Remote sensing-driven soil fertility modelling and summer rice suitability mapping in a floodplain agroecosystem of Assam

Tropical Plant Biology · Springer Nature · 2026-09-24 · DOI: 10.1007/s10333-026-01102-6

Аннотация

Sustainable food production in floodplain agroecosystem is increasingly constrained by variable soil fertility, hydrological instability, and climate variability. Despite advances in digital soil mapping, agricultural suitability assessment by integrating machine learning, remote sensing, and spatial uncertainty analysis remains limited in tropical alluvial floodplains. To address these gaps, this study presents a remote-sensing-driven framework for modelling soil fertility and summer rice suitability across the Brahmaputra floodplain of Dhubri, Assam, India. We analysed 177 stratified soil samples for organic carbon, macronutrients, micronutrients, and pH. We integrated environmental covariates- including Sentinel-2 imagery, terrain attributes, hydrological indicators, climate data, and SoilGrids products- into a Random Forest framework to predict soil properties and map spatial uncertainty. The model moderately predicted potassium (R 2 = 0.54), organic carbon (R 2 = 0.47), and phosphorus (R 2 = 0.45). However, local biogeochemical variability reduced the predictability of nitrogen and micronutrients. A fuzzy Soil Fertility Index (SFI) mapped spatial fertility, and a presence-background Random Forest model with spatial cross-validation estimated summer rice suitability. The SFI classified 40.26% of the study area as very high, 29.87% as high, 19.48% as moderate, and 10.39% as low fertility. However, the high-suitability class was restricted to only 4.21% of the area, with the remainder distributed between moderate-low (40.14%), moderate-high (32.48%) and low (23.17%) suitability. The model demonstrated moderate discrimination ability (AUC = 0.68; Area Under the Receiver Operating Characteristic Curve) with SFI as the primary predictor of summer rice suitability, followed by temperature and soil pH. These results indicate that high soil fertility does not, by itself, guarantee high modelled suitability for summer rice cultivation. This machine learning framework, integrating remote sensing, digital soil mapping, and spatial uncertainty analysis, offers a transferable tool for summer rice suitability assessment in floodplain agroecosystems such as the Dhubri district of Assam.

Читать у издателя →

Похожие статьи