Spatial Statistics And Its Application Study In Mapping And Image Processing Of Forest

Although the research and practice for spatial statistics was less than 50 years, itsapplication has been extended to the analysis of spatial heterogeneity and spatial patternof all kinds of natural phenomena. In the course of growth and development, forest isaffected by the factors (including physical environment, natural disturbance, man-madeinterference, characteristics of tree species, and response of tree species to disturbance),and so it has strong spatial heterogeneity and temporal heterogeneity. In sum, the spatialstatistics should have extensive application prospects in forestry field. However, atpresent, the application of spatial statistics to forestry just began in China. The domesticand foreign documents related to the application of spatial statistics to forestry arescarce. This paper firstly summarizes the basic theory of spatial statistics and reviewsthe current situations of research and application of spatial statistics in forestry, thenconducts the research for using the spatial statistics to resolve 2 important problems (theinterpretation of real information-missing region of remote sensing image and theconditional simulation of spatial distribution map of forest types) related to forestryimage and image treatment, and then develops the corresponding computer proceduresuitable for the characteristics of forestry field. The main conclusions, work andinnovations for this paper are as follows:1) The TM image (resolution of 30m*30m) of Jinggouling Forest Farm of WangqingForestry Bureau of Northeast China in 1997 was selected as data resource, and a patchwith cloud shade from the TM image was adopted as experimental data. The pixelinformation of four patches covered by cloud and cloud shade was restored by applyingthe Kringing interpolation technology of spatial statistics theory. The restored pixelinformation was been reclassified, and then was compared with the results of forestresource survey, thus the results (that the classified precision of the restored informationof each patch was more than 75%), was obtained. The results of precision indicated thatthe Kriging interpolation model was a tool and method for us to interpret the blurregions on image. At present, the results like this paper (that applied the spatial statisticstheory to interpret the problem on real information-missing region of remote sensing iAbstractimage) have not still reported.2) Using the sample plots at the bureau level of the Northeast China's WangqingForestry Bureau as the tested materials, the distribution map of forest types of theWangqing Forestry Bureau was drawn by the sequential indicator simulation (SIS) ofcategorical variables, the improved sequential indicator simulation (ISIS) of categoricalvariables, and the square-drawing method using sampling point as centre (SDMUSPC).The distribution map (obtained by the SIS, by the ISIS, or by the SDMUSPC) of foresttypes was compared with that (obtained by the forest resource survey). The generalsimulation precisions of the SIS, the ISIS and the SDMUSPC were 73.80, 81.43 and69.12% respectively. The simulation precisions of the SIS and the ISIS were higher thanthat of the SDMUSPC, which indicating that the conditional simulation might obtain arelatively high simulation precision. Therefore, using the sample plots as materials, theSIS and ISIS might be used as the effective paths of obtaining forest type distributionmap. The precision analysis indicated that the simulation precision was relatively low inthe regions with relatively abundant forest types, while the simulation precision wasrelatively high in the regions with relatively few forest types. At the same time, thesimulation precision of the forest type (that accounted for the larger percentage amongthe known forest types) was relatively higher in research region.3) In the course of applying the SIS, it was found that when the asc...