University of Twente Student Theses

Login

Improving the classification of different Land and Forest cover types on remotely sensed imagery to support REDD+: A comparative study in the Afram Headwaters Forest Reserve, Ghana

Torgbor, Adjah Benjamin (2012) Improving the classification of different Land and Forest cover types on remotely sensed imagery to support REDD+: A comparative study in the Afram Headwaters Forest Reserve, Ghana.

[img] PDF
15MB
Abstract:Tropical deforestation is an issue of global concern and, in recent times, has been noted to be a major driver of climate change due to the release of CO2 into the atmosphere through the activities of man. There are efforts by the international community to mitigate climate change. One such intervention is the REDD+ programme as proposed under the Kyoto Protocol of the UNFCCC. Ghana ratified this convention on 26th November, 2002 in order to reduce emissions from deforestation and forest degradation and also qualify for financial incentives by way of the so-called “Carbon Credits”. Important criteria for this are the regular monitoring, verification and reporting on the changes in the landcover of a participating nation. This therefore requires an accurate landcover map of the areas earmarked for the REDD+ programme. However, adequate high resolution multi-temporal optical data is often not available in the tropics where cloud cover is inevitable. In this situation, SAR provides a useful alternative. This study seeks to develop a suitable method that will improve the classification accuracy of different land and forest cover types by testing the MLC and OBIA techniques on ASTER and SAR data separately and in combination, to support the REDD+ programme in Ghana. A comparative study was conducted in the Afram Headwaters Forest Reserve in the Ashanti Region of Ghana. Previous studies in the area applied only the MLC algorithm on the above datasets and did not attempt to classify the datasets separately and in combination to compare the results either spatially or statistically or both. When compared spatially, the ASTER alone and combined ASTER+SAR maps showed over 70% agreement between MLC and OBIA. The SAR alone map showed 64% agreement. Statistically, the hypothesis that, there is no significant difference in the classification results (Kappa) between MLC and OBIA was confirmed. Furthermore, there is no significant difference between the kappa of ASTER alone and combined ASTER+SAR maps for both MLC and OBIA. There is, however, a significant difference between the MLC and OBIA classified maps of ASTER /SAR alone and SAR alone/combined ASTER+SAR. The overall best landcover map of the Afram Headwaters Forest Reserve (AHFR) was produced from the MLC of the combined ASTER+SAR data with an accuracy of 82.09% and a Kappa of 0.74. Five classes (Natural Forest (NF), Plantation (P), Agroforestry (AF), Settlement/Bareground (S) and Fallow/Grassland (FG)) and four classes (NFP, AF, S and FG) were identified on the ASTER and SAR data respectively. There is a potential for image combination to improve the classification accuracy of SAR. After the combination, the accuracy of SAR improved by 14% and 21% by applying MLC and OBIA respectively.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/93611
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page