Snow cover mapping using IRS-P6 AWiFS data and the relationships between some climatic factors with snowpack in the northwest of Iran

Water is regarded as the most essential of the world's natural resources (Vorosmarty et al. 2010). Water is perhaps the most valuable natural asset in the Middle East (Nagler et al. 2007). The Zarinerood basin is largely fed from snow precipitation, of which nearly two-thirds occurs in the winter and may remain as snow itself for some of the year. Accordingly, modeling of the snow-covered area in the mountainous regions of Western Iran, one of the major headwaters of the Orumiyeh Lake basin, bears significant importance when forecasting snowmelt discharge, especially for energy production, flood control, irrigation and optimization of reservoir operation. The snow-covered area is a fundamental parameter in the hydrologic cycle and climatology of the earth (Tekeli et al. 2005). Characteristic high albedo reflectance of snow causes the snow surface to reflect much of the incoming solar radiation (Tekeli et al. 2005). The main reason for using a spectral band in the visible part (0.4–0.7 Am) of the reflected solar radiation spectrum is associated with the high reflectance of snow at these wavelengths, thus facilitating its detection (Tekeli et al. 2005). With the high heat capacity of snow, snow cover insulates the soil surface from the atmosphere and slows down the warming process in spring (Tekeli et al. 2005). Thus, the snow has a direct impact on micro and macro atmospheric circulation models by affecting energy absorption and thermal heating of the basin (Tekeli et al. 2005). Snow cover and soil moisture are the most important variables in the heat and water vapor exchange between earth and atmosphere (Tekeli et al. 2005). The presence of snow in a basin strongly affects moisture that is stored at the surface and is available for future runoff (Maurer et al. 2003). Therefore, monitoring of the seasonal snow cover is important for several purposes such as climatology, hydrometeorology, water use and control and hydrology, including flood forecasting (Nagler et al. 2007). Detection of the snow-covered areas from space has been used since the beginning of the 1960s. Attention to snow-covered areas was increased with the studies performed by Martinec (1975), Rango & Martinec (1979) and Tekeli et al. (2005). Maps of dry/wet snow cover in the Indian Himalayas were produced by Gupta et al. (2005) using IRS-LISS-III multispectral imagery, and partitioned dry and wet snowy areas. The attendant problem with the use of observations in the visible wavelengths is the obscuration of snow cover caused by clouds and vegetation cover (Tekeli et al. 2005). One common problem that a scientist encounters in developing nations is the lack of informative data for related study (Tekeli et al. 2005). Even if such data existed, quick, on-line and timely access to the data may be problematic. As Iran is a developing country, quick access to snow data, which is essential in the analysis of the relationships between climatic and topographic factors and snowpack, is not yet possible. Governmental organizations most often perform snow measurements of the snow courses mostly over monthly periods. In these snow courses, the snow equivalent water and snow depths are measured along a track. Furthermore, most of the courses are started mainly in December and they are abandoned by the start of March when the snow ripens and starts to melt, even though some snow may still remain in upper mountainous areas. Iran can be described as a country having an abundance of snow; however, an established continuous operational snow monitoring system is not available. There exist some individual studies such as Ghanbarpour et al. (2007) and Pourhmmat et al. (2004). Thus, the high spatial and temporal variability of snow cover and remote sensing observations are particularly useful for providing spatially detailed input data for snowmelt runoff modeling. In spite of this, satellite-based snow cover information is still strongly under-utilized, in particular for operational hydrology (Tekeli et al. 2005).

In this study, we explored the potential of the multispectral and multi-temporal AWiFS data for mapping the snow cover in the northwest part of Iran. This involved a three-fold objective, viz.: (1) to document the current status of the snow cover and respective area measurements in the northwest part of Iran using the AWiFS data acquired in 2006/2007; (2) to compare the potential of the AWiFS data in identifying the snow cover based on the unsupervised classification (clustering method) with the MODIS data, based on the normalized difference snow index (NDSI); (3) investigation of the relationships between some climatic and topographic factors and the snowpack in the Zarinerood basin. Iran is a country with great diversity, having a wide range of landforms including the major mountain ranges, deserts, rich agricultural plains, and hilly jungle regions. Thus, this study would also help in the efforts to map the mountainous regions and south central Asia, as part of the goal of global snow cover mapping. Besides, the results from the snow cover type maps derived from the analysis of these AWiFS images can be used to improve the estimates of snow and non-snow areas, and support the simulations of snowmelt runoff models that require estimates of the snow area. Also, they support instances of investigating the relationships between some climatic and topographic factors with snowpack researches such as those conducted by Bloschl et al. (1991), in which a relationship between snowpack depth and the elevation was established linearly. However, Balk & Elder (2000) arrived at the relationship non-linearly (Elder et al. 1995). Also, a positive correlation between snowpack depth and elevation was presented by Shaban et al. (2004) although Erickson et al. (2005) revealed it as an adverse correlation.

