Research On Estimating Model Of Atmospheric Optical Turbulence

Atmospheric optical turbulence is a kind of important phenomena in the atmosphere.When it happens,it could have critical effect on the laser propagating in the atmosphere.Therefore,continuous and widely observation of atmospheric optical turbulence could facilitate establishing ground-based optical instruments and astronomical observatories.However because of financial and logical issues,observing atmospheric optical turbulence for a long time is not practical.Therefore,finding a solution which could estimate atmospheric optical turbulence is helpful to settle this problem.In our study,experiments based on Chengdu and Delingha are presented in order to verify two kinds of estimation methods.Three main conclusions are made as the following:(1)The temporal revolution of Cn2 in Chengdu agrees with the results estimated by using Back Propagation neural network after the model is trained well.Besides,forecasts in the nighttime also agree well with the observation.The mean relative error is 3.03%.However when compared with the observation,there is approximate 1 hour earlier in the estimation than what is observed in the experiment.Back Propagation neutral network could also demonstrate basic diurnal variation trend in Delingha.The mean relative error is 3.53%.There are two obvious transition times in the experiment in this area,especially at 18:00 Cn2 often drop dramatically which can be predicted by Back Propagation model.(2)Support Vector Machine is able to estimate atmospheric optical turbulence Cn2 within surface layer after determining several vital parameters.The experiment in Chengdu shows that Support Vector Machine model could demonstrate diurnal variation trend in Chengdu.The mean relative error is 2.81%.Another experiment is conducted in delingha.A 9-day estimation in this area is obtained by utilizing Support Vector Machine model.The results show that predictions agree with measurement well and they can demonstrate diurnal characteristics of atmospheric optical turbulence.The mean relative error is 3.38%.Distribution of SVM-based Cn2 resembles Gaussian function which is similar to the observations.(3)Two experiments in Chengdu and Delingha show that after training,these two models are able to predict Cn2 in 6 days and 9 days respectively by using a set of training data of one day.Relative analysis,mean absolute error,mean relative error and other statistical operators are utilized to demonstrate that these two models could predict Cn2 within surface layer.The mean relative error of Back propagation neural network is larger than that of Support Vector Machine but there isn't too much difference.Both of them could predict atmospheric optical turbulence precisely and demonstrate nonlinear ability.