大范围、及时、准确的灾害损失评估与制图对防灾减灾、农业保险和粮食安全等至关重要。针对传统灾害损失评估方法空间尺度单一、泛化能力差、时效性低，可操作性弱等问题，本文建立了一种遥感产品耦合作物模型的多尺度的灾害损失评估方法MDLA （a Multiscale Disaster Loss Assessment）。该方法利用作物模型的多情景模拟产生大量的灾害样本，结合对应日期的遥感指标构建灾害脆弱性模型，依托Google Earth Engine（GEE）平台将其应用到高分辨率遥感影像和格点灾害指标进行逐象元评估。以鄂伦春自治旗玉米为例，基于精细校准的CERES-Maize模型的模拟，利用两个生长季窗口的LAI和冷积温（CDD）建立统计模型来刻画低温对最终产量的影响，结合Sentinel-2数据逐格点计算完成高精度损失制图。结果显示，校准后的CERES-Maize模拟物候和产量的NRMSE 分别为3.3%和8.9%。冷害情景模拟结果表明不同类型和生育期的低温冷害对玉米产量的影响不尽相同，其中生长峰值期（出苗—吐丝和吐丝—灌浆）最为敏感。回代检验显示，MDLA方法估算精度为11.4%，与历史冷害年份的实际损失相吻合。经评估，鄂伦春2018-08-09的冷害导致玉米减产23.7%，受灾面积1.86×10⁴ ha，其中高海拔地区损失较重（减产率>25%），低温冷害对该区玉米生产构成了严重的威胁。与现有的统计回归、作物模型模拟以及同化等技术相比，其优势在于：（1）结合遥感观测和作物模型模拟技术能更好地刻画了灾害对产量的影响过程；（2）利用GEE平台快速处理海量遥感数据，提高了灾害损失评估的时效性；（3）不受地面实测数据的限制，易操作，可实现动态、多尺度（象元、田块、村，县等）的损失评估，这为防灾减损、维持粮食丰产稳产提供了保障，也为农业保险的业务化运行提供了思路。

Extensive, timely, and accurate mapping of yield losses is critical and prerequisite in disaster prevention and reduction, agricultural insurance, and food security. Given the coarse resolution, poor generalization ability, low timeliness, and weak operability of traditional loss assessment method, we propose a new approach called Multiscale Disaster Loss Assessment (MDLA) by coupling crop model with remote sensing to assess yield loss rapidly with satellite images. A series of disaster scenarios was simulated using a calibrated crop model. Related results, including final yield and crop growing state variable LAI, were inputted into disaster datasets. A susceptibility model of disaster was then constructed. Finally, pixel-by-pixel yield loss was evaluated on the basis of the susceptibility model combined with high-resolution image with gridded disaster indices within the Google Earth Engine (GEE) platform.The new method was used to assess the impacts of chilling injury on maize by applying carefully calibrated CERES-Maize in Oroqen, Inner Mongolia Autonomous Region. We constructed the cold susceptibility model, which properly characterized the cold damage on maize yield, including three independent variables, LAI in two growing season windows and a cold index (cold degree days), and yield loss. We further mapped pixel-based maize yield losses together with Sentinel-2 data. Mapping results showed that CERES-Maize, once calibrated, can appropriately simulate the growth and development state of maize under various management and weather conditions with a phenology bias of < 3.3% and yield NRMSE of < 8.9%. Furthermore, impacts of chilling injury varies in cold type and occurring time due to the high susceptibility of maize at the peak growing period (emergence-silking and silking-graining filling). The MDLA method successfully estimated significant losses during cold years with an accuracy of 11.4%. Moreover, the recent cold event (occurred at 2018/08/09) reduced the maize yield by 23.7% and affected 1.86 × 10⁴ ha of growing areas. The occurrence of more than 25% yield loss in high-altitude regions indicated that low temperature is a major threat on crop production in northeastern China.Our results indicated that MDLA is consistent with statistical regression, crop model simulation, and assimilation technology. Moreover, the advantages of MDLA are presented as follows: (1) The impact of disaster is appropriately characterized by combining remote sensing observation with simulated physiological states in crop models. (2) Processing the satellite image within the GEE platform significantly reduces the computing time of loss assessment. (3) Multiscale losses are mapped in a dynamic and operable way. This type of mapping can be performed not only in large-scale areas but also the county- or even field-scale regions. Our study can help decision-makers in reasonably preventing agricultural disasters and maintaining steady grain production while providing a more practical means for operational agricultural insurance.