Classification Of Rural Settlement By Contextual Information And Its Spatiotemporal Evolution

Multiple policy projects over the past years; such as land exploitation,consolidation and rehabilitation have changed the land-use and land-cover (LULC) in China's rural regions. The change is so great that it has affected rural settlements in amount, architectural form, composition, configuration and also spatial distribution characteristics. Therefore, it is crucial to monitor rural residential area changes, and insufficient supervision may lead to an enormous waste of land resources, severe food problem and degradation of natural ecosystems. The fine approaches for rural LULC classification are required to catch up with the rapid social development. The so called very high spatial resolution remote sensing data (VHR), including the domestic high resolution satellite data, such as Chinese Gao fen Satellites (GF-1 and GF-2) are regarded as tangible information providers in various application.Using GF-2 images as the main data, our study attempted to extract rural settlements accurately in typical rural areas of the northern Zhejiang river plain. a novel approach was presented to extract spatial contextual information by combining hierarchical multi-scale segmentation and landscape analysis. The spatial patterns and the spatial and temporal dynamics of rural settlements in the agricultural land were analysed over the past 10 years. The main research contents and conclusions were summarized as follows:(1) Land cover information in rural area was obtained accurately by using high resolution remote sensing imagery. The whole classification framework was based on the object-based image analysis (OBIA) and multi-scale segmentation. In order to obtain the optimal scale parameters in the multi-resolution segmentation (MRS), the local variance of each object was used as an index to evaluate the segmentation results.Thus, the abundant spatial information in high resolution imagery was utilized effectively. combined with the feature selection methods based on support vector machine classifier (SVM), and parameter optimization, an effective classification framework was established for the extraction and identification of LULC information in rural areas. The accuracy assessment results showed that the precision was beyond 80% for the artificial land cover types, and the overall accuracy was more than 87%.Thus, this result indicated the precision of LULC information extraction in rural area,and provided basic data for distinguishing different rural settlement types and monitoring spatiotemporal dynamics.(2) with the development in the rural areas, various rural settlement types were formed. According to the rural resident classification system proposed in resent years,two distinctly different types of rural settlements spread in study area: the low density dispersed rural settlement and the high density clustered rural settlement. However,conventional approaches which only consider spectral, textural and geometrical information may encounter difficulties in identifying those two settlement categories,because their differences in spectrum, texture and geometry is not significant on single rooftops. This study attempted to discriminate different types of rural settlements by using spatial contextual information extraction method, which integrating hierarchical multi-scale segmentation and landscape analysis. According to the heterogeneous landscape characteristics of these two rural settlements types, a variety of landscape metircs was used to quantify landscape heterogeneity in different types of rural settlement. The contextual information derived was first utilized for discriminating the rooftops between rural settlements and rural enterprises.The dispersed rural settlement and the clustered rural settlement are not only different in building age, but also different in architectural style. Their differences in architectural style are exhibited by three aspects: composition, morphology, and location. In low density dispersed rural settlements, settlement units usually consist of housing, affiliated facilities (livestock pens and barns), gardens, woodlands, etc.Morphologically, dispersed settlement buildings are clustered and have various orientations. There is no identical interval between dispersed settlement buildings, and dispersed settlements are usually located close to rivers and streams for transport and water supply. In contrast, the units of clustered settlements usually encompass neatly arranged standardized buildings which are surrounded by less vegetation cover and these settlements are usually close to roads for easy transportation routes. Therefore,these two types of settlements have entirely different landscape characteristics.At first, based on the multiresolution segmentation algorithm, a finer and a coarser scale segmentation were conducted on GF-2 image respectively. A preliminary LULC map was derived by using only traditional spectral and geometrical features on the finer scale segments. Subsequently, a vertical connection was built between super-objects and sub-objects, landscape metrics information was calculated based on the preliminary LULC map. The vertical connection was utilized for assigning landscape contextual information to sub-objects. Eventually, a final classification phase was carried out in which only multi-scale contextual information was adopted to discriminate two types of rural settlements.Compared with previous studies on multi-scale contextual information, this paper employs landscape metrics to quantify contextual characteristics, rather than traditional spectral, textural, and topological relationship information, from superobjects. Our findings indicate that this approach effectively identified and discriminated two types of rural settlements, with accuracies over 80% for both producers and users. A comparison with a conventional top-down hierarchical classification scheme showed that this novel approach improved accuracy, precision, and recall. The result showed that landscape metrics generated at two-level segmentation could provide valuable information in image classification. Landscape metrics information could be used to characterize spatial contextual information of different settlement types. This approach made fully use of relationship information among segments in OBIA, and showed practicability, applicability and effectiveness in the entire procedure.(3) Based on the classification results on remote sensing image in 2005/2016,using models such as spatial autocorrelation, kernel density estimation, and standard deviation ellipse method from exploratory spatial data analysis and landscape metrics,the paper investigated the spatiotemporal dynamic and landscape pattern of rural settlements in Tongxiang city. The results are shown as follows.The total area of rural settlements reduced by 8.35% in the last 10 years, the dispersed settlements were significantly decreased and the clusterd settlements were growing rapidly. The distribution of two types of rural settlement had evident regional differences. And the increase and decrease in settlement area were also with regional differences. In these districts with negative growth in residential area, scattered settlements were removed, old rural communities were merged, integrated into new centralized communities. As a result, the residential area was reduced, land resouse was protected and conservedIn the aspect of the distribution of rural settlement, building density had high-to-low characteristic from the West to East, and the South to North. In some southern region, the residential density increased regionally. While in some northern part, some clustered settlement centers was formed. With the formation of these high density centers, residential density surrounding declined in different degrees. The changes in multiple landscape aggregation metrics also showed that residential areas had agglomerative effect in the past 10 years.In the hot spots analysis of rural settlement changing, we found two residential growth hot spots in the study region. Another hot spot of reducing settlement area was found in some northern villages. Some villages in the southwest formed a cold spots of settlement area reducing.Based on the case study of a typical region with many newly formed clustered settlements, the standard deviation ellipse of settlement shifted dramatically, the average center became closer to the center of the town, showing obvious agglomeration effect. However in some village in cold spots, the standard deviation ellipse change barely changed, and the average center was relatively far from the town centre.The temporal dynamics of rural resident area, can cause the change in LULC. from the statistical results, the paddy fields and gardens were two major sources for newly formed rural settlements, the removed residential land mainly had becamed nonirrigated farmlands and paddy fields. This phenomenon indicated that a large number of settlement area were used in land reclamation after demolition. Therefore,the residential area was reduced, and the cultivated area was increased at the same time,accomplishing the goal of protecting the red line of arable land.