Artificial Intelligence Statistical Analysis of Soil Respiration Improves Predictions Compared to Regression Methods

Soil respiration (SR) is strongly affected by soil water content and soil temperature. The common methods for SR measurement are costly, time consuming, and laborious. The goals of this investigation were (i) to predict SR by artificial intelligence and regression models as new tools in soil science, (ii) to evaluate which methods are better, and (iii) to determine the most influential factors on SR by sensitivity analysis. SR, moisture content, and diurnal and nocturnal temperatures were determined at four stages throughout growing season. The effect of soil moisture content and diurnal and nocturnal soil temperatures on SR was predicted using artificial neural networks (ANN), adaptive neuro-fuzzy inference system (ANFIS), fuzzy inference system (FIS), particle swarm optimization (PSO), partial least squares (PLS), principal components regression (PCR), ordinary least squares (OLS), and multiple regression (MR). The results showed that artificial intelligence methods (except for FIS) had higher performance than regression methods. Among intelligent models, ANFIS could predict the SR more accurate relative to ANN and PSO in terms of statistical parameters (R2 = 0.93, MSE = 0.001, RMSE = 0.037, MAPE = 12.26, VAF = 93.39, RPD = 3.82). ANFIS model was more efficient than others, and time taken to run ANFIS was shorter than the other models. Sensitivity analysis illustrated that soil moisture content was a more influential parameter on SR compared to diurnal and nocturnal soil temperatures. As a result, soil moisture mainly controls the SR in this study.