Surface Reflectance And Discriminant Analysis For Mapping Of Mangrove Species In Kuala Sepetang Mangrove Forest Reserve, Perak
Loading...
Date
2016-02
Authors
Beh, Boon Chun
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
The identification of mangrove species by using traditional approach has
been commonly developed and studied. Due to the high cost of field work and the
difficulty in assessing the mudflat areas of mangrove ecosystem with conventional
methods, remote sensing techniques have been widely used to examine mangrove
species at either the scale level or leaf level. In this study, two remote sensing
methods have been utilized to identify and discriminate the mangrove species in the
Kuala Sepetang Mangrove Forest Reserve, Perak, Malaysia. The first method is to
develop an algorithm that was based on species’ surface reflectance to map the
mangrove species by using high-resolution CropCam Unmanned Aerial Vehicle
(UAV) data. The airborne images were acquired through the three band channels
(blue, green and near-infrared) of the Normalized Difference Vegetation Index
(NDVI) camera that was mounted on the CropCam UAV. The Top of Atmosphere
(TOA) reflectance was retrieved from the mosaicked airborne image, which covered
the entire study site with an area of 50.12 ha. The distributions of five mangrove
species were successfully mapped by using the retrieved reflectance values. The
scatter plot between the predictions and ground pixels revealed a high correlation
(R2=0.873) with Root Mean Square Error (RMSE) of 0.476 (less than one species per
pixel). The results also indicated that the developed technique was reliable and
produced good results with high accuracy of 85%. The second method used
statistical analysis to analyze the hyperspectral reflectance data of the mangrove
species. Analysis of Variance (ANOVA) and Linear Discriminant Analysis (LDA) tests were applied on the reflectance spectra data. The LDA determined the
influential wavelength, which could be used to distinguish the leaf samples among
the six mangrove species. Twenty-six significant wavelengths (p<0.05) were
obtained in the Very Near Infrared (VNIR), Short Wavelength Infrared I (SWIR I)
and Short Wavelength Infrared II (SWIR II) spectral regions. Sixteen discriminant
functions were generated by using the 26 influential wavelengths. The score range of
each mangrove species in discriminant functions was determined by using the
reflectance spectra. Meanwhile, different combinations of discriminant functions
were compared to determine the most suitable function to classify the mangrove
species. The comparison produced the best accuracy when 11 functions were chosen
for the mangrove species classification. Therefore, the score range that was
established earlier was used to predict the unknown mangrove leaves by using the 11
discriminant functions. The final results showed that even the attained classification
accuracy was lower when identifying certain mangrove species, but the 11
discriminant functions could still determine the correct mangrove species of an
unknown mangrove leaf sample from the study area at the leaf scale. Overall, these
results clearly indicate that the two remote sensing methods that were applied could
successfully discriminate five mangrove species in the Kuala Sepetang Mangrove
Forest Reserve, Perak, Malaysia and accomplished the objective of this research.
Description
Keywords
Mangrove