Integration of Snowmelt Runoff Model (SRM) with GIS and Remote Sensing for Operational Forecasting in the Kırkgöze Watershed, Turkey

Abstract

Accurate snowmelt runoff prediction is critical for water resource management in mountainous regions where seasonal snowpack constitutes the dominant water supply. This study demonstrates operational application of the degree-day-based Snowmelt Runoff Model (SRM) integrated with Geographic Information Systems (GIS) and multi-platform remote sensing for discharge forecasting in the Kirkgoze Basin (242.7 km², 1823–3140 m elevation), Eastern Anatolia, Turkey. Three automatic weather stations spanning 872 m elevation gradient provided meteorological forcing, while MODIS MOD10A2 8-day composite products supplied operational snow cover observations validated against Landsat-5/7 (30 m resolution, 87.3% agreement, Kappa = 0.73) and synthetic aperture radar imagery (RADARSAT-1 C-band, ALOS-PALSAR L-band). Uncalibrated model performance was modest (R² = 0.384, volumetric difference = 29.78%), demonstrating necessity of site-specific calibration. Systematic adjustment of snowmelt and rainfall runoff coefficients yielded excellent calibrated performance for 2009 melt season: R² = 0.8606, correlation coefficient R = 0.927, Nash–Sutcliffe efficiency = 0.854, and volumetric difference = 3.35%. Enhanced temperature lapse rate (0.75 °C/100 m vs. standard 0.65 °C/100 m) reflected severe continental climate. Multiple linear regression analysis identified temperature, snow-covered area, snow water equivalent, and calibrated runoff coefficients as significant discharge predictors (R² = 0.881). Results confirm SRM’s operational feasibility for seasonal forecasting and flood warning in data-scarce snow-dominated basins, with modest requirements (daily temperature, precipitation, and satellite snow cover) aligning with operational monitoring capabilities. The methodology provides a transferable framework for regional water resource management in climatically vulnerable mountain environments where snowmelt supports agriculture, hydropower, and municipal supply.

1. Introduction

Mountain snowpack constitutes a critical natural reservoir in the global hydrological cycle, storing winter precipitation and releasing meltwater during spring and summer to sustain downstream water supply, agricultural production, hydropower generation, and ecosystem services [1,2]. Mountain snowpack provides critical water resources for over one-sixth of the global population [1,2], with mountainous regions serving as “water towers” that supply water far beyond their geographic extent [2]. Climate warming poses particular threats to Asian mountain water resources [3]. Approximately one-sixth of the global population depends on seasonal snowmelt for water resources [1,2], with mountainous regions serving as “water towers” supplying disproportionately large volumes relative to their geographic extent [2,3]. Climate warming disproportionately affects snow-dominated hydrological systems through multiple mechanisms: declining snow accumulation as precipitation shifts from snow to rain, accelerated melt rates, earlier snowmelt timing, reduced snowpack duration, and diminished peak snow water equivalent [1,4,5]. Observational evidence from western North America demonstrates substantial snowpack declines over recent decades despite stable or increasing total precipitation, attributed primarily to temperature-driven phase shifts [5,6]. Observed snowpack declines in western North America demonstrate the severity of climate impacts, despite stable precipitation levels [5,6]. Paradoxically, warming may slow snowmelt rates in certain regions through reduced solar radiation absorption by aging snowpack and increased cloudiness during melt periods [7], though overall trends indicate earlier and more compressed melt seasons with heightened flood risk and water management challenges [8]. These evolving patterns underscore urgent need for accurate operational snowmelt forecasting to support adaptive water resource management, flood risk mitigation, and climate change adaptation planning in snow-dependent regions globally.

The Snowmelt Runoff Model (SRM), originally developed by Martinec in the 1970s for small Alpine basins [9], has evolved into one of the most widely applied operational snowmelt forecasting tools globally. The model employs a degree-day approach relating snowmelt to positive air temperature, balancing physical realism with pragmatic data requirements suitable for operational implementation [10]. Extensive validation through World Meteorological Organization intercomparison studies confirmed SRM’s robust performance across diverse physiographic and climatic settings [11], establishing it as a reference benchmark for snowmelt modeling. To date, SRM has been successfully applied to over 100 basins spanning 29 countries, with drainage areas ranging from <1 km² to >900,000 km² (Ganges River Basin) and elevation ranges encompassing 0–8848 m [12]. Recent comprehensive reviews highlight SRM’s continued relevance for operational forecasting while identifying opportunities for enhancement through data assimilation, uncertainty quantification, and integration with climate projection frameworks [12]. Regional applications demonstrate model versatility across mountain environments, from humid maritime climates to continental settings, including successful implementations in Central Asian highlands [13,14], though performance depends critically on appropriate parameterization for local conditions and availability of representative snow cover observations.

Climate change poses fundamental challenges to snowmelt-dominated hydrological systems globally [4], with widespread snowpack declines observed across major mountain ranges [5,6] and projected warming trends directly affecting snow accumulation patterns, snowpack persistence, and melt timing. Rising temperatures reduce the proportion of precipitation falling as snow, accelerate earlier snowmelt onset, compress melt season duration, and shift peak runoff timing toward winter and early spring months. These changes substantially complicate water resource management, increasing flood risk during early-season melt events while reducing late-summer water availability when agricultural and ecological demands peak. Mountainous regions are experiencing disproportionately rapid warming compared to global averages—a phenomenon termed elevation-dependent warming—with documented temperature increases of 0.3–0.5 °C per decade in high-elevation areas exceeding twice the global mean warming rate. In the Eastern Mediterranean and Middle East region, climate projections indicate 20–40% reductions in snow cover extent and 30–50% decreases in snowmelt-derived runoff by mid-century under moderate emission scenarios, threatening water security for populations dependent on snow-dominated water resources. These projected changes underscore the critical importance of developing robust operational snowmelt forecasting capabilities that can adapt to shifting hydrological regimes and support proactive water resource management under non-stationary climatic conditions.

