|
LAND SURFACE TEMPERATURE RESPONSES TO THE LAND COVER DYNAMICS IN WESTERN GHATS
(PDF)
T V Ramachandra, Srijith A H and Bharath S
Energy and Wetlands Research Group,
Centre for Ecological Sciences,
Indian Institute of Science - 560012
1. Introduction
Land is the most important natural resource, which comprises of soil, water, and the associated flora and fauna. Land constitutes about 21% of the total earth’s surface area and water bodies occupy rest of the area on Earth (ocean, sea, glaciers, lakes, rivers etc.). Land cover (LC) of a region refers to the observed bio-physical features on the Earth surface (Lillesand et al., 2004). Land Use (LU) is defined as human modified earth resources and is an outcome of socio economic factors. LC of a region influences the surface energy and the exchange of moisture, heat and momentum between the atmosphere and land surface (Botkin et al., 1984). As the anthropogenic activities on Earth surface continue to accelerate, the consequent LULC changes have resulted in the alteration in ecosystem functions with changes in the climate. The changes in LULC are issues of global concern due to large scale deterioration of environmental system. The impacts of LULC changes on climate can be divided into two major categories: bio-geochemical and bio-geophysical (Kabat et al., 2002). Bio-geochemical processes affect climate by altering the rate of flow of chemical substances between biotic (biosphere) and abiotic (lithosphere, atmosphere and hydrosphere) components of Earth. Bio-geophysical processes affect the physical parameters that determine the absorption and deposition of energy on Earth surface. The physical parameters are albedo, evapotranspiration, reflective properties, and absorptive properties of Earth (Gordon, 2008). The magnitude of changes are evident from the extent of transformation of forests/woodlands (6million sq. km) and grasslands (4.7 million sq. km) to crop land since 1700 AD (Goldewijk et al., 1997). The LULC changes are so wide spread that when aggregated globally affect the key aspects of Earth’s functioning affecting the climate in a number of ways. Hence, LULC change analysis has become a prerequisite for managing natural resources and monitoring environmental changes.
Forests are large uncultivated tract of land covered with trees and underwood, woody grooves and pasture. Forests influence climate through physical, chemical and biological processes. Forests provide ecological, economic, social and aesthetic services to both natural systems and life forms (Gordon, 2008). The clearing of forests for cultivation, settlements and timber has altered climate, rainfall, temperature etc. Forests perform a series of vital environmental functions at local, regional and global levels. At local level environment services provided by forest include the maintenance of soil from erosion, water and climate. On the global scale, forests have two fundamental functions: the role of carbon sinks in the global carbon cycle and as pools of biodiversity. The trees in forest hold an immense amount of carbon in trunk. When trees are cut, they let out carbon, which joins with oxygen to form CO2, increases carbon content in the atmosphere and results in global warming. Deforestation increases the amount of release of CO2 and other trace gases to atmosphere. When a forest is cut and replaced by cropland and plantations, an excess amount of CO2 is released. In terms of social and economic importance and livelihood support, trees and forest occupy a central position in providing fodder for both domestic and wildlife in addition to being the main source of energy and building material for household (FAO, 2006).
The vegetation cover is one of the important components of the Earth’s surface. It strongly influences evapotranspiration, infiltration, runoff, soil erosion and climate. The vegetation cover has been widely recognized as one of the best indicators for determining land condition (Booth and Tueller, 2003; Bastin and Ludwig, 2006; Wallace et al., 2006). The changing land use, would alter the rainfall pattern affecting the regional climate which may have long term implication like drought (Lin et. al., 2008; Warburton et. al., 2012). The changing land use and decline in rainfall leads to rise of temperature in a region. Tropical deforestation and desertification have been recognized as the most significant cause of land cover change. The reduction in forest cover has paved way to an increase of the mean temperature of Earth from 14.8°C (1980 - 2000) to 15.67°C (2000 – 2016) (Hansen., 2016). The temperature of a region determines the rate of evapotranspiration, climate, water-energy balance etc. With the increase in change in land use from forest to built-up, mono-culture (plantations), agricultural farms etc., the region has witnessed changes in mean temperature. The rise in temperature has direct implications on environment and life forms leading to their death due to heat wave, drought, etc. The rise in temperature of a region is studied through land surface temperature studies over a temporal scale.
