Abstract: It is an important aspect to evaluate the ecological benefits of forestry projects. Using observed seasonal meteorological data of April， July and October （2010） in different density of Pinus tabulaeformis plantations in Yanqing county， Beijing， we established the BP neural network model and MLR model for stereoscopic hydrothermal space of the forest edge-farmland and forest-farmland. The possibility of quantitatively evaluating and forecasting the farmland microclimate by these models were studied. The results showed: （1） The precision of forest edge-farmland BP neural network model was higher than that of forest-farmland BP neural network model in congregate microclimate-gradient （CMG）；（2） The observed-simulated correlation （OSC） of forest edge-farmland BP model was higher than that of forest-farmland BP model during the whole growing season; while the higher OSC of forest edgefarmland MLR model only occurred in October， but forest-farmland MLR model had the higher OSC in April and July. （3） The precision order of these two types of forest edge-farmland model followed as July > October > April; The forest-farmland BP model of density Ⅱ had the highest precision in April， but forest-farmland BP model of density Ⅰ was the highest in July and October. The precision of forest-farmland MLR model of density Ⅱ remained highest during the whole growing season. （4） The BP neural network model had high precision with few parameters， and was able to extrapolate when enough observed data were provided.

Key words: BP artificial neural network, forest ecological benefits, stereoscopic hydrothermal space, microclimate