基于数值预报产品的地面气温精细化预报技术研究

Technology Research of Fine Surface Temperature Forecasting Based on the Numerical Prediction Products

本项目拟采用ERA-Interim再分析资料,T213模式预测资料,结合2152个地面观测站点的温度资料,开展中国地区的精细预测实验。首先采用ERA-Interim数据内插得到的解析场作格点“真值”,采用卡尔曼滤波器降均值法对T213模式下的数据进行了修正,并对修正前和修正后的T213模式数据进行了综合预测。在此基础上,采用降维均值的方法,对修正后的网格预报场进行缩比,并对比其内插方法计算的差别。研究发现:一、用一种新的降均值方法修正格点数据的偏差,不但可以去除偏高的数值中心,而且还可以提高修正之前偏低数据区的预测条件。经格点数据修正,7个时效的绝对误差和相对误差的平均误差分别为1.65˚C和−0.06˚C。第二,采用递减平均统计降尺度方法,可以实现高精度的降尺度预报,显著降低内插偏差,特别适用于西部误差较大的地区。7个时效的绝对误差和相对误差,在经过降尺度预测后,其分别为2.62˚C、0.02˚C。结果表明,采用递减平均统计降尺度方法进行中国地面气温的精细化预报具有一定的可行性。

Based on the summer surface temperature observations data of ERA-Interim reanalysis data, T213 ensemble forecasting information and 2152 ground stations in 2011, the research makes a series of refined forecasting experiments. First, defining the interpolation result of ERA-Interim reanalysis data as “true value” of grids, an revision of T213 numerical forecast products, deviation is made upon the decreasing average technology of Kalman filter, followed by comparation of the deviation between revised T213 ensemble prediction and unrevised products. The second is to make downscaling gridded forecasts referring to the decreasing average downscaling technology, and mark the differences between interpolation products and new prediction. The results show that: firstly, the deviation revision of the adaptive decreasing average method on gridded data is effective. It not only eliminates high value center of the forecast error, but also reduces the model deviation of the area in which the forecast deviation is relatively small before correction. The absolute error and relative error are 1.65˚C, −0.06˚C after deviation revision, respectively. Second, the decreasing average statistical downscaling techniques can effectively make downscaling forecasts, significantly reducing the interpolation error, and also well correct the western regions which have bigger error. The downscaling forecast results indicate that the absolute error and relative error respectively are 2.62˚C, 0.02˚C, with the average absolute error 0.29˚C lowering than the interpolation results. Therefore, it is feasible to predict refined China ground temperature with decreasing average technology.