal dynamics of soil moisture in the Indian tropical Island ecosystem (Andaman and Nicobar Islands) using the fusion of radar (Sentinel-1A C-band synthetic aperture radar (SAR)) and optical/thermal (Sentinel-2A and Landsat 8) data. Field-based soil moisture measurements using the gravimetric method are monitored for different land uses of the study area. The study was carried out in the South Andaman district (E 92°’ to E 94°’ and latitude N 6°’ to N 14°’). A total of 60 surface soil samples (0–10 cm) were collected from four predominant land uses of the study area for 2020-22 years to represent real-time soil moisture status. Soil moisture index (SMI) is assessed based on thermal remote sensing data as Land Surface Temperature (LST) besides, normalized difference vegetation index (NDVI) from red and infrared bands data, and dielectric constants from soil textural analysis. Artificial Neural Network (ANN) based models were developed along with multiple linear regression (MLR) to retrieve the soil moisture accurately using satellite and field-derived input parameters such as backscatter coefficients (σ°: VV & VH), NDVI, SMI, and dielectric constants. The performance of modelled soil moisture is evaluated using different statistical index-based criteria concerning field-based soil moisture measurements (SMCv). It is found that positive correlation among ( σ °: VV+VH) and (SMCv: %) for all land uses and high agreements (R ² values) between them for barren and vegetable fields. The vegetation interferes the backscatter signal and misinterprets the soil moisture estimation solely with only SAR data. However, consideration of NDVI and SMI improves the soil moisture estimation in case of vegetation abundance land uses. The comparative results showed that ANN models surpass the MLR models in soil moisture estimation with high R ² (0.67-0.99) and η (62.6-99.9) and low RMSE (0.05-2.19%) and MAE (0.03-1.74%) values without over or under estimations (R _ratio ≈ 1) in South Andaman area of Andaman and Nicobar tropical ecosystem. This kind of study helps as baseline information for hydrological modelling and policy makers to plan irrigation systems, accurate management of crop quantity and quality in farms, and schedule critical irrigations accurately.