Integration of satellite remote sensing with snowmelt runoff modeling represents a transformative advancement enabling operational forecasting in data-scarce mountainous regions where ground-based snow observations remain sparse or absent. SRM’s explicit incorporation of spatially distributed snow-covered area as a primary input variable uniquely positions it to leverage remote sensing capabilities [15]. The Moderate-Resolution Imaging Spectroradiometer (MODIS) aboard Terra and Aqua satellites provides near-daily global snow cover mapping at 500 m resolution since 2000, with operational products validated extensively against higher-resolution imagery and ground observations [16]. Recent developments include continental-scale validation datasets exceeding 1.86 billion labeled pixels [17], machine-learning-based cloud gap-filling algorithms [18,19], and advanced spatiotemporal reconstruction techniques addressing persistent cloud contamination challenges [20]. Snow exhibits distinctive spectral characteristics in visible and shortwave infrared wavelengths enabling robust optical discrimination [21], with normalized difference snow index (NDSI) approaches widely adopted in operational snow mapping algorithms [22]. Synthetic aperture radar (SAR) provides complementary all-weather observation capability particularly valuable during cloudy periods, though wet/dry snow discrimination and quantitative snow water equivalent retrieval remain active research challenges. The convergence of operational satellite snow products, advancing algorithm development, and validated SRM modeling frameworks establishes a mature technological foundation for operational forecasting systems addressing growing water security concerns in snow-dependent regions.

The degree-day or temperature-index approach underlying SRM methodology represents a pragmatic balance between physical realism and operational data requirements. While simplified relative to comprehensive energy balance formulations [23], degree-day models capture dominant melt controls through air temperature as an integrated proxy for energy inputs [24,25]. Extensive validations demonstrate that appropriately parameterized temperature-index models achieve discharge prediction accuracy comparable to more complex approaches while requiring only routinely available meteorological observations [25]. The degree-day factor, quantifying snowmelt depth per degree-day above freezing temperature, exhibits systematic variations, with elevation, aspect, vegetation cover, and season reflecting spatial and temporal changes in energy balance components [10,24]. Enhanced-temperature-index formulations incorporating radiation and other meteorological variables improve performance particularly in data-rich applications [25], though operational forecasting contexts often prioritize model parsimony and robust parameter estimation over additional complexity. Integration of remotely sensed snow cover observations with degree-day modeling partially compensates for simplified physics by directly observing spatially distributed snowpack presence, enabling accurate discharge simulation despite simplified melt algorithms [26]. Alternative conceptual and physically based models, including HBV [27], provide comparative frameworks, though SRM’s explicit remote sensing integration and extensive operational validation record maintain its prominence for snow-dominated basin applications.

Eastern Anatolia constitutes Turkey’s most important snow water resource region, with mountainous topography (elevations exceeding 3000 m), severe continental climate, and heavy winter precipitation producing extensive seasonal snowpack sustaining critical downstream water supplies. The region experiences pronounced seasonal discharge regimes with 70–80% of annual runoff concentrated during March–June snowmelt period, creating intensive irrigation, hydropower, and municipal water demands concurrent with flood risk management challenges. Previous hydrological modeling efforts in the region established feasibility of remote-sensing-based approaches for snow cover monitoring [28] and demonstrated operational applicability of both SRM and alternative conceptual models [29] for seasonal forecasting applications. Detailed energy balance studies quantified local snowpack processes [30], providing physical insights supporting model parameterization for the region’s severe continental conditions. However, sparse operational monitoring networks, limited real-time data availability, and insufficient integration of emerging satellite products constrain development of robust operational forecasting systems despite demonstrated technical feasibility. Climate projections indicate substantial vulnerability of regional water resources to warming-driven snowpack decline, underscoring urgency of establishing operational forecasting capabilities supporting adaptive water management in this climatically stressed, water-dependent region.

This study demonstrates operational integration of SRM with multi-platform remote sensing and geographic information systems for accurate snowmelt runoff prediction in the Kırkgöze Basin, a representative mountainous watershed in Eastern Anatolia, Turkey. The research addresses critical gaps in operational snowmelt forecasting for data-scarce mountainous regions by combining ground-based meteorological observations with satellite-derived snow cover information. Specific research objectives include (1) establishing an operational meteorological monitoring network across the basin’s elevation gradient (1823–2815 m) to capture spatial temperature and precipitation variability; (2) integrating multi-platform satellite remote sensing (MODIS, Landsat-7 ETM+, RADARSAT-1 SAR, and ALOS-PALSAR) with GIS analysis to derive spatially distributed snow cover depletion patterns at 8-day temporal resolution; (3) calibrating and validating the three-zone SRM formulation for the 2009 snowmelt season using simultaneous ground observations and satellite snow cover data; (4) quantifying model performance through comprehensive statistical metrics, including correlation coefficient, Nash–Sutcliffe efficiency, and volumetric difference; (5) empirically determining elevation-dependent degree-day factors and temperature lapse rates appropriate for continental mountain climate conditions; and (6) developing empirical multiple linear regression relationships between meteorological forcings, snow cover characteristics, and basin discharge to identify dominant hydrological controls. The calibrated model framework and derived parameter relationships provide transferable insights for operational snowmelt forecasting applications in similar ungauged or data-limited mountainous watersheds characterized by continental climate regimes and substantial elevation gradients.

2. Materials and Methods

2.1. Study Area

The Kırkgoze Basin (242.7 km², 40°06′ N, 41°22′ E) is located in the mountainous northeastern region of Turkey, approximately 25 km north of Erzurum city within the Upper Karasu watershed (Figure 1a,b). The basin elevation ranges from 1823 m to 3140 m above sea level (elevation range 1317 m), encompassing alpine terrain with steep slopes and minimal vegetation cover dominated by alpine meadows and sparse shrubland. Basin morphology and hydrological characteristics are summarized in

Table 1. Kırkgöze Basin morphometric and hydrological characteristics. Precipitation data from Turkish State Meteorological Service (DMI) Erzurum station (1975–2009 average).

Characteristic	Value	Unit
Drainage area	242.7	km2
Elevation range	1823–3140	m a.s.l.
Mean elevation	2482	m a.s.l.
Maximum elevation	3140	m a.s.l.
Minimum elevation	1823	m a.s.l.
Relief	1317	m
Mean slope	18.3	degrees
Main channel length	22.4	km
Mean annual precipitation (1975–2009)	448	mm
Snowpack duration (outlet)	150–180	days/year
Peak discharge period	March–June	-

[31]. The Kırkgöze Basin has been the subject of previous detailed investigations of temperature-index snowmelt processes, establishing baseline understanding of local melt rate functions and antecedent temperature effects [32].

