Modeling agricultural soil bulk density using artificial neural network and adaptive neuro-fuzzy inference system

Developing models that can accurately predict the soil bulk density using other soil parameters is of great importance given the arduous task of determining the soil bulk density. We conducted the factorial experiment based on the randomized complete block design with five replications to determine the factors that affect the soil bulk density (BD) in three types of soil texture: loam, sandy loan, and loamy sand. Artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) were applied to predict and model the soil bulk density using several independent parameters that affect it, including the soil cone index (CI), moisture content (MC), and electrical conductivity (EC). The analysis indicated that the developed ANN using the Bayesian tuning algorithm with R-2 = 0.93 is the most suitable model compared to other models that were created. We also performed the soil bulk density modeling using the three effective parameters of soil by ANFIS (adaptive neuro-fuzzy inference system) applying the hybrid method. The coefficient of determination for the ANFIS model was 0.988 (R-2 = 0.98), which indicates the correct choice of the parameters affecting the soil bulk density. The comparison between the artificial neural network models and the neuro-fuzzy model developed in this study shows the complete advantage of ANFIS systems in predicting the soil bulk density as supported by the statistical parameters The results showed that ANN and ANFIS are highly capable to predict the soil bulk density in agricultural lands.

Automated summary

Developing accurate models to predict soil bulk density is important due to the difficulty in determining it. In this study, we conducted experiments to determine the factors affecting soil bulk density in three different soil textures. Artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) were used to predict and model soil bulk density using independent parameters such as cone index, moisture content, and electrical conductivity. The ANN model developed using the Bayesian tuning algorithm with an R-2 value of 0.93 was found to be the most suitable model. The ANFIS model also showed high accuracy with a coefficient of determination of 0.988.