Spectral Features for the Detection of Land Cover Changes

Derivation of more sensitive spectral features from the satellite data is immensely important for better retrieving land cover information and change monitoring, such as changes in snow covered area, forests, and barren lands as some examples from local to the global scale. The major objectives of this paper are to present the potential of water-resistant snow index (WSI) for the detection of snow cover changes in the Himalayas, extant two composite images, biophysical image composite (BIC) and forest cover composite (FCC) for the detection of changes in barren lands and forested areas respectively, and two newly designed composite images, water cover composite (WCC) and urban cover composite (UCC) for the detection of changes in water and urban areas respectively. This research implemented the image compositing technique for the detection and visualization of land cover changes (water, forest, barren, and urban) with respect to local administrative areas where a significant land cover change occurred from 2001 to 2016. A case study was also conducted in the Himalayan region to identify snow cover changes from 2001 to 2015 using the WSI. Analysis of the annual variation of the snow presented in the research demonstrated promising performance for the detection and visualization of other land cover changes as well.


Introduction
The unprecedented changes in land cover triggered by human activities have brought enormous social, economic, and environmental challenges worldwide.
Satellite remote sensing is a suitable technology for large scale monitoring of land cover such as snow covered areas, forests, and barren lands as some examples from local to the global scale. Spectral indices are one of the techniques commonly used for the land cover monitoring of individual classes. Multi-temporal composite images are an alternative approach for simultaneous detection and visualization of spatio-temporal changes of land cover Sharma et al., 2018). However, derivation of more sensitive features from the satellite data is immensely important for better retrieving land cover information and change monitoring.
Major techniques for deriving the snow cover information from satellite data are spectral mixture analysis and spectral indices (Crane & Anderson, 1984;Rosenthal & Dozier, 1996;Hall et al., 2002;Salomonson & Appel, 2004;Painter et al., 2009). For improved discrimination between snow and water areas, the water-resistant snow index (WSI) was developed by . The WSI is the normalized difference between the value and hue obtained by transforming the RGB color composite made up of red, green, and near infrared bands into HSV (Hue-Saturation-Value) color space. For the detection of water bodies and urban built-up areas, normalized difference water index (NDWI, McFeeters, 1996;Gao et al., 1996) and normalized difference built-up index (NDBI, Zha et al., 2003) are available.
The spectral indices such as normalized difference soil index (Rogers & Kearney, 2004;Deng et al., 2015) have been used for the detection of barren lands.
The barren soil indices are derived as the normalized difference between shortwave infrared and near infrared or green reflectance. For improving the detection of barren lands, biophysical image composite (BIC) was designed by . The BIC is a false color composite image made up of the normalized difference vegetation index (NDVI), shortwave infrared reflectance, and green reflectance, which were specially selected from the highest vegetation activity period over an entire year.
The NDVI, the normalized difference between the near infrared and red  (Rouse et al., 1974;Tucker, 1979), is often used for the detection of vegetative areas. For improving the discrimination of forested areas from other vegetative areas, forest cover composite (FCC) was designed by Sharma et al. 2018. The FCC is a false color composite image made up of short-wave infrared reflectance and green reflectance, particularly selected from the period when the NDVI is at a maximum, as the red and blue bands, respectively; and mean NDVI value over an entire year as the green band. The FCC was designed in such a way that the forested areas appear greener than other vegetative areas.
The major objectives of this paper are to present the potential of water-resistant snow index (WSI) for the detection of snow cover changes in the Himalayas, extant two composite images, biophysical image composite (BIC) and forest cover composite (FCC) for the detection of changes in barren lands and forested areas respectively, and two newly designed composite images, water cover composite (WCC) and urban cover composite (UCC) for the detection of changes in water and urban areas respectively.

Satellite Data
The composite images were generated from the Moderate Resolution Imaging In addition, a case study was conducted in the Himalayan region to identify the snow cover changes using the MODIS data.

Detection of Barren Lands Changes
The performance of the BIC was demonstrated to detect changes in barren lands. The BIC is a RGB color composite image made up of Normalized Difference Vegetation Index (NDVI), short wave infrared reflectance, and green ref- lectance, which were specially selected from the highest vegetation activity period (Equation (1)) over an entire year .
An alternative approach for the detection of barren lands was also demonstrated. The barren lands frequently lack the vegetation cover, contain lesser surface moisture, and exhibit higher land surface temperature. To capture these conditions of the barren lands, an index was proposed as the normalized difference between the hue and saturation by transforming the RGB (near infrared, shortwave infrared, red) color into HSV (Hue-Saturation-Value) space. The sensitivity of the MODIS based index was assessed in a transition zone of varied level of vegetation and barren lands concerning the true-color image for 2016.

Detection of Forest Cover Changes
The BIC was designed for the extraction of vegetative areas (forests, crops, shrubs/herbs) from non-vegetative areas (barren, urban, and water/snow). To overcome the difficulty of discriminating forested areas from other vegetative areas such as crops, shrubs, and herbs using the BIC, FCC was designed by changing the green (G) term of the BIC by annual mean NDVI value (Sharma et al., 2018). The composition of the FCC is shown in Equation (2).
Examples of forest cover change detection were demonstrated using the MODIS based FCC concerning the true-color images in areas where substantial changes in forest cover have occurred from 2001 to 2016.