2.2. Meteorological Data Collection Network

Three automatic weather stations (AWS) were strategically installed across the basin’s elevation gradient to capture spatial meteorological variability (Figure 1b,

Table 2. Automatic weather station specifications and locations.

Station	Elevation (m)	Zone	Latitude	Longitude	Key Sensors
AWS-1	2019	A	40°07′ N	41°23′ E	T, P, RH, WS, SR, SD
AWS-2 (Radar)	2454	B	40°06′ N	41°22′ E	T, P, RH, WS, Pressure, SR, SD
AWS-3	2891	C	40°05′ N	41°21′ E	T, P, RH, WS, SR, SD
DSI 21-01	1823	Outlet	40°08′ N	41°24′ E	Discharge gauge

). Station elevations span an 872 m range (Station 1: 2019 m, Station 2: 2454 m, and Station 3: 2891 m), providing observations representative of each elevation zone. Each AWS recorded hourly measurements of air temperature, precipitation, relative humidity, wind speed/direction, atmospheric pressure, incoming shortwave radiation, and snow depth.

Precipitation measurements from heated tipping-bucket gauges required systematic adjustment for wind-induced undercatch, particularly problematic for snowfall. Following WMO guidelines and regional studies, snowfall catch efficiency was estimated at 75–80% (20–25% undercatch) based on Alter wind shield configuration and typical winter wind speeds (3–5 m/s). Rainfall catch efficiency was assumed to be 90–95% (5–10% undercatch). Monthly precipitation totals adjusted using elevation-dependent correction factors were as follows: 1.25 for snowfall (October–April) and 1.08 for rainfall (May–September), with smooth transitions during shoulder seasons. These adjustments increased seasonal precipitation input by approximately 18%, affecting calibrated runoff coefficient values, which partially compensate for input uncertainty. The application of constant correction factors across different seasons, while simplified relative to variable seasonal corrections, represents a pragmatic compromise justified by several considerations. Winter precipitation in the study region falls predominantly as dry snow under consistent meteorological conditions (temperatures −10 to −5 °C; wind speeds 3–5 m/s), producing relatively stable undercatch percentages throughout the accumulation season. Spring rainfall events occur under milder conditions (5–15 °C) with generally lower wind speeds, resulting in less variable catch deficiencies. Regional calibration studies in similar continental mountain climates have demonstrated that seasonal variability in catch efficiency (±5–8% standard deviation within season) is substantially smaller than the mean seasonal undercatch magnitude (20–25% for snow; 5–10% for rain), suggesting that constant seasonal factors capture the dominant bias correction while avoiding overfitting to limited gauge comparison data. More sophisticated approaches employing precipitation-type-dependent and wind-speed-dependent dynamic corrections would require higher-density meteorological networks than available for this study basin and introduce additional parameter uncertainty potentially exceeding the improvement in catch estimation accuracy. Sensitivity analysis indicated that ±5% variation in correction factors produced <3% change in seasonal runoff volume, well within acceptable uncertainty bounds for operational forecasting applications.

2.3. Satellite Remote Sensing Data

Snow cover observations employed a multi-platform satellite remote sensing approach combining operational MODIS products with higher-resolution validation imagery and all-weather synthetic aperture radar observations (Figure 2a–d,

Table 3. Satellite remote sensing data sources for operational snow monitoring.

Platform	Sensor	Resolution	Temporal	Purpose
Terra	MODIS MOD10A2	500 m	8-day	Operational SCA
Landsat-5	TM	30 m	16-day	Validation
Landsat-7	ETM+	30 m	16-day	Validation
RADARSAT-1	C-band SAR	25 m	24-day	All-weather SCA
ALOS	PALSAR L-band	10 m	46-day	All-weather SCA

). MODIS MOD10A2 8-day maximum snow extent composites (500 m resolution) from Terra satellite provided primary operational snow cover time series [32,33]. Landsat-5/7 imagery (30 m resolution) provided independent validation [34]. Synthetic aperture radar from RADARSAT-1 (C-band, 25 m) and ALOS-PALSAR (L-band, 10 m) complemented optical observations [35,36].

SAR image processing involved standard preprocessing, including radiometric calibration converting digital numbers to sigma-naught backscatter values (σ⁰), geometric terrain correction using 30 m SRTM digital elevation model accounting for topographic distortions in mountainous terrain, and speckle filtering (5 × 5 Lee adaptive filter) reducing multiplicative noise while preserving edges and backscatter gradients critical for snow boundary delineation. Wet snow discrimination employed supervised classification based on characteristic backscatter reduction occurring when liquid water content increases snowpack dielectric constant, causing enhanced microwave absorption. The classification algorithm utilized threshold-based decision trees: (1) radiometric normalization to account for incidence angle variations across the swath (local incidence angle correction using DEM), (2) temporal differencing between winter baseline (dry snow) and spring acquisitions to identify backscatter decrease magnitude, (3) threshold determination through training area analysis comparing SAR signatures with contemporaneous Landsat optical imagery snow classifications, establishing wet snow threshold at σ⁰ < −12 dB for C-band and σ⁰ < −10 dB for L-band, corresponding to 4–7 dB reduction relative to dry snow conditions, and (4) spatial filtering to remove isolated pixels and enforce minimum mapping unit (9 pixels = 0.2 ha). RADARSAT-1 C-band (5.3 GHz, wavelength 5.6 cm) provided high sensitivity to surface wetness but limited penetration depth (~2–5 cm in wet snow), while ALOS-PALSAR L-band (1.27 GHz, wavelength 23.6 cm) exhibited enhanced penetration through dry snowpack (10–50 cm depth) enabling improved snow–ground interface detection and wet snow volume estimation beneath surface dry layers. Classification accuracy was validated against Landsat-derived snow maps for cloud-free dates, achieving 87% overall accuracy for wet/dry snow discrimination and 82% accuracy for snow/no-snow boundary delineation. The complementary use of C-band and L-band frequencies addressed limitations of single-frequency approaches: C-band excels at detecting surface melt onset but saturates in deep wet snow, while L-band maintains sensitivity through thicker snowpack but shows reduced surface wetness response.

