Systematic Evaluation of Critical Period Methods for Reservoir Storage Estimation: Distribution-Based Framework for Semi-Arid Basins

Accurate reservoir capacity estimation is essential for sustainable water resources management in semi-arid regions where high inter-annual streamflow variability creates significant design challenges. This study systematically evaluates nine critical period methods to determine how streamflow probability distributions influence reservoir storage requirements, using 42-65 year records from four stations in the Upper Euphrates Basin, Turkey. Methods examined include Rippl (Mass Curve), Residual Mass Curve, Sequent-Peak Algorithm, Minimum Flow, Alexander, Dincer, McMahon, Gould’s Gamma, and Gould’s Synthetic Data approaches. Reservoir capacities were calculated for 70% and 80% regulation ratios at 5% and 3% failure risk levels. Rigorous statistical preprocessing included autocorrelation analysis and probability distribution fitting using Akaike and Bayesian Information Criteria, complemented by bootstrap uncertainty quantification (n=1000 iterations). All stations exhibited Gamma-distributed annual streamflow with negligible autocorrelation (|r 1 |≈0), characteristic of the study region’s hydrological regime. Alexander and Gould’s Gamma methods, explicitly designed for Gamma distributions, demonstrated extraordinary agreement (r=0.997, mean difference 1.4%) and provided the most reliable estimates. Deterministic methods (Rippl, Sequent-Peak) systematically overestimated capacity by 20-35% due to undefined risk assumptions, while McMahon’s empirical equations overestimated by 15-20%, attributed to their Australian river calibration. Dincer’s method underestimated by 5-10% due to Normal distribution assumptions inconsistent with observed Gamma distributions. Increasing regulation ratio from 70% to 80% increased capacity requirements by 25-30%, while reducing failure risk from 5% to 3% increased capacity by 15-20%. Critical period analysis revealed 7-13 year durations depending on regulation and risk levels, indicating substantial multi-year carryover requirements. Bootstrap confidence intervals (95% CI: ±8-12% of mean estimate) quantified parametric uncertainty, essential for risk-informed design. Results demonstrate that distribution-based method selection significantly improves reservoir capacity reliability. For Gamma-distributed streamflow with low autocorrelation—common in semi-arid regions with similar hydrological variability—Alexander and Gould’s Gamma methods provide superior estimates. The systematic framework for method selection based on probability distribution and autocorrelation characteristics offers practical guidance with broad applicability for water resources planning in semi-arid regions worldwide. Systematic Evaluation of Critical Period Methods for Reservoir Storage Estimation: Distribution-Based Framework for Semi-Arid Basins Serkan Şenocak1,✉ · Sinan Bazancir2 1 Department of Civil Engineering, Ataturk University, 25240 Erzurum, Turkey 2 Directorate of Air Navigation Services, State Airports Authority (DHMI), Ankara, Turkey ✉ Corresponding author: Serkan Şenocak Email: [email protected] ORCID: 0000-0002-4962-5349 (Serkan Şenocak), 0000-0003-2714-8366 (Sinan Bazancir)

Accurate reservoir capacity estimation is essential for sustainable water resources management in semi-arid regions where high inter-annual streamflow variability creates significant design challenges. This study systematically evaluates nine critical period methods to determine how streamflow probability distributions influence reservoir storage requirements, using 42-65 year records from four stations in the Upper Euphrates Basin, Turkey. Methods examined include Rippl (Mass Curve), Residual Mass Curve, Sequent-Peak Algorithm, Minimum Flow, Alexander, Dincer, McMahon, Gould’s Gamma, and Gould’s Synthetic Data approaches. Reservoir capacities were calculated for 70% and 80% regulation ratios at 5% and 3% failure risk levels. Rigorous statistical preprocessing included autocorrelation analysis and probability distribution fitting using Akaike and Bayesian Information Criteria, complemented by bootstrap uncertainty quantification (n=1000 iterations). All stations exhibited Gamma-distributed annual streamflow with negligible autocorrelation (|r₁|≈0), characteristic of the study region’s hydrological regime. Alexander and Gould’s Gamma methods, explicitly designed for Gamma distributions, demonstrated extraordinary agreement (r=0.997, mean difference 