Remote Sensing Identification And Evaluation Of Diversity In The Xilin Gol Grassland

Grassland is an important green cover layer of Earth’s land surface just behind the forest cover layer, and as a kind of natural resources which plays an important role and be of great value in livestock carrying, water conservation, biodiversity maintenance, soil and water conservation, ecological balance, climate regulating, leisure and recreation, and nutrient cycling.Grassland resources investigation and monitoring has long been a matter of concern for lawn worker, besides grass recognition and grassland diversity evaluation are the basis of grassland cover condition monitoring and grassland biodiversity protection.Traditional monitoring methods are mainly rely on grass costly, time-consuming, labor-intensive field survey methods, while remote sensing, with rapid development, has become an important means for grassland resources investigation and monitoring in recent years. However, thre are still many probems in terms of the current research situation:For the point of view of grassland recognition, current research stays in the spectral characteristics analysis and comparison, lacking of in-depth research. Many studies built classification tree according to a simple spectral features analysis, which process is cumbersome, universal applicability is poor. In addition, grasses recognition research basing on neural network is less;In the view of grassland diversity research, the traditional monitoring method are mainly base on numerous field survey or aerial photo interpretation, which are time-consuming and laborious, poor update, are not suitable for large area monitoring.Basing on current research analysis and aiming at the existing research problems, the Xilin Gol grassland resources were studied in Inner Mongolia, taking grass recognition and grassland diversity evaluation as the two basic routes. Grass spectral data was sampled in2008summer. First of all, taking spectral characteristics extraction and grass identification for research purposes, taking four common grasses of Xilin Gol Grassland:Chinensis(CH), Stipa(ST), Carex pediformis(CP), Cleistogenes(CL), for the study objects, basing on the analysis of current researches, similar spectral characteristics and distinguish difficulty of the four type grasses, this study used spectral differential technology combined with neural network model powerful data mining capabilities, designed a new type of hyperspectral grass identification process, completed identification for the four categories of grass, and provided a new technology process for grassland classification study. Secondly, method and theory of GIS, remote sensing and landscape ecology were combined to evaluate and compare grassland biodiversity situation of the whole study area and each counties. The main conclusions of this study are as follows:Conclusions on the grass type recognition are as follows:(1) Average spectrum of4grasslands show that,4grasslands have have a similar spectral curve, the spectral reflectance of Chinensis is high than the other three on the whole, especially in the800nm-1300nm and350nm-500nm. Carex pediformis has the lowest spectral reflectance among350nm-500nm, highest spectral reflectance among1100nm-1300nm. The Stipa and Cleistogenes spectrum are very close, especially in visible bands. (2)After the first-order differential of the spectral data was processed,7characteristic parameters, red edge position, red edge slope, red edge area, green peak position, green peak value, red edge position, red valley value, were extracted. Overall, variation coefficient of the red edge, green peaks and valleys position is small, while variation coefficient of red side slope, red edge area, green peak value and red valley value is relatively large, and the variation coefficient of the red valley is greatest among them, especially Stipa variation coefficient is35.98%.correlation analysis and principal component analysis demonstrate that red edge slope and red edge area have a strong correlation, and for the characteristic variables constituting the former three principal components, the red edge position has a small contribution, so in ordering to reduce the data dimension, subsequent analysis will no longer consider red edge area, red valley value and red edge position, only taking red edge slope, green peak value, green peak position and red valley position as neural network model inputs to classify grass species.(3)The study found that neural network model, as the effective means of data mining, can identify different grassland’spectrum, and then identified the different grassland type, the overall precision of training and testing are high and stable, the testing precision is83.30%.Conclusions on the grass diversity evaluation are as follows:(1) According to remote sensing interpret data, GIS method and landscape ecology principle, a total of42kinds of landscape type in the study area were extracted, including16kinds of grassland landscape type, respectively were Meadow steppe, Sparse grass prairie, Drought grassland, Other grassland, Short grass with short semi-subshrub desert grassland, Low wetland vegetation grassland, Grasses with semi-subshrub desert grassland, Bunch grass with rhizome grass typical prairie, Herbal wetland, Desertification grassland, Sandy vegetation grassland, Sandy desert grassland, Grass with semi-shrub grassland, Weedy grasslands at forest edge, Grasses with weeds meadow steppe, grasscluster, Herbaceous green space.(2) Bunch grass with rhizome grass typical prairie is the biggest one of the grassland landscape types, its area is about912912ha, accounting for45.37%of the study area, and it is the dominant landscape for the study area, mainly distributed in the central,north and south area.(3) There are26577grassland patches in Xilin gol, average patch area is756.64ha, on average level there is0.13patch in every100ha; The Largest patch block is Short grass with short semi-subshrub desert grassland, accounting for30.58%of the study area. Landscape richness index, landscape richness density index, Shannon diversity index, Shannon evenness index, landscape dominance index, landscape aggregation index are42,0.0002,2.01,0.54,1.7,63.22.(4) ErLianHot and SuNiTeYouQi have the most abundant grassland types with12types, ZhengXiangBaiQi and ZhengLanQi have the least grassland types, five respectively, and the other counties are in the average level,ablout10; Bunch grass with rhizome grass typical prairie is the domainant grassland type within six counties, followed by Short grass with short semi-subshrub desert grassland, occupying ErLianHot, SuNiTeYouQi and SuNiTeZuoQi with75%60%60%of grassland area, and drought grassland respectively occupied DuoLunXian and TaiPuSiQi with37%and57%of the lawn area, besides Sandy vegetation grassland is the domainant grassland landscape of ZhengLanQi with42%of the lawn area.(5) ZhengLanQi has the largest patch number, is40750, the largest average patch area is in the ABaGaQi with271ha; The largest block exists in WuZhuMuQinQi for68ha, the smallest block exists in DuoLunXian with the size of7ha; The biggest landscape shape index is170in the ZhengLanQi, with the strongest plaque effect, the smallest landscape shape index is in ErLianHot; ZhengLanQi has the biggest patch density,4.008; ABaGaQi has the biggest Landscape richness index; DuoLunXian has the biggest Landscape diversity index, ZhengLanQi has the smaller one, are2.27022.1145; DuoLunXian has the biggest Landscape evenness index, ZhengLanQi has the smaller one, are0.6888、0.6509; XiLinHote has the strongest landscape dominant degree,2.2612; The strongest landscape aggregation degree is in DongWuZhuMuQinQi, the weakest is in ABaGaQi,83and60respectively; DuoLunXian has the largest landscape diversity index and evenness index.In summary, this research provide a scientific basis for the study of high spectral information extraction of grassland resources, grass type classification, grassland diversity evaluation and satellite remote sensing monitoring of grassland resources. Furthermore, in the grass type recognition study, the7extracted characteristic parameters seven are commonly used, follow-up studies will further explore new characteristic parameters, and considering the poor stability and long-time studying of neural network, future study will further optimize neural network model parameters to improve accuracy and stability of grass type recognition; In the grassland diversity study, although high spatial resolution remote sensing data was applied, the grassland classification only reached the landscape level due to the remote sensing data source restrictions, not contain more specific grass species information, therefore, future research can making further efforts to improve rassland classification level in terms of data sources. In addition, follow-up study also can focuses on the dynamic change characteristics of grassland diversity spatial distribution pattern.