Biophysical Spectral Response Modeling Approach for Forest Density Stratification
4. Results and discussion
Satellite remote sensing data have been used to identify vegetation cover and their density (Wu et al. 1995; Roy et al. 1990; McCloy and Hall 1991). The biophysical indices can reduce the effect of bias and assist in the extraction of significant features of a specified ground object (Curran, 1980). These indices also help in categorizing 'mixed pixel' effectively (McCloy and Hall 1991). The present study proposes a set of vegetation indices using linear combinations to isolate vegetation cover characteristics which vary with canopy density. The present approach has been found to be fairly robust and able to provide acceptable accuracy.
4.1 Test area in Evergreen forests
The spectral characteristics of different forest types were analyzed from thirty different sample plots. The average spectral response clearly indicates that the signatures markedly vary. The classifications of forest type and density have been based on these signatures. AVI, BI and SI have been used for preparing combination images and classification. The colour composites using Landsat TM bands and index have been used to highlight forest and density.
Majority of the forests in the study area have canopy closure of 40% to 80%. The density levels could be separated using the present methodology. These density classes are forest> 80%, forest 60-80%, forest 40-60%, mangrove>80%, plantation/regrowth and non-forest/agriculture. The vegetation type classification preformed by using hybrid (unsupervised and supervised) approach has been combined using logical operator (Fig1). The classification accuracy has been evaluated using stratified random sampling approach. The sample point(n=68) have been distributed in the different forest types (inclusive of density). The results (at 95% confidence limit) indicate that the overall mapping accuracy has been 91.4% and 88.2%. Highest accuracy has been observed in the case of evergreen forests of >80% density. Lowest accuracy has been observed in the case of mangrove forest on old tidal flats.
Fig. 1 Classification using Biophysical Spectral Response Model
4.2 Test area in dry deciduous forests
The present approach provided five forest density classes (>80%, 60-80%, 40-60%,20-40% and10-20%) and physiognomic classes like scrub/shrub and grassland. The hierarchical approach of identifying different Landover features and forest density was compiled into seta of rules, which are implemented for the classification results have been compared with permanent sample points (0.1 ha) marked on forest thematic maps. For evaluating classification two maps used are: (i) Landover and forest density classified map, (ii) broad landover classified map (without forest density stratification). The overall accuracy of the generalized landover map has been observed as 91.5%. The accuracy of first three density categories (viz.>80%, 60-80% and 40-60%) has been found to range between 93- 95%. The accuracy of the remaining density classes (viz, 20-40% and 10-20%) ranged between 82-85%. The overall classification accuracy of the map showing landover type and forest density is 89.4%.
The approach has been, so far validated in the dry deciduous forests and evergreen forest of Andaman only. Its extension in other biolimatic conditions need to be tested before terming it a robust methodology. The method may also need suitable adjustments in different vegetation types and terrain conditions.
Reference
-
Anoaymous 1993, Rehabilitation of logged over Forests in Asia/Pacific region. Final Report of Sub Project II International Tropical Timber Organization- Japan Oversea Forestry Consultants Association, pp.1-78.
- Champion H.G. and Seth S.K.1968. A Revised Survey of Forest Types of India (New Delhi: Government of India Publication)
-
Mc Cloy, K.R. and Hall K.A, 1991. Mapping the density of woody vegetative cover
using Landsat MSS digital data; Int Remote Sensing 12 1877-1885.
- Rikimaru A1996. Lnadsat TM Data Processing Guide for Forest Canopy Density
Mapping and Monitioring Model; Project of PD32/93 Rev. 2(F) ITTO
-Japan Overseas Forestry Consultants Association, Pp.1,1-1, 8.
- Roy p.s., Diwakar P.G., Vohra T.P.S. and bahn s.k.1990. Forest Resource management
Using Indian Remote Sensing Satellite data; Asian -Pacific Remote Sensing J. 3
11- 22.
- Wu J.K., Zheng Q.F.W.T. 1985,A forest inventory using Landsat imagery in the
Mao-shan area of China;Int.J. Remote Sensing 61783-1795.
