Spatiotemporal Dynamic Analysis And Prediction Of Groundwater Level And Electrical Conductivity In The Yellow River Delt

The Yellow River Delta region is located in the estuary of the Yellow River.The shallow groundwater in the region is affected by the unique geographical location and the landforming environment,and the shallow groundwater burial dynamic process is complicated,so it is difficult to master the dynamic law of the groundwater system.At present,there are few quantitative analyses on the temporal and spatial dynamics of groundwater in the Yellow River Delta,and the prediction accuracy of groundwater models is not enough.Therefore,the analysis of the spatio-temporal dynamic changes of groundwater level and conductivity and the prediction of the dynamic change trend of groundwater not only contribute to the scientific management of regional groundwater resources,but also effectively protect the terrestrial ecological environment and maintain the balanced development of the ecosystem.Firstly,based on the groundwater level and conductivity data of 18 observation Wells in the region from 2004 to 2010,soft data were constructed by combining environmental factors with Multiple Linear Regression(MLR)and geographically weighted regression(GWR).The spatial interpolation of groundwater level was carried out by combining the Bayesian Maximum Entropy model(BME).The precision was compared with Ordinary Kriging(OK),Co-Kriging(CK),MLR and GWR models.On this basis,a suitable model is selected to analyze the spatial prediction distribution characteristics of groundwater level and conductivity,and reveal the extent and range of regional groundwater affected by seawater intrusion.Secondly,using Seasonal and Trend decomposition using Loess(STL),the time series of regional groundwater level and electrical conductivity is decomposed into seasonal term,trend term and residual term,and through MK trend test and Pearson correlation analysis,the time variation law of groundwater dynamics is discussed.Combined with the time series of precipitation and evaporation,the influencing factors of groundwater dynamic change were analyzed.Finally,based on the results of STL time series,the prediction model of groundwater level and conductivity of STL-LSTM neural network is constructed by using the multi-variable input method,and holt-winters and LSTM models are selected to compare and analyze the prediction results.The results show that:(1)The BME-GWR model has a stronger ability to explain the spatial heterogeneity of groundwater level.The cross-validation determination coefficient is 0.98,the root-meansquare error is 0.15,and the average absolute error is 0.11.The simulation accuracy is higher than that of other models.The interpolation results of groundwater table and electrical conductivity show that the groundwater table in the Yellow River Delta increases gradually from the coastal area to the inland area,and the groundwater table along the banks of the river and in the coastal area is significantly lower than that in other areas.Groundwater conductivity showed a trend of increasing from the middle to the east and west.More than42% of the average groundwater level in the region is lower than the region’s sea level,and more than 96% of the average groundwater conductivity meets the seawater intrusion standard.(2)During the observation period(2004-2010),the groundwater level of the Yellow River Delta showed a slow decline,with a rate of 0.01-0.045 m/a;The variation trend of groundwater conductivity is complicated.About half of the Wells observed showed an increasing trend,with an increasing rate of 1.117～25.6μs/cm.a.The annual variation of groundwater table is in the pattern of "down-up-down",and there is a significant positive correlation between groundwater table and precipitation and evaporation.Rainfall is the dominant factor affecting the seasonal dynamic variation of groundwater.In addition,there is a significant positive correlation between groundwater conductivity and evaporation in most Wells,which is an important factor affecting groundwater conductivity in this region.The dynamic characteristics of groundwater in the Yellow River Delta have obvious spatial differentiation,that is,the groundwater table in the northern part of the region is dominated by long-term trend changes,while the groundwater table in the southern part is dominated by periodic seasonal changes.In the western part of the study area,the groundwater conductivity is dominated by long-term trend changes,while in the eastern part,the groundwater conductivity is dominated by periodic seasonal changes.(3)The multi-variable LSTM neural network model based on STL can be used as an effective tool to predict the dynamic change of groundwater.Compared with the LSTM neural network and holt-winters model,the average RMSE predicted by the STL-LSTM model is increased by 38.5% and 42.58%,respectively.The average RMSE predicted by the test set of groundwater conductivity was increased by 13.76% and 41.71%,which has the advantages of strong learning ability and generalization ability.Choosing precipitation and evaporation as the input layer of the model is beneficial to the accuracy of the prediction of groundwater level and electrical conductivity of the Yellow River Delta,and the combined model has higher stability than the single model.