2.4. Snowmelt Runoff Model Description and Implementation

The Snowmelt Runoff Model (SRM), originally developed by Martinec [9] and extensively validated through World Meteorological Organization intercomparison studies [7], is a conceptual, deterministic, degree-day-based model designed specifically for mountainous basins where snowmelt dominates runoff generation. The model has been successfully applied to over 100 basins across 29 countries, with drainage areas ranging from <1 km² to >900,000 km² (Ganges River) and elevation ranges spanning 0–8848 m [14]. The degree-day approach, while simpler than full energy balance formulations, provides pragmatic operational utility with modest data requirements while maintaining physically meaningful parameter values [6,10]. Previous application to the Kirkgoze Basin by Yerdelen [37] demonstrated model feasibility, which this study extends through integration with modern remote sensing products and enhanced ground observation networks.

For single elevation zone applications, daily discharge Qn (m³/s) on day n is computed as [14]:

Q_n = [C_s × a × (T_n + ΔT) × S_n + C_r × P_n] × (A × 10⁴/86,400) × (1 − k_n) + Q_n−1 × k_n

(1)

where

• Q_n = daily mean discharge on day n (m³/s);
• C_s = snowmelt runoff coefficient (dimensionless, 0–1);
• C_r = rainfall runoff coefficient (dimensionless, 0–1);
• a = degree-day factor (cm/°C·day);
• T_n = degree-days (°C·day above 0 °C);
• ΔT = temperature adjustment to zone elevation (°C);
• S_n = snow-covered area ratio (0–1, dimensionless);
• P_n = daily precipitation contributing to runoff (cm);
• A = basin or zone area (km²);
• k_n = recession coefficient (0–1, dimensionless);
• 10⁴/86,400 = unit conversion factor (cm·km²/day to m³/s).

For basins with elevation range exceeding 500 m, SRM employs multi-zone extension to account for altitudinal variations in temperature, snowpack properties, and melt timing [14]. The Kirkgoze Basin (elevation range of 1317 m) was subdivided into three approximately equal zones (A, B, and C) of ~500 m each, with discharge computed as:

Q_n = {[C_s^A × a^A × (T^A_n + ΔT^A) × S^A_n + C_r^A × P^A_n] × (A^A × 10⁴/86,400)
+ [C_s^B × a^B × (T^B_n + ΔT^B) × S^B_n + C_r^B × P^B_n] × (A^B × 10⁴/86,400)
+ [C_s^c × a^c × (T^c_n + ΔT^c) × S^c_n + C_r^c × P^c_n] × (A^c × 10⁴/86,400)} × (1 − k_n) + Q_n−1 × k_n

(2)

Temperature adjustment ΔT accounts for elevation difference between meteorological station and zone hypsometric mean elevation using site-specific lapse rate γ (°C/100 m) [14]:

ΔT = γ × (H_station − H_γone)/100

(3)

The temperature lapse rate was empirically derived from simultaneous observations at the three automatic weather stations spanning 872 m elevation difference, yielding γ = 0.75 °C/100 m (R² = 0.94, p < 0.001), substantially exceeding the standard environmental lapse rate of 0.65 °C/100 m typically assumed in mountain hydrology [14]. This enhanced gradient reflects Eastern Anatolia’s severe continental climate characterized by low atmospheric humidity and minimal cloud cover.

The recession coefficient k_n was determined from semilogarithmic analysis of hydrograph recession curves following storm events when neither snowmelt nor rainfall occurred [15]. Plotting Q_n+1 versus Qn for recession periods and fitting the lower envelope yields the discharge-dependent relationship k_n = x·Q_n^−y + constant, where x and y are empirically determined constants. For the Kirkgoze Basin, analysis yielded x = 0.7787 and y = 0.036, indicating a recession coefficient ranging from ~0.81 at high flow to ~0.85 at baseflow conditions.

SRM Calibration Methodology

Systematic SRM calibration employed a manual trial-and-error approach guided by visual hydrograph inspection and quantitative performance metrics. The Nash–Sutcliffe efficiency (NSE) served as a primary objective function, supplemented by coefficient of determination (R²), Pearson correlation coefficient (R), volumetric difference percentage, and root mean square error (RMSE) for comprehensive performance assessment. Parameter adjustment followed iterative refinement strategy beginning with degree-day factors (a), progressing through runoff coefficients (CS and CR), and concluding with recession coefficient (k) fine-tuning.

Initial parameter estimates were derived from literature values appropriate for continental mountain climates: degree-day factors 0.20–0.50 cm/°C·day based on elevation and season, snowmelt runoff coefficients 0.50–0.80 reflecting progressive soil saturation, rainfall runoff coefficients 0.30–0.70, and recession coefficient k = 0.70–0.85 from baseflow recession analysis. Parameter bounds constrained within physically realistic ranges as follows: degree-day factors 0.10–1.00 cm/°C·day, runoff coefficients 0.20–1.00, and recession coefficient 0.50–0.95.

Calibration utilized the complete 2009 snowmelt season (30 March–2 June, 65 days) without split-sample validation due to single-year data availability constraint. This limitation acknowledged, parameter values remained within expected physical ranges and achieved performance metrics (NSE = 0.854) consistent with successful SRM applications globally. Future work should validate parameters across multiple contrasting water years to assess temporal stability and transferability.

2.5. Use of Generative Artificial Intelligence Tools

During the preparation of this manuscript, the authors used Claude (Anthropic, version Claude Sonnet 4.5, accessed 16 December 2025, https://claude.ai) to improve language clarity, refine technical descriptions, and enhance the structure of certain sections. Specifically, Claude assisted with (1) improving grammatical accuracy and sentence flow in the Section 1 and Section 4, (2) refining technical terminology and ensuring consistency in the presentation of methodological details, and (3) suggesting organizational improvements for enhanced readability. All content generated or modified using Claude was carefully reviewed, verified for technical accuracy against original research data, and edited by the authors as needed. The authors take full responsibility for the scientific content, interpretations, conclusions, and accuracy of all information presented in this published article. Claude was not used for data analysis, interpretation of results, formulation of conclusions, or generation of original research content.

3. Results

3.1. Meteorological Conditions

The 2009 snowmelt season exhibited representative meteorological conditions (

Table 4. Mean monthly meteorological conditions at Station 2 (2454 m), 2009 snowmelt season.

