Digital processing of landsat thematic mapper satellite data for land cover classification using statistical methods for accuracy assessment: a case study of Ngerengere watershed
Loading...
Date
1998
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Sokoine University of Agriculture
Abstract
This study involved primarily digital image processing and interpretation
land for classification . cover Part of Ngerengere watershed covering about 300 km2 was taken as a case study Land cover which relates to the type
of feature present on the surface of the Earth of the study area
was elaborated. The procedures involved were reconnaissance survey to acquire 'a priori' knowledge about the u n s u pe r v i s e d area, classi f icat ion, field
work classification, for rectification and restoration, ground supervised
truthing, image
ultimately producing
and a land cover map of Ngerengere watershed. ERDAS 7.5 was the
remote sensing (RS) software applied. The
elaborated accuracy, land using cover map statistical was assessed packages.
foric.s Overall classification accuracy (OCA) was about 70% which was below
the recommended minimum level of interpretation accuracy
value of 85%. It was however, revealed that the low OCA was
substantially attributed by low producer's accuracy for the
class of built-up areas which was only 22%. Although the
produced land cover map had relatively low OCA,
it had agood correlation with the present land use and vegetation.
Further statistical analysis, by use of KHAT (K) statistics
gave a value of 0.65 which is above 0.5 and approaches 1.0,
hence acceptance of the land cover map produced from thatiii
classification as primary data for land use planning. It
percentage of also indicated extent the which to correct values of an error matrix used are due to true agreement and not by chance agreement. Chi-square
(X2)distribution to test the goodness of fit of the land cover
map elaborated, concluded absence of significant difference
between the obtained and recommended OCA at significance level reject of
and hence absence of enough evidence to 0.01, the classification for various applications like proposal for land use planning. In order to come up with a sustainable land use planning of the area, category other data are required.
needed its own ancillary Every land cover data which are site
specific to come up with. a proposal or alternatives for
land use planning of an area. For instance, agricultural
lands and rangelands classes needed much more ancillary
data due to their spectral similarity and inseparability
characteristics revealed during this study.
Finally it is recommended for continuation of the study so
as to come up with a sustainable land use planning of an area because thereis
a problem already and timely
intervention is imperative operation for the betterment of
future generation.
Description
PhD Thesis
Keywords
Land Cover Classification, Digital Processing, Landsat Thematic Mapper Satellite Data, Ngerengere Watershed