Machine Learning Based Characterization And Prediction Of OLR Data In East Asia Region

With the rapid development of science and technology,machine learning algorithms are more and more in-depth in today’s social and scientific research.People’s use of machine learning models is more and more widely used in interdisciplinary research.Therefore,this paper mainly uses a machine learning algorithm to analyze and predict the spatial and temporal characteristics of the Outgoing Longwave Radiation(OLR)data retrieved by satellites from September 2002 to February 2022.First of all,the original data is preprocessed to explore the temporal and spatial characteristics of OLR in East Asia.Secondly,the spatiotemporal characteristics and prediction of OLR are studied through a single machine learning algorithm.Finally,the time series of OLR is predicted and analyzed through the combination algorithm model.The main research contents of this paper include.(1)Data acquisition and preprocessing.Firstly,it introduces the source and acquisition method of data sets.Secondly,the preprocessed data set is extracted by writing the script of related batch processing.Then,scale the data and use normalization or standardization to eliminate the differences between different attributes in the sample.Finally,empirically test the data to verify the accuracy of the data.(2)Study on the temporal and spatial characteristics of OLR in East Asia.This paper uses linear fitting,Mann-Kendall(M-K),and Pearson correlation analysis to analyze the temporal and spatial characteristics of OLR in East Asia.The results show that the spatial OLR decreases with the increase of latitude,and the OLR is symmetrically distributed in July,while the temporal OLR changes symmetrically around the equator.Using MK trend analysis,it is found that the absolute value of growth in the region through the significance test in space is about 2.2w/m~2.MK mutation analysis shows that the OLR will be greatly affected by the 2015 strong El Nino"Bruce Lee"around this year.Finally,the correlation analysis shows that OLR has a high positive correlation with Total Column Water Vapor(TCWV),Air Temperature(AT),Cloud Top Temperature(CTT),and a negative correlation with Cloud Top Pressure(CTP).In space,OLR has a negative correlation with TCWV and AT at 0～25°N,and a positive correlation with 30～60°N.OLR and CTP are negatively correlated in South Korea,North Korea,Japan,northern China,and positively correlated in other regions.OLR and CTT have a positive correlation of more than 0.8 in most parts of East Asia,while only a few regions have a low negative correlation.(3)The feature analysis and prediction of OLR based on a single algorithm.This paper uses the empirical orthogonal function(EOF)to analyze the trend of spatial characteristics.The results show that the contribution of the total variance of the four modes decomposed by EOF is more than 70%,the four-time coefficients PC are all decreasing,and the spatial characteristics of the South China Sea type and the Bay of Bengal type are obtained.The LSTM neural network model established by Keras deep learning framework was used to predict the OLR time series data,and the results were good,with low error.The four evaluation indicators,Root Mean Square Error(RMSE),Mean Absolute Error(MAE),Mean Absolute Percentage Error(MAPE),and determination coefficient R~2,were 2.8136,2.1669,0.9043,and 0.6812,respectively.(4)Prediction and analysis of OLR based on LSTM combination algorithm.Based on the LSTM model,this paper proposes two combined models,EEMD-LSTM and VMD-LSTM,to predict the time series of OLR.The results show that the combined algorithm model is better than the LSTM network model,and the VMD-LSTM model is the best of the three.The error evaluation indicators RMSE,MAE,MAPE,and R~2 are0.4238,0.3316,0.1386,and 0.9919,respectively.The error is the smallest of the three and the judgment coefficient is very close to 1,indicating that the predicted value almost fits the true value.Figure 42 Table 16 Reference 96...