The image processing was performed using various enhancement techniques including contrast stretching, band rationing, principal component analysis, spatial filtering, decorrelation stretching and finally color compositing (Jensen 1996). These processed data were then displayed as FCCs of various combinations, and the best combinations, which show maximum discrimination between snow and non-snow in distinct color tones, were selected for visual interpretation (Figure 3).

Unwanted artifacts like additive effects due to atmospheric scattering were removed through a set of pre-processing or clean-up routines. First-order radiometric corrections were applied using a dark pixel subtraction technique (Joshi et al. 2006). This technique assumes that there is a high probability that at least a few pixels within an image should be black (0% reflectance). However, because of atmospheric scattering, the imaging system records a non-zero Digital Number (DN) value at the supposedly dark-shadowed pixel location (Joshi et al. 2006). Therefore the DN value must be subtracted from the data to remove the first-order scattering component (Joshi et al. 2006).

For importing the data sets and, thereafter, processing the data, the PCI Geomatica software Version 9.1 was used. Geometric registration of these data sets was performed using an image-to-image tie down procedure through a second order polynomial fit (Kandrika & Roy 2008). Initially, the IRS AWiFS band 4 (SWIR band of 56 m resolution) was resampled and registered; then, the first three bands (56 m resolution) were resampled and registered with band 4 subsequently. These georeferenced data sets were checked both for temporal as well as spatial geo-referencing accuracies. During georectification an overall absolute positional accuracy of 2 pixels was achieved. However, while carrying out the geo-rectification of temporal AWiFS data sets, a relative accuracy of less than 1 pixel (RMSE <0.5 pixel) has been achieved (Kandrika & Roy 2008). Also, the images were resampled by the nearest neighborhood method. The study area was extracted using digital boundary data. After geometric correction, the DNs of all the data sets were converted into at-satellite reflectance values using the scaling functions and orbit parameters provided in the image header (Kandrika & Roy 2008). These outputs were re-scaled for 10-bit output using a constant scaling function across bands (Kandrika & Roy 2008). The study area was extracted using digital boundary data.

One of the possible solutions for the estimation of snowpack spatial distribution is in relation to the changes in the snowpack specifications due to changes in the factors that affect them. The climatic factors are more effective factors on the snowpack (Elder et al. 1995). As climate was identified as an effective factor on the snowpack, thus temperature and rainfall were selected from among the climatic factors, as they are of greater importance than the other factors. Therefore, in this study, the relationships between daily rainfall (cm) and components of daily temperature (cM) including the daily medium, maximum medium, minimum medium, absolute maximum and absolute minimum temperature (cM) were obtained from synoptic stations (Figure 1). They were explored along with snowpack specificities which included snow depth (cm), snow equivalent water (cm) and snow density (gr/cm³) for 3 months (from December, January and February) of 1996–97 until the years 2005–06 in SPSS 11.5 software. Also, the relationships between daily rainfall (cm) and components of daily temperature (cM) were explored along with the discharge of snowmelt runoff (m³/s) for 3 months (December, January and February) of 1996–97 until the years 2005–06 in SPSS 11.5 software.

This study was conducted in the Zarinerood basin due to its importance and the strategic location to direct the water of the Orumiyeh province into Orumiyeh Lake. This was the reason for selecting the Zarinerood basin for investigation. Data on snow measurements in this study are obtained from the snow courses of the Iran Water Resources Management Organization (IWRMO) for the monthly periods of December until February in 1996–97 until the years 2005–06 (Figure 1). Due to the importance of the elevation factor among the topography factors, the influence of the elevation (m) on snowpack depth (cm) was studied, therefore the correlation analysis and relationship was obtained between them from this basin, in the years 2004–05. As a primary correlation, an analysis was performed between snowpack depth (cm) and elevation (m) in the eight synoptic stations using the SPSS 11.5 software. Then the elevation of the basin was classified into six classes. To enhance the significant surface, with an increase of every 200 m of height a new class was defined. This stage provided a survey of the relationship between the snowpack depth and the elevation for different ranges of height using the SPSS 11.5 software. At this stage the primary homogeneity test of variances was carried out using the one-way analysis of variance. Providing the rejection of the equivalent assumption for variance groups in 1% of surface we used Dunnett's T3 test; otherwise the Duncan test was used for credibility. Finally, for the investigation of the different changes in snowpack depth for different heights in each class, correlation analysis was performed using the SPSS 11.5 software.