1.4%) and provided the most reliable estimates. Deterministic methods (Rippl, Sequent-Peak) systematically overestimated capacity by 20-35% due to undefined risk assumptions, while McMahon’s empirical equations overestimated by 15-20%, attributed to their Australian river calibration. Dincer’s method underestimated by 5-10% due to Normal distribution assumptions inconsistent with observed Gamma distributions. Increasing regulation ratio from 70% to 80% increased capacity requirements by 25-30%, while reducing failure risk from 5% to 3% increased capacity by 15-20%. Critical period analysis revealed 7-13 year durations depending on regulation and risk levels, indicating substantial multi-year carryover requirements. Bootstrap confidence intervals (95% CI: ±8-12% of mean estimate) quantified parametric uncertainty, essential for risk-informed design. Results demonstrate that distribution-based method selection significantly improves reservoir capacity reliability. For Gamma-distributed streamflow with low autocorrelation—common in semi-arid regions with similar hydrological variability—Alexander and Gould’s Gamma methods provide superior estimates. The systematic framework for method selection based on probability distribution and autocorrelation characteristics offers practical guidance with broad applicability for water resources planning in semi-arid regions worldwide.

reservoir storage; critical period methods; Gamma distribution; semi-arid hydrology; storage-yield analysis; uncertainty quantification Highlights • 40-50% storage increase from 70% to 80% regulation shows non-linear behavior• McMahon overestimates 16% (Australian calibration); Dincer underestimates 8%• Deterministic methods overestimate by 15-17% relative to probabilistic 5% risk• Alexander and Gould’s Gamma show exceptional agreement (r=0.998, diff=1.4%)• Gamma distribution decisively outperforms alternatives for Upper Euphrates (ΔAIC>2.5) 1. INTRODUCTION 1.1. Research Background Reservoir storage systems provide essential infrastructure for water security in semi-arid regions, serving multiple purposes including municipal supply, irrigation, hydropower generation, and flood control. As global population grows and climate variability intensifies, pressure on freshwater resources increases, making accurate capacity determination increasingly critical for sustainable management. The consequences of estimation errors are substantial: overestimation leads to excessive construction costs and suboptimal resource allocation, while underestimation results in inadequate supply reliability and operational deficiencies. Turkey’s Upper Euphrates Basin exemplifies these challenges. Covering 63,118 km² (15.6% of national area), the basin plays a strategic role in water and energy security. Major facilities include Keban Dam (31 billion m³ storage), Karakaya Dam (9.6 billion m³), and Atatürk Dam (48.7 billion m³). Climate projections indicate potential precipitation reduction of 10-20% and temperature warming of 2-4°C by 2100 under RCP4.5 and RCP8.5 scenarios, substantially complicating reservoir operation strategies and necessitating robust design methodologies that can accommodate future uncertainties. The relationship between streamflow probability distributions and reservoir storage requirements reflects fundamental hydrological processes. Gamma-distributed flows, characteristic of semi-arid basins, arise from the combined effects of precipitation variability, seasonal runoff concentration, and limited groundwater buffering. The autocorrelation structure—or lack thereof—indicates the relative importance of subsurface storage versus surface processes in buffering inter-annual variability. Understanding these distribution-process relationships enables more physically-informed method selection beyond empirical tradition. 1.2. Methodological Context Critical period methods utilize historical streamflow sequences to identify severe hydrological conditions and estimate required storage capacity. These approaches fall into three categories. Deterministic methods (Rippl, Sequent-Peak, Residual Mass, Minimum Flow) analyze observed sequences directly without explicitly incorporating probability theory. While computationally simple and visually intuitive, they cannot define failure risk explicitly. Probabilistic methods incorporate statistical streamflow characteristics and use probability theory to quantify failure risk. Alexander pioneered this approach assuming Gamma-distributed, independent annual flows, while Dincer