1979年—2014年三峡库区月平均气温的时空变化分析

Spatial-temporal analysis of monthly air temperature changes from 1979—2014 in the Three Gorges Dam region

三峡区域气温变化长期以来受到科研人员和公众的关注。受三峡复杂地形的影响,仅仅基于气象站点观测数据很难准确获取区域气温变化的空间格局,遥感技术则可以通过提供空间连续的地表观测数据来辅助气温变化分析。以广义加性模型GAM(General Additive Model)为插值算法,以高程和夜间地表温度(LSTnight)遥感产品为辅助变量,估算三峡库区1979年—2014年1 km空间分辨率的月气温数据,在此基础上分析了气温变化趋势的时空特征及其与高程和森林覆盖率的关系。研究表明,(1)在插值算法中引入遥感产品LSTnight作为辅助变量可以明显改善气温估算精度,冬春季的改善幅度高于夏秋季;(2)三峡库区年平均气温在1997年后明显上升,但在2003年库区蓄水后无明显变化趋势,几乎所有月(除12月以外)的气温都呈现上升趋势,增温趋势最显著是3月和9月,3月增温主要来自于库区东部山区的贡献,而9月增温主要来自于库区西部平原的贡献;(3)多数月份(除7月、8月、9月以外)的低温上升速度超过高温上升速度,导致区域气温的动态变化范围缩小;(4)三峡库区年平均气温上升速度与高程呈正相关,即海拔越高,升温越快,但在同一海拔高度处,森林覆盖率越高,年均气温上升速度越慢,暗示森林具有抑制增温的作用。

The near-surface air temperature（Ta） change in the Three Gorges Dam region（TGD） has long been a popular topic in public and research fields. However, fully capturing the spatial pattern of Ta change in TGD is challenging because of the sparse observation net and complicated topographic conditions. Thermal remote sensing technology can obtain spatially contiguous observations of Land Surface Temperature（LST） in a synoptic manner, thus providing invaluable information for spatial pattern analyses of Ta change given the fact that LST and Ta are closely related. This study aims to obtain the monthly Ta from 1979-2014 and determine its trend at a spatial resolution of 1 km.We used the satellite product of LST at nighttime（LSTnight） as a covariate in the general additive model（GAM）, which incorporates spline interpolation and linear regression, to ensure high quality of Ta data. First, monthly Ta estimation accuracies estimated with and without LSTnight were compared to evaluate the contribution of LSTnight. Second, the pixel-wise Ta trend was calculated with the Mann-Kendall method,and the spatial-temporal features of the Ta trend were analyzed. Finally, the effects of elevation and tree cover on the Ta trend were assessed.The main results were as follows.（1） When LSTnight was used as a covariate in GAM, temperature interpolation accuracy dramatically improved. The improvement in the cold season was more obvious than that in the warm season because Ta in the cold season is mainly influenced by LST through a strong radiative cooling effect.（2） Inter-annual variation analysis of regional mean annual Ta in TGD revealed that pronounced warming occurred after 1997, and no significant change in Ta was observed after the water level rose to 135 m in 2003.（3） Temporal-spatial analysis of monthly Ta showed that warming occurs in almost every month（except for December）, and the most dramatic warming occurs in March and September. In March, pixels with significant warming trends are mainly located in the eastern mountainous TGD, whereas in September, they are mainly located in the western TGD with a relatively flat terrain.（4） The Ta range for most months has been decreasing because the minimum temperature increased at a faster speed than the maximum temperature. Consequently, the lapse rate of Ta showed a decrease.（5） The enhanced warming trend over high elevations indicated a strong positive correlation between the trend of annual Ta and elevation（r = 0.76）. However, when the elevations are similar, the warming trend is less pronounced in regions with dense tree cover, suggesting that forests can restrain warming. We conclude that LSTnight information is beneficial to Ta estimation and that the change trend of Ta in TGD shows various features depending on season, region, land cover properties, and temperature metric. Further indepth analysis of the driving factors of the Ta trend, such as land use/cover or forest cover change, should be implemented in the future to be fully prepared to meet the challenges of climate change in TGD.