Land Use/Land Cover Classification And Change Detection In Mountainous Areas Using Multi-resolution Remotely Sensed Data

Land use/land cover change is a key content and a hotspot issue of global environmental change and sustainable development research currently. But in extremely complex mountainous terrain characterized by high heterogeneity, it's very difficult to get accurate land surface information. Doing some research on the adaptability of remote sensing images with different resolution for mountainous terrain, is vital to the improvement of remote sensing classification accuracy in mountainous areas and understanding the temporal and spatial variation of land use/land cover in Manwan.This work takes Manwan as a study case. Landsat TM acquired in 2009 and Landsat OLI acquired in 2014, ALOS acquired in 2009 and ZY3 acquired in 2013, DEM together with field measurements with geolocation, are employed to land use/land cover classification in a SVM classifier integrating spectral, textural, and topographic data Then, evaluate the classification accuracy with confusion matrix and analyze the land use structure. At last, analyze the dynamic change of land use/land cover using change detection method of comparison after classification.The results show that:(1) The overall accuracy of land use/land cover classification of Landsat-5 TM, Landsat-8 OLI, ALOS and ZY3 images is 93.00%,90.27%,89.23% and 84.06%, respectively with a the Kappa coefficient of 0.92,0.88,0.87 and 0.81. The classification accuracy of Landsat is better than that of ALOS/ZY3. Landsat images perform better in land use/land cover classification and change detection. Totally, for each land use type, the classification accuracy of water is the best, followed by construction land and forest, whereas farmland and garden land the last. (2) During the study period, the land use change is characterized by the decrease of forest land, the increase of garden land and construction land. The area of decreased forest land is about 100km~2 and it's mostly converted to garden land and farmland. The area of construction land is in a large increase. The contribution rate of agricultural land to the growth of construction land is more than 39% and surpasses that of construction land itself. The area of increased garden land is about 96 km~2, except itself, with the biggest contribution rate of 17.10% from farmland. That's as a result of encouraging the mixed planting of crops and forests and planting economic forests of tea and walnut tree on the farmland. Some farmland was converted to garden land and the area of garden land increased. (3) Each land use type is under the influence of altitude, slope and aspect. Broad-leaved forest lives at higher altitude interval than coniferous forest. Broad-leaved forest lives at the altitude of 1600 to 2860m, while coniferous forest lives at the altitude of 1000 to 2200m. Broad-leaved forest and coniferous forest mainly live in regions of steep slope, sharp slope and shady slope. Garden land is distributed at the altitude higher than 1600m, while farmland lower than 1800m. Garden land and farmland are mainly distributed in regions of steep slope and sunny slope. Construction land is evenly distributed at the altitude lower than 2000m, decreasesing gradually with higher slope. Water is almost distributed at the altitude lower than 1000m. (4) High spectral resolution will help to improve classification accuracy of forest land, garden land and farmland, meanwhile high spatial resolution can better identify surface features with simple spectrum, like water. The using of topographic data as an additional auxiliary band for classifier, can improve the accuracy of image classification. Creating interpretation signs at different slopes for broadleaf forest and coniferous forest and drawing samples of forest at different slopes, can improve the classification accuracy of forest land.