Sensitivity Analysis of Observation Data in Numerical Weather Prediction over East China

In order to deeply understand the impact of observation data on numerical prediction in China, based on the adjoint-based forecast sensitivity to observation (FSO)method, this paper constructs a FSO system and evaluates the impact of national basic automatic weather station (AWS),radiosonde soundings and wind profiler radar observations on the regional 12-hour-forecast for the East China region in July and August of 2018.

(1) All observations types play a positive role in reducing the 12-hour forecast error. Among all types, sounding observation, with the largest number, is the most influential, AWS makes the smallest normalized contribution to forecast on average.Wind profiler radar has the highest beneficial observation fraction.
(2) For vertical distributions,AWS has a very limited impact on forecast above 850hPa. Wind profile observation has a greater impact below 500hPa, and the impact above 500hPa is rapidly reduced. The effect of sounding observation is more obvious on the vicinity of 100-200hPa and 500-700hPa.
(3) For horizontal distributions, automatic weather stations,which obviously reduced 12-hour forecast error of East China, are mainly concentrated in the target region itself and in regions with sparse stations. Only three of the 120 sounding stations show a significant cumulative negative impact, and the prositive impact of stations in the plateau area are weaker than that in other areas. The wind profile stations along the coasts of Jiangsu, Zhejiang and Fujian have a stronger impact on the forecast improvement,and the stations that have negative contribution are mainly concentrated in Beijing and Hebei.