Detection of Water Cover Changes
A new image composite called the water cover composite (WCC) is proposed in the research. The WCC is made up of annual minimum green ( min Green ) reflectance, annual minimum near infrared ( min Nir ) reflectance, and annual maximum superfine water index ( max SWI ) values as the red (R), green (G), and blue (B) bands respectively. The SWI, the normalized difference between the saturation (Hue-Saturation-Value transformation of the RGB reflectance) and near infrared, was designed for better extraction of surface water at global scale (Sharma et al., 2015). The composition of the WCC is shown in Equation (3).
The composite images are designed by conceptualizing the spectral characteristics and temporal/phenological variations of the land cover types by harnessing multi-temporal satellite images of an entire year.

Detection of Urban Cover Changes
For the detection and visualization of urban built-up areas, a new image composite called the Urban Cover Composite (UCC) was also designed in this research. The UCC is made up of Enhanced Urban Built-up Index (EUBI; , annual mean NDVI (NDVI mean ) values, and annual minimum green (Green min ) reflectance as the red (R), green (G), and blue (B) bands respectively. The EUBI  is the ratio between annual median composite of the nighttime lights values and annual maximum composite values of the normalized difference between hue and value obtained from the Hue-Saturation-Value transformation of the RGB (near infrared, shortwave infrared, and red bands) composite of the MODIS data. The composition of the UCC is shown in Equation (4).

Detection of Snow Cover Changes
The MODIS data available from 2001 to 2015 were used for detecting snow cover changes in the Himalayas. For each 8 days data, the WSI was calculated, and monthly median composites were prepared. The monthly true-color composite images were used as the reference data to determine the threshold between snow and non-snow areas on the WSI images. The Himalayas, also known as water towers of Asia, provides freshwater to large rivers of Asia including Mekong, Ganges, and Brahmaputra. These downstream rivers are the backbone of water supply, agriculture, and livelihood in many Asian countries including Vietnam.

Water Cover Changes
The potential of the water cover composite (WCC) was assessed for the detection and visualization of spatio-temporal changes of the water bodies in two administrative regions: Aralskiy, Qyzylorda, Kazakhstan and Andamarca, SudCarangas, Oruro, Bolivia. In both locations (Figure 1 and Figure 2), a large area of water body lost gradually from 2001 to 2016. Almost all water bodies lost in Andamarca, SudCarangas, Oruro, Bolivia in 2016. It should be noted that, only inter-annual variation of the land cover changes have been presented in this paper. Though the water bodies exhibit seasonal changes intra-annually, the WCC is designed in such a way that it can visualize even a minimum level of the water body. Therefore, the loss of the water body as detected by the WCC conveys severe environmental concerns in those regions.

Barren Lands Changes
The potential of the biophysical image composite (BIC) was assessed for the detection and visualization of the spatio-temporal changes of the barren lands in  In addition, Figure 5 demonstrates the potential of the proposed index for the detection of barren lands. The proposed index has shown different index values between the barren lands (high values) and vegetative areas (low values). However, it is the subject of future research to see its applications in other geographic regions.

Forest Cover Changes
The forest cover composite (FCC) has successfully detected the forested areas and their temporal changes as shown in Figure 6.
In addition, the potential of the forest cover composite (FCC) for the detection and visualization of spatiotemporal changes of the forested areas was assessed in two administrative regions: PreaekPrasab, Kraches, Cambodia and Moreno, Santiago del Estero, Argentina. In both locations (Figure 7 and Figure  8), a large forested area lost gradually from 2001 to 2016. The loss of forested areas has alarmed the environmental concerns in those regions.

Urban Cover Changes
We assessed the potential of the urban cover composite (UCC) for the detection and visualization of spatio-temporal changes of the urban built-up areas in two

Discussion and Conclusions
Moderate Resolution Imaging Spectroradiometer (MODIS) has provided precious assets of multi-temporal data suitable for monitoring land cover types such as snow covered areas, forests, and barren lands as some examples. Spectral indices derived from remote sensing data are crucial to the monitoring of land cover changes. The multi-temporal image compositing approach provides an opportunity for simultaneous detection and visualization of spatiotemporal changes of land cover. In this research, image compositing technique was implemented for the detection and visualization of different land cover (water, barren, forest, and urban) changes with case studies in several local administrative regions where significant land cover changes have occurred from 2001 to 2016. The special composite images were designed by exploiting the time-series of the multi-spectral satellite imagery on an annual basis. However, the application of the spectral index and composite images presented in this research in different study areas is a subject for future research. It is also important for validating the performance of the spectral index and composite images with the support of ground truth data. Environmental issues require not only national measures but also local level measures. Local level environmental data are important for the assessment of global-local sustainability (Tateishi et al., 2017). This research presented an idea of monitoring land cover changes with respect to the unit of local administrative area.