Vegetation type mapping using decision tree classification(DTC)
MetadataShow full item record
Remote sensing provides area coverage, and aids the mapping and classification of land cover features, such as vegetation, soil, water and forests. However, with remote sensing using medium to coarse spatial resolution imagery, some difficulties are encountered in the definition of vegetation type classes based on their spectral responses alone, thus, it is a challenging task to use such images for mapping vegetation at species level. However, when these images are integrated with other ancillary data, it becomes possible to map vegetation at species level. This research endeavored to demonstrate the possibility of integrating spectral with ancillary data in mapping vegetation types of Zambia and, for that, ENVI Decision Tree classifier was used. The nature of the vegetation in an area is determined by a complex combination of effects related to climate, soils, history, fire and human influences. Therefore, this mapping method takes advantage of the relationship that these features, vegetation types, have with their environmental factors, such as soil type and elevation. The study focused on three vegetation types, namely Miombo, Mopane and Munga, and thus the factors that influence their spatial distribution were studied and identified. Based on the literature reviewed and the GIS desktop analysis, it was found that, for the area of study, Miombo had the following factors: Ferralsols as the dominant soil type where it thrives; at 900m-1600m elevations; Band4 reflectance of 0.0 - 1.0; NDVI values of -1.0 to 1.0; maximum Band4/Band3 ratio value of 16.235294; and EVI of -1.0 to 1.0. While for Mopane: Luvisols as the dominant soil types where it thrives; at 700m- 900m elevations; Band4 reflectance of 0.0 - 0.5198; NDVI value of -1.0 to 0.857818; maximum Band4/Band3 ratio value of 12.707866; EVI of -1.0 to 1.0; and also significantly occurs in Agro-ecological zone I. For Munga: Luvisols phaezom as the dominant soil types where it thrives; at 580m-1320m elevations; Band4 reflectance of 0.0 - 0.7249; NDVI value of - 0.536278 to 0.872132; maximum Band4/Band3 ratio value of 14.641149; and EVI of -1.0 to 1.0. These parameters were used to develop the decision tree classifier binary rules and executed for the final produced map of the three vegetation types. With the decision tree map produced, the study demonstrated the possibility of mapping vegetation at sub-nation level by combining spectral response with other geographic parameters via the use of a decision tree classifier.
University of Zambia
Master of Engineering in GeoInformatics and Geodesy