Wildlife Institute of India
Dehra Dun, India
e-mail: [email protected]
The availability of high-resolution Indian satellite data from IRS P-6 LISS IV is expected to have a significant impact on the mapping potential of varied natural and man-made resources. The present study is aimed to evaluate the efficacy of IRS P-6 LISS IV in natural resource mapping for a wildlife protected area. A comparison of qualitative and quantitative assessments of IRS P-6 LISS IV data with Landsat ETM+ and IRS 1D LISS III data sets was attempted for a part of Katerniaghat Wildlife Sanctuary in India. The boundaries of polygon features, linear features (roads) and contrast with the surroundings were more precise on LISS IV data as compared to other data sets. It was found that 82% and 112% more forest roads could be delineated with LISS IV data compared to LISS III and ETM+ data sets, respectively. The level of vegetation classification and accuracy improved as well with LISS IV data compared to LISS III data. The study concluded that the LISS IV is better source of data for natural resource mapping over commonly used satellite imagery and can serve the diverse application needs of the protected area managers.
World over, natural areas continue to decline in extent and their quality, resulting into loss of biodiversity. These losses have been attributed to expanding human footprint on natural ecosystems. In the past four-five decades, the establishment of wildlife protected areas (PAs) has received global recognition as a key step towards conservation of biodiversity. Effective and scientific management of protected areas is not only the need of the present time for sustainability of remaining natural resources but it is challenging, particularly when such PAs are located amidst human dominated landscapes. One of the prerequisites for sustainable management of a protected area is the availability of fine detailed land use/ resource maps with high communication value, accuracy and details on geophysical and biological attributes, infrastructure (e.g. forest roads, firelines), plantations, encroachment, location and boundaries of small, ecologically sensitive habitats to the resource managers. The baseline information provides basic record of resources and is vital for long term monitoring and preparation of wildlife/protected area management plans. (Kent and Cooker, 1992; Welch et al., 2002; Mathur and Midha, 2008).
Advancements in remote sensing technology offer unparalleled technique to monitor urban, green, and non-green land changes at regional and local scales from space (Al-Bilbisi and Tateishi, 2002). According to Kerr and Ostrovsky (2003), it generates a remarkable array of ecologically valuable measurements which include the details of habitats (land cover classification) and their biophysical properties (integrated ecosystem measurement) as well as the capacity to detect natural and human induced changes within and across landscapes. Forestry sector is one of the main application areas wherein remote sensing technology has been used from the beginning of its emergence (Porwal and Roy, 1982; Roy et al., 1985; Porwal and Pant, 1989).
In addition, remote sensing technology has made an enormous progress in the last four decades and a variety of sensors now deliver medium and high resolution data on an operational basis (Hussin, 1999; Clark et al., 2004). New sensors have enabled evaluation of impacts of specific management policies on biodiversity (Innes and Koch, 1998; Mahito and Takeshi, 1998). The likelihood of ‘pure’ pixels being collected for specific land cover type increases with finer spatial resolution (O’hara, 2002). The current and upcoming high resolution satellite imagery is expected to have a significant impact on the mapping potential, its applications and especially on the updating of databases. Presently, India has an impressive array of high resolution remote sensing satellites meeting the national need for management of its diverse natural resources. IRS P-6, launched in October 2003, has opened up new dimensions of applications by virtue of its multispatial and multispectral capabilities. The LISS IV sensor on-board this satellite with high spatial resolution of 5.8 m and multispectral capability with three spectral bands in the green, red and near infrared regions has now replaced the IRS-1C/D LISS-III + PAN datasets. It can be tilted up to ± 26° in the across-track direction, thereby providing a revisit period of 5 days and 70 km 70 km stereo pairs. This has opened a new field of micro-level applications (Gupta and Jain, 2005).
Initial studies done on LISS IV data have demonstrated its applications in various fields with enhanced level of detailing. Kumar and Martha (2004) assessed the capability of LISS IV for landslide damage assessment in Uttarakashi region of Uttaranchal. The results revealed that major fault zones, minor joint trends, old landslide zones, and different levels of river terrace were better discernible in LISS IV as compared to LISS III-PAN merged data. Ramesh et al. (2004), after studying urban land use/cover of parts of Delhi regions using LISS IV data, remarked that it is comparable to IKONOS (4m; Mx) data for field level mapping. Bahugana (2004) evaluated the usefulness of LISS IV for coastal zone studies in the region of Okha, Gulf of Kachchh and found that LISS IV data is valuable in getting information on build-up areas and high tide lines with greater precision. LISS IV was also found to meet some of the essential requirements of the precision farming technology. Sesha Sai et al. (2004) confirmed its potential to capture intra field variability in crop fields of size 1 ha in ICRISAT farm near Hyderabad, India.
