Extraction And Dynamic Monitoring Of Grass Resource Information Based On RS Data

China has a vast area of grass resource which is not only the material basis for developing the national economy, but also a natural protection of terrestrial ecological environment. At the same time it is vital important to learn about space-time configuration status of grass resource timely and accurately for it is a relatively fragile ecosystem and Chinese natural grass resource faces such problems as area decreased, the quality declined, grass resource and livestock's contradiction sharpened, and so on. Remote sensing monitoring with the features of large-scale of the area, small-scale of the time and grasping the situation of grass resource in real-time, could provide strong support for the dynamic monitoring and analysis management of the grass resource.Taking Habahe County, Xinjiang Uygur Autonomous Region as an example, the study made use of multi-temporal MODIS, LANDSAT and HJ star multi-spectral image data. After preprocessing the data and by analyzing the spectral characteristics of grassland resources, the study extracted the information of grassland resources with the methods of decision tree classification and supervised classification. Determined the grading information and monitored the temporal and spatial development of the study area in 2007 growth period and annual and inter-annual of 2000-2010.(1) Remote sensing data preprocessing Remote sensing data of different pre-treatment is slightly different. MODIS data is mainly use software of MRT to take projection conversion so that all research data under the same projection information; TM / HJ satellite data preprocessing include: band data calculation of the HJ star, transform the DN values into radiance density values to strike a spectral response function, and then use FLAASH model to take atmospheric correction of TM / HJ satellite data respectively. Taking the TM data as reference data, reposition satellite data on the TM/HJ with polynomial correct model and re-sampling with cubic convolution interpolation to complete the geometric exact correction in which the RMS error is 0.37 pixels. The results showed: radiation correction can eliminate the interference like atmosphere to remote sensing data, restore object spectrum information high-fidelity to ensure the accuracy of NDVI index; exact geometric correction laid the foundation for analyzing the latter part of the annual temporal and spatial dynamics development of grassland.(2) Extraction of grass resource informationTaking the 2007 current land use map and grass level map in the scale of 1:20 million of Habahe County as reference, analyze temporal characteristics of the MODIS's NDVI in different grassland growth periods. Take combined treatment on multi-temporal remote sensing data. According to the temporal differences of main land types and typical grassland's NDVI, extract the main land type Habahe County and typical grassland by using the decision tree method. Overprinting comparison between the classification results and the reference data were made to obtain classification accuracy. In order to better analyze temporal and spatial characteristics of the grassland resources in the growing period, equidistant graduation on the MODIS-NDVI were made to prepare for dynamic monitoring of the next step.Select the training sample area from the preprocessed TM / HJ satellite data, analyze the separable degree, and choose the maximum likelihood classifier to make the supervised classification. Extract the grass land types and obtain the NDVI index by logic operation from the classification results data after accuracy validation. Classify by natural inflection point method and extract the information of the grass grade distribution based on the frequency distribution information.The results show that after combination of multi-temporal remote sensing data, the accuracy of decision tree classification on grassland information extraction of MODIS data was 83.37%. This method provides a new method for remote sensing study on grass resource based on MODIS data. Accuracy of maximum likelihood supervised classification on the remote sensing images with the 30m resolution was 88.36% and 87.18% respectively.(3)Dynamic monitoring of grassland resourcesMake statistics on the MODIS data of grass resource in two sides that are the area of each level and each period and the mean value of NDVI in the different stages and different grass types. From the perspective of the mean value of grassland resources NDVI, the variation and trend of NDVI was consistent in various grassland types, which was first increased and then decreased. The index peak happened in the summer. The peak value of mountain meadow type and temperate meadow steppe type happened from the end of June to the early July. The peak value of temperate steppe type and temperate desert steppe type happened from the end of May to the early June. The peak value of alpine meadow type and lowland meadow type happened from the end of July to the early August. From the perspective of the grass area level, the period of high coverage grassland happened from the end of July to the early August. This three areas which NDVI> 0.6 were 1125.87 km~2, 1507.97 km~2 and 1418.96 km~2, significantly higher than in other months.Analyze grass resource information of TM / HJ satellite data dynamically. Compute the area of grassland of 2000 and 2010. Analyze the transfer matrix and the transition situation around each land classes. Count the area of grass resource in each grade. Calculate two NDVI data grid to get the information of the dynamic changes. Results showed that: the area of the grassland from 2000 to 2010 reduced by 407.86 km~2, of which winter pastures reduced by 142.10 km~2, summer pastures reduced by 205.80 km~2.Desirable agricultural land area increased by 183.75 km~2, residential land increased by 3.27 km~2. Desert area of further increase, reaching to 852.59 km~2, increased to 149.45 km~2. Because of the reuse of Spring Ranch, pastures were overstocked and grassland degradation was serious. Outside of winter pastures, resident increased and human activities exacerbated. This led the grassland degradation.Habahe County located in the Altai Mountain. It is mountainous and the spatial distribution of grassland resources affected by the altitude significantly. Divide the DEM equal interval at 100 m. Count the average grassland index of each grade. The results showed that: the NDVI index and DEM have a certain correlation in grass growing season. The NDVI index was stable in the plain area. Affected by water restrictions, the piedmont region of 700-800 m altitude produced a decline in the inflection point. With the rise of high and increasing precipitation, the NDVI index and DEM were positively correlated in the hilly piedmont region of 800-1800 m altitude. The NDVI index was negatively correlated with the DEM above1800 m, limited by temperature. Overall, except the fitting degree of 0.8977 from the end of May to early June, the remaining time is above 0.90, NDVI was correlated with the DEM. Analyze the mean NDVI in grasslands of DEM level quantitatively.The results showed that the index in 2000 was significantly higher than in 2010. This indicated that the quality of grass tend to downward.Taking Habahe County as an example , this paper adopted multi-source remote sensing data, making full use of the advantages of grassland remote sensing. Integrated knowledge of GIS and mathematical statistics, the paper extracted grassland resource information and analyzed the spatial characteristics during growing season in 2007 and inter-annual of 2000 to 2010. The study provides a feasible technology for extraction and spatial dynamic analysis of the grassland resource information. This study can present temporal distribution conditions of grassland resource information timely and accurately. It is of important reference value and significance for reasonable utilization of grassland, insurance of a steady livestock production development and improvements of ecological environment.