(1. Key Laboratory of Land Use, Ministry of Land and Resources, China Land Survey and Planning Institute, Beijing,100035; 2. China Academy of Surveying and Mapping, Beijing, 100039)
Based on MODIS 16-day NDVI time series data, 8-day LST data,1∶ 50,000 DEM data and other auxiliary data, this paper mainly discusses the investigation of land resources in northwest Beijing and the study of land use and vegetation cover change for many years. Firstly, the land cover classification system suitable for MODIS data classification is selected, and the NDVI time series data is enhanced and compressed by principal component analysis. Combined with LST data, DEM data and rainfall temperature data, the land cover classification of the study area is carried out by fuzzy K-means unsupervised classification method, and the present situation of land resources is obtained. Then, using CVA analysis method, this paper analyzes the changes of land use and vegetation cover in northwest Beijing for many years. The results show that MODIS data can be well applied to a wide range of land resources monitoring and achieve good results.
Keywords: northwest of Beijing; Moddis; Present situation of land resources; Land use and vegetation change
With the increasingly acute contradiction between population, resources and environment, in order to achieve the strategic goal of sustainable development, governments all over the world are vigorously strengthening the construction of resources and ecological environment monitoring systems. China's "National Ecological Environment Construction Plan" and "National Ecological Environment Protection Outline" also clearly put forward to improve the ecological environment monitoring and information service system. The Outline of the Tenth Five-Year Plan for Science and Technology Development of the Ministry of Land and Resources emphasizes vigorously promoting the informationization of land and resources management and striving to modernize the investigation and evaluation of land and resources. Among them, land resources investigation and monitoring is one of its main contents.
With the rapid development of modern remote sensing technology, various remote sensing data suitable for land resources investigation and monitoring at different spatial scales have appeared one after another. At present, SPOT5, Landsat TM/ETM ++ remote sensing data are the main data sources for land resources investigation, which are suitable for different scales such as township, county and district, but when applied to land resources investigation on a larger spatial scale, it consumes a lot of manpower and material resources and is not economical. In recent years, MODIS data with medium spatial resolution and high temporal resolution provide a good data source for large-scale land resources survey. Tucker et al.' s research shows that [1], normalized vegetation index NDVI actually reflects the biophysical and chemical properties of vegetation biomass, coverage and chlorophyll content, and NDVI series under different phase conditions can accurately reflect the seasonal variation law of vegetation growth. This has become the basic idea of mapping large area vegetation and land cover using remote sensing data [2 ~ 4].
This paper attempts to use MODIS NDVI time series data set as the main data source, combined with MODIS LST, DEM, rainfall and temperature and other auxiliary data, firstly select a suitable land cover classification system, and use fuzzy K-means unsupervised classification method to study the automatic land cover classification in northwest Beijing. Then the change vector (CVA) analysis method is used to analyze the changes of land use and vegetation cover in this area for many years, which provides a quick and convenient method for large-scale land resources investigation and monitoring.
1 Overview of the study area
The northwest of Beijing is the focus of attention and investment by the ecological environment construction department of China. The study area mainly includes the sandstorm source area in the northwest of Beijing, involving 8 cities (prefectures, leagues) in Hebei, Shanxi and Inner Mongolia (5 1 county (city, flag), with a total land area of 22.83× 104 km2. The area starts from Siziwangqi in Inner Mongolia in the west, reaches Aohan Banner in Inner Mongolia in the east, reaches Daixian County in Shanxi Province in the south and reaches Arukerqin Banner in Inner Mongolia in the north. The geographical coordinates are110 20' ~121kloc-0/'and 38,565,438 north latitude.
The study area is located in the middle of Inner Mongolia Plateau, at the northern end of Loess Plateau, at the junction of Inner Mongolia, Shanxi and Hebei provinces. The surface morphology in the area is mainly composed of plateaus, mountains, hills and basins, with high terrain in the middle and low in the north and south. The study area spans the middle temperate zone and the cold temperate zone, and belongs to the arid and semi-arid continental monsoon climate with obvious climate change. Winter and spring are controlled by cold high pressure in Siberia and Mongolia, with dry climate and little rain, and the dominant wind direction is northwest wind, with strong wind force and strong geological effect of wind erosion. The ecological environment of Hunshandake Sandy Land and Horqin Sandy Land in the north of the study area is extremely fragile, which are the main sandstorm source areas in the northwest of Beijing. Regional summer and autumn are controlled by Pacific subtropical high, with southeast wind, weak wind, less water vapor supply, hot climate and little rain. The average annual temperature in the region is below 12.6℃, and the annual rainfall is 200 ~ 750 mm, but the rainfall is concentrated, the rainfall intensity is high, the terrain slope in the additional area is large, the soil is loose, the water erosion type external stress geology and gravity erosion are strong, the soil erosion is serious, and landslides and mudslides are prone to occur.
