Simulation of soil CO2 efflux under different hydrothermal conditions based on general regression neural network

Soil respiration (Rs) is an important component of global carbon (C) cycle and represents the second largest C exchange between atmosphere and geosphere. Regression models have been widely applied to describe Rs process and its relations to environmental factors in terrestrial ecosystems. However, the development of these semi-empirical regression model needed a large number of observation data in order to chives a reliable result. The successful performance of the regression model was highly dependent on data quality. In this study, a general regression neural network (GRNN) model and six validated two-factor-semi-empirical regression models were compared to stimulate changes of Rs under the influence of soil temperature (T-soil) and soil moisture (W-soil) alone and combination in camphor forests in subtropical China. The results showed the GRNN model produced greater accuracy than the regression models in predicting Rs. The R-2 ranged 0.773-0.809 for the six two-factor regression models, but 0.84 for the GRNN model, with calculated RMSE of 0.404-442 in the regression models compared to 0.20 in the GRNN model. The dataset expanded by GRNN model could better fit the semi-empirical model than the observation dataset, which indicated the GRNN model had satisfactory generalization properties. Additionally, the GRNN model revealed the non-linear relationship between Rs and W-soil when W-soil was not a limiting factor, while the regression models were hard to detect the internet linkage. Therefore, GRNN model can not only be considered as a method to provide more accurate predication of Rs in forest ecosystems, but also provide an optional scheme for studying Rs under extreme and long-term climate change.

Automated summary

This study compared a general regression neural network (GRNN) model and six validated two-factor-semi-empirical regression models to simulate changes in soil respiration (Rs) under the influence of soil temperature (T-soil) and soil moisture (W-soil) in camphor forests in subtropical China. The GRNN model outperformed the regression models in predicting Rs, and it also revealed the non-linear relationship between Rs and W-soil.