基于Google Earth Engine和作物模型快速评估低温冷害对大豆生产的影响

作为中国商品粮的主要生产基地,东北地区频发的低温冷害给中国粮食安全带来了严重的影响,及时、准确和大范围评估低温冷害灾损是降低损失、快速恢复生产的重要前提。本文以鄂伦春为例,提出了一种快速评估低温冷害对大豆生产影响的新方法。首先诊断出该地区典型冷害事件发生的年份为1989年、1995年、2003年、2009年和2018年;然后基于本地化后CROPGRO-Soybean模型设置512组低温冷害和田间管理组合模拟情景;其次构建了1600组包括3个变量（有效积温距平值（CDD）、模拟的叶面积植被指数（LAI）和产量）的冷害脆弱性模型;最后依托Google Earth Engine（GEE）平台逐像元提取大豆关键生育期早晚窗口内最大的宽动态植被指数（WDRVI）及对应的日期（DOY）,将WDRVI转化为大豆种植格点的实际LAI,结合构建的冷害脆弱性模型逐像元计算出产量和减产率。主要发现如下：① 校准后的CROPGRO-Soybean模型能较为准确地模拟不同冷害情景下的大豆生长发育过程;② 大豆遭受全生育期的降温情景（1~3 ℃）的减产幅度比局部降温情景（4个生育期随机生成连续5日温度为0 ℃）的减产幅度大;③ 历史冷害年1989年、1995年、2003年、2009年平均减产率约为9.6%、29.8%、50.5%和15.7%,与实际6.4%、39.2%、47.7%和13.2%的减产率相比,冷害灾损评估结果具有较好的精度且误差均在一倍方差以内;④ 运用该方法评估2018年冷害田间尺度的产量损失,并利用Sentinel-2A影像进行10 m高精度制图。结果显示,该方法能够快速、准确地评估不同尺度的低温冷害灾损情况,为作物估产以及农作物灾害损失评估的业务化运行提供了新的思路。

Damage evaluation on soybean chilling injury based on Google Earth Engine (GEE) and crop growth model

The frequent occurrence of chilling injury has serious impacts on national-level food security, and it mainly affects the grain yield in Northeast China. Timely and accurate measures are desirable to assess the large-scale impacts, which are the prerequisites for disaster reduction and production recovery. Therefore, we propose a novel method to efficiently assess the impact of chilling injury on soybean. Inner Mongolia is taken as a case study. The specific chilling injury events was diagnosed to occur in 1989, 1995, 2003, 2009, and 2018. The 512 combinations of cold and field management simulation scenarios were established based on the localized CROPGRO-Soybean model. Furthermore, we constructed 1600 cold vulnerability models including CDD (Cold Growth Days), simulated LAI (Leaf Area Index) and yields from the CROPGRO-Soybean model. Finally, we extracted the maximum wide dynamic vegetation index (WDRVI) and corresponding date of the critical windows of early and late growing seasons in the GEE (Google Earth Engine) platform, converted the WDRVI into actual LAI of soybean pixel, and calculated the pixel yield and losses according to the corresponding vulnerability models. The findings show that the localized CROPGRO-Soybean model can accurately simulate the growth and development processes of soybean under different cold scenarios. The soybean yields were reduced due to changes in cold stress during the whole growth period (a decrease of 1-3 ℃), which were greater than those from the local cooling treatments (the temperature of 0 ℃ for 5 consecutive days which are randomly generated during four growth periods). Moreover, simulated historical yield losses in 1989, 1995, 2003, and 2009 were 9.6%, 29.8%, 50.5%, and 15.7%, respectively, which were very close (all errors were within one standard deviation) to the actual losses (6.4%, 39.2%, 47.7%, and 13.2%, respectively). The above proposed method was applied to evaluate the yield loss of 2018 at a pixel scale. Specifically, sentinel-2A image was used for 10 m high-precision yield mapping, and the estimated losses well characterized the actual yield losses from 2018 cold event. The results highlighted that our proposed method can efficiently and accurately assess the chilling injury impact on soybean at different spatial scales. The novel method is also effective for efficient assessment of the impacts of different disasters on other crops.