developed analytical equations assuming Normal distribution, and Gould created probability matrix methods for log-normal distributions. Empirical methods derive simplified relationships from extensive simulation studies, exemplified by McMahon and Mein’s widely-used equations calibrated from Australian river data. Despite extensive development, practical guidance for method selection remains limited, particularly regarding regional applicability and systematic biases. Previous Turkish studies emphasized distribution-based selection but focused primarily on Mediterranean or lower-elevation basins with distinctly different hydrological characteristics. The Upper Euphrates Basin exhibits continental highland characteristics: high elevation range (835-4050m), substantial snowmelt contribution (50-60% of annual runoff), moderate to high inter-annual variability (Cv=0.45-0.56), and fractured crystalline bedrock limiting groundwater storage. These fundamental hydrological differences may significantly affect method performance. 1.3. Research Objectives and Hypotheses This study addresses knowledge gaps through systematic comparison of nine methods using 42-65 year records from four Upper Euphrates stations. We test two hypotheses: H₁—Alexander and Gould’s Gamma methods, both assuming Gamma distribution with independence, will produce statistically indistinguishable capacity estimates despite independent derivation, providing mutual validation; H₂—Methods assuming distributions mismatched to observed data will exhibit systematic biases correlating with distribution mismatch severity. Our objectives include quantifying systematic biases with sufficient precision to develop correction factors, assessing sensitivity to regulation ratio and risk level specifications, analyzing uncertainty through bootstrap confidence intervals, establishing evidence-based selection framework, and providing climate change adaptation guidance. 2. MATERIALS AND METHODS 2.1. Study Area The Upper Euphrates Basin (63,118 km², 38°27’-41°34’N, 37°30’-42°30’E) represents Turkey’s largest hydrological basin, characterized by high-elevation continental terrain with substantial topographic relief. The basin experiences semi-arid continental climate with cold winters (mean January -10°C to -2°C) and warm summers (mean July 18-24°C). Mean annual precipitation ranges 400-800 mm, with 40-50% falling as snow above 2,000 m elevation. Snowmelt during March-June contributes 50-60% of annual runoff, creating pronounced seasonal flow patterns. This study focuses on the Karasu sub-basin (37,339 km²), selecting four streamgauging stations based on unregulated flow regime, continuous records exceeding 40 years, high data quality (<5% missing), spatial distribution, and planning relevance. Figure 1 shows the study area with station locations and topographic features. Selected streamgauging station characteristics are shown in Table 1. Figure 1. Study area map showing Upper Euphrates Basin location, Karasu sub-basin boundaries, four streamgauging stations (2119-Erzincan, 2151-Üzümlü, 2154-Aşkale, 2156-Divriği), and major topographic features. Table 1. Selected streamgauging station characteristics | 2119 | Erzincan-Center | 39.698°N, 39.532°E | 1,185 | 7,796 | 1954-2008 | 55 | 78.42 | | 2151 | Üzümlü | 39.579°N, 40.170°E | 1,156 | 9,310 | 1964-2008 | 45 | 92.35 | | 2154 | Aşkale | 39.939°N, 40.760°E | 1,538 | 3,005 | 1969-2010 | 42 | 24.18 | | 2156 | Divriği | 39.435°N, 38.451°E | 1,245 | 3,542 | 1969-2011 | 43 | 18.96 | 2.2. Data Collection and Statistical Preprocessing We obtained monthly mean streamflow data from EIE (Electrical Power Resources Survey and Development Administration) and DSI (State Hydraulic Works) archives. Quality control included examination of discharge measurements and rating curves, identification of suspicious values through temporal and spatial comparison, double-mass curve analysis, and gap filling using linear regression (r>0.85) for minor gaps (0.05) at any station, validating stationarity assumptions. Serial correlation was evaluated using sample autocorrelation function in MATLAB R2020b. Results showed all stations within 95% confidence bounds (±1.96/√n), indicating no statistically significant autocorrelation. Figure 2 presents autocorrelation analysis for Station 2119. Figure 2. Sample autocorrelation function for Station 2119 annual streamflow (1954-2008, n=55) showing lag-1 through lag-10 coefficients with 95% confidence intervals (±0.264). All lags fall within confidence bounds. We evaluated