Artificial Neural Networks in Soil Quality Prediction: Significance for Sustainable Tea Cultivation

Abstract

In today's era artificial intelligence is quite popular, one of the most effective algorithms used is Artificial Neural Networks (ANN). In this study, the determination of soil quality using the Soil Management Assessment Framework (SMAF) model in areas where tea cultivation is carried out at the micro-watershed scale and the predictability of soil quality using ANN were evaluated. According to the results, the soil quality indices of teagrowing areas were generally classified as "medium " between 55 and 70 %. Among the evaluated features for determining soil quality, the highest relative importance value was for soil organic carbon content (13 %) and potential mineralizable nitrogen (13 %), whereas the lowest values were for exchangeable potassium (4 %) and sodium adsorption ratio (SAR) (4 %). In addition, when comparing the actual and predicted values for soil quality prediction using ANN, the Lin's concordance correlation coefficient (LCCC), ratio of performance to deviation (RPD), and R 2 values were found to be 0.93, 2.95, and 0.89, respectively. Significant properties for the determined values within a 90 % predicted interval were found to be organic matter, microbial biomass carbon, bulk density, and aggregate stability of the soils. Moreover, the uncertainty values (standard deviation) in the model predictions were determined to be within the range of 1.01 -4.56 %. Consequently, the Soil Quality Index (SQI) obtained from the SMAF model using 12 soil properties in tea-growing areas could be accurately predicted using ANN. As a result of this study, digital maps showing the spatial distribution of SQI and the predicted uncertainties can help monitor SQI levels in this area.