The strong relationship between spatial and spectral resolution of the imagery and level of applications it enables requires an essential step of undertaking an assessment of capabilities of any sensor to fully understand its potential and the level of possible applications, so that user can select appropriate imagery for relevant applications (Gupta and Jain, 2005; Collins and van Ganderen, 1976). Gupta and Anil Kumar (2002) evaluated IRS LISS III, IRS LISS III+PAN merged and IKONOS MS+PAN merged products to assess the level of information available in each for the purpose of urban planning. Similarly, Forghani (2002) has carried out an assessment of KOMPSAT-1 vs SPOT-2/4 satellite imagery for maintenance of Geoscience Australia topographic databases. With this background, we made an attempt to evaluate the enhanced capabilities of the LISS IV sensor for natural resource mapping and its implications for improved protected area management. The present study is specifically aimed to compare the potential of IRS P-6 LISS IV, India’s newest operational high resolution satellite and Landsat ETM+ and IRS 1D LISS III, the most widely used satellite data for natural resource mapping, particularly protected area management.
We selected a study site covering a part of Katerniaghat Wildlife Sanctuary in the Bahraich district of the state Uttar Pradesh in India. The site lies between latitude N 28˚ 15′ 44.7’’ and 28˚ 17′ 13.4” and longitude E 81˚ 10′ 38.7” and 81˚ 15′ 17.2” and encompasses an area of 2260 ha (Fig.1). The tract experiences very gentle slopes towards the south-east. The average elevation is 160 m above mean sea level. The soil consists of Gangetic alluvial formations. The sanctuary is a mosaic of moist and dry deciduous sal (Shorea robusta) forests, riparian forest, and plantations of khair (Acacia catechu), sissoo (Dalbergia sissoo), and teak (Tectona grandis). These wooded areas are interspersed by extensive tall grasslands and wetlands (Jha, 2000).
Fig. 1. Location of the study site in Katerniaghat Wildlife Sancatuary, Uttar Pradesh, India
Data used and methodology
In order to assess the level of information in the study site, we carried out a qualitative and quantitative comparison of digital data from LISS IV with Landsat ETM+ and IRS 1D LISS III. The details of digital data used in present investigation are presented in Table 1. All three datasets were georeferenced and the study site was delineated from each dataset. Firstly, the qualitative assessment was carried out which included the visual analysis of features on imagery so as to observe the level of sharpness, clarity, and reliability of information. Secondly, we focused on quantitative assessment which involved the estimation of extent of the mapping in three datasets.
Table 1 Data sets used for comparison of information content
Using visual interpretation at the scale of 1: 25,000 for LISS IV and at the scale of 1: 50,000 for remaining datasets, four categories were extracted for quantitative comparison. The categories included: metalled road, forest road, railway line, and vegetation types (Fig.2). For the first three features, length was measured and compared and for the last category, only comparison between LISS IV and LISS III was performed and ‘concordance area’ (mutual agreed area i.e. area of vegetation type classified correctly by both the datasets) was estimated. The data on the dominant plant life form at 70 point locations was collected during 2006 to validate maps.
Fig. 2 Linear features (metalled road, forest road, and railway line) extracted from Landsat ETM+, IRS 1D LISS III, and IRS P-6 LISS IV data sets
The qualitative and quantitative assessments evaluated by three data sets and their comparison provided the following insight.
Visual analysis of images revealed more contrast among features in LISS IV data as compared to two other datasets. The boundaries were more precise and easy to delineate in LISS IV. Few examples of accurate boundary delineation and identification of small important patches are presented in Fig. 3. In case of LISS IV data, presence of contrast and discernible bank line was evident (Fig. 3a). LISS IV data allowed better demarcation of grassland boundaries and delineation of a plantation patch within, which was otherwise invisible in ETM+ and LISS III data sets (Fig. 3b). Similarly, contrast tone and texture of Dense Sal Forest was conspicuous within other forest types in case of LISS IV data (Fig. 3c). Delineation of boundaries of Dense Sal Forest in ETM+ and LISS III data sets was confusing.
