Water Level Forecast Of Dongting Lake Based On A Long Short-Term Memory Network And The Study Of The Impact Of The Three Gorges Project On The Water Level Of Dongting Lake

Geography studies the geographical environment on which human beings depend,including natural resources such as atmosphere and water.For years,geographers have devoted themselves to studying the fluctuation of natural systems and analyzing the changing patterns of natural processes with time,in order to predict the results of geographical processes and determine the impact of human activities on the natural environment.A representative case is studied in this paper,which focuses on the water level prediction of Dongting Lake and the influence of the Three Gorges Project on the water level of Dongting Lake.As Dongting Lake is the second largest freshwater lake in China,accurate and reliable prediction of the water level is of great significance to water resources management,wetland protection and flood prevention.On the other hand,the Three Gorges Project(TGP)is the largest comprehensive water conservancy project in the world and its operation inevitably affects the lower reaches of the Yangtze River.By studying the influence of the TGP on the water level of Dongting Lake,this paper can provide a reference for the water dispatch and regulation of the TGP as well as the water resource protection of Dongting Lake.Previous research methods on the lake water level prediction can be divided into two categories: physics-based methods and data-driven methods.Given the fact that physics-based methods require a vast number of data and physical parameters for model building and also entail complex solving processes,data-driven methods are preferred in this paper.There are different kinds of data-driven methods,many of which though are incompetent to deal with non-linear data or large amounts of data.As a result,this paper innovatively introduces a deep learning method based on a Long Short-Term Memory(LSTM)network to establish a model that predicts the daily water levels of Dongting Lake and achieves excellent prediction results.Moreover,this model is applied to investigate the influence of the TGP on the water level of Dongting Lake.This paper elaborates the whole research process following the logical order of data collection,model construction and model application.Firstly,according to the regional characteristics,the paper analyzed the main influencing factors of the water level of Dongting Lake,based on which 11 years(2003-2013)of data were collected.This period marks the beginning of the TGP operation and covers the three-stage impoundment process of the Three Gorges Reservoir(TGR).Because the data presents an unusual missing condition,this paper delves into the missing data interpolation methods and proposes a cubic spline interpolation method using skeleton points based on periodic characteristics,which has better success at interpolating missing data.Secondly,according to the characteristics of the data and the idea of multivariate regression,this paper proposes a LSTM network model to predict the daily water levels of Dongting Lake.The whole model constructing and perfecting process is described in detail.Eight years of daily data(from 2003 to 2010)were used to train the model and then the model was tested on the daily data for the next three years(from 2011 to 2013).The experiment shows that the root mean squared error is within the range of [0.08,0.10],and the coefficient of determination is as high as 0.999,which indicates superb prediction accuracy.Moreover,the proposed model proves to have better forecast accuracy compared to the support vector machine model.Furthermore,the proposed model can simultaneously predict the water levels of Dongting Lake for any number of days in the future,thus shattering the glass ceiling that traditional regression models can only have one output value.Finally,as the proposed model has already captured the relationship between the Yangtze River discharge and the water level of Dongting Lake,it can be used to study how the operation of the TGP influence the water level of Dongting Lake.In order to explore the impact of the TGP,this paper simulates the natural runoff state where the TGP does not exist.According to the different time scales of the simulation,the daily impact and the cumulative impact among the time period of 2003-2013 were analyzed respectively.The experiment shows that:(1)The impoundment of the TGR caused the water level of Dongting Lake to drop acutely;(2)In dry seasons,the water level of Dongting Lake had a mild increase due to the replenishment of the TGR;(3)The predischarge operation of the TGR made the water level of Dongting Lake conspicuously higher than that of the natural state;(4)The impact of the TGR resulted in a water level decline in Dongting Lake during flood peaks and a subsequent lagged rise,suggesting a peak staggering effect.What's more,the sensitivity analysis of how the water level of Dongting Lake will react to the change of the TGR discharge is carried out using the LSTM model,in the hope of providing a reference for the water regulation of the TGP.