Evaluation and Mapping of Snow Characteristics Using Remote Sensing Data in Astore River Basin, Pakistan

Abstract

Being an agricultural country, Pakistan requires lots of water for irrigation. A major portion of its water resources is located in the upper indus basin (UIB). The snowmelt runoff generated from high-altitude areas of the UIB provides inflow into the Indus river system that boosts the water supply. Snow accumulation during the winter period in the highlands in the watershed(s) becomes a source of water inflow during the snow-melting period, which is described according to characteristics like snow depth, snow density, and snow water equivalent. Snowmelt water release (SWE) and snowmelt water depth (SD) maps are generated by tracing snow occurrence from MODIS-based images of the snow-cover area, evaluating the heating degree days (HDDs) from MODIS-derived images of the land surface temperature, computing the solar radiation, and then assimilating all the previous data in the form of the snowmelt model and ground measurements of the snowmelt water release (SWE). The results show that the average snow-cover area in the Astore river basin, in the upper indus basin, ranges from 94% in winter to 20% in summer. The maps reveal that the annual average values of the SWE range from 150 mm to 535 mm, and the SD values range from 600 mm to 2135 mm, for the snowmelt period (April–September) over the years 2010–2020. The areas linked with vegetation experience low SWE accumulation because of the low slopes in the elevated regions. The meteorological parameters and basin characteristics affect the SWE and can determine the SD values.

1. Introduction

In natural water reservoirs, snow plays a critical role in the high-latitude hydrological cycle [1]. Snow accumulation in high-latitude regions during winter is characterized by measurable properties, like the snowmelt water depth, density, and snowmelt water release (SWE), defining the snowpack characteristics [2]. It is crucial to evaluate snow properties on a spatiotemporal scale in order to calculate the amount of runoff produced by snowmelt and in order to distribute water for various purposes [3]. Snowmelt runoff recharges groundwater aquifers and surface water reservoirs, providing a vital water source for urban and agricultural areas [4,5]. In the winter, the snowmelt water depth and snowmelt water release are essential factors to consider when characterizing the attributes of accumulated snow that leads to springtime water discharge [6]. SWE is the liquid depth of melted snow and, thus, is critically important in water resources management. SWE is difficult and time consuming to measure [7]. The snowmelt water depth can be quickly and easily assessed in the field compared to SWE. Although a precise global count of snowmelt water depth and SWE measurements is lacking, the available data indicate a significantly higher number of snowmelt water depth measurements being conducted in comparison to SWE measurements [8]. The Astore river basin, a sub-catchment of the upper indus basin, is a critical region regarding water resources in Pakistan. Precipitation in the basin is primarily driven by westerly disturbances during winter and the monsoon system in summer, contributing significantly to snow accumulation in high-altitude areas. Snowmelt from the basin is a vital source of water for irrigation, agriculture, and domestic use, particularly during the dry season [9]. However, climate change poses a significant threat to snow dynamics in the region, with potential impacts on the extent of snow cover, melt timing, and water availability. However, snow cover data alone cannot provide precise information on crucial factors like snowmelt water release (SWE), snowmelt water depth (SD), and snow density.

There are various methods in the literature for estimating the spatiotemporal distribution of SWE [5,7,8,9,10]. A combination of field measurements, remote sensing, and modeling was used to map the spatiotemporal variability of SWE and the snow-cover area in the Langtang Valley of Nepal. The study showed that remote sensing data can provide accurate information on the snow-cover area, while modeling is useful for estimating the SWE. Sturm et al. [8] developed a novel approach to estimating the SWE and snowmelt water depth, using ground-based cosmic ray sensors. The study demonstrated the effectiveness of the method in the Hindukush Karakoram Himalaya (HKH) region, where traditional snow measurement techniques are limited. Zheng et al. [11] used a combination of remote sensing and machine learning techniques to map the snowmelt water depth in the Tianshan Mountains of China, which shares similar characteristics with the HKH region. The study showed that the method can accurately estimate the snowmelt water depth, with potential for application in the HKH region. Li et al. [12] developed a hybrid approach to map the SWE and snowmelt water depth using remote sensing data, machine learning algorithms, and ground-based observations, in the upper indus basin of Pakistan. The study demonstrated the effectiveness of the approach when applied to a complex terrain like the HKH region. Li at al. [13] developed a data assimilation framework to map the SWE and snowmelt water depth in the upstream region of the Yellow River in China. The study showed that the method can significantly improve the accuracy of snow-melt water depth estimates, with potential for application in the HKH region.

