甘肃省降水量自动观测与人工观测对比分析

Comparative Analysis of Automatic Observation and Manual Observation of Precipitation in Gansu Province

摘要: 利用甘肃敦煌、酒泉、民勤、榆中、西峰和岷县6个气象站2004~2007年自动观测与人工观测的降水资料，对比分析了2种观测序列的差值特征，并基于BP神经网络建立了自动降水观测订正模型。研究表明，自动观测的降水量与人工观测的标准差为0.254 mm，平均偏大0.033 mm (相对偏高2.24%)；76%的自动观测与人工观测的日降水量差值在±10%以内，近11%的样本日降水量测量差值在±20%以上；日降雨量相对差值较大的样本，其日降雨量均比较小；利用BP神经网络模型订正后，自动观测的日降水量拟合差为0.05~0.52 mm，平均相对误差0.5%~8.2%，平均相对均方根误差为0.10~0.70 mm，相关系数为0.75~0.99。

Abstract: Using the precipitation data from 6 meteorological stations in Dunhuang, Jiuquan, Minqin, Yuzhong, Xifeng and Minxian in Gansu Province from 2004 to 2007, the difference characteristics between automatic and manual observations were analyzed. Based on BP neural network, a correction model for automatic precipitation observation was established. Studies have shown that the standard deviation between the automatically observed precipitation and the manually observed precipitation is 0.254 mm. The average of automatic observation is 0.033 mm (2.24%) larger than that of manual observation. 76% of the difference in daily precipitation between automatic observation and manual observation is within ±10%, and at the same time, the measurement difference of daily precipitation close to 11% of the samples is more than ±20%. For samples with relatively large daily rainfall differences, the daily rainfall is relatively small. After the correction using the BP neural network model, the fitting difference of automatically observed daily precipitation is 0.05 to 0.52 mm, the average relative error is 0.5% to 8.2%, the average relative root mean square error is 0.10 to 0.70 mm, and the correlation coefficient is 0.75 to 0.99.