Research On The Correction Algorithm Of Conventional Climate Elements Based On Neural Network

Climate forecasting operations are becoming increasingly important as the global climate warms.Among the conventional climate forecasting elements,precipitation and temperature are the two most concerning,especially since summer precipitation and winter temperature are closely related to our lives.Precipitation and temperature are the main factors affecting droughts and floods.Accurately forecasting precipitation and temperature is an important task in climate prediction.At present,there are still errors in climate models as the main forecasting method.Therefore,it is necessary to improve the accuracy of climate model predictions.On the basis of summarizing the characteristics and status quo of precipitation in different regions of the country,this paper conducts an in-depth study of machine learning-related methods.According to the requirements of the project,it mainly carried out research on the correction technology of the regional climate model reporting summer precipitation errors and completed the error correction of the model reporting summer temperature and winter temperature in North China.Research content includes:Firstly,learned the principles and sources of errors of regional climate forecast models(CWRF)for predicting atmospheric states.Through the summary and analysis of the correction methods for precipitation and temperature errors at home and abroad,it is shown that the model correction of the machine learning algorithm is a reasonable and effective way of thinking.Therefore,an improved artificial neural network model is proposed for mode error correction.Secondly,the improvement of neural network model structure is studied,and the artificial neural network(ANN)structure is improved by introducing dendrites(DD),and an artificial dendritic neural network(ADNN)is proposed for model error correction.The Random Forests(RF)algorithm was used to explain the correlation between different climate elements in the CWRF model.At the same time,according to the law of precipitation in different regions of my country,the main continent of China is divided into eight different climatic regions for precipitation error correction.Thirdly,the regional and national summer precipitation forecast results based on the ADNN model were evaluated.Experiments show that the overall correction effect based on the ADNN algorithm is better than the model return,in which the anomaly correlation coefficient has increased by about 0.10,most areas have passed the 90% significance test,the mean square error has decreased by about 26%,and the abnormal score has increased by 6.55.The experimental results verify the validity and feasibility of the proposed algorithm applied to the post-processing of summer precipitation forecast by climate models.Finally,in order to verify the scalability of the algorithm in this paper,the ADNN model is applied to the CWRF model to report the error correction of summer temperature and winter temperature in North China.After correction by ADNN,the anomaly correlation coefficient of the summer temperature increased by about 0.10,and some regions passed the90% significance test;the anomaly correlation coefficient of the winter temperature in North China increased by about 0.13,and most grid points passed the 95% significance test.Effective results are obtained by applying the ADNN algorithm to CWRF mode temperature error correction,which shows that the algorithm in this paper has good scalability and universality.