Applying An Artificial Neural Network To Simulate And Predict Chinese Fir(Cunninghamia Lanceolata) Plantation Carbon Flux In Subtropical China

C flux and meteorological data were collected over a four year period between January2008and December2011at the Huitong National Research Station of Forest Ecosystem. This study aimed to determine important non-redundant input variables to quantify C flux and to develop a new application of Genetic Neural Network (GNN) model that accurately simulates C flux. Results showed that:(1) CIV8(grouping atmospheric CO2concentration, air temperature, Par, relative humidity, wind speed, and soil temperature) performed best, yielding a correlation coefficient (R2) of0.87, Outlier of0.79%, and an RMSE of0.11. Prec has weak positive correlation with carbon flux. Par with the strongest correlation between carbon flux (R=-0.880, p=0.000). Based on the C flux and meteorological data were collected over three months period between July to September in2008at the Huitong National Research Station of Forest Ecosystem, six inputs (Ta, Par, pc, Rh, Ws, Ts), the number of hidden layer generates automatically by system, one output (carbon flux) was the best model in current environment.(2) For the entire year, R2ranged from0.77to0.83,Ivf ranged from0.92to0.97, Outliers ranged from1.40%to1.78%, and RMSE ranged from0.10to0.11. GNN has not show significant difference between2008and2009. Some simulated error come from data quality by several periods.(3) C flux data during July-Sept. provided the performance with R2ranged from0.79to0.86,Ivf ranged from1.00to1.02, Outliers ranged from1.19%to1.40%, and RMSE ranged from0.11to0.13. C flux data during Jan.-Mar. provided the performance with R2ranged from0.64to0.67,Ivf ranged g from0.64to0.84, and Outliers ranged from1.81%to2.85%, and RMSE ranged from0.08to0.10. Except the RMSE, July-Sept. generally the best performance(R2、 Ivf、and Outlier showed best), but Jan.-Mar. showed the opposite result (R2、Ivf、 and Outlier showed worst).(4) There was no significant change in the daily mean annual Changes of CO2fluxes. However, the peak of the carbon emissions in June turns into July and September into October, peak of carbon sequestration in the June and fall. There was a single waveform in monthly mean diurnal changes of CO2fluxes, the peaks of carbon sequestration and carbon emissions occurred on July or August. December and January have min carbon sequestration and carbon emissions.