Simulations Of Runoff And Evapotranspiration In Chinese Fir Plantation Ecosystems Using Artificial Neural Networks

The hydrological cycle is a key ecosystem function that links biological and biogeochemical processes in forest ecosystems. Because the forest structures and functions have been changed by natural and human disturbances, hydrological cycles and water balance in forests are probably altered. An improved understanding of the relationships between forest structure and the water cycle and budget in forest ecosystems has important implications for the effects of forest dynamics and management on water resources. Chinese fir is a fast-growing native species with valuable timber attributes in subtropical area of southern china. Because of it is own characteristics and unreasonable managements, it is land productivity has extremely declined.Runoff and evapotranspiration are two key variables of water budget in forest ecosystems. Modelling runoff and evapotranspiration dynamics play a vital role in assessing the hydrology cycle and function of forest ecosystems. Based on the hydrological and meteorological data collected over 20 years from January of 1988 to December of 2007 at Huitong National Forest Ecosystem Research Station, we used back propagation neural network (BPNN) and genetic neural network (GNN) models to simulate runoff and evapotranspiration of Chinese fir plantations for NO.Ⅱand NO.Ⅲwatersheds located in Huitong county of Hunan Province, China.The results shows that:(1) For different training periods, the R2 range from 0.8523 to 0.9651, however, for validation periods the R2 drop from 0.796 to 0.8577 in BPNN model. The results of training and simulation are relatively stable with GNN model and the R2 ranges from 0.6572 to 0.8724. (2) We compared the two ANN models (including BPNN and GNN) with traditional statistical regression to examine the effect of simulations of both ANN and multivariate statistics methods on evapotranspiration and runoff. The average error was estimated between 17.5 and 17.8 by multivariate statistics (M-slat) model and between 11.9 and 14.3 by GNN model, respectively. The larger error variation (12.6-19.2) was found with BPNN model. (3) The monthly evapotranspiration was the highest in June (147.76mm for observed,174.36mm for GNN simulated,192.97mm for BPNN simulated) and lowest in December (23.15mm for observed,27.70mm for GNN simulated,23.29mm for BPNN simulated). The annual evapotranspiration was 667-1006mm and the average annual evapotranspiration was 831mm by observed. The annual evapotranspiration was 653-1077mm by GNN simulated and 759-1195mm by BPNN simulated. The average annual evapotranspiration was 828mm by GNN simulated and 909mm by BPNN simulated. (4) The monthly runoff was the highest in June (75.76mm) by GNN simulated that was slightly lower than the observed value (97.52mm). The lowest monthly runoff was 13.29mm in December. The results compared with the observed value (10.96mm) were slightly high.