Month	Temp (°C)	Precip (mm)	Snow Depth (cm)	Radiation (W/m2)
March	1.2	58	142	185
April	5.8	67	89	235
May	9.4	43	24	285
June	11.4	19	0	310
Season Mean	5.8	187	-	254

). Temperature lapse rate averaged 0.75 °C/100 m (R² = 0.94), exceeding standard value by 15%.

3.2. Snow Cover Depletion Patterns

Basin-wide snow cover depletion exhibited systematic progression from lower to upper elevation zones throughout the 2009 melt season (Figure 3). Initial basin-wide snow cover approached 97.2% on 1 March, declining progressively through spring and early summer. Zone A (1823–2262 m) reached 50% coverage by April 8 and complete snow-free conditions by April 26. Zone B (2262–2704 m) achieved 50% coverage on 1 May and snow-free status on 28 May. Zone C (2704–3140 m) maintained >95% coverage through late April, reaching 50% coverage on 18 May, with 15–20% residual snow persisting through early June.

3.3. Uncalibrated SRM Performance

Initial simulation with a priori parameters yielded unsatisfactory performance (

Table 5. Uncalibrated SRM model performance metrics.

Metric	Value
R2	0.384
Correlation R	0.620
Nash–Sutcliffe Efficiency	0.371
Volumetric difference (%)	29.78
RMSE (m3/s)	3.42

).

3.4. Calibrated SRM Model Performance

Systematic iterative calibration of snowmelt and rainfall runoff coefficients within physically realistic bounds yielded dramatic performance improvement (Figure 4,

Table 6. Calibrated SRM model performance and optimized parameters.

Parameter/Metric	Zone A	Zone B	Zone C	Basin
Degree-day factor (cm/°C·day)	0.20	0.27–0.30	0.40–0.48	-
C_S (snowmelt coef.)	0.50–0.70	0.55–0.70	0.60–0.70	-
C_R (rainfall coef.)	0.40–0.75	0.45–0.75	0.50–0.75	-
R2	-	-	-	0.8606
Correlation R	-	-	-	0.927
NSE	-	-	-	0.854
Volumetric diff. (%)	-	-	-	3.35

). Calibrated simulation achieved R² = 0.8606, correlation coefficient R = 0.927, Nash–Sutcliffe efficiency NSE = 0.854, and volumetric difference of only 3.35%. This nine-fold improvement over uncalibrated performance (R² = 0.384) demonstrates critical importance of site-specific parameter adjustment for operational forecasting applications.

3.5. Multiple Linear Regression Discharge Prediction

Stepwise multiple linear regression analysis identified temperature, snow-covered area, snow water equivalent, calibrated snowmelt runoff coefficient, and shortwave radiation as statistically significant discharge predictors (p < 0.05), collectively explaining 88.1% of discharge variance (Figure 5,

Table 7. Multiple linear regression model statistics and predictor variables.

Predictor Variable	Coefficient	Std Error	t-Value	p-Value	Partial r
Temperature (T)	0.289	0.042	6.88	<0.001	+0.485
Snow-covered area (S)	−15.724	2.134	−7.37	<0.001	−0.544
Snow water equiv. (SWE)	0.647	0.089	7.27	<0.001	+0.312
Runoff coef. (C_S)	15.218	3.451	4.41	<0.001	+0.278
Radiation (R)	−0.009	0.003	−3.00	0.004	−0.089
Model R2	0.881	-	-	<0.0001	-
RMSE (m3/s)	1.42	-	-	-	-

). Temperature exhibited the strongest partial correlation (r = +0.485), confirming a dominant role in energy-limited snowmelt environment, while snow-covered area showed a strong negative correlation (r = −0.544) reflecting an inverse relationship as decreasing extent accompanies peak discharge. The empirical model (Equation (4)) achieved prediction accuracy approaching physically based SRM while requiring minimal calibration.

Q = −1.116 + 0.289T − 15.724S + 0.647SWE + 15.218C_S − 0.009R

(4)

4. Discussion

4.1. SRM Model Performance and Calibration Effectiveness

The calibrated SRM achieved excellent performance metrics (R² = 0.8606, R = 0.927, NSE = 0.854, and volumetric difference = 3.35%) that place this application among the most successful documented implementations globally. These results compare favorably with previous SRM applications in similar continental mountain environments; Martinec and Rango [10] reported R² values ranging from 0.70 to 0.92 across diverse basins with optimal calibration, while recent applications in Central Asian mountains achieved R² = 0.75–0.88 [13,14]. The nine-fold improvement from uncalibrated (R² = 0.384; volumetric difference = 29.78%) to calibrated performance demonstrates both the model’s sensitivity to parameter specification and the critical importance of site-specific calibration for operational forecasting applications. This dramatic improvement underscores that, while physically reasonable a priori parameter estimates provide useful first-order approximations, achieving prediction accuracy suitable for water resource management decisions requires systematic calibration against observed discharge. The necessity for temporal variation in calibrated runoff coefficients (CS: 0.50–0.70 and CR: 0.40–0.75 across five sub-periods) reflects genuine physical evolution of runoff generation mechanisms throughout the melt season, as documented in comprehensive snow hydrology studies [38].

The temporal variation observed in the calibrated snowmelt and rainfall runoff coefficients reflects the progressive evolution of hydrological conditions during the melt season. Increasing runoff efficiency over time is consistent with snowpack ripening, soil saturation, and reduction in infiltration capacity. Although the calibration was conducted using a single snowmelt season, the achieved model performance demonstrates the feasibility of applying SRM as a proof-of-concept for operational discharge forecasting when supported by remotely sensed snow cover information.

The substantial improvement in model performance following calibration highlights the sensitivity of the Snowmelt Runoff Model to site-specific parameterization in snow-dominated mountainous basins. The increase in performance metrics from an uncalibrated R² of 0.384 to a calibrated value of 0.8606 indicates that physically reasonable a priori parameter estimates alone are insufficient for reliable operational forecasting without adjustment to local hydro-climatic conditions. Similar levels of performance enhancement through calibration have been reported in previous SRM applications in continental and alpine environments, emphasizing the importance of basin-specific tuning.

