基于常规气象资料估算南方地区日辐射总量方法比较

Comparison of Total Radiation Estimation Methods in South Area Based on Conventional Meteorological Data

摘要:

日地表总辐射量（Rs）是作物生长模型和参考作物蒸发蒸腾量估算的重要基础数据，但我国只有约1/20的气象站能够直接观测Rs。由于气温资料很容易获得，使用基于基本气象资料的经验模型是估算Rs的常用方法。以1982—2014年南方20个气象站的气象资料为基础，对Bristow—Campbell (B—C)方法和Hargreaves（Harg）方法各6种不同形式重新进行了参数率定，并对以上方法和支持向量机15种参数输入形式进行了适用性评价，结果表明：支持向量机模型整体好于B—C方法和Harg方法。其中，以最高温度（Tmax）、最低温度（Tmin）、相对湿度（RH）和降水量（P）为输入变量的支持向量机模型精度最高，其20站平均R2达到0.80、RMSE平均为3.20MJ/(m2·d)，且在包含降雨量资料后，不存在Rs为负或大于地外总辐射量(Ra)的问题。仅有温度资料时，支持向量机模型的20站平均R2为0.74，RMSE为3.72MJ/(m2·d）。不同输入变量对支持向量机模型预报Rs的精度影响不同，输入变量为Tmax和Tmin优于输入变量为ΔT；而除温度资料外，当拥有相对湿度和降水量资料时，模型优劣依次表现为RH+P、RH、P。经验模型中B—C方法的M1和M3以及Harg方法的M10和M12模型精度较好，其R2为0.69～0.70、RMSE在4.00MJ/（m2·d）左右，但M10和M12模型对气象资料要求更高，除日温度差外，需要降水量资料，同时还存在有降水时日Rs严重高估或负值问题。

Abstract:

Global solar radiation (Rs) is an important elementary datum for crop modeling and reference evapotranspiration (ETo) estimation, but only 1/20 of Chinese weather stations can observe it directly. It is a common method for estimating Rs to use empirical model based on temperature data, which are easy to get. Based on the temperatures of 20 weather stations in south of China from 1982 to 2014, parameters of six different forms of Bristow—Campbell (B—C) and Hargreaves (Harg) methods were calibrated, and the applicability of abovementioned methods and fifteen support vector machine (SVM) parameter input forms were evaluated. The results showed that SVM model was better than B—C method and Harg method as a whole. The SVM model with maximum temperature (Tmax), minimum temperature (Tmin), relative humidity (RH) and precipitation (P) as input variables had the highest precision. On average, R2 and RMSE from the twenty weather stations were 0.80 and 3.20MJ/(m2·d), respectively, even when it included precipitation data, Rs was not negative and even greater than the extraterrestrial total radiation (Ra). R2 from the twenty weather stations was 0.74 on average, and RMSE was 3.72MJ/(m2·d) when based on temperature data. Different input variables had different influences on the SVM model forecasted Rs, the input variables of Tmax and Tmin were superior to ΔT. In addition to temperature data, when the model had the relative humidity and rainfall data, it was showed that RH+P > RH > P. Among the empirical models, the B—C model’s M1 and M3, and the Harg models’ M10 and M12 were preferable, their R2 were 0.69~0.70, RMSE was about 4.0MJ/(m2·d). While the M10 and M12 had higher request to the meteorological data, which needed the data of dayly temperature and precipitation. There existed the dayly Rs overestimation or negative problems when it rained.