There are many techniques, such as remote sensing, ground-based measurements, snow modeling, and machine learning, which have been used to estimate the SWE and SD [9,12,14,15,16,17]. However, the limitations of remote sensing techniques include difficulty in distinguishing between snow and ice, as well as challenges related to quantifying the density and water content of the snowpack [18]. The limited spatial coverage and accessibility, and the need for intensive labor and the high cost, are major limitations of ground-based measurements [18,19,20]. Model performance is highly dependent on the accuracy of the input data, and model uncertainty can be considerable, due to the complex interactions between snow processes and the environment [21]. Machine learning techniques, while powerful, come with several limitations. These include the requirement for large amounts of high-quality training data to ensure accurate model performance, the difficulty in interpreting model outputs due to their black-box nature, and their sensitivity to the selection of the input variables, which can significantly impact the results. Additionally, machine learning models may struggle with generalizing new or unseen data, especially in complex environments like those involving snow hydrology, wherein the spatial and temporal variability is high [1,20,22]. The data assimilation technique used in the current study is not explored in the studies mentioned above. However, data assimilation techniques have the potential to improve the accuracy of SWE and snowmelt water depth mapping by integrating remote sensing data with in-situ measurements and models [2,9,23,24].

This study introduces a novel data assimilation framework to improve SWE and SD estimation in the Astore river basin, wherein observational data are limited. This innovative approach provides deeper insights into snow dynamics and aids water resource management amidst evolving climate conditions. The main objectives of this study are: (a) to determine the snow-cover area using the normalized difference snow index (NDSI), (b) to create snowmelt water release (SWE) and snowmelt water depth (SD) maps during the snow-melting period, and (c) to assess the relationships between heating degree days (HDDs) and solar radiation (R) and snowmelt water release. The objective of analyzing snow cover was to identify the snowmelt period. The accurate estimation of the SWE and SD parameters will be helpful to understand the hydrological process in high mountain regions.

2. Materials and Methods

2.1. Study Area

The Astore river basin is one of the upper indus basin’s sub-catchments and is located on the western side of the Himalayas, with geographic coordinates ranging from 74°25′06″–75°14′16″ E to 34°46′34″–35°38′38″ N, and a height varying between 1198 and 8069 m (Figure 1). Only 5% of the Astore river basin is above 5000 masl. The total drainage area of the Astor river basin is 3990 km². In the Astore river basin, three meteorological stations, named Rama, Rattu, and Astore, are installed at various heights. According to the available data, the average annual rainfall measured at the Rama, Rattu, and Astore stations is 794 mm, 723 mm, and 501 mm, respectively [25]. The Astore river basin is temperate and experiences low temperatures all year round, apart from July and August, when temperatures can reach 27 degrees Celsius. Due to northern winds during the winter, the temperature drops even further, frequently to as low as −25 °C [26]. During the period 1998–2012, the mean average temperature at the station in Rattu valley (2718 m) was 9.9 °C, whereas the temperature at the station in Rama, which is located at a higher elevation, was 2.9 °C (3179 m). The precipitation gradient in the Astore river basin is 19.8 mm per 100 m [26].

2.2. Data Collection

In this study, the remote sensing data from the MODIS products were downloaded, and the observational data were gathered from Water and Power Development Authority (WAPDA) and the Pakistan Meteorological Department (PMD). A description of the data is provided in

Table 1. Description of the data used in the study.

Source	Data	Description
NASA	MODIS 8-day snow cover images, with a 500 m spatial resolution (MOD10A2).	These images were used to calculate the snow-cover area.
NASA	MODIS 8-day land surface temperature (LST), with a 1 km spatial resolution (MOD11A2).	This product was used to calculate the heating degree days.
WAPDA	Solar radiation on a daily basis for the period of 2010–2020 at the Rama and Rattu stations.	This information was used to generate SWE and SD maps and develop them in relation to the HDDs.
PMD	Solar radiation on a daily basis for the period of 2010–2020 at Astore Station.
Astore river basin	Snow weight and snow depth data of fresh and settled snow samples at 26 locations.	These data were used to calibrate and validate the model.