4.2. Degree-Day Factors and Altitudinal Snowmelt Patterns

Calibrated degree-day factors exhibited systematic increase with elevation (Zone A: 0.20 cm/°C·day, Zone B: 0.27–0.30 cm/°C·day, and Zone C: 0.40–0.48 cm/°C·day), consistent with enhanced energy inputs at higher elevations. This altitudinal gradient reflects fundamental physical processes governing snowmelt energy balance in mountainous terrain. At higher elevations, several interacting mechanisms enhance melt efficiency per degree of positive air temperature [25]: (1) increased incoming shortwave radiation due to reduced atmospheric path length and lower aerosol concentrations, with clear-sky solar radiation increasing approximately 7–10% per 1000 m elevation gain; (2) enhanced diffuse radiation from surrounding snow-covered terrain creating higher surface albedo in the basin’s upper reaches, paradoxically increasing absorbed radiation on south-facing slopes through multiple reflections; (3) reduced atmospheric longwave radiation at higher elevations due to lower atmospheric temperature and water vapor content, but this deficit is typically offset by increased solar radiation receipts; (4) stronger wind speeds at exposed high-elevation sites (mean wind speeds 15–25% higher in Zone C compared to Zone A based on AWS measurements) enhancing turbulent heat exchange and accelerating energy transfer to the snowpack surface; and (5) thinner, less dense seasonal snowpack at wind-exposed ridges and upper slopes resulting in faster thermal response to positive air temperatures compared to deeper accumulations in sheltered mid-elevation zones. The observed degree-day factor range (0.20–0.48 cm/°C·day) corresponds well with documented values for continental mountain climates: Alps 0.25–0.60 cm/°C·day, Himalayas 0.30–0.80 cm/°C·day, and Rocky Mountains 0.20–0.50 cm/°C·day, demonstrating that our empirically derived parameters capture representative snowmelt physics. Temperature lapse rate analysis revealed enhanced vertical temperature gradient (−0.85 °C/100 m) exceeding standard atmospheric values (−0.65 °C/100 m), attributable to strong radiative cooling in snow-covered valleys creating persistent temperature inversions during early spring months—a characteristic feature of continental mountain climates that substantially influences spatial melt patterns and requires explicit accounting in operational forecasting applications.

4.3. Study Limitations and Uncertainty Sources

Several important limitations constrain interpretation and transferability of results, with single-year calibration representing the most significant constraint on model robustness and generalizability. The 2009 calibration period, while exhibiting representative snowmelt season conditions, prevents rigorous assessment of parameter stability and model performance across the full range of hydrological variability characteristic of continental mountain climates. Inter-annual variations in winter precipitation totals (regional coefficient of variation 30–45%), snowpack accumulation timing and spatial distribution, spring temperature progression, and resulting melt season characteristics (early versus late onset; rapid versus gradual ablation) substantially influence optimal parameter values and model accuracy. The strong performance achieved after calibration (R² = 0.8606; NSE = 0.854) reflects parameter optimization for 2009 conditions specifically and may not fully represent model behavior during contrasting years. Dry years with below-normal snowpack could yield different optimal degree-day factors due to altered snow density and surface characteristics, while wet years with deep accumulations might require adjusted recession coefficients to capture extended baseflow recession. Early snowmelt years characterized by rapid spring warming could demand different temperature–melt relationships compared to gradual melt progressions. Most critically, the calibrated parameters’ transferability to future conditions under non-stationary climate cannot be confidently assessed from single-year data. Standard practice in hydrological model development recommends minimum 3–5-year calibration periods spanning diverse conditions (wet/dry cycles and early/late snowmelt seasons) with independent validation on separate time periods to establish robust parameter ranges and quantify prediction uncertainty. The present single-year approach, while necessitated by data availability constraints and common in initial operational forecasting implementations for ungauged basins, should be interpreted as demonstrating methodological framework feasibility rather than providing definitive parameter values for long-term operational application. Future research priorities include multi-year validation across contrasting conditions, split-sample testing to assess parameter transferability, and evaluation of calibrated versus uncalibrated model performance to establish confidence bounds for operational forecasting applications.

Meteorological forcing uncertainty substantially impacts model performance and parameter estimation. Precipitation measurement uncertainty (±20–25% for snowfall and ±5–10% for rainfall) documented in this and previous regional studies [39,40] propagates through modeling chain, with calibrated runoff coefficients partially compensating for systematic input biases. The three-station automatic weather network, while spanning an 872 m elevation gradient, provides limited spatial representation of precipitation variability across the 242.7 km² basin, particularly for convective summer storms and wind-redistributed winter snowfall. Temperature interpolation assumes linear lapse rate (0.75 °C/100 m) derived from three elevations, neglecting potential micro-scale topographic effects (valley inversions, cold air pooling, and aspect-driven variations). Discharge measurements contain inherent uncertainty (±5–8% under normal flow conditions and ±10–15% during peak flows and ice-affected periods), affecting performance metric interpretation, parameter estimation precision, and volumetric balance assessment.

Model structural limitations warrant explicit acknowledgment. Degree-day approach, while pragmatic for operational forecasting with limited data availability [24,25], simplifies complex energy balance processes by neglecting spatial and temporal variations in net radiation, turbulent heat fluxes, and snowpack thermal properties. This simplification may reduce accuracy under rapidly changing meteorological conditions (warm-rain-on-snow events and intense solar radiation during clear periods) or limit applicability to climate change scenarios where energy balance component relationships shift. Evapotranspiration was not explicitly parameterized and, instead, was absorbed into effective runoff coefficients; this simplification is reasonable during the snow-dominated high-flow period (March–May) but may introduce systematic errors during late-summer low-flow conditions when evapotranspiration represents substantial loss. Groundwater dynamics were represented only through lumped recession coefficient without spatial differentiation, neglecting heterogeneity in subsurface storage properties, spring contributions, and groundwater–surface water exchange processes.

Remote sensing observations carry inherent uncertainties affecting snow cover area determination and subsequent discharge simulation. MODIS 500 m resolution, while suitable for basin-scale applications, cannot resolve sub-pixel heterogeneity or small patches of persistent snow in topographic depressions. Cloud contamination during storm periods creates data gaps precisely when snow accumulation and melt conditions change most rapidly, requiring temporal interpolation that introduces additional uncertainty. Binary snow classification (snow/no-snow) using optical imagery cannot quantify snow water equivalent variability or distinguish shallow snow from bare ground under certain illumination conditions. SAR-based wet snow detection, though providing all-weather capability, remains qualitative rather than quantitative due to complex interactions between backscatter, snow wetness, grain size, and layering.

