Spatiotemporal Prediction Analysis Of GDP Based On Night-Light Remote Sensing Data

Night-light remote sensing satellite captures human social activities by photographing the information of visible brightness at night,and indirectly and effectively reflects the social and economic situation.Therefore,night-light remote sensing plays an important role in exploring urban economic development,analyzing and prediction urban GDP changes.However,at present,the analysis and prediction of the spatial correlation between night-light remote sensing and GDP are mostly focused on the macroscopic development of urban economy,and lack of detailed analysis of urban industrial economic development.At the same time,in most of the studies on the time correlation analysis and prediction of night-light remote sensing and GDP,a variety of complex factors affecting the economy are ignored,leading to the limitation of the prediction accuracy.Therefore,the spatio-temporal correlation analysis and prediction method of GDP based on night light remote sensing is studied in this paper..The specific research contents are as follows.(1)The correlation and prediction of night-light remote sensing and GDP in spatial scale are studied.Firstly,According to the characteristics of Luojia1-01push-broom imaging mode,the sensitivity of the urban primary industry to the noise in the night-light images is analyzed,and a multi-frame threshold denoising method is proposed.Considering the spatial distribution characteristics of the industrial structure,it is proposed to use the OSM road network and POI reclassification method to divide the industrial structure.Finally,a linear regression model of economic parameters of night-light and industrial structure was constructed,and GDP analysis and estimation of 16 cities of China based on industrial structure were carried out.The experimental results show that the multi-frame threshold denoising method can effectively improve the overall GDP prediction accuracy,among which the prediction accuracy of Fuzhou and Zhoushan are increased by 10.64% and 10.55% respectively;compared with the NPP-VIIRS night-light data,it is found that the Luojia1-01night-light data has more advantages in the economic forecast of the tertiary industry structure.At the same time,the GDP prediction accuracy based on luminous data will be affected by the development of the secondary and tertiary industries.(2)The correlation and prediction between night-light remote sensing and GDP in time series are studied.Pearson correlation coefficient is used to study the temporal correlation between night-light data and GDP and the expression method of the optimal index is proposed in this paper.Seven optimal indicators such as total lighting value and fixed asset investment were constructed.Considering that economic development is affected by a variety of factors,based on the above seven optimal indicators,three GDP prediction models of random forest,support vector machine and LSTM were designed and constructed,and their applicability was analyzed.Taking Beijing as an example,the experiment shows that the prediction accuracy of GDP by using night-light indicators such as total light value is low,but the prediction accuracy is significantly improved after adding economic indicators such as fixed asset investment.The LSTM model which is sensitive to multi-temporal characteristics achieved the best prediction accuracy in the experiment,and the MAPE evaluation index was 2.79%.The experimental results show that the support vector machine and LSTM prediction models are sensitive to economic indicators and are more suitable for the combination of night-light and economic indicators.However,the sensitivity of random forest to the two indicators is slightly poor,and the prediction error is relatively large.