2 data and preprocessing
2. 1 remote sensing data
The remote sensing data used in this paper are MODIS images provided by EROS data center in the United States. The NDVI data is the time series data synthesized from 200/kloc-0 to June 6, 2004, with 23 phases and a spatial resolution of 250m ... The land surface temperature (LST) is the synthetic time series data of 8 days in 2002, with 46 phases and a spatial resolution of 1 km.
In MODIS data processing, MRT geometric correction and mosaic software is used to complete the geometric correction and mosaic of images. Then the multi-temporal data of the same area, such as vegetation index and land surface temperature, are synthesized and preprocessed by the maximum synthesis method (MVC), that is, every pixel in the image is replaced by the maximum pixel value in J days. The purpose of this treatment is to reduce the influence of atmospheric clouds, particles, shadows, viewing angles and solar altitude angles (Brent, New Hampshire, 1986). Although the maximum synthesis process (MVC) reduces the influence of clouds and particles in the atmosphere, cloud pollution still exists. Then, the improved optimal exponential slope extraction method is used to process the NDVI multi-temporal phase difference cloud. Although the LST data of MODIS are all 8-day synthetic data, the quality of Ts data is very poor. In order to solve the problem of incomplete data, we use linear regression to simulate these data. The surface temperature has a strong correlation in space, and Ts in the same area has some same correlation in space when it is adjacent, so this relationship is fitted by linear relationship. Use the same size template to slide on the restored image and reference image at the same time. If the central value of the template in the restored image is zero or abnormal, the linear regression coefficient between the valid data in the two templates is obtained by the least square method, and then a new value is obtained by using the coefficient and the central value of the reference image template to replace the original zero or abnormal value.
2.2 Other auxiliary data
The auxiliary data mainly include the current land use map of northwest Beijing in 2002, the DEM data of1∶ 50000 in northwest Beijing, and the precipitation and temperature data in northwest Beijing. According to the data of meteorological stations in northwest Beijing, the annual average accumulated temperature and annual average precipitation of each station are calculated first, and then the annual average temperature and annual average precipitation distribution map of the grid in northwest Beijing is obtained by kriging interpolation method.
3 research methods
3. 1 land cover classification
(1) Select a land cover classification system suitable for MODIS data classification. In this study, a land use/land cover classification system based on remote sensing data is adopted [5]. The most important feature of this classification system is that there are corresponding classifications for different spatial scales and corresponding remote sensing data sources, and the classification types are gradually refined. For primary classification and secondary classification, land cover classification is emphasized, that is, for remote sensing data with medium and low spatial resolution, land cover classification is the main method.
(2) The information of NDVI time series data is enhanced and compressed by principal component analysis, so as to eliminate all kinds of interference factors and improve the classification accuracy. Using PCA transform, most of the useful NDVI information in the original 12 months can be compressed into a few first principal component, and some noise caused by data quality can be eliminated. Therefore, PCA transform can effectively ensure that the classification accuracy is not lost. The research on the actual results also shows that PCA plays an important role in suppressing the influence of noise and ensuring the classification accuracy [6].
(3) Combining the LST data, DEM data and rainfall temperature data, the land cover classification of the study area is carried out by using the fuzzy K-means unsupervised classification method [7], and after the classification processing, the map spots with obvious classification errors are corrected to obtain the land cover classification map of northwest Beijing.
3.2 Analysis of land use and vegetation cover changes over the years
Change vector (CVA) analysis is a very potential method for vegetation comparative analysis. According to the intensity and direction of the change vector, the change area and type can be determined [8]. The change vector analysis technology takes each data value in the annual time series of index parameters as a point in the time series space, and connects the points in the time series space of several years to form a change vector. The direction of the change vector determines the progress of the change, and the size of the vector represents the intensity of the change.
For example, if the data of many years 12 months are used to analyze the change vector, and the change vector space consists of images of 12 change monitoring indicators every year, then the indicators of the whole year correspond to a time vector of 12 dimension:
Innovation of Land Information Technology and Development of Land Science and Technology: Proceedings of the 2006 Annual Conference of china land science Institution.
P (i, x) represents the vector of pixel I corresponding to X years, x (t) is the index value of pixel I from time t 1 to tn, and n represents the time dimension. The modulus ‖P‖ of the vector represents the accumulation of index factors throughout the year, and the direction of the vector is a comprehensive reflection of the time curve shape of index factors throughout the year.
