Research On Status And Change Of Vegetation In Beijing City Based On Remote Sensing Imagery Analysis

Vegetation cover is one of the primary indicators for ecosystem. As a metropolis and the capital of China, ecosystem and environment are crucial to political, economic and cultural development of Beijing.Therefore, it is very meaningful to understand the recent vegetation change and its status in Beijing comprehensively and accurately for ecological management, urban planning and development of the economy. This study aims at finding the following vegetation aspects in Beijing using vegetation analysis methods, change detection technologies and vegetation extraction algorithms based on the multiple remote sensing imageries:vegetation change trends and their major causes in Beijing from1998to2010, land use and land cover change from2006to2010and vegetation distribution pattern using AdaBoost algorithm based on LandSat TM remote sensing imagery in Beijing in2006. The main work and conclusions are as follows:1. Aiming at the problem which the manual threshold in linear regression vegetation change analysis method affect dramatically by the human factors, linear significant test was employed to settle the classification of vegetation change trend method. The significant test can divide the vegetation change trend into four classes (significant degradation, degradation but not significant, improvement but not significant and significant improvement) or five classes (significant degradation, degradation but not significant, no trend, improvement but not significant and significant improvement). The F test and correlation coefficient were used to classify the vegetation change trend. The experimental results show that the classification difference between F test and correlation coefficient is very little, only0.01%. At the same time, these two methods are very same in amounts of calculation, so they can substitute for each other.2. The significant test only can divide the vegetation change trends into four or five classes. The more subdivision, for instance seven classes, is return to use the method of manual threshold. A clonal selection clustering algorithm was introduced to subdivision in vegetation change trends. The experimental results show that the DBI of the proposed approach is significantly less than the compared algorithms OTSU, K-Means and FCM. The results are feasible because it is consistent with the data released by the Beijing Gardening and Greening Bureau.3. For the shortcomings of the parameter test, which the data should be satisfied the normal distribution and it is also sensitive to the noise, Sen+Mann-Kendall approach was used. This method has good property in resistance of the noise and it is also no requirement in data distribution. The experimental results indicate that the classification difference between Sen+Mann-Kendall and linear regression vegetation change analysis is only2.36%, so Sen+Mann-Kendall is more suitable for the noise existing commonly in remote sensing imagery and unknown distribution of the data than the other approaches.4. The overall vegetation change trends are increasing in Beijing from1998to2011and the increasing regions are76%of the total areas. The vegetation’s change is significantly increased in the following areas:the urban of Beijing city, Yanqing county, Huairou and Pinggu district. On the contrary, the vegetation’s decreasing areas locate at the north, east and south of the urban district in Beijing and surrounds like a Horseshoe-shaped, especially in the neighbor on urban area regions of ChangPing, ShunYi and DaXing.5. Temperature and precipitation are not the major driving factors on vegetation change. Vegetation change was affected by human factors remarkably. This affection is showed in the following two aspects:plants and green increase the vegetation cover such as the urban area of Beijing, in contrast to the rapid urbanization destroyed the vegetation on the earth surface, vegetation degraded significantly.6. Interannual changes in vegetation conditions in Beijing were also studied using the vegetation index approach based on the SPOT VEGETATION NDVI data from August,212006to August,212010. The experimental results indicated that vegetation change was stable from2006to2010. The difference between degradation and improved regions is only2%.7. Vegetation extraction approach based on AdaBoost algorithm was proposed combined the property of spectral of vegetation (NDVI and combination of bands) and weaker classifiers of decision tree using high resolution remote sensing data. The experimental results showed that the overall accuracy and Kappa reached96.33%and0.93respectively. The proposed algorithm is robust according to the more testing data. Therefore, the proposed approach can be applied to other areas of vegetation information extraction.