Deep learning algorithm-based carbon emission assessment of agricultural soils

Agriculture is the second largest source of carbon in the world, and in the context of global climate change, studying the spatial and temporal variability of carbon emissions from agricultural soils is an important indicator for predicting future changes in atmospheric CO2 concentrations. This study establishes a soil CO2 emission prediction model for agricultural soils based on RBF neural network algorithm, taking soil water content, temperature, organic carbon, ammonium nitrogen and nitrate nitrogen content as input signals and soil CO2 emission fluxes during the crop growing period as output signals. The effectiveness of this prediction model was evaluated by choosing multiple linear regression (MLR) and BP neural network models. The correlation coefficient of the neural network prediction model ( R2 = 0.975 ) was higher than other models (linear and nonlinear regression models), and its mean square error ( RMSE=0.091 ), mean absolute error ( MAE=0.048 ) are lower than the other models. The study shows that the RBF neural network algorithm prediction performance is significantly better than other prediction models, and the accuracy is higher, which can better predict soil CO2 emission fluxes.