基于BP神经网络的油松林小气候的模型研究

Microclimate Model of Pinus tabulae formis Plantations Based on BP Neural Network

为了检测林缘、农田的小气候是否可定量预测林分小气候,从模型角度分别研究林缘、农田与林内小气候的相互关系,为油松人工林小气候预测、林分生态效益的定量评价及森林资源经营管理提供科学依据,本研究利用延庆不同密度油松人工林林分、林缘及农田的小气候监测数据,构建了林缘-林分、农田-林分立体空间的BP神经网络模型和多元线性回归模型,借助回归估计标准误差对2种模型的预测精度进行了比较。结果显示:对于集合小气候环境梯度,林缘-林内的BP模型预测精度整体高于农田-林内的BP模型预测精度;BP神经网络模型对气温、相对湿度及光照强度小气候要素测精度明显高于利用多元线性回归模型。结论:选用BP神经网络构建的林缘-林分模型实用性强,模拟精度高,可达到评价林分小气候、定量预测的目的。

From the perspective of model,the interrelationship of forest mieroclimate with forest edge microclimate and farm mi-eroclimate was studied to explore whether forest edge microclimate or farm microclimate would predict forest microclimate so as to provide scientific basis for facility meteorological service in plantations,quantitative evaluation for ecological effect and man-agement regulation of plantations. Back propagation neural network model （BP） and multiple linear regression model were em- ployed for forest edge-forest microclimate and farm-forest microclimate prediction by using observed meteorological data for the study site in Pinus tabulaeformis plantations of Yanqing county, Beijing. Predictive accuracy was compared by Root mean squared error （RMSE） between these two models. The results showed that, for congregative microelimate-gradient, the preci-sion of forest edge-forest BP neural network model was higher than that of farm-forest BP neural network model obviously based on RMSE;the predictive accuracy of air temperature, relative humidity and light intensity via BP neural network model were higher than that of the multiple linear regression model obviously. Conclusion ： BP neural network model is of strong practi-cality and higher predictive accuracy,which could evaluate and quantitatively predict forest microclimate.