基于时间序列多源遥感数据的旱灾损失精细评估模型研究

Drought damage assessment is a difficult problem in natural disasters relief. According to drought being one delayed-disaster and drought temporal and spatial pattern in Yunnan province or in north China, we put forward a quantitative and high accuracy model to make drought disaster evaluation. The model combines the MODIS satellite more useful bands and the environment and disaster monitoring and forecasting constellation (HJ) and ZY-3 satellite having higher spatial resolution ability, and to focus the two key parameters of crop yield loss rate and crop yield area. Firstly,we use supervised classification method and HJ CCD data to obtain 30meters crop yield distribution data, and object-oriented method and ZY3 CCD data to obtain 30meters crop yield distribution data. Secondly, we set up MODIS vegetation index to reveal the characters of vegetation greenness and leaf water content influence by drought and using Savitzky-Golay filtering algorithm to obtain time series MODIS-VI and remove cloud and noise effect. We also set up HJ drought index using CCD data and thermal infrared data to reveal the characters of vegetation-soil system water stress by drought. Furthermore, we use local government report data and field measurement of spectral and leaf water content and soil moisture to choose the suitable MODIS-VI and HJ-DI and set up agricultural comprehensive index F. Then we establish the Variety Drought Index (VF) by using the drought period and historical normal period agricultural comprehensive index F. Then we simulate the relationship between time series VF and crop yield loss rate and the damaged class. Therefore, it can assess the drought crop affected areas (loss rate≥10%) and total crop failure areas (loss rate≥80%) by integrating with GIS spatial analysis and statistics analysis methods on the unit of county throughout Yunnan province. For the severe damaged area, the HJ land surface temperature and be down scaled to 30 meters by mean of Modified HUTS algorithm and using 5.8meters ZY3 vegetation fraction ratio. Therefore, it can assess the drought crop affected areas and total crop failure areas on the unit of town. Our study is a useful exploration to integrate multi-resource remote sensing and GIS technologies. The models can be used in drought disaster relief work.

损失评估是旱灾救助的难点问题。根据云南或北方旱灾旱灾特点，发挥MODIS波段多时间分辨率高、环境减灾和资源三号空间分辨率高的互补作用，重点提高区域和田间尺度损失率和面积两个关键指标精度。首先构建表征植被长势和叶片水分影响的MODIS植被指数MODIS-VI、表征植被-土壤系统水分胁迫的HJ干旱指数HJ-DI，利用历史灾情、地面测量和农气数据对其进行时间空间优化，采用Savitzky-Golay滤波形成时间序列MODIS-VI，进而构建多源遥感农业干旱综合指数F并通过历史同期对比计算差异干旱指数VF，利用时间序列VF结合历史灾情拟合作物损失率，综合HJ-CCD提取30米农作物分布信息，区域上以县为单元评估受灾面积和绝收面积，最后针对重灾区，利用资源三号植被覆盖度及MHUTS算法对HJ地表温度降尺度到30米，结合5.8米农作物信息，以乡为单元评估开展重灾县损失评估。研究对于指导政府旱灾救助决策意义重大。

旱灾是我国发生频率最高、影响范围最广、灾害损失最大的灾害类型之一，开展灾害风险分析和损失评估对于旱灾救助具有重要意义。项目总结了国内外干旱发展趋势和研究方法，从灾害成因、产品形式两种类别建立了旱灾产品目录，收集整理了风云卫星、MODIS、环境减灾小卫星、资源三号等高中低分辨率遥感卫星数据，以及气象、土壤、基础地理、社会经济、灾情上报和野外调查等数据。项目重点针对2009年至2010年云南特大干旱进行了风险分析和损失评估：首先，研究了基于环境减灾小卫星热红外数据改进的地表温度反演宽通道算法，针对资源三号的数据融合及地形信息提取方法，利用每日FY-1C卫星遥感数据，基于云参数模型开展了西南地区2009年下半年至2010年上半年区域旬干旱风险分析，对研究区干旱风险范围和等级进行了初步识别；其次，根据MODIS数据特点，采用Savitzky-Golay滤波方法构建了时间序列归一化植被指数、垂直植被指数、地表温度等遥感数据集，基于干旱指数VSWI采用加权法建立了遥感旱灾损失指数；最后，通过结合遥感数据与生物地球化学模型（DNDC），考虑作物种植面积和灌溉因素等，获得研究区5km栅格的作物产量旱灾损失量，统计结果表明，此次云南特大干旱造成作物产量损失达320万吨。此外，项目还在山东、吉林等旱灾中进行了相关示范应用。项目研究的区域干旱风险分析方法对于指导旱灾早期应对与预警有一定意义，项目通过耦合遥感数据和DNDC机理模型，建立了分布式作物旱灾损失评估方法，首次获得了省级行政区5km分辨率空间分布的作物产量旱灾损失数据，相关结果可作为灾情数据的重要补充，研究成果对于国家和地方灾害管理部门定量评估灾害损失、核对地方上报数据、制定科学合理的旱灾救助措施，具有较好参考意义。