Prediction of Soil Organic Matter in Black Soil Region Based on BP Neural Network with Genetic Algorithm Variable Selection

Soil organic matter (SOM) serves as a crucial indicator for assessing soil fertility and cultivated land quality. This study developed a SOM prediction method based on a back propagation neural network with genetic algorithm variable selection. The experimental area was in Hailun City, a typical black soil region of Heilongjiang Province, China. A total of 157 sampling points were selected at a depth of 0-20 cm. The Landsat 8 OLI images from the bare soil period were used for calculating the spectral indexes. The Shuttle Radar Topography Mission (SRTM) data was utilized to extract topographic factors. Then the multiple stepwise regression (MSR) and genetic algorithm (GA) methods were used to screen the variables as input, respectively. Multiple linear regression (MLR) and back propagation neural network (BPNN) methods were utilized to build the SOM prediction model. The root mean square error (RMSE) and correlation coefficient (R2) were employed to evaluate the accuracy of SOM prediction. Finally, the optimal model was used to predict the SOM in the black soil area. The results indicated that the GA outperformed MSR in variable selection. The BP neural network (with GA-selected variables GA_BP) has the highest accuracy for the SOM prediction, with an R2 of 0.94 and an RMSE of 0.14, followed by the BP neural network with a stepwise regression-selected factor (MSR_BP) with an R2 of 0.91 and an RMSE of 0.24. The R2 of BPNN was 0.84 and the RMSE was 0.29, while MSR showed the lowest accuracy, with an R2 of 0.69 and an RMSE of 0.57. This study not only proved the advantages of GA in significantly improving model accuracy by decreasing the dimensionality of the inputsbut also provided effective methods for improving the precision of SOM remote sensing prediction, showing potential applications in predicting SOM in black soil areas.