Research On Dynamic Evaluation Model For Quantity Of Groundwater Resources Based On Remote Sensing Big Data And Machine Learning Method

Quantity of groundwater resources is the dynamic water amount updated year by year in saturated zones beneath Earth's surface,in other words the groundwater recharge amount by rainfall and surface water infiltration.Evaluation for the quantity of groundwater resources is beneficial for providing a basis for the protection and reasonable allocation of groundwater resources and a significant technological support for reasonable and efficient development of groundwater resources,for mitigating the occurrence of ecological and environmental geology disasters caused by irrational development and utilization of groundwater resources,and for sustainable development of society and economy and ecoenvironmental construction.When modeling for the quantity of groundwater resources based on remote sensing data,there is a problem that remote sensing data and quantity data of groundwater resources served as reference data derived from Water Resources Bulletins are at different scales as remote sensing data with multi-time and multi-space scales belongs to big data while reference data volume is limited due to its high cost and long cycle.A conventional method for solving the problem is reducing the scale of remote sensing data to that of reference data,which results in underutilization of remote sensing data and limited accuracy of evaluation model.Thus,an approach for building a dynamic evaluation model for the quantity of groundwater resources was proposed in this thesis,based on remote sensing big data and a machine learning method.In this approach,in order to make a dynamic evaluation of quantity of groundwater resources,two sets of data are at a scale same with that of remote sensing data,and a machine learning method is utilized to reveal the association between remote sensing data and reference data of quantity of groundwater resources.A dynamic evaluation model for the quantity of groundwater resources in Guangdong Province was built,and its accuracy was evaluated.Furthermore,natural regions were generated based on the grading criteria of quantity of groundwater resources and real-time change monitoring was achieved.Several results and new insights have been obtained as follows:1.A method for building a dynamic evaluation model for the quantity of groundwater resources based on remote sensing big data and a machine learning method was proposed and tested in Guangdong Province in this thesis,which has a high accuracy.After extracting the static and dynamic factors related to groundwater from multi-source(MODIS,TRMM,Landsat ETM+/OLI and SRTM),multi-time(2004-2015)and multi-scale remote sensng data,thematic maps and documents,and rasterizing the reference data of quantity of groundwater resources,factor data and reference data were at the same quantity-scale,thus mass training data could be obtained.Based on the training data,a dynamic evaluation model for quantity of groundwater resources in Guangdong Province was built by using multiple back propagation neural network learning and removing samples with large errors.The reference data of quantity of groundwater resources of administrative regions(cities and counties)were used to test the model and the test results show that the model has a high accuracy: 66.4% and 79.5% of relative errors are respectively smaller than 20% and 30% for cities;60.9% and 79.6% of relative errors are respectively smaller than 20% and 30% for counties.2.The approach for building a dynamic evaluation model for quantity of groundwater resources has solved the problem that factor data and reference data of quantity of groundwater resources are at different scales by rasterizing the reference data.In order to improve the accuracy of the evaluation model,training data with large error are removed after the machine learning being done everytime,and finally,the evaluation model is built through several times of learning using the remainder training data.A conventional method for the problem is reducing the scale of remote sensing data to that of reference data,on which a evlauation model for quantity of groundwater resources can be built based with limited training data(only 252 samples).And the accuracy is much lower than that of the evaluation model built based on our proposed approach.3.The dynamic evaluation model for quantity of groundwater resources can be utilized to calculate the quantity of groundwater resources based on piexls in the study area,which can reflect the uneven distribution of groundwater resources in a region to make up for shortcomings that each value of reference data of quantity of groundwater resources represnets the quantity of groundwater resources in a whole region.4.After calculating the quantity of groundwater resources in the study area based on the dynamic evaluation model and defining the grading criteria of the quantity of groundwater resources,the study area can be divided into several natural regions(taking 2007 and 2009 as examples).Natural regions of the quantity of groundwater resources can reflect the relationships between the factors and the quantity of groundwater resources better,can evaluate the quantity of groundwater resources more objectively,and thus can provide better support for scientific researches than administrative regions of the quantity of groundwater resources.5.Real-time change monitoring of the quantity of groundwater resources can be achieved by using the dynamic evaluation model for quantity of groundwater resources.The dynamic evaluation model was used to calculate the quantity of groundwater resources at different times in the study area in three short time periods.And changes of the quantity of groundwater resources in each short time period was analyzed.