干旱区农业生产依赖于灌溉。在灌溉可用水量不断减少的背景下，作物水分生产率的定量评价是提高干旱区作物产量、保障粮食安全的基础。本文以我国干旱区最大的灌区—内蒙古河套灌区为典型区，将遥感数据和田间调查、实测数据相结合，通过建立灌区遥感蒸散发模型、作物分布识别模型及估产模型，对灌区主要作物水分生产率及种植适宜度进行评价，并在不同优化目标和尺度下对灌区主要作物种植分布进行优化。 首先利用环境一号(HJ-1A／1B)卫星数据反演得到的空间分辨率为30m的归一化植被指数(NDVI)序列，建立了基于植被指数与物候指数特征空间的作物识别模型，对2009至2015年河套灌区主要作物(玉米和向日葵)的种植分布进行识别。结果表明识别效果较好，验证点的总体精度达到70％以上，与统计面积的相对误差小于15％。 在作物分布识别的基础上，利用随机森林(RF)回归算法，建立了基于植被指数与物候指数的作物估产模型，对河套灌区玉米和向日葵的多年产量分布进行估算。结果表明，玉米的最优估产因子为第120到210天、时间间隔为10天的NDVI时间序列，而向日葵则为NDVI和物候特征值的组合；RF估产模型可以较好地估算作物产量，玉米和向日葵产量估算值的均方根误差分别为0.93和0.34 t／ha。 将混合双源遥感蒸散发模型(HTEM)与大气边界层(ABL)模型相结合，建立了一个新的双源遥感蒸散发模型(HTEM-ABL)，模型利用一日内两个卫星过境时刻的遥感陆面温度来估算冠层和地表温度，在瞬时蒸散发的估算中避开了受下垫面类型影响较大的气温，在地面气象观测资料较少的地区具有更好的适用性。 在以上模型估算结果的基础上，对30m像元尺度下灌区玉米和向日葵的水分生产率进行估算，并基于作物水分生产率的频率分布构建作物种植适宜度指数，进而对灌区玉米和向日葵的种植适宜度进行评价。结果表明玉米主要适宜种植在磴口县和杭锦后旗，而向日葵则主要适宜种植在临河区和五原县。 分别以经济效益和节水效益最大化为目标，建立了两种作物种植结构优化模型，并分别在县级、3000 m和300 m网格尺度下对灌区玉米和向日葵的种植结构进行优化。结果表明，大多数年份优化后的玉米种植比例小于现状，向日葵则大于现状；优化尺度越小，对灌区净收入的提升和灌区作物生育期内总耗水量的降低效果越显著。研究结果可以为干旱区灌区水土资源合理利用提供技术支撑。 关键词：遥感；蒸散发；作物水分生产率；作物种植适宜度；作物种植结构

 In arid region, agricultural production relies on heavily irrigation. The base for ensuring food production with decreasing agricultural water use is the quantitative assessment of crop water productivity. In this paper, Hetao Irrigation District (HID) in Inner Mogolia, the largest irrigation district in arid area of China, is taken as the study area. By combining remote sensing data with field investigation and measurements, remote sensing (RS) based models for regional evapotranspiration, crop identification, and yield estiametion are developed in the study region. Furthermore, water productivity and planting suitability of main crops in HID are assessed, and crop planting structure of main crops in HID are optimized with different optimization objectives and spatial scales. Firstly, using the Normalized Difference Vegetation Index (NDVI) time series with the spatial resolution of 30 m retrieved from HJ-1A/1B satellite data, the planting distributions of main crops (maize and sunflower) in HID from 2009 to 2015 are identified based on the crop identification model of vegetation and phenological indexes space. The results show that the crop identification precision is high. For verification points, the overall accuracy is more than 70%. For the whole irrigation district, the relative error is less than 15% compared with the statistical crop planting area. Based on the results of crop identification, crop yield estimation models using vegetation and phenological indexes are established using random forest (RF) regression algorithm, and then multi-year yield distributions of maize and sunflower in HID are estimated. The results show that the optimal yield estimation model for maize is NDVI time series from 120 to 210 days with the interval of 10 days, while that for sunflower is the combination of NDVI and phenological indexes. The root mean square error (RMSE) of the estimated maize and sunflower yields is 0.93 and 0.34 t/ha, respectively. A novel dual-source remote sensing evapotranspiration model (HTEM-ABL) is developed by coupling the remote sensing evapotranspiration model (HTEM) with an atmospheric boundary layer (ABL) model. HTEM-ABL model uses remote sensing land surface temperature (LST) at two transit times of satellites in a day to estimate the canopy and soil temperature. In the estimation of instantaneous evapotranspiration, the air temperature that is greatly affected by underlying surface type is avoided, and the HTEM-ABL model has better applicability in areas with sparse ground meteorological observation staions. Based on the identified crop distributions, the estimated crop yield and water consumption during the crop growth periods, multi-year water productivity of maize and sunflower in HID is estimated at 30 m pixel scale. A crop planting suitability index is constructed based on the frequency distribution of crop water productivity, and then the planting suitability of main crops in HID is assessed. The results show that maize is mainly suitable for planting in Dengkou and Hangjinhouqi, while sunflower in Linhe and Wuyuan. Two optimization models of crop planting structure aiming at maximizing economic and water-saving benefits are developed. The crop planting structure of maize and sunflower in HID is optimized at the scales of county, 3000 m and 300 m grids, respectively. The results show that the proportion of maize planted in most years after optimization is less than the current situation, while sunflower is greater than the current situation. The smaller the optimization scale, the more significant improvement of net income and the reduction of total crop water consumption during the crop growth periods in HID. These results can provide technical support for rational use of water and land resources. Key words: Remote sensing; evapotranspiration; crop water productivity; crop planting suitability; crop planting structure