基于En-KF的土壤水分多源遥感数据同化

发布时间：2018-12-15 07:36

【摘要】：土壤水分作为水文、大气、陆面的重要内容,是地表水与地下水的重要纽带,更是描述陆地、大气和植被生长能量交换的重要参数。精确的估测土壤水分含量,对于干旱区的粮食安全和水土保持具有重要意义,而真实情况下准确估计土壤中的水分含量是一项十分困难而复杂的工作,而遥感技术使得长时间、大面积的估测土壤含水量成为可能。利用遥感技术采用水热传输模型反演土壤水分信息也已经得到了广泛应用。数据同化技术已经日趋成为了生态水文过程研究与遥感反演研究的前沿问题,在协调和融合遥感技术与生态建模起到了很好的桥梁作用。它能够有效地融合多源遥感数据以及模型的模拟结果,提高土壤水分的预测精度。本文利用以MODIS与Landsat TM数据为数据源,利用其反演的条件温度植被指数TVDI(Temperature-Vegetation Dryness Index)作为观测算子,将集合卡尔曼滤波同化方法应用于一维的水文模型HYDRUS-1D进行了表层土壤水分的模拟,主要结论如下:1利用不同的植被指数得到的特征空间都能反映区域土壤的含水量状况。RVI(Ratio Vegetation Index))指数相比常用的NDVI(Normalized Difference Vegetation Index)指数来说,更适应高植被覆盖的区域,对植被覆盖度大于50%的地区有较为灵敏的反应和指示作用,对植被覆盖较小的地区反应不够。2地表温度、土壤水分、植被指数之间有着显著地互动关系,但需要较长时间序列的植被数据。反演土壤水分所用的模型TVDI与表层土壤的相关性更高,这说明TVDI对表层的土壤水分比较敏感。3针对研究区环境以及前人积累的经验,结合经验参数与SCE-UA方法估测了模型的初始边界条件,并经过多次的调试确定了初始边界条件,利用模型模拟后输出了土壤水分的模拟结果。HYDRUS能够较为便利的模拟土壤含水量,反映研究区土壤含水量状况,对估算更大区域土壤含水量起到了一定的借鉴作用。4集合卡尔曼滤波能够较好地处理非线性问题,当遥感数据可用时,将模型经验参数与预报的土壤水分含量作为输入,利用TVDI模型反演的土壤水分作为初始的输入值更新模型算子,利用集合卡尔曼滤波将更新的土壤水分重新初始化模型参数,进行下一时刻的同化。在遥感反演数据可用时,再进入下一时刻的观测。5土壤体积含水量与残留含水率以及Ks都在同化过程中有了一定的变化,各个参数的调整量随着时间调整变化趋势不一致。在0-10cm,参数值的调整对同化有一定的效果。同化的效果的产生在于遥感数据对土壤表层含水量的反演值得到分析集合后,对状态变量持续更新,对HYDRUS模型的参数持续更新,即对模型的状态参数与状态变量都进行了调整,减少模型的预报误差。与单独使用HYDRUS-1D模型相比,同化得到的表层土壤水分含量精度有了明显的提高,其中均方根误差缩小了0.0153,平均误差缩小了0.0426,多源遥感数据对表层土壤水分的同化研究中具有较大的潜力。
[Abstract]:As the important content of the hydrology, the atmosphere and the land surface, the soil moisture is an important link between the surface water and the ground water, and it is an important parameter to describe the energy exchange between the land, the atmosphere and the vegetation. The accurate estimation of soil moisture content is of great significance to the food security and soil and water conservation in the arid area, and it is a very difficult and complicated work to accurately estimate the water content in the soil, and the remote sensing technology makes for a long time. It is possible to estimate the soil moisture content in a large area. The use of remote sensing technology to invert the soil moisture information by using the water heat transfer model has been widely used. The data assimilation technology has become the leading problem of the research of the ecological hydrological process and the remote sensing inversion, and has played a very good bridge role in the coordination and fusion of remote sensing and ecological modeling. The method can effectively fuse the simulation results of the multi-source remote sensing data and the model, and improve the prediction precision of the soil moisture. In this paper, using MODIS and Landsat TM data as the data source, using the condition temperature vegetation index (TVDI) as the observation operator, the ensemble Kalman filter assimilation method is applied to the one-dimensional hydrological model HYDRUS-1D for the simulation of the surface moisture of the surface layer. The main conclusions are as follows: 1. The water content of the soil can be reflected by the different vegetation index. Compared with the common NDVI (RVI) index, the RVI (RVI) index is more suitable for the area covered by high vegetation, has more sensitive response and indication to the area with more than 50% of the vegetation coverage, and is not enough for the area with small vegetation coverage. The surface temperature and soil moisture, The vegetation index has a significant interaction, but it takes a long time series of vegetation data. The model TVDI used in the inversion of soil moisture is higher than that of the surface soil, which indicates that the TVDI is more sensitive to the soil moisture of the surface layer. The initial boundary condition of the model is estimated by combining the empirical parameters with the SCE-UA method for the study area environment and the experience accumulated by the predecessors. and the initial boundary condition is determined through multiple debugging, and the simulation result of the soil moisture is output by using the model simulation. HYDRUS can simulate the water content of soil more conveniently, and reflect the soil moisture content in the study area, which can be used for reference for estimating the soil moisture content in the larger area. The 4-set Kalman filter can better deal with the non-linear problem, and when the remote sensing data is available, the model empirical parameters and the predicted soil moisture content are taken as input, and the soil moisture inverted by the TVDI model is used as the initial input value to update the model operator, and the updated soil moisture is re-initialized to the model parameters by the set Kalman filter, and the assimilation at the next time is carried out. When the remote sensing inversion data is available, the observation of the next time is carried out. The water content of the soil and the residual water content and Ks have some variation in the process of assimilation, and the adjustment amount of each parameter is not consistent with the time adjustment trend. At 0-10cm, the adjustment of the parameter value has a certain effect on the assimilation. The effect of assimilation is that the inversion of the water content of the soil surface by the remote sensing data is worth to the analysis set, the state variable is continuously updated, and the parameter of the HYDRUS model is continuously updated, that is, the state parameter and the state variable of the model are adjusted, and the prediction error of the model is reduced. Compared with the model HYDRUS-1D alone, the soil moisture content of the surface layer is obviously improved, and the root mean square error is reduced to 0.0153, the average error is reduced to 0.0426, and the multi-source remote sensing data has great potential in the assimilation study of the surface water of the surface layer.
【学位授予单位】：新疆大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S152.7;S127

【相似文献】