Study On Difference And Homogeneity Of Meteorological Data Between Automatic Observations And Manual Observations

Recently, with the improvement of Automated Surface Observing System, Automatic Observations have gradually achieved mono-way operation and replaced Manual observation.Observations difference between Automatic Observations and Manual Observations are somewhat inevitable. This challenges the continuity of historical weather data and influences atmospheric researches and applications. Therefore, based on the understanding of influence caused by the change of Observations instruments, homogeneity correction of Observations data is really necessary. In this paper, data of temperature, relative humidity, wind and precipitation from six weather stations in Gansu Province from 2004 to 2007 are compared and analyzed; Meanwhile, annual variation, daily variation and frequency distribution of each element are calculated by two kinds of Observations data. In order to ensure successive use of Observations data, based on characteristics comparisons, we use linear regression model and BP Network to establish and analysis revised model of Automatic Observations data and Manual Observations data. The results are as follow:Differences between Automatic and Manual temperature Observations show annual variation characteristic. The maximal differences occur in summer, while the minimums occur in winter. From Spring through Winter, differences between two kinds of temperature instruments show similar daily variation seasonally. Daily variations are stable during night, and become unstable before or after sunrise.Differences between Automatic and Manual humidity Observations from 6 stations show inconspicuous annual variation characteristic. The average of differences(Automatic Observations minus Manual observation) is negative mostly. Daily variations of difference have wider ranges in summer period than winter period. The absolute value of the summer period is larger. The difference distributions of 6 stations show peak values all between-2%-2%. Furthermore, differences are larger in both lower and high temperature areas.The difference of wind speed of 6 stations has large annual variations. Monthly differences (Automatic Observations minus Manual observation) of wind speed of 4 stations are mostly positive, while negative of other two stations. Daily difference of wind speed is stable. Most difference of Dunhuang, Jiuquan, Yuzhong and Minxian stations are positive, while mostly negative of Xifeng and Yuzhong.By analyzing hourly and daily precipitation data, the result shows that hourly precipitation data of Automatic Observations averagely is about 0.033mm more than Manual observation, with a standard deviation of 0.254mm. About 22% samples of daily precipitation from Automatic Observations are less than Manual observation. 76% hourly precipitation biases between Automatic and Manual Observations are less then 10%, while about 11% are over 20%. The samples whose hourly precipitation biases are large have small hourly precipitation data.By using F test (confidence interval=0.01), we can see that revised models for temperature, humidity, wind speed and daily precipitation will fit observed data very well. By fitting analysis,fitting differences are between 0.1 -1.2℃,0.1-3.7%,0.2-2.6m/s,0.01 - 0.18mm in turn,and unadjusted differences are between 0.2-1.9℃,0.04-5.2%,0.9-5.2m/s,0.02-0.62m/s in turn.By using linear regression model, statistical average standard errors are between 2.0%-4.2%,3.6%-9.0%,10.1%-14.7%,2.4%-6.8%; standard deviations are between 3.1-4.2,3.0-5.5,2.9-4.1,0.1-0.5; correlations are between 0.96-0.99;0.96-0.98;0.92-0.96,0.98-0.99.Based on analyses of differences of Meteorological variables, we use BP network revised model for artificial observation data. After parameter correction and sample data testing, we find that fitting differences are between 0.1-1.2℃,0.1-3.8%,0.2-2.0m/s,0.05-0.52mm in turn,and unadjusted differences are between 0.2-1.9℃,0.04-5.2%,0.9-5.2m/s,0.02-0.62m/s in turn. By using BP network model, statistical average standard errors are between 1.9%-4.0%,3.6%-9.5%,8.1%-16.3%,0.5%-8.2%; standard deviations are between 3.1-4.2,3.0-5.6,2.6-3.8,0.10-0.70; correlations are between 0.99-0.99,0.96-0.98,0.93-0.96,0.75-0.99.