Study On The Attribution Of Lake Surface Water Temperature Change And The Key Technology And Theory Of Predicting Its Spatio-Temporal Change Trend ——A Case Study Of Nine Plateau Lakes In Yunnan Province

The lake surface water temperature(LSWT)is the lake's water temperature within 1 meter,and is the water quality parameter and important physical parameter of the lake.Some studies have shown that under the stress of global warming,the surface water temperature of global lakes also shows an upward trend.With the continuous intensification of human activities,the impact of the rapid expansion of impervious surface on lake surface water temperature in the process of urbanization cannot be ignored.With its special geographical location,the change of lake surface water temperature will have a very important impact on the diversity of climate,biology and culture in the plateau mountainous region.Therefore,it is necessary to analyze the temporal and spatial change process of lake surface water temperature from the macro scale,and explore the driving factors that affect the change of lake surface water temperature,so as to monitor and warn the water environment emergencies such as the outbreak of cyanobacteria bloom,and then control and improve the lake water ecological environment from the source.Taking the nine plateau lakes in Yunnan(Chenghai,Dianchi,Erhai,Fuxian,Lugu,Qilu,Xingyun,Yangzonghai and Yilong Lake)as the research area,this dissertation mainly explores the image fusion and downscaling of driving factors,the attribution analysis of lake surface water temperature and the prediction of future trends.In terms of image fusion and downscaling,it mainly includes three parts:downscaling of LSWT and natural factors,and spatial estimation of human factors.(1)For the downscaling of LSWT,based on Landsat TM/TIRS,MOD11Q3 and MOD11A1/A2 remote sensing images,combined with statistical Distrad and SRCNN based on deep learning,the downscaling of 1 km to 50 m resolution is realized.Finally,the monthly average LSWT data set with 50 m spatial resolution is obtained,so as to avoid the impact of nearshore surface pixels on LSWT to the greatest extent.The results show that the error of downscaling results is smaller than that of MOD11A2,and LSWT-day is better than LSWT-night(MAE_day=1.10±0.13,RMSE_day=1.36±0.15,MAE_night=1.62±0.11,RMSE_night=1.95±0.11).The downscaling results are reliable and can be used for further research and analysis of LSWT.(2)For natural factor downscaling,Delta and geographic weighted regression(GWR)methods are used for comparison.Natural factors include near surface air temperature(NSAT),surface solar radiation(SSR),wind speed(WS)and water vapor pressure(WVP).The results show that the spatial downscaling and estimation results are reliable and can express the spatial distribution characteristics of various factors.NSAT and WVP use GWR method(MRE_NSAT=8.33%,MRE_WVP=13.03%);Delta method is used for WS and SSR(MRE_WS=11.64%,MRE_SSR=12.78%).(3)For the spatial distribution estimation of human factors,Logistic and Random Forest(RF)regression methods are used for comparison.Human factors include population(POP)and gross domestic product(GDP).The results show that the random forest method is more suitable for estimating the spatial distribution density of human factors(MRE_POP=0.04%,MRE_GDP=0.12%).In the aspect of attribution analysis of LSWT,it mainly includes two parts:lake type division and driving factor analysis.The driving factor analysis includes trend analysis,correlation analysis,contribution rate analysis and threshold analysis.(1)For the classification of lake types,a new classification method that can quantify the impact of human activities on lakes is proposed based on K-Means clustering method,and the lakes are divided into three types at the watershed scale,which are natural lakes(Chenghai,Fuxian,Lugu,and Xingyun Lake),semi-urban lakes(Erhai,Yangzonghai,and Yilong Lake),and urban lakes(Dianchi and Qilu Lake).(2)For the analysis of driving factors,five indicators such as NSAT,SSR,WS,WVP and total precipitation(TP)are taken as natural factors,three indicators such as impervious surface area(ISA),POP and GDP are taken as human factors,the change trend of each driving factor is analyzed by Theil-Sen change rate,and the correlation between each factor is analyzed by Pearson correlation coefficient,the interpretability of regression analysis is used as a method to measure the contribution rate of factors,and the threshold of LSWT is calculated by regression analysis and kneedle analysis.The results show that:For trend analysis,LSWT-day and LSWT-night showed an upward trend in general from 2001 to 2018(CR_day=0.06?/10a,CR_night=0.02?/10a),natural factors except WVP showed an upward trend in general,and human factors showed a significant upward trend in general;For correlation analysis,the correlation between natural factors and LSWT is higher than that of human factors,and NSAT has the highest correlation;For the contribution rate analysis,the eight driving factors can explain the contribution of 79.82%and 76.65%of the LSWT-day and LSWT-night.The main influencing factors of LSWT-day and LSWT-night vary according to the lake types,but it can be determined that NSAT is the most important influencing factor,followed by POP;For threshold analysis,LSWT-day is 18.07?18.59?,and LSWT-night is 14.03?14.36?.The prediction of LSWT's future trend is mainly divided into two parts:short-term and long-term prediction.Based on machine learning theory,this dissertation proposes a SHANN model suitable for short-term prediction of LSWT and a WTFLSTM model suitable for long-term prediction.Both SHANN and WTFLSTM models can better predict the future change trend of LSWT,but both models have the disadvantage of slow convergence and inaccurate prediction of extreme values.Compared with WTFLSTM model,the prediction effect of WTFLSTM model is better(RMSE_LSWT-day=0.1773,RMSE_LSWT-night=0.1788,R~2_LSWT-day=0.8323,R~2_LSWT-night=0.8452);SHANN performs well on local features in short-term prediction(RMSE_LSWT-day=0.1924,RMSE_LSWT-night=0.1851,R~2_LSWT-day=0.8262,R~2_LSWT-night=0.8171).Finally,the change characteristics of LSWT from 2001 to 2028 are analyzed,which further reveals the impact of future human activities on the future change trend of the nine plateau lakes'LSWT in Yunnan.The results show that:(1)The warming trend of urban lakes'LSWT is more obvious than that of semi urban and natural lakes,indicating that human activities have a more significant impact on the urban lakes'LSWT,and the main driving factors are the expansion of impervious surface and the increase of population;(2)The surface water temperature of semi urban lakes is mainly affected by human factors and near surface air temperature;(3)The surface water temperature of natural lakes is mainly influenced by human during the day,and the main driving factor of the LSWT's change at night is the near surface temperature;(4)The influence of human activities on the surface water temperature of the nine plateau lakes in Yunnan is becoming more and more obvious,and it is also the main factor causing the deterioration of the lake water environment in Yunnan Guizhou Plateau.