. Snow depth and snow weight measurements were taken at 26 locations from different areas (Rama meadows, Rattu meadows, Garikot, Challam, and Parjot) of the Astore river basin to calculate the snow density (Figure 2). A GPS and snow density kit were used to measure the snow depth and weight of the sample. Ground-based measurements of the snow depth and snow weight were used to calculate the observed SWE to calibrate and validate the model. The observed SWE was calculated as a function of the snow depth ($d_{snow})$ and snow density (${\rho }_{snow})$, using the following gravimetric method [5,9]:

$$\mathrm{S}\mathrm{W}\mathrm{E}=d_{snow}\left({\displaystyle \frac{{\rho }_{snow}}{{\rho }_{water}}}\right)$$

(1)

2.3. Methodology

An overall schematic diagram of the methods employed for the modeling conducted in this paper is presented in Figure 3.

2.3.1. Snow Cover

The Moderate Resolution Imaging Spectroradiometer (MODIS) Terra product MOD10A2 provided the available snow-cover images, with a 500 m spatial and 8-day temporal resolution [27]. The data set consists of 455 images that were downloaded from the National Snow and Ice Data Center (NSIDC) https://nsidc.org/data/data-access-tool/MOD10A2/versions/61/ (accessed on 25 June 2021). The MODIS data were converted to GEO—TIFF format from HDF-EOS format, so that they could be further processed using ArcGIS 10.2. The MODIS tiles with a snow count percentage greater than 70%, relative to cloud cover, were selected for further analysis [25]. The area of one snow count was multiplied by the total number of counts to get the total snow-covered area in the basin [28]. The processed snow-cover images were also used to determine the snow-melting period in the Astore river basin for the SWE and SD maps.

2.3.2. Heating Degree Days (HDDs)

The land surface temperature (LST) is the key factor regarding calculating the heating degree days (HDDs). The LST 8-day data was retrieved with a resolution of 1 km pixels. The 450+ images downloaded from MODIS LST version 6 provide a pixel-by-pixel temperature of the area in Kelvin. The heating degree days (HDDs) were calculated using Moderate Resolution Imaging Spectroradiometer (MODIS)-based land surface temperature images in the ArcGIS environment. The following equation was used in the Raster calculator:

$$\mathrm{H}\mathrm{D}\mathrm{D}=T-T_{base}$$

(2)

where T is the satellite-based temperature, which is calculated by multiplying the LST images with a conversion factor of 0.02 [29]. The temperature 273.15 K (0 °C), over which snowmelt begins, is denoted as T_base [2,30]. The heating degree days were calculated for the snow-melting period, which starts in April and lasts until September, determined via the processing of snow-cover images.

2.3.3. Model Setup

The model (Equation (3)) was used to generate SWE and SD maps, by combining the images of the LST and snow cover, ground-based measurements, and solar radiation [14,31].

$$\mathrm{S}\mathrm{W}\mathrm{E}=\mathrm{a}\times \mathsf{\Sigma }\mathrm{H}\mathrm{D}\mathrm{D}+\mathrm{b}\times \Sigma R$$

(3)

where the heating degree day coefficient ‘a’ was obtained from the ground-based measurements of the snow density, HDDs, and solar radiation; and ‘b’ is the energy to water depth conversion ratio (0.026 cm W⁻¹ cm⁻² day⁻¹) [2,32]. R is the solar radiation during the snowmelt period. In Equation (3), SWE represents the water equivalent of the melted snow, reflecting the total amount of water released during the snowmelt process.

To calibrate and validate the model, the value of ‘a’ was calculated for the year 2020 from the observed SWE, HDDs, and R. The model (Equation (3)) was calibrated and validated using a total of twenty-six ground-based measurements. Ten measurements were used to calibrate the model (Equation (3)), while the remaining sixteen measurements were used to validate it. For the value of 0.35 of the HDD coefficient ‘a’, the coefficient of determination (R²) for the observed and modelled SWE was 0.76 and 0.88 during the calibration and validation, respectively, as shown in Figure 4.

The snowmelt water release was divided by the snow density (Equation (4)) to get the snowmelt water depth for the Astore river basin.

$$d_{snow}={\displaystyle \frac{SWE\ast {\rho }_{water}}{{\rho }_{snow}}}$$

(4)

where ${\rho }_{water}$ is the density of water and ${\rho }_{snow}$ is the average snow density in the study area, which is 0.25 gcm⁻³. The snow density is not constant and varies spatially depending on the snow conditions. To account for this variability, twenty-six (26) in situ measurements of the SWE and SD were used to determine the average snow density for the study area. The SWE maps were generated using Equation (3) for each year and these maps were then converted into snowmelt water depth maps by applying the calculated average snow density. This approach ensures that the spatial variability in the snow conditions is appropriately represented.

