基于多源遥感数据的马铃薯收入保险应用研究

APPLICATION OF POTATO INCOME INSURANCE BASED ON MULTI-SOURCE REMOTE SENSING DATA

关键词:  农业保险  马铃薯  估产  遥感  图像分割  时序分类

Key words:  agricultural insurance  potato  yield estimation  remote sensing  image segmentation  time series classification

摘要:

目的 研究面向创新型农业保险业务中缺少及时准确的第三方作物产量结果用于灾损理赔的问题。引入多源卫星遥感测产技术，识别测产关键因子，构建产量模型。方法 文章运用多元线性回归分析方法，选取山西省马铃薯主产县岚县为研究区，计算基于Sentinel2影像的植被指数，结合气象卫星数据与实测单产数据，筛选关键因子，建立马铃薯单产遥感测产经验模型。结果 采用GF-2影像分割与Sentinel2长势时序识别岚县马铃薯种植面积为8 477.65hm²，精度检验Kappa值为0.72。保险公司岚县承保马铃薯面积2 476.37hm²，承保覆盖率为29.21%。测产结果显示，马铃薯单产与区域关键期地表温度参数相关性较好，岚县遥感测产获得平均单产为13.76 t/hm²，实地测产获得平均单产为14.06 t/hm²，误差百分比为2.13%，分乡镇平均误差百分比为22.97%，基本满足理赔业务需求。在2018年保险期结束后一周内，保险公司启动快速赔付，支付赔款125.29万元，赔付率48.46%。结论 遥感测产具有大范围、时效性好、可靠性高等特点，能够迅速为创新型保险产品提供测产理赔结果，提高理赔效率，保障农民收入。

Abstract:

Although innovative agricultural insurance products have been increased in China, timely and accurate evaluation of crop yield data around crop harvest time are still lack. The satellite remote sensing production estimation technology were employed to identify the key factors and establish the yield model. Taking Lan county of Shanxi province as an example, the potato planting plots were extracted based on GF-2 images in 2018, and the vegetation index was calculated based on sentinel2 images in 2018. By combining to the TRMM 3B42 precipitation data from TRMM precipitation-monitoring radar satellite and the MODIS land surface temperature data form NASA MOD11A2 data, we established a multi-parameter linear regression model to estimate the potato yield per unit area. The result showed that potato planting area in the study was 847 7hm² by using supervised segmentation model and the time-series products of vegetation parameters. By using Kappa-test, the remote sensing identification accuracy of planting distribution plots reached 0.72, which was in good agreement with the reported data by the local agricultural bureau. The coverage area of insurance company was 3 416 hm², which was 40.31% of the total area. The results of yield estimation showed that, using the early stage of seeding time and maximum LST data, the yield could be accurately estimated. The average estimated yield of the whole study area was 13.76 t/hm², compared to 14.06 t/hm² in actual measurement. And it showed the mean relative error between estimated yield data and actual measured data reached 22.97%, which met the application requirement of agricultural insurance. Within a week of the underwriting period, the insurance company paid 1 252 900￥ to the farmers. Our study suggests that the crop yield estimation of remote sensing has advantages, such as obtaining images at large area, timely, high reliability, which is an effective way to optimize the insurance company's claim work.