Relevance of Hyperspectral Data for Natural Resources Management

Relevance of Hyperspectral Data for Natural Resources Management

SHARE


T. V. Ramachandra,


Uttam Kumar
Centre for Ecological Sciences, Indian Institute of Science
Centre for Sustainable Technologies, Indian Institute of Science
[email protected]

Abstract
Land cover mapping relates to identifying the types of features present on the surface of the earth. It deals with discerning the extent of land cover features namely vegetation, geologic, urban infrastructure, water, bare soil or others. Variations in land cover and associated physical characteristics do influence weather and climate of our earth and hence, it is considered an essential element for modelling and understanding the earth as a system for many planning and management activities. Thus, understanding of land cover dynamics plays an important role at the local/regional as well as global level. Identifying, delineating and mapping land cover on temporal scale provides an opportunity to monitor the changes, which is important for planning activities and sustainable management of the natural resources.

Land cover mapping can be done most effectively through space borne remote sensors of various spatial, spectral and temporal resolutions. Due to the spectral resolution limitations of conventional multispectral imageries, hyperspectral sensors, which collect numerous bands in precisely defined spectral regions were developed. Hyperspectral images have ample spectral information to identify and distinguish spectrally unique materials that allow more accurate and detailed information extraction. These imageries are classified into different land cover categories using various algorithms. The genesis and the underlying principle behind each of these algorithms are different and essentially produce different output maps. This paper discusses the various efforts made for land cover and land use mapping with an emphasis on the hard classification algorithms for hyperspectral image processing at a regional scale. Neural network algorithm for classifying MODIS data has been implemented for Kolar district, Karnataka. The accuracy assessment is done using ground truth data and classified multispectral map on a pixel to pixel analysis.

Introduction
Land cover is the discernible vegetation, geologic, hydrologic or anthropogenic features on the planet’s land surface. Broadly speaking, land cover describes the physical state of the earth’s surface and immediate surface in terms of the natural environment (such as vegetation, soils, groundwater, etc.) and the man-made structures (e.g. buildings). These land cover features can be classified using the data of different spatial, spectral and temporal resolutions acquired through remote sensors mounted on space borne platforms. Land cover changes induced by human play a major role in patterns of the climate and biogeochemistry at a regional scale [1].

Land cover mapping using high spectral resolution has several advantages, because it aids in numerous mapping applications such as soil types, species discrimination, mineral mapping, etc. Hyperspectral data processing poses both challenges and opportunities for land cover mapping. Land cover mapping can be performed using various algorithms by processing the remotely sensed data into different themes or classes.

The terms land use and land cover are often used in natural resources management, meaning types or classes of geographical determinable areas. Land cover provides the ground cover information for baseline thematic maps. In contrast, land use refers to the various applications and the context of its use. This involves both the manner in which the biophysical attributes of the land are manipulated and the intent underlying that manipulation (the purpose for which the land has been used). Identifying, delineating and mapping land cover on temporal scale provides an opportunity to monitor the changes, required for sustainable management of natural resources. .

Recent exercise on global LULC (Land Use Land Cover) for vegetation mapping was the use of MODIS data as one of the most critical global data sets. The classification included 17 categories of land cover following the International Geosphere-Biosphere Program (IGBP) scheme. The set of cover types includes eleven categories of natural vegetation covers broken down by life form; three classes of developed and mosaic lands, and three classes of non-vegetated lands [2].

In India, land use and land cover (LULC), an important study from national perspective on annual basis using data from the latest Indian Remote Sensing Satellite – Resourcesat has been initiated by ISRO (Indian Space Research Organisation) and NRSA (National Remote Sensing Agency), Department of Space in coordination with several RRSSCs (Regional Remote Sensing Service Centres). Spatial accounting and monitoring of land use and land cover systems was carried out on a national level on 1:250,000 scale using multi-temporal IRS (Indian Remote Sensing Satellites) AWiFS (Advanced Wide Field Sensor) datasets to provide on an annual basis, the net sown area for different cropping seasons and the integrated LULC map. The AWiFS data covered Kharif (August – October), Rabi (January – March) and Zaid (April – May) seasons to address spatial and temporal variability in cropping pattern and other land cover classes. Decision tree classifier method was adopted to account the variability of temporal datasets and bring out reliable classification outputs. Legacy datasets on forest cover, type, wastelands and limited ground truth were used as inputs for classification and accuracy assessment [2].

The most significant recent breakthrough in remote sensing has been the development of hyperspectral sensors. The ‘Hyper’ in hyperspectral means ‘too many’ and refers to the large number of measured wavelength bands. Hyperspectral images are spectrally over determined, which means that they provide ample spectral information to identify and distinguish spectrally unique materials. Hyperspectral imagery provides the potential for more accurate and detailed information extraction than is possible with any other type of conventional remotely sensed data.

Moderate Resolution Imaging Spectroradiometer (MODIS) is a major instrument on the Earth Observing System EOS-AM1 and EOS-PM1 (termed AQUA) missions [3]. The ‘heritage’ of the MODIS comes from several space-borne instruments. These include the Advanced Very High Resolution Radiometer (AVHRR), the High Resolution Infrared Sounder (HIRS) unit on the National Oceanic and Atmospheric Administration’s (NOAA) Polar Orbiting Operational Environmental Satellites (POES), the Nimbus-7 Coastal Zone Colour Scanner (CZCS), and the Landsat Thematic Mapper (TM). MODIS is able to continue and extend the databases acquired over many years by the AVHRR, in particular, and the CZCS/Sea Star-Sea WiFS series.