The LST is the temperature of the skin surface of the land derived from satellite information or direct measurements. LST is a measure of the energy balance between Earth and atmosphere (Zhenming et al., 1989). They show a spatial heterogeneity and vary with differing land uses. They are different from air temperature, which is taken at a height of 2 m from the ground. The degree of LST is influenced by elevation, slope, aspect which exert control on incoming solar radiation (Dubayah, 1990). Variation in LST also may be subject to seasonality, time of day, sea breeze, surface air temperature, humidity, wind speed, and land use (Wang et al., 2008). In remote sensing terminology, it can be just called the surface radiometric temperature emitted by the land surface, observed by sensor at instant viewing angles (Prata et al., 1995). The studies on land surface temperature have helped us understanding the effects of rising temperature like the melting of snow in the Polar Regions, rising sea water levels etc. The low lying islands are threatened by rising water level which is a huge concern. Studies have shown forest cover helps in regulating the temperature of a region over other land use classes.
Landscape comprises of prime natural resources: soil, water and the associated flora and fauna interacting with each other and forming ecosystem. Globally, Land use Land cover (LULC) change is as old as human kinds (Helmer et al., 2000). The abrupt changes in landscape alterations are evident from increase in rate of change in past few centuries threatening humankind globally (Lambin et al., 2006). The rapid LULC changes exert detrimental and adverse impacts on environment and livelihood. The recent focus is on forest conversion because of its adverse impacts on global and regional climate change, soil degradation, loss of biodiversity and goods and services provided by natural system (Lambin et al., 2006). The change in climate can be attributed to changing LU pattern (Dutt et al., 2005). The monitoring of forest cover is essential for developing and implementing appropriate biodiversity conservation and carbon emission reduction policies (Defries et al., 2007). The quantification of forest cover is essential for forest resource management, land use planning, bio-mass estimation etc. Land cover mapping and monitoring has been done with the spatial data acquired through space borne sensors at regular intervals (i.e. Remote sensing data).
The assessment of forest cover is essential to understand LULC changes of the region. Prior to remote sensing technology forest cover changes have been quantified through field visits by measurement of canopy cover (Jennings et al., 1998), identification of faunal diversity (Peterken et al., 2003), permanent sample plot techniques (Synott et al., 1992) etc. The traditional methods of assessing forest cover are too complicated and demand huge resources and time consuming. The use of satellite data and GIS has now gained momentum largely due to the advantages like synoptic coverage, consistency, global reach, precision etc. (Lambin, 1997). Remote Sensing is the process of deriving information about the earth’s land and water surfaces using images acquired from an overhead perspective, using electromagnetic radiation in one or more regions of the electromagnetic spectrum, reflected or emitted from the earth’s surface (Campell, 1996). LULC studies are usually carried out using sensors with spatial resolution between 30 to 1,000 m or large extents (Defries et al., 2007). Hansen et al., 2008 highlighted how earlier the global LULC mapping has been carried out using data acquired by AVHRR at coarser resolution, but with the launch of MODIS sensor in Terra (1999) and Aqua (2002) platform, the land cover mapping has become easier using time series data. The Moderate Resolution Imaging Spectro radiometer (MODIS) satellite data products are reliable and useful for monitoring land cover changes. Even though it has a minimum spatial resolution of 250 m, the advantages include: high temporal resolution (ie., daily), data availability at no cost, wide range of data products etc. These characteristics allow land use mapping not only at global and national scales, but at regional and sub-regional scales as well.
Remote sensing has a long history with the aerial photographs taken from top of hills, air balloons, aircrafts etc. initially used for mapping. With the improvement in technology, development of platforms and sensors gained the thrust. Different platforms were developed to acquire data employing different methods. With the launch of LANDSAT – 1 in 1972, there has been a continuous large supply of remotely sensed data of Earth’s surface. Post this era, the development in both sensors and mapping techniques reached a new level. The digital data acquired through remote sensing platforms are to be processed to acquire information. This process involving manipulation and interpretation of the digital spatial data is known as Digital Image Processing (Keifer et. al., 2014). The satellite data acquired have varying pixel values which are allotted to different land use classes and the whole process is known as classification. When the rules of classification are solely based on spectral radiances, the process is known as spectral pattern recognition and if the rules are based on geometric shapes, size, pattern etc. the process is called spatial pattern recognition. The temporal pattern recognition uses different periods of time as an aid in feature identification. Land use changes are identified through temporal pattern recognition of the same area. The classification schemes are usually based more on the spectral pattern of the satellite data. The classification schemes broadly based on spectral pattern of recognition are supervised, unsupervised and hybrid approaches.
