Research Of Water Area Extraction Methods Based On MODIS Satellite Data

With the continuous progress and development of remote sensing technology,the useof remote sensing data to extract water information has become an effective means ofmonitoring water resources and statistical water region area.How to deal with large amountof remote sensing data quickly is also a problem encountered in the application of remotesensing technology.In this paper,based on the rapid processing of remote sensing data byprogramming,the water index method and random forests algorithm are used to extractwater bodies information.This paper mainly takes MODIS image as the data source,and takes the global 30 meterssurface coverage data as the verification data.Water index method and random forestsalgorithm are used to extract the water body information in Hulun Lake region,and theaccuracy of the two extraction methods was calculated.The water body information of 32 land protection priority areas in the 2001-2016 year was extracted by using water indexmethod,and the area statistics and analysis were carried out.The research work of thispaper mainly includes:1.The current development of waterbody extraction technology is summarized andanalyzed,and the advantages of random forests algorithm and its application in Remotesensing field are introduced.2.The water index method is combined with satellite data for a whole year to select thethreshold value.By programming the code to realize the NDWI calculation quickly,usingeach point's annual data accumulation and selection threshold,the extraction result iscompared with the validated data,and the appropriate threshold value is selected throughthe experiment.3.Water extraction of MODIS remote sensing image based on random forests.Thewater index was calculated according to the difference of reflectance characteristics ofwater and non-water in different bands and the characteristics were constructed with thepixel values greater than zero.Then the 30 meters spatial resolution land cover productswere selected as training dataset and validation dataset.Then select specific appliedclassification features according to the importance of classification features in randomforests,and random forests model parameters with better classification results wereselected by a certain amount of experimental statistics.4.Statistics on the water area of 32 land protection priority areas nationwide.The areaof the high water-level period and perennial water area of the 32 priority regions werecalculated by using the area of the high water-level period and the year-round waters asindicators of surface water changes,and the trend of change from 2001 to 2016 wasanalyzed.Through a large number of experiments analysis,the new improved water index methodand random forests algorithm adopted in this paper proposed has relatively good precisionin water extraction.