Climate non-stationarity implications for long-term operational application require consideration. Model parameters calibrated under 2009 temperature and precipitation regimes may exhibit temporal instability as climate warms, particularly as rising temperatures shift precipitation phase (more rain and less snow), alter melt timing (earlier peaks and compressed seasons), modify energy balance components (increased longwave radiation and changing cloud cover patterns), and potentially trigger threshold responses (mid-winter melt events and rain-on-snow flooding). Regular model recalibration or development of adaptive parameter adjustment schemes accounting for evolving climate may prove necessary for sustained operational forecast accuracy. Uncertainty quantification frameworks (GLUE, ensemble approaches, and Bayesian methods) would provide probabilistic forecasts and confidence bounds valuable for risk-based water management decisions but remain unimplemented in current deterministic framework, limiting utility for decision-makers requiring probability-of-exceedance information.

The multiple linear regression model warrants explicit acknowledgment of potential circularity in predictor variable selection. The calibrated snowmelt runoff coefficient (Cs), included as a predictor variable in the MLR formulation (Equation (4)), was itself derived through optimization against observed discharge during the calibration process. This creates an element of circularity where a parameter calibrated to minimize discharge prediction error is subsequently used to predict that same discharge through regression relationships. While this approach enhances explanatory power (R² = 0.881), it may overstate the model’s independent predictive capability and could lead to overfitting to the specific 2009 calibration conditions. The high correlation between predicted and observed discharge (R = 0.939) partially reflects this self-referential relationship rather than purely independent prediction from meteorological and snow cover forcings alone. A more robust approach for independent validation would exclude calibrated model parameters as MLR predictors, relying instead solely on external forcing variables (temperature, precipitation, snow-covered area, and SWE) and empirical coefficients. However, the current formulation serves its intended purpose of identifying dominant hydrological controls and quantifying relative importance of different factors in explaining discharge variability, even if absolute prediction accuracy metrics should be interpreted cautiously given the parameter circularity. Future applications for independent prediction should either (1) use regionally transferred a priori parameter values rather than basin-calibrated values as MLR predictors or (2) develop separate regression models trained on independent time periods or basins to avoid circular parameter usage.

The MLR model’s high R² value (0.881) raises legitimate concerns about potential overfitting, particularly given the relatively small sample size (n = 65 days) and number of predictor variables (five predictors plus intercept). The ratio of observations to predictors (65:5 = 13:1) exceeds the commonly cited minimum threshold of 10:1 but remains at the lower end of recommendations for robust regression modeling. Several diagnostic approaches suggest overfitting risk is moderate but manageable: (1) variance inflation factors (VIF) for all predictors remained below 3.5, indicating minimal multicollinearity that would inflate coefficient standard errors and reduce model stability; (2) residual analysis showed approximately normal distribution without systematic patterns across the discharge range, suggesting the model captures underlying relationships rather than fitting noise; (3) the predictor variables represent physically meaningful hydrological controls (temperature, snow cover, snow water equivalent, runoff coefficient, and rainfall) rather than spurious correlations, lending mechanistic credibility to the statistical relationships; and (4) cross-validation through leave-one-out resampling yielded similar R² values (R²cv = 0.86), indicating the model generalizes reasonably well within the calibration dataset. However, the true test of overfitting—validation on independent data from different years or basins—remains impossible with single-year calibration. The model should be interpreted as demonstrating relationships between forcings and response for 2009 conditions specifically, with extrapolation to other years or locations requiring validation. Alternative regularization approaches (ridge regression and LASSO) that penalize model complexity could reduce overfitting risk but were not explored given limited data availability for training–validation splits. The current MLR formulation serves its intended purpose of identifying dominant controls and quantifying explained variance, but prediction uncertainty likely exceeds that suggested by the high R² value alone.

Future research should employ more comprehensive uncertainty assessment methodologies beyond the single deterministic calibration approach used here. Recommended approaches include (1) Monte Carlo simulation propagating meteorological forcing uncertainties (precipitation ±20–25%; temperature ±1–2 °C) through the model to quantify output uncertainty envelopes, enabling probabilistic discharge predictions rather than deterministic point estimates; (2) Generalized Likelihood Uncertainty Estimation (GLUE) framework identifying behavioral parameter sets through likelihood-weighted sampling, providing confidence bounds on both parameters and predictions while accounting for equifinality (multiple parameter sets producing similar performance); (3) Bayesian calibration explicitly quantifying parameter uncertainty through posterior probability distributions conditioned on observed data, particularly valuable for data-limited applications where prior knowledge supplements sparse observations; (4) ensemble modeling combining multiple model structures (degree-day, enhanced temperature index, and simplified energy balance) to assess structural uncertainty beyond parametric uncertainty within single model formulation; (5) split-sample and differential split-sample testing using contrasting climatic periods (wet/dry and warm/cool) for calibration–validation to assess parameter transferability across conditions, particularly critical given single-year limitation; and (6) spatial uncertainty analysis evaluating sensitivity to elevation zone delineation, demonstrating whether three-zone discretization adequately captures spatial heterogeneity or whether finer zonation would improve accuracy. These approaches would provide more realistic characterization of prediction uncertainty essential for risk-based water resource management decisions, moving beyond deterministic “best estimate” predictions to probabilistic forecasts with quantified confidence intervals. The current limitation to a single calibration parameter set without formal uncertainty quantification represents the most significant gap for operational forecasting applications.

5. Conclusions

This study demonstrates the operational feasibility of integrating the degree-day-based Snowmelt Runoff Model with geographic information systems and multi-platform satellite remote sensing in a snow-dominated mountainous basin. While limited to a single hydrological year, the results should be interpreted as a proof-of-concept rather than a fully transferable modeling framework. Future work should focus on multi-year calibration and validation, formal uncertainty assessment, and evaluation under contrasting hydro-climatic conditions. Key findings and contributions are summarized as follows:

• Model Performance and Calibration

Calibrated SRM achieved excellent performance for the 2009 snowmelt season (30 March–2 June, 65 days), with coefficient of determination R² = 0.8606, correlation coefficient R = 0.927, Nash–Sutcliffe efficiency NSE = 0.854, and volumetric difference of only 3.35%. These metrics place this application among the most successful documented SRM implementations globally and demonstrate the model’s capability for operational water resource forecasting in similar continental mountain basins. The dramatic nine-fold improvement from uncalibrated (R² = 0.384; volumetric difference = 29.78%) to calibrated performance underscores the critical importance of site-specific parameter adjustment, while confirming that physically reasonable a priori estimates provide useful first-order approximations.

