| The tree species of Poplar is beloved by the mass because of its features of rapid growth, short rotation and because it can be cultivation-oriented easily. Recently, the objective to improve the ecological situation greatly was presented by Jiangsu provincial government. To realize this goal, forest cover percent will be a very significant indicator to judge whether it was improved or not. Because of its unique biologically features, Poplar is chosen as one of the principal tree species for forestation in Jiangsu province. The forestation of Poplar has a significance for heightening the forest cover percent, increasing the income of local farmers and enriching its financial revenue of local authorities. To attain a healthy and sustainable development of Poplar industry in Jiangsu province, it is necessary to execute a scientific and proper planning with respect to Poplar industry. However, the precondition for this planning is to acquire the present status of Poplar resources. The conventional or classical ways to master its spatial distribution of Poplar resources are fargoing and intensive field surveys, obviously, this method for data collection will be time-consuming, labor intensive, and furthermore, the inventory fruits gained from the traditional methodologies will be a bit subjective. Howecer,due to its merits of macroscopic quality, objectivity and efficiency, and dynamic monitoring, remote sensing technique integrated with GIS and GPS will be an effective means for modern resources inventory. On the basis of the upper background, a study aiming at the extraction of Poplar resources by means of remote sensing, GIS and GPS was carried out in Subei plains and this research was funded by a project issued by State Forestry Administration, the thesis of this project was "3S techniques application on the forest resources inventory". Through statistical analysis of ASTER raw datasets and characteristic spectral analysis of principal objects(Land cover types) within the research area, the spectral separability of various land cover types derived from the original 9 ASTER images was intensively mastered. several image fusion algorithms were implemented to produce new feature images according to their spectral coverage of original 9 bands, these fusion methods include principal components analysis, inverse principal components transform ,HIS transform, MNF transform, wavelet-based transform, vegetation index transform and LBV transform respectively. the output of previous transforms was 37 new feature images, plus original ASTER 9 bands, 46 bands were totally acquired and these 46 bands placed a firm foundation for the following remote sensing research regarding the modeling of forest measuration factors and spatial mapping of Poplar. After that, the standard deviation, entropy and optimum index factors for each band were calculated and compared ,especially, the average separability based on the J-M distance and transformed divergence among the dominant land cover types was figured out and compared as well, the results showed that the combination of NDVI, RVI, PVI and MSAVI had the highest separability, so it was assigned to the optimal combination for classification. Depending on the coordinates of sample plots collected by GPS, numerous pixels within the imagery were chosen as training data for each land cover type to mine its statistical characteristics of each category . After all training data were refined by an algorithm designed to eliminate the exotic pixels by using mean and standard deviation of training datasets, the purified features for each land cover can be available. Based on the refined training data, maximum likelihood classifier , artificial neural network classifier and decision tree classifier were developed, executed and evaluated. The assessment results indicated that decision tree classifier had the best classification accuracy, its overall classification accuracy was 84.4% and its corresponding Kappa coefficient was 0.822, furthermore, the producer's accuracy and user's accuracy of Poplar was 93.6% and 83% respectively. The total area of Poplar category derived from the classified image within research region was approximately 5805 hm2 and accordingly the spatial distribution pattern map of Poplar was established. In order to resolve the difficulty of mixed pixels, the reason for mixed pixels to exist is its disperse distribution of Poplar and its ground area is smaller than the corresponding ground area of IFOV of sensor , linear mixed spectral decomposition algorithm and fuzzy clustering algorithm were implemented to extract area fraction and membership grade of Poplar within a mixed pixel , and the Poplar total area on a scale of sub-pixel was converted to 6770 hm2, which was very significant for the calculation of forest cover percent but was less meaningful for mapping of Poplar pixels. By making use of correlation analysis between remote sensing spectral data and the measured biophysical parameters on the ground, the optimal bands for inversing or retrieving biophysical parameters by means of remote sensing image data were discovered or revealed ,and consequently, the estimation models only with one independent variable for stock volume, average height and age groups were constructed .The thematic Maps of forest measuration factors of Poplar were produced by utilizing the constructed estimation models and the informative support to a wise decision-making can be interpreted from those thematic maps. To improve the accuracy of estimation, regression tree models for estimating 3 forest measuration factors were also developed by the aid of cubist and cart software, meanwhile, another series of thematic maps of 3 factors were produced according to regression tree models. By the comparison of two types of models ( models only with one independent variable and regression tree models) and validation of ground truth data, regression tree models with a higher model simulation accuracy and predictive accuracy was testified to be more reliable for inversion of biophysical parameters. Consequently, the spatial pattern distribution thematic maps of forest measuration factors developed by regression tree models had a higher accuracy. after its completion of the upper research work, the conclusions were as follows: 1 After much calculation and comparison of original ASTER 9 bands, the fact that ASTER 3 and 4 were abundant in information deposits so they should be the focus for following remote sensing research was clear. 2 through comparison and analysis of characteristic spectral curves derived from different age Poplar pixels, their characteristic spectral curves were similar and very close to each other. So, there was no need to subdivide Poplar category into 3 age groups. conversely, all Poplar with different ages should be regarded as one class to conduct classification and modeling research. because remote sensing spectral data was insensitive to the variation of Poplar age, the constructed remote sensing spectral estimation models will have a low accuracy ofprediction. 3 The band combination of 4 vegetation indices had the highest average separability for principal land cover types, this band combination was the optimal combination for land cover classification in this study. 4 The method to eliminate exotic pixels by using mean combined with standard deviation of training data had a theoretic foundation of normal distribution, meanwhile, it was easy to operate and it had a reliable performance of eliminating exotic pixels. 5 After much validation of ground truth, a conclusion that decision tree classifier had the highest classification accuracy, followed by neural network classifier, the worst one was maximum likelihood classifier can be made. 6 In regard to the traditional estimation models with one independent variable, HIS_H band was the optimum band for biophysical modeling of Poplar stock volume and average height, but NDVI band was the best band for estimation of Poplar age. 7 From the results of the models simulation and validation, totally, it could be confirmed that regression tree models were superior to the traditional models only with one independent variable because of its higher accuracy . 8 According to the classification and modeling results derived from remote sensing data and processes, the fact that the total area of Poplar within Jianhu county was 5805 hm2,the total stock volume was 348986.74m3(note : the scattered trees were exclusive) can be manifested 。...
|
PDF Full Text Request