The selection of wavelength region and number of bands is a research problem for remote sensing experts for utilization of data provided by the sensor system. The present study proposes to make an evaluation for optimum band selection and classification accuracy.
The entropy, brightness value overlap index (BVOI), optimum index factor (OIF) and spectral separability analysis, i.e. Euclidean distance (ED), divergence, transformed divergence (TD) and Jefferies‐Matusita (JM) distance and accuracy of MLC classification were carried out. For the present study Terra ASTER, Landsat ETM+ and IRS 1D LISS III dataset has been used. The first three methods were for the spectral evaluation of the three satellite data used and for determination of information content, variance and spectral overlap among the classes present in the natural and man‐made landscape. The fourth method is for selection of spectral band combinations with highest separability of classes using divergence matrices. These band combinations are selected for the classification and subsequent accuracy assessment.
The OIF values are clearly indicating that the performance of ASTER data is the best, having the lowest correlation between the bands; hence the separability of the feature is also highest, while LISS III have shown high correlation between the bands, with the poor separability of the features. Landsat ETM+ data are in between these two sensors, better than LISS III but poorer than ASTER. The BVOI outputs of the three datasets of man‐made landscape show that band 3 of ASTER has the least overlap of the classes, followed by band 4 of ETM+. Very high overlap of the classes has been found in LISS III data. It has been found from spectral separability analysis of all the three datasets for the man‐made landscape that ASTER data with band combination of spectral bands 123468 contains the highest value of all the measures of spectral separability, i.e. ED (291.72), divergence (2,133.37), TD (2,000.00) and JM distance (1,414.10).
It can be inferred from the present study that spectral resolution plays a very important role in discrimination of vegetation features. ASTER data which are with the highest number of the bands amongst the satellite data used had shown highest classification accuracy, while LISS III data with lowest number of bands had shown lowest accuracy, and Landsat ETM+ stood in between the two sensors.
It is important to evaluate the sensor systems and their spectral regions for discrimination of vegetation features. The number of bands present in a particular sensor and the spectral regions used in it are some of the crucial factors which decide the usefulness of the data for different applications, including vegetation‐related studies. The selection of spectral wavelength region, i.e. spectral bands and the sensor system, presents the research problem for remote sensing experts to suggest the best spectral regions and satellite sensor for the discrimination of the vegetation features in different landscapes, namely man‐made and natural.
In the present study all the three datasets are extensively examined and tested for their vegetation discrimination capabilities using well‐established methodologies. All the parameters applied on the datasets revealed that spectral resolution definitely plays a role in the performance of the data as far as discrimination of features is concerned both in natural and man‐made landscape with desirable accuracy.
Joshi, P., Gupta, B. and Roy, P. (2008), "Spectral evaluation of vegetation features using multi‐satellite sensor system (Terra ASTER, Landsat ETM+ and IRS 1D LISS III) in man‐made and natural landscape", Sensor Review, Vol. 28 No. 1, pp. 52-61. https://doi.org/10.1108/02602280810850035Download as .RIS
Emerald Group Publishing Limited
Copyright © 2008, Emerald Group Publishing Limited