The correlation between the different bands of the IRS-P6 AWiFS from 2006 to 2007 was investigated by considering diagrams of different compositions of bands. Then the most suitable bands that could separate the snow from non-snow areas were identified from among the IRS AWiFS. Finally, the snow cover maps were obtained from the IRS AWiFS by unsupervised classification (clustering method). The diagram of the IRS AWiFS combination of binary bands obtained showed that a combination of three bands included B2 (green, 0.52–0.59 μm), B3 (red, 0.62–0.68 μm) and B4 (NIR, 0.77–0.86 μm). Among the four bands of the IRS, the AWiFS enhanced the gray degree scattering of pixels of the images in the possible range (0 to 255) and enhanced images in varying amounts. Also, the greater the spread, the more the number of pixels of gray degree (in the range of 0 to 255), and therefore the greater the volume of information and image contrast generated. However, a combination of the other bands with B5 (SWIR, 1.55–1.70 μm) showed an inverse conclusion. Therefore, using the SWIR band along with the composite bands to separate the snow versus non-snow areas is not useful. Also, the correlation between the four bands showed a high correlation between the B2–B3, B3–B4 and B2–B4 bands and a low correlation between the B2–B5, B4–B5 and B3–B5 bands of the IRS AWiFS, respectively. Therefore, a combination of the three bands including the B2 (green, 0.52–0.59 μm), B3 (red, 0.62–0.68 μm), and B4 (NIR, 0.77–0.86 μm) and the four bands of the IRS AWiFS, generates the best way to separate snow from the other phenomena in these images. The green and red bands are especially sensitive to snow, and help in improving the segmentation of the snow and other land cover types. These conclusions have been presented by Philip & Sah (2004), where they showed that the satellite image characteristics of the IRS-1C/1D data in the visible and near-infrared (NIR) regions of the electromagnetic spectrum are useful for obtaining information on the parameters of snow features. They also revealed that for snow studies, the combination of the green (0.52–0.59 mm) and red (0.62–0.68 mm) in the visible and NIR (0.77–0.86 mm) regions are found to be the most useful. Spectral reflectance curves of snow in the visible and NIR wavelength regions show that in the visible region fresh snow has very high reflectance, and as it begins to age, the reflectance slightly decreases (Zheng et al. 1984). However, in the NIR region, the reflectance of snow decreases significantly compared to that of fresh snow (O'Brien & Munis 1975). The combination of the aforementioned spectral bands of satellite images, particularly of the Indian remote sensing satellite, generates false color composites (Philip & Sah 2004). These provide the basis for the snow study. In general, the utility of the specific satellite data for snow study depends upon the defined objective of the project in question. The SCA in these maps was calculated and compared to calculate the SCA of MODIS image by NDSI in the same year.

A comparison of the SCA from the MODIS with IRS AWiFS showed a decrease in the SCA from December until last February. However, in the IRS AWiFS, the earth, water and cloud were classified as non-snow. The results confirm the unsupervised classification and clustering method for estimating the SCA of the IRS satellite images. Melting and further crystallization of the snow crystals were intensely affected by temperature changes, and thus the snow density affected the melting and further crystallization. Therefore, in the area under study, the values recorded for the components of daily temperature showed a significant correlation with the snowpack density in 1% of surfaces, whereas significant correlation was observed with the other snowpack specificities. In the Zarinerood basin, from the 3 months of study in these years, only 5 rainy days were recorded; therefore, the amount of rainfall had no significant correlation with the snowpack specificities. Also, in this study the results showing a linear and positive relationship existing between snowpack depth (cm) and elevation (m) is in accordance with the results of Bloschl et al. (1991) and Shaban et al. (2004). Lastly, because of global warming and thus increasing temperature, most of the snow melts in the winter from the Zarinerood basin. As a result, snow had disappeared from the basin by March, immediately prior to the beginning of the agricultural season.

The authors are grateful to the spatial organization and army geographic organization of Iran for providing the satellite images. Also, the authors are grateful to IWRMO and Orumiyeh Water Resources Management Organization for providing the climate and hydrometry information.