首页>
外文学位
>Using Remotely Sensed Data and Decision Tree Classifiers to Determine if the Changes in Accra, Ghana are Concentrated in the Most Vulnerable Areas.
【24h】
Using Remotely Sensed Data and Decision Tree Classifiers to Determine if the Changes in Accra, Ghana are Concentrated in the Most Vulnerable Areas.
Accra is a rapidly changing West African city. A lack of reliable census data presents challenges in monitoring this change. Remotely-sensed imagery can be used to assess changes in population based on the presence of roads, buildings, and other human built, impervious surfaces. Decision tree classifiers assign land cover values based on input imagery and training sites. Decision trees were used to identify impervious surfaces from Quickbird imagery from 2002 and 2010. Over 200 classifications were first performed to determine what quantity and variety of training data and input imagery yielded the maximum classification accuracy. Gray level co-occurrence matrix texture and anisotropic texture, which measure the relationships between neighboring pixels, at several window sizes were tested as well as different vegetative indices. The bands selected as a result of these tests were used to classify both Quickbird images. Change in impervious surface coverage was then calculated at the enumeration area (EA) and neighborhood scale. Data from the 2000 Ghana census and principal component analysis were then used to measure vulnerability to health risks at both scales. Linear regression analysis was used to compare changes to the vulnerability scores.;Results indicate that the most accurate classifications were obtained from a combination of Quickbird spectral bands and vegetative indices. Large amounts of change were found in the outer areas of the city but little change was seen in inner city neighborhoods that had almost total impervious surface coverage in 2002. Vulnerability scores differed at the EA and neighborhood scales with EA level scores emphasizing immigration and living in the peri-urban fringe. At the neighborhood level, crowding was a major component of vulnerability. At both scales, poverty and lack of education were major components. Vulnerability was not found to be correlated with change in impervious surface coverage at either scale
展开▼