3. Results and Discussion

3.1. Snow-Cover Dynamics in Astore River Basin

Snow cover is an important variable regarding freshwater availability in the dry season, particularly for high-altitude areas. Snow-cover dynamics play a vital role in the hydrological characteristics of the basin. The snow-cover area was estimated from the MODIS-derived snow-cover images during the period 2010–2020. The accuracy of MODIS snow-cover products is reported to be more than 95% for mountainous regions [33,34]. The average snow-cover area varies from ≈95% in winter to ≈20% in summer [35]. It is also reported that the extent of the snow cover is high in winter and spring and low in summer. The annual cycle of the snow-cover area (SCA) for the data period (2010–2020), with a line plotting the average values, for the Astore river basin, is shown in Figure 5. The accumulation of snow starts in September and the maximum snow cover reaches the range of 90–94% in January or February. The melting of the snow starts in April, and the minimum snow cover was 15–20% of the basin area during August (Figure 6).

Refs. [36,37] also presented the same results, namely that the extent of the snow-cover area increases, particularly in the months of January and February, and achieves its minimum value in the month of August, in the Astore river basin and the northern areas of Pakistan. Similar findings were reported by [25,26,38], highlighting the climatic conditions, such as lower temperatures and increased precipitation during the winter months, that significantly contribute to the expansion of snow cover in high-altitude regions.

Table 2. Mean and percentage monthly snow-cover area.

Sr. No.	Month	Snow-Cover Area (km2)	Snow-Cover Area (%)
1	January	3718	93
2	February	3767	94
3	March	3767	94
4	April	3657	91
5	May	3206	80
6	June	2608	65
7	July	1591	39
8	August	800	20
9	September	1591	39
10	October	2740	68
11	November	3540	88
12	December	3695	92

presents the monthly average snow-cover area and % of the SCA in the Astore river basin for the data period (2010–2020). The maximum SCA is 94% during the months of February and March and the minimum SCA is 20% in the month of August.

3.2. Snowmelt Water Release and Depth Maps

The HDD coefficient ‘a’ was iteratively adjusted to achieve the best agreement between the modeled and observed SWE values, with an optimal value of 0.35 determined through the calibration process. The variation in ‘a’ across different years reflects inter-annual variability in the snow properties (e.g., density and depth) and climatic conditions (e.g., temperature regimes, cloudiness, and precipitation patterns). These factors underscore the sensitivity of empirical parameters in regard to local snow and climate dynamics, which were carefully accounted for during the model calibration process. The SWE and SD maps were generated in this study through the implementation of the energy-inflow based snowmelt modeling method. The HDD coefficient ‘a’ was determined using the observed SWE, HDDs, and R from Equation (3). This coefficient, along with the solar radiation and HDDs, was then used to generate the SWE map. The snowmelt water release within the Astore basin varies from <200 mm at low altitudes to >500 mm for higher altitudes, which are mostly glaciated. Figure 7 shows the snowmelt water release distribution during the snowmelt period, with higher values (400–500 mm) concentrated in glaciated, high-elevation regions (6.9% of the basin), while the majority of the basin (over 45%) falls within the 200–300 mm range, consistent with lower elevations. More than 95% of the total basin area lies within this range. The SWE that occurs below 200 mm is ≤1% of the total, whereas the SWE that occurs at 500 mm is <0.5% of the total in every year investigated. The aerial distribution of the snowmelt water release in the Astore basin is almost the same every year, i.e., low SWE at low elevations and high SWE at higher elevations. The accumulation of SWE is affected more strongly by the elevation and slope of the terrain. While lower elevations are snow free in late summer, the averaging process captures the cumulative effect of snowmelt contributions throughout the period, leading to apparent high values in these regions. This is consistent with the dynamics of snowmelt processes, where meltwater from higher elevations can accumulate in lower areas during the early melt season. The areas with vegetation experienced low SWE accumulations because of the low slopes in these elevated areas. The spatial and temporal variations in the SWE and SD are influenced by climatic factors, such as the amount of snowfall and temperature. Increased snowfall during winter enhances SWE accumulation, while rising temperatures during the snowmelt period accelerate snow depletion, particularly at lower elevations. Additionally, topographic factors, including the slope aspect and elevation, play a significant role in snow distribution. North-facing slopes retain snow longer due to reduced solar radiation, while south-facing slopes experience faster melt rates. The snowmelt water release depth was estimated as a function of the SWE from Equation (4). The snowmelt water depth in the Astore basin varies from less than 600 mm at lower altitudes to >2000 mm at higher altitudes. The snowmelt water depth (SD) maps (2010–2020) reveal that the Astore basin is predominantly characterized by shallow-to-moderate snowmelt water depths, with >95% of the basin area falling within the 200–500 mm range annually (Figure 8). Extreme SD values (<600 mm) are spatially limited, covering <5% of the basin, and depths exceeding 2000 mm are exceptionally rare (with <0.5% coverage) and typically confined to high-elevation accumulation zones.

