FOREST CROWN CLOSURE ASSESSMENT & TREE SPECIES CLASSIFICATION USING MULTISPECTRAL IMAGERY

Abstract

Forest inventories have traditionally been used for acquiring quantitative and qualitative information of forests in Pakistan. With the advent of technology, remote sensing has been found to provide an alternative for forest mapping and monitoring in less time and low cost. This research involves the study of forest crown closure, type of tree species and their spatial distribution in Ayubia National Park with SPOT 5 (2.5m spatial resolution) & ALI (30m spatial resolution) imageries based on the collaboration of Geographic Information System and Remote Sensing. In this study the optical properties of tree species were used to classify the study area into different vegetation categories. Being different and unique for each species, optical properties can be used as a source of species identification. Species classification was performed using various supervised classification methods, both on high spectral (ALI) and spatial (SPOT) resolution I multispectral space bome imageries

Pre-processing and normalization of bands was carried out for SPOT and ALL, prior to the application of Indices and Principal Component Analyus (PCA) for forest crown closure. Raw data was used for estimating the spatial distribution of various tree species in Ayubia National Park Advanced Vegetation Index, Bare Soil Index, Shadow Index and Scaled Shadow Index were used to assess forest crown clonure for each pixel of SPOT and ALI imageries. The results from SPOT showed that crown cloure generally fell between 20% and 65%, in the stady area The analysas further releaved that maximum ares bed under crown closure range of 35 to 63% Sumalarly, results from ALI umagery showed that crown cloure. generally fell between 40% and 60% Majority of the arra hed under classes ranging in percentage om 45 to 63% (30.30 sq. km) with a maximum at 50 to 53%

Movreves, the results showed that the greatest value for crown cloure was 79%, whereas, the maximum area bed under crown cloure 45 to 63%. The results obtined from SPOT magery were nave precise giving details of lower and upper crown cloure classes, which were not evident in the results of ALI imagery, the reason being that SPOT imagery has high spatial resolution of 2.5 m, wheress ALI has spatial resolution of 30m, and thin was unable to capture intricate details Though ALL imagery has high spectral resolution as compared to SPOT magery. yet the greater spectral resolution alone was not suffice for the study. Thaan, it is concluded that unless the spatial resolution of satellite imagery is not high, it will give generalized results for CC, even if it had high spectral resolution. The remits of CC derved from SPOT and ALI were father venfied through securacy assessment. It was found that the overall accuracy of CC results obtained from SPOT was 75%, whereas from ALL, the overall accuracy wat 58%Species clasufication was performed uting various methods on SPOT and All imageries. Among the various methods, Decision Tree Classification method was found to be the most natable for the shaly ares, while using both SPOT and ALI imagery. The percentage of comferous and broadlerved tree species was about 91% and 9% through SPOT imagery, and 88% and 12% through ALL imagery of the total vegetation cover. Decimon Tree Analyus (DTA) method performed on both SPOT and ALI showed highest aren occupancy by Bharpши пресия. Ассигису esment cactied out for all the methods of tree species classification showed. overall accuracy of 78% for SPOT and 72% for ALL DTA showed best results, nevertheless, a number of pixels were left unclassified that did not meet any criteria given since DTA is a resultant of a miltistage classifier that involves series of binary decisions. ALI imagery did not give satisfactory results through any method fixcept DTA.

While comparing the results of SPOT and ALI, SPOT imagery gave much better results for species classification than ALL FMinD, MinD, PMinD and Hybrid classifiers gave unsatisfactory results for both SPOT and ALI imagery Results for SPOT were, however, more accurate in case of DTA and Hybrid methods compared to ALI imagery, although ALI showed higher accuracy through results of PMinD. FMinD and MinD classifications.

It is concluded that the best suited approach would be to have a high spatial resolution imagery coupled with a better spectral resolution for any meaningful classification of forest tree species, to be used for the management of this resource that is so scarce in our country.