Pythagorean Fuzzy Swara Weighting Technique for Soil Quality Modeling of Cultivated Land in Semi-Arid Terrestrial Ecosystems

Abstract

Currently, the assessment of soil quality and creating digital soil maps are crucial for sustainable land management. In the present study, the main objective is to evaluate soil quality around Lake Van's agricultural areas using Pythagorean Fuzzy SWARA (PF-SWARA) weighting for soil indicator assessment. Additionally, the predictability of soil quality is demonstrated through spatial distribution maps using random forest (RF) and artificial neural network (ANN) algorithms. PF-SWARA weighting assigns higher weights to indicators of physical quality. Soil quality index (SQI) values for the study area range between 0.36 and 0.74, classified as "from very low to high." RF and ANN models provide Lin's concordance correlation coefficient (LCCC) values of 0.93 and 0.87, respectively, for soil quality prediction. The RF model exhibits the lowest error rate (root mean square error (RMSE): 0.03; mean absolute percentage error (MAPE): 4.51%). The RF algorithm identified pH, available phosphorus, organic matter, CaCO3 and electrical conductivity as the most effective soil properties for estimating SQI. Ordinary Kriging geostatistical interpolation is identified as the interpolation method with the lowest RMSE value based on observed and predicted values' spatial distribution maps using Gaussian semivariogram from the geostatistical model. The study concludes that machine learning algorithms can be utilized alongside PF-SWARA approaches for digital soil quality mapping.