The supervised learning is based on the sample training data. Using this method, the analyst has sufficient knowledge on pixels to generate representative parameters for each class of interest. This process is called training. Once trained the classifier is then used to attach labels to all the image pixels according to the trained parameters. The effectiveness of supervised classification largely depends on accurate estimation of pixels of each class during training. In an unsupervised classification, the prior knowledge of the classes is not required. The clustering is based on the spectral classes present in the image data. The spectral classes or clusters are compared with reference data to determine the identity and informational value of the spectral classes i.e. the information utility is defined by determining spectrally separable classes. There are numerous algorithms to determine natural spectral groupings present in data set: K-means, Fuzzy C means, hierarchical clustering, mixture of Gaussians etc. K-means is an exclusive clustering algorithm, Fuzzy C means is an overlapping clustering algorithm, mixture of Gaussian is a probabilistic clustering algorithm, hierarchical clustering clusters dataset into two kinds: similar and dissimilar (Guiliano et al., 2000). There are many hybrid approaches combining the advantages of both supervised and unsupervised classification like decision tree, neural network classification etc. The hybrid approach of clustering combines the supervised and unsupervised clustering techniques. The hybrid classifiers are useful in analyses where there is complex variability in spectral response patterns. It may be carried out by performing unsupervised classification initially to identify the numerous spectral classes that needs to be defined. This is followed by supervised classification wherein the spectral classes are classified using the training areas, which forms the hybrid approach. Even though both these approaches are common, the suitable approach of classification is based on the kind of data required to classify. Datasets that are too large are usually classified through unsupervised classification method followed by identification whereas smaller dataset are classified through supervised classification.
The first global LULC map through remote sensing data was prepared by DeFries et al. (1994) using maximum likelihood classification of monthly composited AVHRR NDVI data at 1° spatial resolution. Following this study, DeFries et al. (1998) used a decision tree classification technique to produce global land cover map at 8km resolution from AVHRR data. The launch of MODIS aboard Terra with 36 spectral bands provided a great improved basis for mapping and monitoring land cover data. Friedl et al. (2002) carried out global land cover mapping at 1km spatial resolution using data acquired from MODIS. The mapping was done using an algorithm containing two parameters global land cover at 1km and the land use dynamics at 1 km resolution. The analysis was carried out using supervised classification technique to classify into 17 land cover class globally following the IGBP standards. The global land map was classified at an accuracy of 75%. Even though there were some inadequacies in the map, the mapping remains the foremost in global modelling.
Chen et al. (2015) studied the human impacts on land system at a finer scale of 30 m resolution using images acquired by Landsat for 2001 and 2010. Since the analysis is carried on a global scale a large volume of data is involved which involved the use of automated approaches towards mapping. A pixel and object based method was used for classification. The land cover classification was carried out on the basis of split and merge strategy into 10 land cover classes. During classification, each land use class has been identified and extracted before proceeding further classification. The classification was based on pixel object knowledge wherein the pixels along with relief feature were taken into consideration. The classification could achieve an overall accuracy of 78.6%. Henderson and Gornitz (1984) studied the impacts of land cover transformation on climate change with emphasis on tropical deforestation. The deforestation of forests has been identified to be the major driver to land cover changes (Bolin, 1977). Of all the kinds of forest, the destruction of tropical forest is environmentally undesirable. The dense tropical forests have low infra-red emissivity which tends to absorb the incident short wave and inhibits the emission of long wave, thereby maintaining a positive net radiation balance. The inhibited thermal energy is used during photosynthesis and evaporation of water. The reduction in tropical forests also cause massive increase in run-off and soil erosion leading to flash floods and river sediment deposit (Delwaulle, 1973; Eckholm, 1973). The Western Ghats is one the tropical forests listed as global hotspots of biodiversity which is under a tremendous pressure of land use change. Menon et al., (1998) estimated the annual rate of deforestation in Western Ghats to be 0.57% in the period 1920-1990. The annual decline in forest cover of Kerala as assessed by Prasad et al., (1998) was found to be 0.9% annual for the period 1961-88.
