Detection of the Snow Cover Area Using NOAA-AVHRR in Shahcheraghi Dam Basin

Extended Abstract
Introduction
Snow, as one of the basic factors of water supply, plays an important role in water resources
management, especially in areas with cold winters and warm summers. The data obtained from
snow gauges as well as temperature and precipitation time series data are generally being used
to develop experimental models in order to estimate the spatial and temporal distribution of
snow in watersheds. However, when reliable snow or other necessary climatic data records do
not exist, using proper substitutes becomes essential. Hence, the snow cover area (SCA) derived
from satellite images can be used as a representative of the amount of snow in a basin.
Moreover, Remote Sensing (RS) is a useful tool in identifying snow and calculating SCA in
mountainous regions with low accessibility and deficiency of snow gauges. Accordingly, the
SCA time series data can then be used as input dataset in flow forecasting by hydrologic
models.
This paper aims to study the snow cover area of Shahcheraghi Dam basin in order to collect
the necessary input data for developing dam inflow forecasting models. The basin is located in
the north of Semnan province, Iran. The area of the basin is 1373km2 and the annual
precipitation and mean temperature of the basin are 124mm and 12°c, respectively. Since there 
is no active snow gauges within the basin and also there is only one weather station with reliable
temperature records in the region, NOAA satellite images have been used for defining the SCA.
Methodology
In this paper snow cover area detection in Shahcheraghi dam basin has been studied using
NOAA-AVHRR images in a 22-year period from 1986 to 2007. In order to improve the
precision of calculated monthly SCAs, an image per 10 days was processed (3 images per
month). The highest value of SCA among the three calculated values in each month is selected
as the final SCA data of the month. Since during this period of time two different sensors of
AVHRR-2 and AVHRR-3 have recorded data in different spectral bands, it is necessary to use
different algorithms in separating snow from other phenomena including cloud and land cover.
By employing the differences between the spectral characteristics of snow compared with other
phenomena, the snow covered area can be separated. Therefore, two threshold algorithms are
used to separate SCAs. These algorithms are based on grouped conditions of comparing albedo
of bands 1 and 2 and brightness temperature values of thermal bands. The most significant
difference between the conditions in these methods is using the albedo of band 3A (1.6μm) in
AVHRR-3.
On the other hand, it is necessary to evaluate the numerical difference among the snow
separation methods as they may significantly affect the statistic parameters of the time series.
Moreover, two trend detection methods are used to examine whether significant trends in the
time series exist. The hypothesis-based linear regression and non-parametric Mann-Kendall
methods are applied to the maximum annual SCA data.
Results and Discussion
Based on the NOAA-AVHRR image properties, snow cover area is detected by the
aforementioned threshold algorithms. The results show that the maximum amount of SCA
occurs in January. Generally the snow settlement in the basin is from December to April while
there is no record of snow from May to September, which is due to the abrupt air temperature
rise in spring. Furthermore, the difference between the snow separation methods is analyzed by
comparing two successive images of the basin, taken by different sensors on 5th November
2003. One of the images contains channel 3B which includes thermal infrared band and the
other contains channel 3A that scans near infrared wavelengths. Accordingly, the SCA of
AVHRR-3 sensor which contains channels 3A has been calculated 4% more than the SCA of
AVHRR-2 which records channel 3B. Moreover, the result of applying trend detection tests
shows that the SCA time series has no evident linear or monotonic trend.
Conclusion
The trend analysis on the SCA dataset has demonstrated that no significant statistic trend exists
in the SCA time series. Moreover, the difference between calculated values of the SCA derived
from two different AVHRR-2 and AVHRR-3 sensors does not affect the reliability of the SCA
dataset, considering the area of the basin. Hence, as a representative of the snow in Shahcheraghi basin, it is possible to consider the calculated snow cover area data as an appropriate input for hydrologic flow forecasting models.