Analysis Of Runoff Evolution Based On Data Mining Under Changing Environment

Water resources are the most basic natural resources to support the development of human society,also are the source of sustainable development of social economy and ecological environment.Runoff is a very important part of water resources,and it is the origin of drinking water,domestic water and industrial water.Recent years,the evolution of runoff has been increasingly affected by human activities and climate change,and human activities mainly affected the evolution of runoff by changing the land cover/use,while climate change has been mainly affected by precipitation,temperature,and evapotranspiration processes.The evolution of runoff and the hydrological sequence become unstable under the interaction and influence of various factors such as climate change and human activities.The variation of hydrological factors has already triggered a series of hydrological problems such as water resources assessment mistakes,extreme hydrological disasters,flood control and drought decision errors,and hydrological simulation errors.The evolution among runoff,climate change and human activities has become a vital research direction in the fields of documented hydrology and ecology.Taking Qingliu River as a case,characterize climate change with historical observations of precipitation,temperature,and surface evaporation.Continuous land use/cover classification for the study area to characterize land use/cover change.Expressing vegetation change by whole catchment Enhanced Vegetation Index(EVI).And treat all the above sequence data as data stream.Combining data stream mining methods to propose a dynamic runoff simulation model.Use the online XGBoost algorithm to dynamically build relationships among runoff,human activities and climate change.Meanwhile,this paper combines concept drift detection algorithm based on Shewhart control chart to detect concept drift.The Root Mean Square Error(RMSE)of the runoff simulation method proposed in this paper is 8.4,and the Nash-Sutcliffe Efficiency coefficient(NSE)is 0.73.Outperforming the traditional machine learning methods(Support Vector Regression and Regression Tree)and the hydrological models(Sim Hyd?Sacramento?SMAR and TANK).The main innovation of this article are:(1)The data stream mining method is introducted,and combined with the concept drift detection method,dynamic capture the relationship between human activities,climate change and runoff,to achieve dynamic simulation of runoff.(2)This paper through continuous change detection and classification to characterize human activity data,and at the same time realizes more accurate analysis of vegetation spatial changes.This article is an interdisciplinary study,integrating remote sensing and data stream mining techniques into hydrology,provides new research ideas and perspectives for the analysis of runoff evolution under changing environment.Also provides scientific basis and technical support for the management and sustainable use of water resources.