Fig. 3 – Images for visual comparison between Landsat ETM+, IRS 1 D LISS III, and IRS P-6 LISS IV data sets. a: Arrow indicates contrast and discernible bank line in LISS IV data. b: Circle indicates distinctive grassland boundary and added information of Eucalyptus plantation within grassland as indicated by arrow. c: Arrow indicates contrast tone and texture of Dense Sal Forest
The linear features such as metalled road, forest road, railway line, etc were very clear and easy to extract in LISS IV data, except in some places where the contrast was relatively low due to thick canopy cover. In case of other data sets, sometimes it was difficult even to identify the railway line adjacent to a metalled road. Regrettably, point features such as water wells and single trees were impossible to be detected in any of the data sets.
The results demonstrated that the extent of linear features mapped from LISS IV data was much more than other data sets. Statistics of the features mapped is given in Table 2. The comparison revealed that the extent of the railway line mapped from three data sets was almost identical. Likewise, the length of metalled road extracted from LISS IV and LISS III data sets was also almost equal. On the contrary, a significant difference in the extent of the metalled road mapped from LISS IV and ETM+ was recorded (Table 2). The metalled road in ETM+ data got merged with adjacent railway line. Forest roads mapped using three data sets allowed remarkable distinction in length (Table 2). Fig. 2 also illustrates the distinction in extent of extraction of forest roads. The length of forest (dirt) roads extracted from LISS IV data was 26.2 km greater in comparison to ETM+ data i.e. 112% enhancement. The enhancement of such extraction was to the extent of 22.4 km (82%) from LISS III to LISS IV data (Table 2; Fig. 2).
Table 2 Length of linear features (km) extracted from Landsat ETM+, IRS I D LISS III, and IRS P-6 LISS IV data sets
We compared the vegetation maps derived from LISS III and LISS IV data sets. Seven vegetation classes were delineated (dense sal forest, moderately dense sal forest, Terminalia alata forest, mixed deciduous forest, teak plantation, tropical seasonal swamp forest and upland grassland). All seven classes were recorded in LISS IV data but upland grassland could not be deciphered in LISS III data (Fig. 4). The overall accuracy computed for the vegetation map was 91.3% for LISS IV and 85% for LISS III data. In order to compare the ‘concordance area,’ we generated a confusion matrix (Table 3).
Fig. 4 Vegetation maps derived from IRS 1D LISS III and IRS P-6 LISS IV data sets
The major diagonal of the matrix indicated concordance. For example, out of 483.9 ha area of dense sal forest delineated by LISS IV data, the concordance area with LISS III data was 148.8 ha i.e. 30.7% coincidence. The remaining area (335.1 ha) of dense sal forest was misclassified by LISS III data into three different classes (moderately dense sal forest, Terminalia alata forest and teak plantation). The maximum mismatch was with moderately dense sal forest, suggesting that LISS IV data was able to segregate two most closely related vegetation classes accurately. The values of per cent coincidence for six other vegetation classes ranged from 57.6% to 100%; the lowest value was registered in case of Terminalia alata forest while the highest value was obtained in case of upland grassland. The values of percent coincidence were found to be high for mixed deciduous forest and teak plantation, being 89.7% and 89.2%, respectively. Higher values of coincidence indicated that both data sets classified them near equally due to their distinct tone and texture. The overall per cent coincidence was found to be 66.4%.
Vegetation Classes from LISS III
Table 3 Concordance area (ha) of vegetation classes based on IRS 1D LISS III and IRS P-6 LISS IV data sets
The visual comparison of features among three studied data sets indicated the distinctive superiority of LISS IV data in delineation of boundaries as compared to two other data sets. Interestingly, LISS IV and LISS III sensors have the same spectral resolution. However, the major difference lies in their configuration. LISS IV data have unique pixel of 5.8 m, whereas in LISS III, spectral information comes from the 23.5 m spatial resolution pixel. Due to this reason, there is more contrast among features in LISS IV data compared to LISS III. The inherent advantage of high resolution data, coupled with multispectral capability, assisted in identification and easy delineation of small plantation patch and otherwise confusing dense sal forest. Thus, results indicated that it can prove to be highly useful in natural resource mapping in any environmentally gradient environment. Similarly, Sudhakar et al. (2004) also found LISS IV data helpful in delineation of small patches of semi evergreen forests associated with riverine and moist deciduous tracts in forests of Mudumalai Wildlife Sanctuary in Tamil Nadu.