Any change of the index factor in any two years will be shown in this 12 dimensional space, which can be described by the change vector as follows:
Innovation of Land Information Technology and Development of Land Science and Technology: Proceedings of the 2006 Annual Conference of china land science Institution.
Δ p (I) is the change vector of pixel I from X year to Y year. Δ p (i) contains the change information of pixel I in each time dimension within (Y-X) years. The modulus ‖Δp (I) ‖ of the change vector is determined by the Enclidean distance, which indicates the change intensity of the index.
Innovation of Land Information Technology and Development of Land Science and Technology: Proceedings of the 2006 Annual Conference of china land science Institution.
When ‖Δp (i) exceeds a certain threshold, it often corresponds to the change of vegetation coverage type from one type to another. The direction of Δ P (I) is defined by a series of angles, which determines the changing process of the index.
According to the histogram characteristics of image and ground data, threshold segmentation method can be used to divide different vector change intensity levels.
The type of vector change is judged by the change rate of the cumulative value of the indicator factor. The rate of change is defined as follows:
Innovation of Land Information Technology and Development of Land Science and Technology: Proceedings of the 2006 Annual Conference of china land science Institution.
4 Conclusion and discussion
4. 1 Investigation on the present situation of land resources
The first four principal components of NDVI time series data of the study area in 2002 12 months were obtained by principal component analysis, and the annual average LST data of the study area were obtained based on the 8-day synthetic LST data of MODIS and DEM data with the resolution of 1∶50000. Then, combined with the rainfall temperature data, the land cover classification results in northwest Beijing were obtained by using the fuzzy K-means unsupervised classification method. Then the classification results are classified and post-processed, and the patches with obvious classification errors are corrected, and finally the land cover classification map of northwest Beijing in 2002 (figure 1) is obtained.
Figure1Land cover classification map of northwest Beijing in 2002
As can be seen from Figure 1 and Figure 2, grassland accounts for the largest proportion of land cover types in northwest Beijing, accounting for about 53% of the total area, and the Hunshandake sandy land and hilly area in Inner Mongolia Plateau and the southern edge of Horqin grassland in the east of the study area are concentrated. The low-lying areas of Bashang Plateau in the central and western regions, the surrounding areas of rivers, lakes and beaches, and the hilly areas in the east of Yinshan Mountain are also relatively concentrated. More than 60% of the total grassland area is distributed in sandy and even sandy arid grassland areas. Agricultural land accounts for 265,438+0% of the total area of the study area, mainly distributed in plateaus and basins in the southwest of the study area, and mostly distributed along river valleys and river alluvial plains. Woodland accounts for 65,438+03% of the total land area in the study area, mainly distributed in Daxing 'anling, Yanshan, Hengshan and Yinshan areas. Forest land is concentrated in the eastern and southwestern mountainous areas of the study area, and most of them are distributed in the upper part of the mountain. Bare land accounts for 8% of the total area of the study area, mainly distributed in Hunshandake sandy land in the north and Horqin sandy land in the east. In the study area, wetlands, waters and construction land account for the smallest proportion.
Figure 2 Proportion of land cover types in northwest Beijing in 2002
4.2 Changes in land use and vegetation cover over the years
The change of NDVI in northwest Beijing from 2001to 2004 was monitored by using the change vector analysis method. The data used are the monthly maximum NDVI time series from 200 1 to 2004 in northwest Beijing. Firstly, the change vector modulus of NDVI is calculated, and then the change intensity of NDVI is generated by using the image segmentation technology of change vector modulus. Image segmentation meets the following requirements: ① similarity principle, that is, pixels in the same area should be similar; (2) Discontinuity principle, that is, when searching from one area to another, some variable characteristics of pixels (such as gradient) must suddenly change, so as to determine the boundary.
4.2. 1 change intensity
The change intensity of NDVI reflects the change of vegetation cover. Considering the histogram, mean and variance of the change vector module of vegetation coverage comprehensively, each segmentation point is determined, and the change vector module is segmented to get the change intensity.
Table 1 thresholds of different levels of vector change intensity
As can be seen from Figures 3 and 4, during the four years from 200/kloc-0 to 2004, the land use/vegetation cover in most parts of northwest Beijing changed little, and the ecosystem remained basically balanced. The area with no change and low change accounts for 92.3% of the total area in northwest Beijing.