Jha et al. (2000) estimated the changes in forest cover of Southern Western Ghats between 1973 and 1995. The study was conducted using satellite images acquired from Landsat MSS, 1973 and IRS 1B LISS-I, 1995. The supervised classification technique used for classification and categorized LU as forest and non-forest with an accuracy of 80%. It highlights a huge loss in the forest cover (40,000 sq. km) and biological diversity of Southern Western Ghats due to anthropogenic pressure. The study revealed an estimated a loss of 25.6% of forest cover in 22 years.
Ramachandra et al. (2014) conducted a study to analyse the land surface temperature and rainfall dynamics changing landscape of Western Ghats of India. The study was conducted using MODIS image derived NDVI (Kreigler et al., 1969) at a spatial resolution of 250 m.
The study was carried out for the period 2003 and 2012. The study assessed the effect of land cover change on temperature and rainfall. The land cover classes were assigned on the basis of NDVI threshold and the land cover change during the period was assessed. The study revealed a consistent decline in dense vegetation cover in all 3 regions of Western Ghats i.e. north, central and south. The study revealed a significant correlation between NDVI and climatic parameters (rainfall and temperature). The study revealed a relationship between rainfall and land use pattern. The areas that receive lesser rainfall in the leeward side of the Western Ghats are often plains where agriculture is practised whereas the coastal side which has a tropical climate with annual rainfall around 1800mm supported plantation crops like rubber, coconut, tea etc. The study also revealed there is a decreasing trend in rainfall in the regions affected by loss of forest cover. Also another study by the same author reveals a massive diversion of forest land in Idukki district of Kerala at the rate of 3.6% in the period between the period 1980 and 2016. The study reveals a massive increase in the area under plantations, agriculture and urban settlements and reduction in open areas and forests.
Bharath and Ramachandra, (2012) conducted a study in Uttara Kannada district, Karnataka to analyse the spatio-temporal land use dynamics from 1973 to 2010 and the impacts of land use change. The study was conducted by analysing NDVI (Normalized Difference Vegetation Index) to assess vegetation at temporal level. The study was conducted by considering the agro-climatic zones of the district. The study revealed a trend of declining forest cover in the district, with forests giving way to monoculture plantations, agricultural activities and developmental projects. The study revealed an increase in number of patches and decline in class area of forest cover which clearly shows the declining trend of forest.
Kale et al. (2016) conducted a study to assess the pattern of LULC change in Western Ghats of India during 1985 to 2005 and project the future (2025) spatial distribution of forest using logistic regression and Markov model. The land use has been classified on the basis of IGBP classification scheme. The classification has been carried out using the land use vector layer of 2005 acquired from ISRO. The land use of 2005 was compared with the land use raster data set of 1995 and the land use change was updated. Similarly, the land use of 1995 was compared with the raster data set of 1985 and the land use change was updated. The land use transitions from 1985 to 1995, from 1995 to 2005 and from 1985 to 2005 were investigated. Other auxiliary data like population, temperature, slope, soil, road etc. that have been outsourced were used as drivers of land use change. The modelling of land use has been made possible using different modelling codes i.e. CLUE model based in Arc GIS, IDRISI, Markov Chain, Landis, urbanism etc. The spatial allocation of forest demand was predicted through logistic regression model based on Akaike Information criterion and area under receiver operating characteristic curve. There has been an increase in shrub land from 1985 to 2005. The study states there has been an increase in built-up at the cost of crop land in the period 1995 to 2005. The results for 2025 show a trend of forest degradation with greater probability of conversion of dry forest to non-forest. The forest cover of 2025 in Western Ghats is estimated to be about 37794 sq. km.
Setiawan et al. (2012) conducted a study to assess the change detection in land use and land cover dynamics at regional scale using MODIS Time-Series Imagery. The analysis of multi-year time series of land surface attributes (NDVI, EVI, LST etc.) and their seasonal change can indicate complexity land cover. The study was carried out using the wavelet transformed vegetation index EVI (Enhanced Vegetation Index), at 250m resolution from MODIS Terra platform between the periods January 2001 to December 2007.
The change detection between successive years was assessed on the basis of distance comparison as described by Bouman, (2009). The pixels identified with change in class were validated using Google Earth and high resolution aerial photograph. This study characterized the conversion of forested lands to agricultural lands; barren land to urban land etc. The study had identified climatic effects like drought that occurred during the study period by analysing the EVI data. This study recognizes the importance of studying multi-temporal data sets to identify possible changes in LC and vegetative phenology. The study revealed the LU dynamics occurring in forest land such as deforestation, reforestation and forest regrowth during the study period.
