Modeling and Predicting Carbon and Water Fluxes Using Data-Driven Techniques in a Forest Ecosystem

Accurate estimation of carbon and water fluxes of forest ecosystems is of particular importance for addressing the problems originating from global environmental change, and providing helpful information about carbon and water content for analyzing and diagnosing past and future climate change. The main focus of the current work was to investigate the feasibility of four comparatively new methods, including generalized regression neural network, group method of data handling (GMDH), extreme learning machine and adaptive neuro-fuzzy inference system (ANFIS), for elucidating the carbon and water fluxes in a forest ecosystem. A comparison was made between these models and two widely used data-driven models, artificial neural network (ANN) and support vector machine (SVM). All the models were evaluated based on the following statistical indices: coefficient of determination, Nash-Sutcliffe efficiency, root mean square error and mean absolute error. Results indicated that the data-driven models are capable of accounting for most variance in each flux with the limited meteorological variables. The ANN model provided the best estimates for gross primary productivity (GPP) and net ecosystem exchange (NEE), while the ANFIS model achieved the best for ecosystem respiration (R), indicating that no single model was consistently superior to others for the carbon flux prediction. In addition, the GMDH model consistently produced somewhat worse results for all the carbon flux and evapotranspiration (ET) estimations. On the whole, among the carbon and water fluxes, all the models produced similar highly satisfactory accuracy for GPP, R and ET fluxes, and did a reasonable job of reproducing the eddy covariance NEE. Based on these findings, it was concluded that these advanced models are promising alternatives to ANN and SVM for estimating the terrestrial carbon and water fluxes.