The unchanged area is the largest, accounting for 53.7% of the total area, mainly distributed in Chifeng City, Aohan Banner, Wengniute Banner, Bahrain Right Banner and Arukerqin Banner in the northwest of Beijing, and Xilingol League and Siwangzi Banner in the north, indicating that the vegetation coverage in the Three Gorges reservoir area has not changed much in the horizontal direction in recent five years.
Low-variable area accounts for 38.6% of the total area, mainly distributed at the junction of chahar right middle banner, Chahar Right Front Banner, Hexigten Banner, Zhengxiangbai Banner, Zhenglan Banner and Taibus Banner in the northwest of Beijing.
The medium-variable area is relatively small, mainly concentrated in Liangcheng area in the northwest of Beijing.
The drastic change areas are mainly concentrated in the northwest of Beijing, around Hunshandake sandy land, north of Hexigten Banner and Siwangzi Banner.
Fig. 3 Variation intensity of vegetation index (NDVI) in northwest Beijing in 2001-2004.
Fig. 4 Intensity-area ratio of vegetation coverage vector change
Change type
The intensity of NDVI vector change in northwest Beijing in the past four years is calculated above. The vector change intensity reflects the change degree of NDVI in northwest Beijing from 200/kloc-0 to 2004, but it is impossible to judge whether the vegetation coverage has increased or decreased in these four years. Therefore, the change type of ND-VI vector can be judged by the change intensity of NDVI and the change rate of NDVI cumulative value [9].
Threshold m is the threshold of no change and low change in change intensity. Pixels with varying intensities less than m are considered to be stationary. When the change intensity is greater than m, the change type is determined according to the cumulative change rate. The specific parameters are as follows:
Innovation of Land Information Technology and Development of Land Science and Technology: Proceedings of the 2006 Annual Conference of china land science Institution.
According to the calculated cumulative change rate of NDVI and considering Formula (5), the NDVI vector change type map can be obtained.
As can be seen from Figures 5 and 6, from 200 1 to 2004, the vegetation coverage in northwest Beijing showed a steady increase trend. To sum up, the change characteristics are as follows:
Fig. 5 Vegetation coverage index (NDVI) in northwest Beijing from 200/kloc-0 to 2004.
Fig. 6 Intensity-area ratio of vegetation coverage vector change
(1) The vegetation cover change type is mainly fixed, accounting for 56.9% of the total area, mainly distributed in the northeast and northwest of Beijing.
(2) The increasing type accounts for a large proportion, accounting for 30.2% of the northwest area of Beijing, mainly distributed in the south-central part of the northwest area of Beijing.
(3) The reduction of vegetation coverage has been reduced to a certain extent, accounting for a small proportion, mainly distributed in sporadic areas north of northwest Beijing.
(4) The wave type accounts for 10.9% of the northwest area of Beijing, mainly distributed in the northeast of northwest Beijing. The fluctuation of vegetation coverage is a normal natural phenomenon, which is the result of natural effects such as normal vegetation growth, long-term climate change and various human economic activities.
refer to
[1]Tucker, C.J., Townshend, J.R.G and Goff, T.E. Classification of land cover in mainland China using meteorological satellite data [J]. Science, 1984 (227):369~375.
Cihlar J. Satellite Land Cover Mapping: Present Situation and Research Emphasis [J]. International Journal of Remote Sensing, 2000,21(6&7):1093 ~11/kloc-0.
[3]Townshend J.R.G, et al. Classification of global land cover by remote sensing: current capacity and future possibility [J]. Environmental remote sensing, 199 1 (35):243~255
Lloyd D. Phenological classification of terrestrial vegetation cover using short-wave vegetation index images [J]. International Journal of Remote Sensing, 1990,1(12): 2269 ~ 2279.
[5]J.Wang, T.He. Development of land use/cover classification system in China based on remote sensing data. SPIE proceedings-application of remote sensing in environmental monitoring, GIS application and geology IV, 2004, (5574):52~60.
Liu, Wang Jing. Study on forest mapping in China based on PCA transform and neural network classification. Yangtze River Basin Resources and Environment, 2006,15 (1):19 ~ 24.
Liu, Wang Jing, et al. Study on land cover classification in northwest Beijing based on MODIS data. Progress of geographical science, 2006,25 (2): 96 ~102
Chen, Jin Chen and Shi. 1983 ~ 1992 Variation vector analysis of NDVI evolution characteristics of terrestrial vegetation in China. Journal of Remote Sensing, 2002,6 (1):12 ~18.
[9] China Academy of Surveying and Mapping. Study on ecological environment monitoring technology in Three Gorges reservoir area. Project acceptance summary report .2005, 10