Enhancing the soil quality index model based on neutrosophic fuzzy-AHP integrated with remote sensing and artificial intelligence technique

Abstract

Intensive agricultural practices to meet food demand have led to a decline in soil quality and agricultural productivity, posing significant challenges to environmental sustainability. Consequently, the present research focused on the development of models based on artificial intelligence techniques to predict the soil quality index (SQI) for soybean (Glycine max) cultivation using a total of 89 soil samples taken at 300-m grit system at depths of 0–20 cm. A set of 28 parameters categorized into main physical, chemical (organic matter, pH, EC, etc.), fertility (macro- and micronutrient elements), and biological (soil respiration, metabolic coefficient, and microbial biomass carbon) parameters were used for the total dataset (TDS). The minimum dataset (MDS), which consisted of the most sensitive parameters, was selected using principal component analysis. In this study, SQI was calculated for both TDS and MDS using a neutrosophic fuzzy analytic hierarchy process and standard scoring function. The resulting SQI_TDS and SQI_MDS values were then predicted using machine learning approaches, including multiple linear regression (MLR) and random forest regression (RFR). The accuracy of these predictions was then examined using various metrics such as mean absolute error, mean squared error, and root mean square error. The results show that MLR outperforms RFR for both SQI_TDS and SQI_MDS with significantly lower error indices and higher R² values than RFR through 10-fold cross-validation. In addition, this study statistically compared the obtained SQI_TDS and SQI_MDS values with normalized difference vegetation index (NDVI) values derived from the Sentinel-2A satellite for May 2021. The same satisfactory R² values (0.84) were obtained by statistically comparing both SQI_TDS and SQI_MDS with NDVI values. Furthermore, this study demonstrates the effective integration of advanced techniques such as machine learning models with remote sensing and geographic information system technologies, for the analysis and processing of both original and generated information in the vast domain of SQI.