LISS IV imagery was also found extremely suitable for identification and delineation of linear features such as road and railway network. Although in case of metalled roads, the difference was comparatively less between LISS IV and LISS III data sets, but still, LISS IV data demonstrated better clarity of features. Interestingly, LISS IV data was able to distinctively map more number of forest roads from within forested area as compared to two other data sets. The extent of mapping was enhanced by about 112% as compared to ETM+ while 82% high as compared to LISS III data. Analogously, Gupta and Jain (2005) demonstrated that almost 100% more information on secondary, tertiary and village roads of Dehradun city, Uttarakhand was available with LISS IV data set as compared to PAN sharpened LISS III data. In another study on urban infrastructure mapping, Ramesh et al. (2004) found the LISS-IV image to be highly sharp compared to PAN sharpened LISS-III data and the infrastructure (road network) could be delineated upto level II. In contrast, Singh (2004) found mapping of roads using LISS IV data feasible in plain areas but not in the hilly areas of Gujarat, India.
The percent concordance between LISS III and LISS IV data sets helped in understanding how much, in quantitative terms, two maps deciphered from different sensors matches with each other. The results indicated overall agreement of 66% between vegetation classes delineated using LISS III and LISS IV data sets. As already discussed in visual comparisons, owing to high resolution, a small upland grassland patch was quite distinct in LISS IV image. Similarly, this high resolution data, with its enhanced multispectral capability, was able to segregate two closely related vegetation classes i.e. dense sal forest and moderately dense sal forest. Thus, our above findings endorse that LISS IV data set can prove to be quite useful for mapping of protected area with varying land cover/vegetation types. Furthermore, the overall accuracy of vegetation map prepared using LISS IV data was also found to be high too.
Summary and conclusions
The study enabled us to conclude that multispectral high resolution data of LISS IV distinctly gives better results for identification of features than the two other compared data sets of medium resolution. The LISS IV imagery gives crisper boundaries of polygon features which make their delineation more accurate. Additionally, LISS IV imagery has been found extremely suitable for identification and delineation of linear features such as road network, especially forest roads. About 82% and 112% more forest roads could be delineated with LISS IV data compared to LISS III and ETM+ data sets, respectively. Furthermore, the study also showed that with LISS IV data the level of classification of vegetation classes and accuracy improved and was found to be highly useful in mapping ecologically gradient environment.
The study demonstrated the additional quantifiable spatial and spectral advantages of LISS IV data over commonly used medium resolution satellite imagery and validated it to be better source of data for natural resource mapping. Moreover, with this high resolution data, large fine scale map at the scale of 1:25,000 can be generated. These maps could serve the diverse application needs of the natural resource managers for proper planning, management, monitoring, and evaluation of management effectiveness of protected areas.
The research was conducted at the Wildlife Institute of India (WII) with funding support from the Ministry of Environment and Forests, Government of India for the WII-MoEF-NNRMS Project. Thanks are due to P.R. Sinha, Director, WII and V.B. Mathur, Dean, Faculty of Wildlife Science, WII for advice and support. Special thanks are due to senior forest officials of Uttar Pradesh Forest Department for their help in various ways and frontline staff for field logistics.
- Al-Bilbisi H and Tateishi R (2002) A study on land use/cover classification with textural analysis using multi-temporal JERS-1 (SAR) L-band in arid and semi-arid (A case study in Northeastern Jordan). GIS Development. Available at www.gisdevelopment.net/aars/acrs/2002/luc/luc0115pf.htm
- Bahuguna A (2004) Evaluation of high resolution Resourcesat-1 LISS IV data for coastal zone studies. NNRMS bulletin. National Remote Sensing Agency, Hyderabad. India
- Clark DB, Read JM, Clark ML, Cruz AM, Dotti MF and Clark DA (2004) Application of 1-m and 4-m resolution satellite data to ecological studies of tropical rain forests. Ecological Applications 14: 61-74
- Collins W G and van Ganderen JL (1976) Land use studies by remote sensing, Remote Sensing Society, New York
- Forghani A (2002) Evaluation of KOMPSAT-1 versus SPOT-2/4 PAN for maintenance of Geoscience Australia topographic databases, Proc. of 11th Australasian Remote Sensing and Photogrammetry Conference held at Brisbane, Queensland from Sep. 2–6, 2002, pp. 1–12
- Gupta K and Jain S (2005) Enhanced capabilities of IRS P6 LISS IV sensor for urban mapping. Current Science 89 (11):1805-1812
- Gupta K and Anil Kumar G (2002) Evaluation of multispatial/spectral/temporal satellite data for urban planning and development exercises – scientific report. LPPD/FLPG/RESA, Space Application Centre, Ahmedabad
- Hussin YA (1999) State of the art: remote sensing applications for sustainable management of forests. GIS Development. Available at www.gisdevelopment.net/aars/acrs/1999/ts6/ts6125pf.htm.
