基于FY-3D MERSIⅡ数据的辽宁省作物生长季日平均气温估算方法比较

Comparison of Daily Mean Temperature Estimation Methods for Crop Growing Season in Liaoning Province Based on FY-3D MERSIⅡData

作为反映气候特征的重要指标之一,日平均气温在农业气象灾害监测和气候变化研究等领域承担着至关重要的作用。与传统日平均气温的监测和估算方式相比,遥感技术具有全方位、宏观、动态等不可比拟的绝对优势,能够准确地描述日平均气温的空间异质性。为提高农业服务质量,保证农业健康、可持续发展,探索作物生长季日平均气温遥感反演方法,提高农业气象灾害监测精确度,以FY-3D MERSIⅡ遥感数据为基础,提取研究区日间地表温度(LST_(day))、夜间地表温度(LST_(night))、归一化植被指数(NDVI),同时还考虑了高程(DEM)、坡度(Slope)两个变量,结合气象站日平均气温数据,分别构建多元线性回归和随机森林日平均气温遥感反演模型,开展辽宁省2021年作物生长季(5—9月)日平均气温遥感监测的应用研究。结果表明:(1)基于多元线性回归模型反演的日平均气温均方根误差(RMSE)为1.71℃,平均绝对误差(MAE)为1.45℃;基于随机森林反演误差RMSE为1.17℃,MAE为0.96℃;整体上,随机森林的日平均气温反演结果更好,适用性更强。(2)实验当天和前1 d的降水总量对日平均气温的估算结果具有很大影响,降水量随时间的变化曲线与日平均气温的反演误差散点分布情况基本一致,呈现降水总量越大,日平均气温的反演误差越大的趋势,日平均气温反演结果受大气水汽含量的影响很大。(3)对输入的气温影响因子的重要性进行动态的统计分析,发现LST_(day)和DEM是日平均气温反演时两个最重要的变量,且LST_(day)对日平均气温反演的影响最为重要,但是随着作物的生长,DEM的重要性也越来越凸显。

As one of the important indicators reflecting climate characteristics,mean daily temperature plays an essential role in monitoring agro-meteorological disasters and researching global change.By comparing with the traditional methods of monitoring and estimating daily mean temperature,remote sensing technology has the incomparable absolute advantages of omni-directional,macroscopic and dynamic,and can accurately describe the spatial heterogeneity of daily mean temperature.In order to improve the quality of agricultural services and ensure the healthy and sustainable development of agriculture,remote sensing inversion methods of daily mean temperature in the growing season are explored to improve the accuracy of agrometeorological disaster monitoring.Based on FY-3D MERSIⅡremote sensing data,we extract the daytime surface temperature(LST_(day)),nighttime surface temperature(LST_(night)),normalized differential vegetation index(NDVI),and by considering the two variables of elevation(DEM)and Slope(Slope),combined with weather stations measured,a multivariate regression model and random forest remote sensing inversion model are built to investigate the application of remote sensing monitoring of daily mean temperature in Liaoning Province from May to September 2021.Results shew that.(1)The random forest method has good applicability in the daily mean temperature retrieval,the inversion results of random forests model(RMSE:1.17℃,MAE:0.96℃)are better than those of multiple linear regression model(RMSE:1.71℃,MAE:1.45℃).(2)The total precipitation on the experimental day and the previous day has a great influence on the estimation results of the daily mean temperature,the variation curve of precipitation with time is basically consistent with the scatter distribution of inversion error of daily average temperature,showing that the inversion error of daily average temperature increases with the increase of precipitation,In summary,the inversion results of daily mean temperature are greatly affected by the content of atmospheric water vapor.(3)The temperature of the input factors thought statistical analysis found that LST_(day) and DEM of daily mean temperature of the two most important factors,and the influence of LST_(day) on daily mean temperature inversion is the most important,but with the growth of crops,the importance of DEM is becoming more and more prominent.