Predicting Soil Properties of Goshchi Rangeland in Urmia: A Comparison of Adaptive Neuro-Fuzzy Inference System (ANFIS) and Response Surface Methodology (RSM) Approaches

Understanding the spatial distribution of vegetation and its relationship with ecological factors, particularly soil properties, is crucial for effective rangeland management. This study aimed to evaluate soil properties, including organic carbon, nitrogen, phosphorus, potassium, electrical conductivity, and acidity, in Goshchi rangeland of Urmia, representing the Azerbaijani vegetation climate in West Azerbaijan province. The study utilized the adaptive neuro-fuzzy inference system (ANFIS) and response surface methodology (RSM) approaches to predict these soil properties and compare the results.

Soil properties were assessed in two sites, one under exclosure and the other under grazing, using ANFIS and RSM approaches. The collected data were used to train ANFIS in MATLAB software and RSM in Design Expert software.

The correlation coefficients (R2) between the predicted data from the ANFIS model and RSM method, and the measured data for organic carbon, nitrogen, phosphorus, potassium, electrical conductivity, and acidity were 0.75, 0.93, 0.44, 0.95, 0.68, and 0.95, respectively, and 0.84, 0.93, 0.24, 0.96, 0.54, and 0.98, respectively. The results indicated that grazing conditions (exclosure vs. grazed) and distance from plants (near vs. between plants) significantly influenced acidity, nitrogen, potassium, and organic carbon, while they had no significant effect on electrical conductivity and phosphorus. The exclosure conditions and proximity to plants resulted in higher levels of organic carbon, nitrogen, and phosphorus, highlighting the sensitivity of these factors to grazing. The RSM results showed that grazing conditions and distance from plants had no significant effect on phosphorus and electrical conductivity but had a strong significant effect on other soil properties. The lower R2 values observed for phosphorus and electrical conductivity in both ANFIS and RSM methods suggest their limited accuracy in predicting non-significant outputs.

The findings from both ANFIS and RSM methods demonstrate their effectiveness in accurately predicting soil properties under the investigated conditions. However, their performance is more accurate for factors that exhibit a significant effect on the input variables, while their accuracy is limited for non-significant outputs.