| This study revealed the landscape spatial pattern of the comprehensive management in loess hilly -gully watershed, to supply the reference for further optimization of the watershed landscape. Also to explore the accurate efficient landscape map mapping approach. This study took the typical loess hilly and gully Yan'gou watershed as an example, the field delineation and the remote sensing imagery based mapping approaches were employed to get the landscape map of Yan'gou watershed. Based on the landscape maps, analyzed the status landscape pattern, landscape patch diversity and heterogeneity, landscape function types, land use suitability and the land use change during near the past 20 years.Determined the landscape pattern features, the land use characteristics of the landscape pattern and land use evolution trends, explored the object-based classification as the accurate and efficient approach to classify the loess hilly and gully watershed satellite imagery. The way of further optimize the landscape pattern of Yan'gou watershe was proposed.The main results were as follows:(1) The current landscape pattern of Yan'gou watershed. The total landscape patche was 2679 and the patch fractal dimension was 1.0855, indicating that the shape of the watershed landscape was simple and stability. The main landscape pattern were bush and forest, took 43.32% and 19.17% of the total watershed area respectively. Robur (Quercus liaotungensis Koidz.) forest was the only remaining natural woodland, located at the palm of the watershed. The average area of the grass patch was 0.0041 km2, and the large number of the grass patch played a major contribution for the landscape fragmentation in the watershed. The patch density of the farmland and the orchard were 51 and 25 respetively which played a buffer role for the watershed landscape fragmentation.(2) The dominant landscape type was ecological protection landscape. Its patches took more than 3/4 of the total landscape patches, and it took 70% of the watershed area. Economic production landscape patches scattered in the watershed, and its patches took 1/4 of the total patches and area took 1/3 of the total area. Living-service landscape type took about 3% of the watershed area. The area ratio of the three landscape function types, ecological protection, economic production and living-service landscape, was about 23:9:1, and the mean patch perimeter were about 663m, 1028m and 472m respectively, which means that the landscape patches of ecological protection and living-service were small and broken, while the economic production patches presents better unity under more human intervention.(3) The suitability of the overall landscape distribution in Yan'gou was relatively reasonable. The land-use suitability of Yan'gou was 0.76. The suitability of parts of Yan'gou area were: the highest land use suitability was the shaded aspect with value of 0.94. The land use on the semi-shady aspect was the most unreasonable with the suitability of 0.74; the suitability was 0.02 in the area which had altitude less than 900m, and the highest value in all the elevation levels was 0.998 for altitude of 900～1000m. When the altitude was more than 900m, the suitability of the elevation decreased with altitude increasing.(4) The main character of the changing in the past 20 years is significant increases occurred in bush, forest and orchard.Forest and the orchard nearly doubled the area, and bush increased by about 11%. Grass was significantly reduced the distribution area. The residential area increased by nearly 1 percentage point. The dominant landscape type of Yan'gou was bush in both the initial (2000) and late (2007) stages of the management. In 2000, the wild protection area was the dominant landscape type on the slopes with more than 25 degrees, however, bushes and woodlands were dominant in 2007.(5) The object-based classification could efficiently abstract the information in the imagery with high accuracy, it's the suitable approach for mapping the landscape map in hilly-gully area. The object-based classification results got the hightest accuracy among the approaches, the object-based classification approach got the highest classification accuracy was up to77.73%, Kappa value was 0.7144; Yan'gou land use/cover maps obtained from the field survey and the supervised classification approaches got relative lower classification accuracy and kappa values, which were 37.89 %, kappa 0.1935, and 59.38%, kappa 0.4937 respectively. The firld survey results had as many as 24 categories of land cover types, however, other classification approaches that employed in this study such as visual interpretation, supervised classification and object-based classifications all abstracted 7 landscape types which were bush, forest, grass, farmland, orchard, residential area and water.(6) The approaches affected the landscape patch and function differently. The total landscape patches of the results from the supervised classification was 10,447, which was about 4 times more than the results that was from visual interpretation (2679), and much more than the landscape patches that was from the field survey and the object-based classification (920 and 900 respectively), The area ratio of the three landscape function types which were ecological protection, economic production and living-service in the difference classification approach were: for the field survey, 21:10:1; the visual interpretation, 23:9:1; supervised classification, 8:7:4.6; object-based classifiction, 23:13:1.(7) Auxiliary data of DEM and the NDVI of Yan'gou watershed were employed to add to the object-based segmentation to improve the classification accuracy. The auxiliary data information affected the land use/cover in the watershed. The panchromatic imagery was weighted in the object-based segmentation and classification. With the land use/cover classification threshold values and suitable segmentation scales of 150, and the color/Shape value was 0.9/0.1, compactness/Smoothness value was 0.5/0.5, got the best segmentation results.(8) Different classification approaches affected the expansion of Yan'gou land use/cover. Results from the object-based classification and visual interpretation had smaller area differences between the land use/cover respectively. Area variance in supervised classification methods was mainly for the bush, farmland, residential area and water. Field survey was mainly for woodland patches with large area, orchard and grasslands. Whereas all of the land use types obtained from the four approaches had shown spatial area extension (less than 2%), all the methods had been separated the natural land use/cover types and there were no huge differences of the spatial area among the approaches. All the methods could classified the bush as the dominant land use/cover type in Yan'gou watershed.
|
PDF Full Text Request