Research Of Remote Sensing Drought Monitoring Model Considering Spatial Non-stationary

Changing in global climate leads to frequent drought events,serious impact on economic development,industrial and agricultural production and ecological environment etc.How can we develop a timely,accurate and efficient drought monitoring approach is very urgent for the social,economic,ecological and environmental.Traditional monitoring methods generally rely on the ground monitoring system for meteorological stations,which is a direct approach for drought monitoring with rich history datum,research data and strict theoretical basis.However,it faces the matter of low resolution and complicated processes in obtaining and processing.With the rapid development of satellite remote sensing technology,the advantages of high spatio-temporal resolution,large-scale coverage and abundant data sources provide great possibility for a real-time and rapid monitoring for drought information.At present,the domestic and foreign research scholars developed a series of drought monitoring models integrated ground and remote sensing data,such as SDCI(Scaled Drought Condition Index,SDCI),SDI(Synthesized Drought Index,SDI),OMDI(Optimized Meteorological Drought Index,OMDI),MIDI(Microwave Integrated Drought Index,MIDI)ect.However,most of them ignore the changing of spatial model in large scale causing by differences in innate attributes of different spatial areas,and making the sumpition that it has constant relationship between standard station data and remote sensing inversion data.Aiming at the problem of fluctuation in spatial relationship,causing by changes of geographical data attributes,naming spatial non-stationary,many researches put forward the solution of'regionalization' to detect spatial non-stationary relations of regionalized variables.Geographically weighted regression technology can insert location features of regionalized variables into the process of model optimization.It can not only solve the local relationship between regionalized variables and ground data,but also detect the changing process of spatial non-stationary and achieve visual expression.Compared with other non-stationary detection technology,GWR based on the idea of local smooth,has simple process,low computational complexity,high degree of automation,it is widely applied to the direction of geology,environmental science,property characteristics,landscape ecology,health research and crime research.In this study,a new model of remote sensing drought monitoring is constructed by using the non-stationary detection technology,GWR,combining the neighborhood characteristics of geographic information.The continental United States,CONUS,is selected as the main research region,using 2002-2011 month scale multi-source remote sensing data as the main data source.We completed CONUS's 10 years drought monitoring process,and selected various widely used approaches,comparing to our work.The main research conclusions are as follows:(1)As the other geospatial events,the drought data also have the characters of spatial non-stationary and spatial autocorrelation.In addition,the drought monitoring owns its specific complexity.The traditional global experience models have much difficulty in detecting spatial non-stationary and expressing heterogeneity of local drought models.Global drought results are not good enough in large-scale research area cause global parameters are difficult to adapt the local complex surface condition.(2)GWR could detect the spatial non-stationary characteristics of drought data effectively,and the local optimization model obtained has better local fitting results when compared with global empirical model.Meanwhile,local model parameters could explain the local drought pattern in detail,and have the advantage of regional visualization.GWR technology has a good application effect in drought monitoring,and has a broad development space.(3)Each monitoring method has advantages and disadvantages:single index model has advantage of simple process,but ignore the complexity of drought development process.It is difficult to accurately fit the real surface condition.Models based on spectrum space are difficult to adapt to the long time series data.Models combined multi characters have the best performance,expressing multi factors caused drought,and solve the problem discrete distribution.In future,it is an important way for better drought monitoring by using multi source remote sensing data developing a comprehensive drought monitoring model.