Abstracts Geography

Potential of the kNN method for estimation and monitoring off-reserve forest resources in Ghana

by Christian Kutzer

Continuously increasing population and concomitant demand for food have been devastating forests by exploitation and shifting cultivation at an alarming rate. With the awareness the exhaustion of fossil fuels and recent price peaks for crude oil, the tendency towards forest destruction has been accelerated with the transformation of forest lands into oil palm plantations. In the same way, non-timber forest products like rattan, bamboo or raphia palms are disappearing. Local populations are more and more dependent on the illegal gathering of rattan from protected forest reserves and chainsaw lumbering practices in Ghana’s off-reserve forests – prohibited since 1997 – serve as a source of livelihood for a good number of Ghanaians. The monitoring of forest lands is vital to enable the sustainable management and development of the area. The mission of the internationally acting organisation Tropenbos, where this study is imbedded, is to generate scientific input for sustainable management of natural resources of tropical countries. The application of the k nearest neighbour (kNN) method in the combination of terrestrial data with remotely sensed data for forest attribute estimation and mapping has become an integral part of forest inventory methods. The object of this study is to assess the potential of the kNN method for the development of a cost efficient monitoring system of specific non-timber forest products, forest resources, and different land use types, in the off-reserve forests of the Goaso forest district in Ghana. For this purpose ten land use types and land resources, were identified, these being the following; bamboo, banana plantations, bush fallow, cocoa plantations, elephant grass, grassy vegetation, herbaceous vegetation, oil palm plantations, raphia palms, and forests. Based on selectively chosen sample circles with variable diameters, 3360 pixels were distributed within the off-reserve forests, covering an area of about 1710 km2. For the classification assessment, the registrations were each divided respectively into a collective of training and test pixels. The method was applied to ASTER data and out of it mathematically generated indices were deduced to identify how various spectral band combinations, sample sizes, and sample distributions contribute to the overall accuracy of the various land use types. The kNN classification achieved overall accuracies ranging from 75 % for plantain, 78 % for tree resources and raphia palms, 81 % for cocoa plantations and bamboo, up to 83 % for oil palm plantations. With increasing sample sizes or simulation of a particular sample plot distributions, the results could be improved. Likewise, selection of optimal parameters for the kNN programme settings were found. K-values between 3 and 6 showed the highest values for the land resource bamboo. For a multi-attribute classification of the ten land use types, overall accuracy and overall agreement in the classification could be increased by limiting the occurrence probability from the results of…