基于光学与被动微波遥感的青藏高原地区土壤水分反演

摘要

青藏高原地区高精度的土壤水分反演对高原能水循环、全球大气循环研究有着极大的影响.因此,获取青藏高原土壤水分时空布信息是一个迫切需要解决的问题.温度植被干旱指数(TVDI),是基于光学与热红外遥感通道数据反演土壤水分的重要方法,但在研究区域较大、地表覆盖格局差异显著时,TVDI模型反演精度会受到地表温度(Ts)等因素的影响.被动微波AMSR-E数据精确记录了像元内的土壤水分信息,但空间分辨率低.本文利用同时期的MODIS与被动微波数据,发展了针对青藏高原地区高精度土壤水分反演算法.首先,在TVDI模型中,利用修正型土壤调整植被指数(MSAVI)代替归一化植被指数(NDVI),以改进NDVI易饱和的缺点;其次,利用ASTERGDEM数据,对地形高程和纬度差异引起的地表温度变化进行了校正;然后,通过神经网络训练建立基于TVDI、被动微波以及辅助气象数据的土壤水分反演模型,并应用该模型反演了青藏高原地区三个观测网(CAMP/Tibet、玛曲和那曲)的土壤水分;最后,利用实测土壤水分数据对反演结果进行验证,结果表明该模型的精度均方根误差(RMSE)数值可达到0.031～0.041 m3·m-3.本文还应用该算法反演了青藏高原连续的土壤水分的空间分布,并比较了土壤水分的变化趋势与实测降水变化趋势,结果表明二者变化量的正负关系一致.%High-precision soil moisture estimation in Tibetan Plateau plays an essential role in the terrestrial water cycle and its impact on the global weather and climate.The Temperature Vegetation Dryness Index (TVDI) is widely used to estimate soil moisture on a large scale,but the accuracy of the TVDI can be influenced by some factors (eg.the surface temperature and NDVI).The AMSR-E data accurately records the soil moisture information,but the spatial resolution is low.This study aims to develop appropriate methods to estimate the soil moisture with high accuracy over the Tibetan Plateau.Firstly,the NDVI was replaced by MSAVI to correct the saturation characteristic of NDVI,and the terrain-induced variations in the land surface temperature (Ts) were removed using the ASTER GDEM data.Then,a soil moisture estimating model was built from a neural network combining the improved TVDI data,AMSR-E data and TRMM data and applied to obtain the soil moisture of three networks (CAMP/Tibet,Maqu,and Nagqu) in Tibetan Plateau.Finally,the method was validated using in situ soil moisture measurements.The results show that the soil moisture retrieved by the present algorithm has a higher accuracy (RMSE=0.031～0.041 m3 · m 3).This study also has applied the algorithms for the Tibetan Plateau continuous soil moisture spatial distribution retrievals,and the trend of the soil moisture was compared with the trend of measured precipitation.The result showed the trends of the two variables had a strong correlation.