摘要
提高气象数据空间分辨率对水文、气象和生态等领域的流域尺度研究至关重要。青藏高原气候变化在全球气候研究中占有重要的位置,并且对局域降水分布的研究在大气科学中处于基础地位。为获取青藏高原地区准确、有效、更高空间分辨率的降水数据,基于随机森林算法,引入植被和地形因子,采用热带降水测量计划卫星(Tropical Rainfall Measuring Mission,TRMM) 3B43降水数据(0. 25°×0. 25°)、NOAA-AVHRR归一化植被数(normalized difference vegetation index,NDVI)数据(8 km×8 km)、航天飞机雷达地形测绘任务(Shuttle Radar Topography Mission,SRTM)数字高程模型(digital elevation model,DEM)数据(90 m×90 m)以及经纬度信息,建立了非线性空间统计降尺度模型,最终获得8 km分辨率降水降尺度结果。另外,采用将时间序列分析和非线性回归分析融合的方法,基于2000—2012年TRMM年均降水数据和NDVI数据,建立降水量时间尺度预测模型。分析结果表明,综合考虑植被和地形因子对青藏高原地区降水空间分布的影响,基于随机森林算法建立的降尺度模型,其降尺度结果与地面站点测量值拟合系数为0. 89,高于TRMM数据与地面站点测量值的拟合系数0. 81,说明降尺度结果提高了卫星遥感降水数据的空间分辨率。另外,降水预测模型能够较好地描述青藏高原地区的年际降水变化趋势和数量级,2006—2012年的预测降水量与TRMM降水数据拟合系数均高于0. 80。
So far,precipitation products with high spatial resolution have been crucial for the basin scale hydrology,meteorology and ecology.The climate in the Tibetan Plateau is of vital significance to global climate variation.So,the study of the distribution of precipitation with high spatial resolution is in the basic position of environmental science.Based on random-forest algorithm,the authors introduced environmental factors such as topography and vegetation,which was developed for downscaling the remote sensing precipitation products accurately and effectively.The non-linear spatial statistical downscaling model was demonstrated with the Tropical Rainfall Measuring Mission(TRMM)3B43 dataset with the spatial resolution of 0.25°,the Normalized Difference Vegetation Index(NDVI)from NOAA-AVHRR with the spatial resolution of 8km,the Digital Elevation Model(DEM)from Shuttle Radar Topography Mission(SRTM)with the spatial resolution of 90 m and the information of slope,aspect,longitude and latitude.And the model based on time series and vegetation factor,which was demonstrated with TRMM3B43 annual data in order to forecast the precipitation,was introduced in this paper.The downscaling results were validated by applying the observations from the rain gauges in the Tibetan Plateau and the coefficient of determination R^2 is 0.89.The analytical results showed that the downscaling results improved the spatial resolution and accuracy by applying the random-forest algorithm and introducing environmental factors.And the model,which was developed for forecasting the precipitation,captured the trends in inter-annual variability and the magnitude of annual precipitation with the R^2 ranging from 0.81 to 0.87.
作者
徐彬仁
魏瑗瑗
XU Binren;WEI Yuanyuan(Institute of Remote Sensing and Digital Earth,Chinese Academy of Sciences,Beijing 100101,China;University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《国土资源遥感》
CSCD
北大核心
2018年第3期181-188,共8页
Remote Sensing for Land & Resources
关键词
青藏高原
降水量
降尺度
预测
随机森林
时间序列
Tibetan Plateau
precipitation
downscale
forecast
Random-forest
time series
作者简介
徐彬仁(1990-),男,硕士研究生,主要从事大气遥感方面的研究。Email:xubr@radi.ac.cn。;通信作者:魏瑗瑗(1992-),女,硕士研究生,主要从事大气遥感方面的研究。Email:weiyy@radi.ac.cn。
