Abstract:Atmospheric electric field instruments can record the entire process of thunderstorms for the creation,development and extinction,based on the principle of conductor induced charges in the electric field.As a result,the instrument scan continuously measure the strength and polarity of atmospheric electric fields for a long period time.In addition,the thunderstorm process is a complex nonlinear chaotic system,which leads to the measured electric field values having strongly nonlinear oscillation characteristics.At the same time,Empirical Mode Decomposition is a new type of nonlinear and nonstationary signal processing method which does not have a pre-determined basis function,and is able to smoothly process the signal based on its own characteristics,which leads to a new concept for ground electric field research.However,atmospheric electric field instruments are susceptible to the effects of the surrounding environment,the observation data quality are severely affected by noise,and model aliasing problems often exist.Fortunately,the Ensemble Empirical Mode Decomposition(EEMD) method has been developed,and is able to resolve these problems by enhancing signal continuity to weaken the mode aliasing effect,and by better reconstructing the amplitude and frequency characteristics of the signal component,based on the Gaussian white noise statistical characteristics.Therefore,this paper analyzes the atmospheric electric field data in the Nanjing area from June to August,2010 and 2011 by means of EEMD,and respectively discusses the oscillation characteristics of the atmospheric electric field in fair,weak and severe thunderstorm weather.In addition,the energy of thunderstorm is found to be far higher than the fair weather,which mainly contains low frequency components.Meanwhile,with the further development of thunderstorm,the electric field energy is mostly concentrated in the high frequency parts.Clearly,the intrinsic mode function(IMF) component variances are significantly different in the three types of weather conditions,and show some dynamic characteristics,which may be a result of the thunder cloud charge accumulation,charge and discharge,and the complex oscillation of internal charge.Although one station observation is influenced by the distance and location of the thunderstorm,the IMF component simply expresses the oscillation characteristic of the thunderstorm interior,which means that the observation data can intuitively reflect the thunderstorm development stages and ignore the position relation between the station and thunder cloud.Therefore,the variance characteristics of the IMF components can be studied to provide the basis for thunderstorm forecast.Moreover,according to the analysis results of fair weather and two different thunderstorms,some energy dynamic changing characteristics of fair weather,weak and strong thunderstorm may also exist.Thus,we take fair weather as the background with only one station,and establish the two-level thunderstorm nowcasting model,based on the dynamic characteristic between the maximum variance related layers and total number of decomposition layers of IMF.As for the weak thunderstorm,the related IMF layers exhibit obvious oscillation before the actual thunderstorm time above 1 hour.Furthermore,the severe thunderstorm development can be divided into the three sections of unrelated,energy storage and discharge.In the energy storage section,the decline of related IMF layers indicates rapid accumulation of electric charge before the thunderstorm arrives,and the four consecutive decline points are taken as the forecast standard,which can effectively increase the forecast time to more than 1 hour.Finally,the independent samples are selected to forecast and verify the proposed method.The results show that energy is concentrated in the low-frequency component in fair weather,and the high-frequency component in thunderstorm.The maximum variance related layers of IMF change stably before weak thunderstorm weather,and fiercely before severe thunderstorm.According to these characteristics,92 independent samples are tested,and the results show that the detection probability of warning is 73.3%,and the false alarm rate is 14.5%.This proves that the established thunderstorm nowcasting model meets the forecast basic requirements,and has a certain degree of application value.