Simulation Of Groundwater Level Based On CNN-GAN Coupled Network

Groundwater is an important fresh water resource.Its abstraction accounts for more than33% of the global water consumption,and more than 50% of the global irrigation water is groundwater.With the intensification of climate change and human activities,the factors affecting regional groundwater dynamics are becoming increasingly complex,and the difficulty of simulating and predicting groundwater level dynamics is increasing.Simulating and predicting the spatiotemporal changes of regional groundwater level is an important prerequisite for formulating scientific and reasonable groundwater development,utilization,and protection plans,and can provide important technical support for the sustainable,efficient,and safe utilization of groundwater resources.Traditional groundwater simulation method based on physical processes requires a large amount of meteorological,hydrogeological data,etc.It needs to be established on the basis of a clear understanding of the groundwater system and its replenishment,runoff,and discharge laws.There are great challenges in the application in areas where there is a lack of data and insufficient information,and there are also problems in the application process,such as long data collection and modeling cycles,poor portability,etc.Difficulty in utilizing multi-source remote sensing data with high spatiotemporal resolution.A data-driven deep learning model can to some extent compensate for the shortcomings of physical models.It has significant application value in areas lacking hydrogeological data,but requires a long series of representative data for network training and structural adjustment.This work is based on deep learning algorithms and constructs a CNN GAN coupled network model to conduct groundwater level simulation research.It can enrich groundwater simulation methods and provide technical support for sustainable development,utilization,and protection of groundwater,which has important theoretical significance and application value.This work generates a large amount of high-precision spatiotemporal data through conducting indoor controlled experiments on seepage channels,compares the simulation effects of the CNN-GAN coupled network model and the MODFLOW numerical simulation model,and further evaluates the impact of different training data lengths on the simulation performance of the CNN-GAN coupled network model.Based on this,it is applied to simulate the groundwater level in a project area in the plain area of Guangrao County,and generates groundwater level dynamic data in batches based on Flo Py,Evaluated the ability of coupled network models to solve practical problems.The following main conclusions were obtained:(1)Based on indoor controlled experiments,a large number of high-precision spatiotemporal groundwater level observation data were obtained,and simulated using the MODFLOW numerical simulation model and the CNN GAN coupled network model,respectively.The results indicate that the MODFLOW model and the CNN-GAN coupled network model have good simulation results,with R2 and RMSE values of 0.89 and 1.57 for the MODFLOW model,and 0.96 and 0.62 for the CNN-GAN coupled network model,respectively.The simulation results of the coupled network are better than those of the MODFLOW model.(2)We quantitatively evaluated the impact of data length on the simulation performance of the CNN-GAN coupled network model by setting a fixed test length of 200 and a fixed data segmentation ratio of 3:1:1.We constructed simulation models for five scenarios with data volumes of 600,800,1000,1200,and 1395.The R2 mean values of the fixed test length scheme under five scenarios are 0.59,0.80,0.90,0.92,and 0.96,respectively.The RMSE mean values are 2.13,1.52,1.11,0.97,and 0.88,respectively;The R2 mean values for fixed data segmentation ratios are 0.63,0.82,0.93,0.96,and 0.97,respectively.The RMSE mean values are 2.50,1.75,1.03,0.86,and 0.69,respectively.The results indicate that simulation performance increases with increasing data length,but the improvement in model simulation performance significantly decreases when the data volume exceeds 1000.(3)The CNN-GAN coupling network model constructed in this study was applied to simulate the groundwater level in a project area in the plain area of Guangrao County,and the ability of the coupling network model to solve practical problems was evaluated.Using the only measured groundwater level data from 2019 to 2020 in the project area,a MODFLOW numerical model was constructed based on GMS.After exporting it,Flo Py growth series water level simulation data was used to increase the amount of data available for constructing the network model.The training set,validation set,and test set were divided in a 3:1:1 ratio.The results indicate that the accuracy indicators R2 and RMSE of the coupled network model for water level simulation are 0.94 and 0.12,respectively,indicating that the CNN-GAN coupled network model can effectively simulate the groundwater level in the project area and can simulate the groundwater level in areas with limited hydrogeological data.