2.

Elevation Zonation and Degree-Day Factors

Three-zone elevation stratification (Zone A: 1823–2262 m, 42.40%; Zone B: 2262–2704 m, 40.18%; and Zone C: 2704–3140 m, 17.42%) successfully captured systematic altitudinal variations in snowmelt timing and energy balance. Degree-day factors increased with elevation (Zone A: 0.20 cm/°C·day, Zone B: 0.27–0.30 cm/°C·day, and Zone C: 0.40–0.48 cm/°C·day), consistent with enhanced solar radiation and turbulent heat transfer at higher elevations in thinner atmosphere. Temporal increases throughout the melt season (early: 0.20–0.40; late: 0.30–0.48 cm/°C·day) reflected snowpack aging and albedo reduction. These values align with physically realistic ranges established by extensive international SRM applications.

3.

Site-Specific Temperature Lapse Rate

Empirically derived temperature lapse rate (γ = 0.75 °C/100 m, R² = 0.94) substantially exceeded the standard environmental lapse rate (0.65 °C/100 m), reflecting Eastern Anatolia’s severe continental climate with low atmospheric humidity and minimal cloud cover. This 15% enhancement has important operational implications; employing standard lapse rates would systematically bias temperature estimates and degrade model performance. The finding emphasizes the necessity of deriving site-specific lapse rates from local observations rather than assuming globally averaged values, particularly in continental mountain environments with extreme climatic characteristics.

4.

Remote Sensing Integration

MODIS MOD10A2 8-day composite products provided operationally robust snow cover observations with 87.3% pixel-level agreement (Kappa = 0.73) against Landsat validation imagery. Systematic snow cover recession exhibited clear altitudinal progression (Zone A snow-free April 26, Zone B May 28, and Zone C residual through early June), enabling anticipation of discharge timing for water resource allocation. SAR observations (RADARSAT-1 C-band and ALOS-PALSAR L-band) offered valuable all-weather capability during cloudy periods and successfully discriminated wet snow through characteristic 4–7 dB backscatter reduction, though complex topographic interactions precluded fully automated classification. Optical sensors remain preferable for routine basin-wide mapping, with SAR providing complementary information during cloud-obscured periods.

5.

Precipitation Measurement Challenges

Systematic precipitation undercatch averaging 20–25% was identified through comparison with regional gridded products and water balance calculations, attributed to wind-induced measurement errors [40,41] during snowfall events despite heated gauge design and wind shields. This pervasive challenge in high-altitude precipitation measurement explains precipitation’s exclusion from the final multiple linear regression model despite physical importance; systematic bias introduced sufficient noise to preclude statistically significant discharge prediction. SRM’s successful calibration despite precipitation limitations demonstrates model robustness through implicit error correction via adjusted runoff coefficients, though this compensation potentially limits parameter transferability to years with different measurement characteristics.

6.

Empirical Discharge Prediction Model

Multiple linear regression analysis revealed temperature, snow-covered area, snow water equivalent, snowmelt runoff coefficient, and radiation as statistically significant discharge predictors, yielding the empirical model: Q_calc = −1.116 + 0.289T − 15.724S + 0.647SWE + 15.218C_S − 0.009R (R² = 0.881, R = 0.938, p < 0.0001). This high explanatory power (88.1% of discharge variance) using readily available variables suggests potential operational utility complementing physically based modeling. Temperature’s dominant positive influence, snow-covered area’s strong negative coefficient (reflecting inverse relationship during active melt), and SWE’s positive contribution provide physically interpretable insights into fundamental discharge controls in snowmelt-dominated systems.

7.

Operational Water Resource Management Implications

Results demonstrate operational feasibility for seasonal water supply forecasting and flood warning in Eastern Anatolian basins where 70–80% of annual discharge derives from snowmelt. Modest data requirements (daily temperature, precipitation, and satellite snow cover) align with operational monitoring capabilities. Volumetric accuracy (3.35% error) supports reservoir operation, irrigation scheduling, and hydropower planning. Peak discharge prediction accuracy (4.3–2.5% error) enables flood forecasting, though inherent 1–3-day lead times limit warning utility compared to seasonal climate prediction approaches. Extension to additional regional basins would provide integrated water resource management capability in this climatically vulnerable, water-stressed region where snowmelt constitutes critical resource supporting agriculture, municipal supply, and ecosystem services.

Recommendations for Future Research

Future research directions should address identified limitations and extend operational capabilities:

• Multi-year validation across diverse meteorological conditions to assess parameter stability and quantify inter-annual performance variability;
• Enhanced precipitation measurement through dual-gauge installations (shielded accumulation gauge plus heated tipping-bucket) or wind-correction algorithms to reduce systematic undercatch bias;
• Integration of higher-resolution snow cover products (Sentinel-2 20 m and Landsat-8/9 30 m) with cloud-gap-filling algorithms to resolve fine-scale heterogeneity while maintaining temporal frequency;
• Comparison with physically detailed energy balance models to quantify trade-offs between process representation complexity and operational data requirements;
• Regional extension to additional Eastern Anatolian basins to develop transferable parameter relationships enabling ungauged basin prediction;
• Climate change impact assessment using downscaled global climate model projections to anticipate future snowmelt timing shifts and water resource availability changes;
• Real-time operational forecasting system development integrating automated satellite data acquisition, model execution, and decision support visualization for water resource managers.

In conclusion, this research demonstrates that integrating the degree-day-based SRM with modern GIS and remote sensing capabilities provides accurate, operationally feasible discharge prediction in data-scarce mountainous regions. The methodology’s successful application to Kirkgoze Basin establishes a foundation for regional operational water resource management in Eastern Anatolia and similar snow-dominated mountain environments globally, addressing critical needs for improved seasonal forecasting supporting agriculture, hydropower generation, flood risk management, and climate adaptation planning.