- Innes JL and Koch B (1998) Forest biodiversity and its assessment by remote sensing. Global Ecology and Biogeography Letters 7 (6): 397 – 419
- Jha RN (2000). Management Plan of Katerniaghat Wildlife Sanctuary (2000-2001 to 2009-2010). Wildlife Preservation Organisation. Forest Department. Uttar Pradesh
- Kent M and Cooker P (1992) Vegetation description and analysis. A practical approach. London: Belhaven press, WC2E 9DS. 355 p
- Kerr JT and Ostrovsky M (2003) From space to species: ecological applications for remote sensing. review. TRENDS in Ecology and Evolution 18 (6):99-304
- Kumar VK and Martha TP (2004) Evaluation of Resourcesat-1 data for geological studies. NNRMS bulletin. National Remote Sensing Agency, Hyderabad. India. Available at www.nnrms.gov.in.nnrms/bul2004/2004_11.pdf
- Mahito K and Takeshi O (1998) Vegetation mapping with the aid of low altitude aerial photography. Applied Vegetation Science 1: 211- 218
- Mathur PK and Midha N (2008) Mapping of National Parks and Wildlife Sanctuaries, Dudhwa Tiger Reserve. WII-NNRMS-MoEF Project. Final Technical Report. Wildlife Institute of India. Dehradun, India. 216 p
- O’Hara CG (2002) Remote sensing and geospatial application for wetland mapping, assessment, and mitigation, Proc. of ISPRS Commission 1 Mid Term Symposium with Emphasis on Integrated Remote Sensing at the Global, Regional and Local scale held at Denver, USA from Nov. 10-15
- Porwal MC and Roy PS (1982) Vegetation types discrimination on Landsat TM data in heterogeneous forested landscape of Western Ghats – accuracy evaluation from large scale aerial photo maps. Journal of Indian Society of Remote Sensing 20 (1):21-33
- Porwal MC and Pant DN (1989) Forest cover type mapping using Landsat TM FCC. A case study for Chakrata in Western Himalayas (U.P.). Journal of Indian Society of Remote Sensing 17 (1):33-40
- Ramesh B, Kumar R, Neelakantan K and Sunde, BS (2004) Land use/ land cover studies using Resourcesat-1 data. NNRMS bulletin. National Remote Sensing Agency. Hyderabad, India
- Roy PS, Kaul RN, Roy MR and Gabryl SS (1985) Forest type stratification and delineation of shifting cultivation in the eastern part of Arunachal Pradesh using Landsat MSS data. International Journal of Remote Sensing 6 (3&4):411-418
- Sesha Sai MVR, Narasimha Rao PV, Ramana KV and Raju PV (2004) Utilization of Resourcesat-1 data for improved crop discrimination and surface soil wetness zonation. NNRMS bulletin. National Remote Sensing Agency, Hyderabad. India
- Singh TP (2004) Applications of Resourcesat-1 data in Gujarat. NNRMS bulletin. National Remote Sensing Agency, Hyderabad. India
- Sudhakar S, Pujar GS, Rajeshekar G, Gharai B and Murthy MSR (2004) Utilization of Resourcesat-1 data for advanced forestry applications. NNRMS bulletin. National Remote Sensing Agency, Hyderabad. India
- Welch R, Madden M and Jordan T (2002) Photogrammetric and GIS techniques for the development of vegetation databases of mountainous areas: Great Smoky Mountains National Park. ISPRS Journal of Photogrammetry and Remote Sensing 57:53-68