Guo et al. (2007) conducted a study to compare and evaluate the ability of two vegetation indices: NDVI (Normalized Difference Vegetation Index) and EVI (Enhanced Vegetation Index) in a diverse range of biome of North West China. The vegetation of this biome includes broadleaf forests, needle leaf forests, meadows, grasslands, steppes, scrub, desert and agricultural farms. The satellite results are validated using CE 313 that has five filters between 450 and 1650nm to calculate vegetation indices over different vegetation types in different seasons. The results from satellite data proved NDVI shows a higher value than EVI and the difference between the two increased from deserts, steppes, cultivated vegetation, meadows to forest. There has been saturation of NDVI in high biomass types and the study recommends for EVI data in complex heterogeneous high biomass regions.
Dugarsuren et al. (2011) conducted a study to assess the LC change in Mongolia for a period of 10 years between August 2001 and August 2009. The land cover classification and change detection was carried out using a single stacked image of MODIS vegetation products and visible bands. The study followed a homogeneous classification wherein the complex land cover consisting of different types of forest cover, grasslands, and non-vegetated lands could be identified and categorized into 13 classes. The study recommends for homogeneous classification in complex terrains so as to reveal the spatio-temporal changes between different types of land cover. The study recommends to analyse the relationship of LC with temperature, precipitation etc. for further understanding of dynamics.
Setiawan et al. (2014) conducted a study on the seasonal dynamics of paddy field to analyse the cropping intensity of Java Island. The study enables to understand the effect of climate change on cropping pattern and postponement of cropping in non-irrigated lands. The study was conducted out using vegetation index EVI (Enhanced Vegetation Index) acquired from MODIS Terra 250 m platform. The data was analysed by looking at the time series trend of the cropland. The continuous fluctuation in EVI values all through the year determines if the land is cropland. The study was able to identify 8 different classes of cropland on the basis of intensity (single, double, triple etc.) and type of actual land class converted to cropland (upland, barren land etc.).
Wang et al. (2009) conducted a study to detect land cover change based on MODIS 250 m vegetation index (EVI). The study was conducted for the Dongjiang River basin in South Eastern China. The study was carried out by unsupervised isodata classification technique and decision tree classification method for land use land cover analysis in the period between 2001 and 2008. The 23 band EVI data are classified into 50 clusters which are evaluated using field survey data and high resolution images. Then rule based decision tree is built based on slope of terrain, land surface temperature etc. which are used to reduce confusion existing between different classes. The study has revealed a trend of declining forest cover, quantitatively indicated through the rate of change and transfer matrix. The study also reveals that classification carried out through hybrid classifiers have better accuracy and greater advantages.
Aide et al. (2012) conducted a study to identify the forest recovery trend in the Andes range of Columbia. The region was studied by mapping the land use land cover maps from 2001 to 2010 using MODIS (MOD13 250m) product coupled with reference data from QuickBird imagery and Google Earth to visually interpret the land use classes and for accuracy assessment. The study revealed that there has been a net gain in woody vegetation at national scale from 2001 to 2010. This increase in forest cover has been attributed to the establishment of protected areas. The study also reveals the advantage of MODIS data and the benefits of long term conservation planning. The regions prone to deforestation showed improvement with the establishment of eco-regions and protected areas.
Li et al. (2016) studied the potential and actual impacts of deforestation on land surface temperature (LST). The changes in forest cover modifies water, energy and carbon cycle of land surface in turn affecting the climate of the region are investigated. The study was conducted by calculating LST between forest and non-forest land features. The actual impact on temperature is quantified by analysing the trend of LST between deforested and non-forested land over several years. The study was conducted using the reference MODIS land cover data (Friedl et al., 2010) and Global Forest Change data (Hansen et al., 2013). The pixels were identified to be either forest or non-forest on the basis of a threshold set with at least 50% of pixel covered with trees, considering it to be forest. The forest change maps were created for the period 2000-2006 and 2008-2012. The LST data acquired from MODIS Aqua for the period from July 2002 to December 2013 was used for the study. The study revealed that deforestation increases Tmax (maximum temperature) and Tmin (minimum temperature) values in tropical regions and decreases both Tmax and Tmin in boreal regions. The high resemblance of air temperature with LST enables this kind of studies and this indeed helps us to understand the effects of land cover change.
|
