阿尔泰山森林土壤温度动态变化及其预测

阿尔泰山作为干旱区典型的山地系统，其土壤温度的日、月、季节和年际动态及其影响因素研究，是深入理解干旱区山地森林生态系统能量循环过程的关键所在。基于阿尔泰山森林生态站2014年11月-2019年7月的气象因子和土壤温度数据，应用相关分析、回归分析和BP人工神经网络分析了阿尔泰山5、10、20 cm和30 cm深度土壤温度的动态变化及其对气象因子的响应，同时，应用多元线性回归和BP人工神经网络对土壤的温度进行了模拟。结果表明:1）近5 a各层土壤温度月均值年际变化一致，最低最高温度和日较差最大值均出现在20 cm，仅30 cm土壤温度的月变化出现自表层至深层滞后现象，年内月较差最大值出现在30 cm深度；各土壤层温度在春夏秋季变化较大，冬季变化较小；2）空气温度、气压和太阳辐射等与土壤温度的相关性达到极显著水平，其中与空气温度的相关性最强；3）回归模型和BP人工神经网络对20 cm土壤层的模拟结果最好，且BP人工神经网络模型的性能总体上优于回归模型。

Dynamic Change and Prediction of Forest Soil Temperature in the Altai Mountain

The Altai Mountain is a typical mountain system in arid area,the daily,monthly,seasonal and interannual dynamics of soil temperature and its influencing factors are the keys to understand the energy cycle process of the mountain forest ecosystem in arid area,so it is also an issue worthy of further discussion.Based on the meteorological factors and soil temperature data collected by the Altai Mountain Forest Ecological Station from November 2014 to July 2019,the present study used correlation analysis,regression analysis and BP artificial neural network to analyze the variations of the soil temperatures at 5,10,20,and 30 cm depths in the Altai Mountain and their responses to meteorological factors.The soil temperature was simulated by multiple linear regression and BP artificial neural network.The results showed that 1) the interannual variation of monthly mean soil temperature in each layer was consistent in the past five years.The lowest and highest temperature and maximum daily range appeared at 20 cm layer,while the monthly variation of 30 cm soil temperature lagged from surface to deep layer,and the maximum value of monthly range appeared at 30 cm depth.The temperature of each soil layer varied greatly in spring,summer and autumn,but little in winter.2) The correlation between air temperature,air pressure and solar radiation with soil temperature were significant.Among the variables,the correlation between soil temperature and air temperature was the strongest.3) The simulation results of the two models (regression model and BP artificial neural network) for 20 cm soil layer were the best,and the performance of BP artificial neural network model was better than regression model in general.