Simulation And Prediction Of Diving Water Level During Seasonal Freezing And Thawing Based On Machine Learning

The phase change of water during seasonal freezing and thawing causes the interconversion between diving and soil water,which in turn affects the dynamic change of diving water level.In-depth analysis of submerged water level dynamics under the influence of seasonal freezing and thawing is important for accurate evaluation of groundwater resources and prevention of soil salinization and other environmental problems induced by submerged water level dynamics.The complexity of the mechanism and parameters has limited the practical application of mechanismbased physical models for simulation and prediction of submerged water levels under the influence of freeze-thaw.In recent years,the rapidly developing machine learning approach has become an important tool for studying groundwater simulation and prediction because of its characteristics of not needing to consider internal physical mechanisms and being able to better approximate the input-output relationship in the model.In this paper,taking the western plain area of Jilin Province as the study area,using the daily diving water level monitoring data from 75 dynamic observation holes in the national groundwater monitoring project from 2018 to 2021,and based on an in-depth analysis of the characteristics of diving water level dynamics during the freeze-thaw period and its influencing factors,a machine learning-based simulation model of diving water level dynamics during the seasonal freeze-thaw process was constructed using a machine learning approach,and based on the Sixth International Coupled Model Comparison Program(CMIP6)climate model to predict the trend of submerged water level during the freeze-thaw period under different climate change scenarios in the future.The following main insights were obtained from this study:(1)The annual freeze-thaw period in the area is 146 days long,and the depth of permafrost is 0.9～1.4m.The diving water level dynamics has the characteristics of first falling and then rising,continuous falling and continuous rising.The main factors controlling the freeze-thaw effect on diving water level dynamics are the initial groundwater depth,the lithology of the air pocket,the temperature and the snow depth.Among them,when the initial water table burial depth during the freezing period is less than the sum of the maximum depth affected by freeze-thaw and the maximum capillary rise height,the diving water level dynamics is influenced by the freeze-thaw action.(2)Based on an in-depth analysis of the type of diving water level dynamics during the freeze-thaw period and its influencing factors,a time series model of diving water level dynamics during the freeze-thaw period was constructed using the extreme gradient boosting tree(XGBoost)method and compared with the simulation accuracy of the long short-term memory neural network(LSTM)model,and the results showed that the XGBoost model was higher than the LSTM model in terms of simulation accuracy and efficiency.The results show that the XGBoost model has higher simulation accuracy and efficiency than the LSTM model,and the R2 of the validation period is greater than 0.87 and the RMSE is less than 0.09 m.The input variables in the XGBoost model are interpreted by applying the Shapley summation and interpretation(SHAP)method,and the results show that the temperature has the greatest influence on the dynamics of the submerged water table with different initial groundwater depths during the freeze-thaw period,and the characteristic importance values are 0.271～0.521;the snow depth and snow water equivalent are negatively correlated with the submerged water table depth.The snow depth and snow water equivalent were negatively correlated with the depth of submergence,and the influence of both on the dynamics of submerged water level decreased gradually with the increase of initial groundwater depth,and the characteristic importance value decreased from 0.162-0.264 to 0.048-0.055.(3)Using the climate model data from the sixth phase of the International Coupled Model Comparison Program(CMIP6),we predicted the future freeze-thaw submergence depths under three climate scenarios: low radiative forcing sustainable development scenario(SSP 1-2.6),medium radiative forcing moderate development scenario(SSP 2-4.5)and high radiative forcing conventional development scenario(SSP 5-8.5)in the study area.The prediction results show that the dynamics of dive water level during the freeze-thaw period under the three future scenarios of SSP 1-2.6,SSP 2-4.5,and SSP 5-8.5 all show a decreasing trend followed by an increasing trend,and the variation of water level during the freeze-thaw period is 0.15-1.25 m,and the variation of water level during the thaw period is 0.11-0.52 m.Due to the different predictions of future climate change,the variation of water level during the future freeze-thaw period under the three climate scenarios The magnitude and dynamic patterns of the future freeze-thaw period under the three climate scenarios are different.