当前位置: X-MOL 学术Ecol. Indic. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Establishment of agricultural drought loss models: A comparison of statistical methods
Ecological Indicators ( IF 6.9 ) Pub Date : 2020-01-13 , DOI: 10.1016/j.ecolind.2020.106084
Xiufang Zhu , Chenyao Hou , Kun Xu , Ying Liu

Agricultural drought loss models provide services for the rapid risk assessment of agricultural disasters, and regional disaster prevention and mitigation efforts. This paper takes wheat as an example, and chooses counties dominated by rain-fed farmland in Henan Province as the study area. Counties dominated by rain-fed farmland are determined by setting a rain-fed threshold that is related to the proportion of the effective irrigation area to the cultivated land area. Modeling samples are screened by considering both drought occurrence time and wheat yield reductions. Under different thresholds (30%, 40%, 50% and 60%), we use the yield loss ratio as the dependent variable and 24 standardized precipitation evapotranspiration index parameters as independent variables to build drought loss models using both a multivariate stepwise regression model and a random forest model. Yield loss ratio from 1990 to 2015 is calculated by decomposing historical wheat yield time series. 24 standardized precipitation evapotranspiration index variables are 1–3 months’ time scale standardized precipitation evapotranspiration index during the growth period (from October to May of the following year) of winter wheat in Henan Province. The results show that the random forest-derived model outperforms the stepwise regression model in all tests. The accuracy of all the models increases with an increase of the proportion of the rain-fed threshold. When the rain-fed threshold is 60%, the R2 values of the random forest model and the multivariate stepwise regression equation are 0.720 and 0.523, respectively. The validation results show that the mean absolute error and the root mean square error of the multivariate stepwise regression are 1.38 times and 1.31 times larger than the mean absolute error and the root mean square error from the random forests model. Moreover, both models identify that standardized precipitation evapotranspiration indices in October (sowing/planting stage) and February (overwintering stage) are important variables. However, the multivariate stepwise regression model fails to recognize the importance of standardized precipitation evapotranspiration indices during April–May (filling stage).



中文翻译:

建立农业干旱损失模型:统计方法的比较

农业干旱损失模型为快速评估农业灾害风险以及区域防灾减灾工作提供了服务。本文以小麦为例,以河南省雨养耕地为主的县域为研究对象。通过设置与有效灌溉面积占耕地面积的比例有关的雨养阈值,可以确定以雨养农田为主的县。通过考虑干旱发生时间和小麦减产来筛选模型样本。在不同的阈值(30%,40%,50%和60%)下,我们使用产量损失比作为因变量,并使用24个标准化降水蒸散指数参数作为自变量,以使用多元逐步回归模型和随机森林模型构建干旱损失模型。通过分解历史小麦单产时间序列来计算1990年至2015年的单产损失率。河南省冬小麦生育期(次年10月至次年5月)的1〜3个月时间尺度标准化降水蒸散指数是24个标准化降水蒸散指数变量。结果表明,在所有测试中,随机森林模型均优于逐步回归模型。所有模型的准确性都随着雨水供给阈值比例的增加而增加。当雨水供给阈值为60%时,随机森林模型和多元逐步回归方程的2个值分别为0.720和0.523。验证结果表明,多元逐步回归的平均绝对误差和均方根误差比随机森林模型的平均绝对误差和均方根误差大1.38倍和1.31倍。此外,两个模型都确定10月(播种/播种阶段)和2月(越冬阶段)的标准降水蒸散指数是重要变量。但是,多元逐步回归模型未能认识到在4月至5月(充水阶段)标准化降水蒸散指数的重要性。

更新日期:2020-01-13
down
wechat
bug