Northern Inland Temperate Grasslands With The Grasslands Of Remote Sensing Classification

Grassland resource was one of the most important natural resources in our country. It was helpful to use grassland resource sufficient to grasp the number,quantity and distribution rule of grassland resource in our country effectively and timely. Achievement of the grassland type classification in the grassland resource research was the important technical means of the investigation of the situation of grassland resource. In this paper, taking temperate grassland zone in north land that range was the temperate zone except Northeast three Provinces and Xinjiang Province as research area to carry out the research of grassland types classification and information extraction. There are five typical grasslands of mountain meadow, temperate meadow steppe, temperate steppe, temperate desert steppe, lowland meadow in the research area. The main purpose of the research in this paper is to classify the five typical grasslands and extract grassland information by the way of the knowledge classifier and spectral angle mapper using MODIS data in 2005 as the main information resource combine with the multi-source data such as accumulated temperature data and precipitation data and to test and evaluate the precision of classification result.First, this paper discusses the variation situation of EVI of the main grassland types in the research area all over the year in order to know the growth period and main features of the different grassland types and get EVI range of the different grassland types during the growing season. Take the difference of the EVI range of the different grassland types as one of the main basis of building grassland classification model after comparing with the difference of the EVI range of the different grassland types. Analysis the distribution of moisture index of the different grassland types and spectral Characteristics of MNF image, and put them into the research of grassland classification. By quantitative analyzing the features of EVI, MNF1 eigenvalue and moisture index to build the rule of grassland extract information and expert classification model. It makes preliminary grassland classification and get a preliminary grassland classification result by the modules (Expert Classifier) of ERDAS software.The low-land meadow couldn't be extracted by the way of expert classifier because the low-land meadow belongs to the non-belt latent vegetable and has no the obviously feature to distinguish with the other grassland types all over the whole research area. Because the low-land meadow distributes border upon the other grassland types all over the research area, this paper choose the way to extract the information of the low-land meadow that classified the low-land meadow with the other grassland types on the little area that contained the low-land meadow and one of the other grassland types. It was to further realized the classification of the low-land meadow and the other grassland types by the way of expert classifier and spectral angle mapper through further analyzing the difference of the grassland types'characteristics on the basis of the division of grassland. The grassland classification result in the whole research area was made by merging all of the grassland classification results in little regions and the preliminary grassland classification result of the whole research area.Finally, this paper makes an accuracy assessment for the grassland classification result by the two ways of stratified random sample and simple random sample. The eventual results showed that the overall precision of the two way were 60.4% and 67.23%and the Kappa coefficient of the two ways were 0.611 and 0.636.The classification result was excellent and could show the distribution of the grassland types. It has very important meaning for the reasonable development and utilization of grassland resources that the result of the grassland types classification and information extraction could provide the main evidence of the grassland dynamic change detection.