Figure 9 shows the average annual SWE and SD, including the percentage of the basin area, based on the data for the years 2010–2020. The average SWE ranged from 150 mm to 535 mm and the SD ranged from 600 mm to 2135 mm. The distribution in terms of the study area in regard to the various ranges for the SWE and SD was almost the same, allowing it to be easily observed that the snowmelt water depth was a function of the SWE (Figure 10 and Figure 11). The relationship between the SWE and SD is mathematically straightforward when using a constant snow density (0.25 g/cm³), as in Equation (4). The generation of SD maps remains important for practical applications. The snowmelt water depth (SD) is a directly measurable parameter in the field, making it valuable for validation and operational use, such as snowpack monitoring and water resource management. The difference in the coverage of snowmelt water release is affected by the plant type and cover, gradient, wind speed, atmospheric temperature, relative humidity, and precipitation in the basin. Figure 10 and Figure 11 show the snowmelt water release (SWE) and snowmelt water depth (SD) during the snow-melting period. The white areas show the region with no snow. The monthly variation in the SWE and SD shows that snow accumulation starts in the winter months, but it is prolonged until the snow-melting period. In the early months of summer, snow accumulation at a higher altitude generated high amounts of SWE and SD. Moreover, not all snowmelt necessarily occurs during the snow-melting period (April–September). The snowmelt water depth is a more practical parameter for field applications, such as avalanche forecasting and snowpack monitoring, despite being directly proportional to the SWE. The relatively uniform spatial distribution of the snowmelt water depth values in April and September can be explained by examining the meltwater release dynamics. During these transition months, when heating degree days (HDDs) are minimal, even at lower elevations, snowmelt is primarily driven by solar radiation, which tends to be more uniformly distributed across the basin. This results in less spatial variability in the snowmelt water depth compared to peak-melt months.

To determine the correlation between HDDs and solar radiation, a linear regression analysis was employed. This research demonstrated that throughout the snow-melting season, there was an almost direct positive relationship between the heating degree days (HDDs) and solar radiation (R). In this study, the HDDs and solar radiation increased during the months of April–June and then started to decrease from July-September, with a correlation coefficient (R²) of 0.74, as shown in Figure 12. The snowmelt water release was directly affected by the increase or decrease in energy inputs. The increase in heating degree days and solar radiation initiate snow melting and generate more runoff. Snow disappearance may be visible on the SWE and SD maps because of the increase in incoming energy in the form of HDDs and R and snow remains at higher elevations.

4. Conclusions

The SWE plays an important role in the simulation and forecasting of spring runoff and probable floods from snowmelt. However, due to the difficulty in measuring snow data in cold weather and at high elevations, the application of snow estimates provided by assimilated systems involving hydrological modelling has gained interest. In this study, we demonstrated the construction of the SWE and snowmelt water depth of snow cover in the Astore river basin, in the upper indus basin, by integrating MODIS-based observations, ground-based observations, (snow weight and depth), and solar radiation levels. In this study, we evaluated the application of the energy-based snowmelt method, with suitable calibration and validation processes. Then, this calibrated and validated model was used to generate maps of the SWE and SD for the study area. In regard to this application, the snow properties during the field measurements were analyzed in detail and the density of the snow in the study area was correctly estimated. In this research, we assumed a uniform single layer of snowpack for the snowmelt water depth estimation. From the abovementioned results, conclusions are drawn that the average snow cover area in the Astore river basin ranges from 94% in winter to 20% in summer. The SWE and SD maps were generated for the snowmelt period (April–September) over the years 2010–2020. The average SWE values for this period range from 150 mm to 535 mm, while the average SD values range from 600 mm to 2135 mm. The study area-specific average values of the SWE and SD during the snowmelt period were found to be 42—3 mm and 1691 mm, respectively. These results reflect the spatial and temporal variability of the snow across the Astore river basin, with higher values typically observed in elevated regions. The methods described in this study have potential to be used to construct SWE maps at the landscape level. A good estimation of SWE is of great importance for short- or long-term spring runoff forecasting to support water management, flood forecasting, and reservoir operations.