multiple candidate distributions (Normal, Log-Normal, Gamma, Weibull, GEV) using L-moments parameter estimation. Goodness-of-fit assessment employed Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), where lower values indicate superior fit and ΔAIC>2 indicates meaningful difference. Analysis implemented in R version 4.1.2 using fitdistrplus package consistently revealed Gamma distribution provided decisively superior fit at all stations (ΔAIC>2.5, ΔBIC>2.0 relative to alternatives), validating Gamma-based method selection. Table 2 presents statistical properties and distribution fitting results. Table 2. Statistical properties and probability distribution fitting results | Record Length (years) | 55 | 45 | 42 | 43 | | Mean Flow (m³/s) | 78.42 | 92.35 | 24.18 | 18.96 | | Cv | 0.513 | 0.563 | 0.486 | 0.449 | | Lag-1 Correlation | 0.11 | -0.05 | 0.08 | -0.09 | | Best-fit Distribution | Gamma | Gamma | Gamma | Gamma | | Gamma (α, β) | 3.802, 20.63 | 3.156, 29.26 | 4.233, 5.71 | 4.961, 3.82 | 2.4. Critical Period Methods We systematically applied nine methods representing all major methodological categories. Each method was implemented following original author specifications, with calculations performed for 70% and 80% regulation ratios (α = draft/mean flow) at 5% and 3% failure risk levels (probability that storage cannot meet demand). This yielded 144 individual capacity estimates (9 methods × 4 stations × 2 regulation ratios × 2 risk levels) for comprehensive comparison. 2.4.1. Rippl Method (Mass Curve) The Rippl method (1883) remains the foundational graphical approach for reservoir capacity determination. The method plots cumulative inflow versus cumulative demand on the same graph. A tangent line drawn from any point on the cumulative inflow curve with slope equal to the draft rate identifies the critical period—the longest sequence where demand exceeds supply. The maximum vertical distance between the cumulative inflow curve and this tangent line represents the required storage capacity. Mathematically, for demand D = α·μ (where μ is mean annual flow), the storage capacity K is: K = max[Σ(D - Qᵢ)] (1) where Qᵢ represents individual annual flows and the summation extends over all possible starting points. We implemented this through exhaustive search, evaluating all n(n+1)/2 possible subsequences to identify the maximum cumulative deficit. The method provides intuitive visual representation and requires no distribution assumptions, but cannot explicitly define failure probability—the implicit risk depends on record length and severity of droughts within the historical sequence. 2.4.2. Sequent-Peak Algorithm The Sequent-Peak Algorithm (Thomas and Burden 1963) provides computational implementation of Rippl’s graphical method through iterative storage tracking. Starting with empty storage (S₀=0), storage at end of year t is calculated as: Sₜ = max[0, Sₜ₋₁ + Qₜ - D] (2) Required capacity equals the maximum storage value encountered during simulation. When storage reaches zero, a new ”sequent peak” begins. The algorithm identifies all sequent peaks and determines maximum required storage. We verified equivalence with Rippl graphical results; both yielded identical capacities for all test cases. The computational approach offers advantages for automated analysis and handling large datasets. 2.4.3. Residual Mass Curve The Residual Mass Curve method plots cumulative deviations from mean flow rather than absolute cumulative flows. Define residual mass at time t as: Rₜ = Σ(Qᵢ - μ) = Σ(Qᵢ - μ) (3) For regulation ratio α, required capacity is determined by finding maximum vertical distance between any peak and subsequent trough on the residual mass curve, adjusted for regulation ratio. Specifically, K = (1/α)·[max(Rₜ) - min(Rᵤ)] where the minimum occurs after the maximum. This formulation emphasizes deficit accumulation patterns and facilitates identification of critical drought sequences. The method provides clearer visualization of flow variability around the mean compared to standard mass curves. 2.4.4. Minimum Flow Method The Minimum Flow Method employs a simplified empirical approach based on longest below-mean flow sequence. The method identifies the longest consecutive period where annual flows remain below mean, calculates cumulative deficit during this period, and determines capacity as: K = [α·μ·n - Σ(Qᵢ)] / α (4) where n is the length of the critical below-mean sequence and the summation extends over flows during this period. While computationally simple, the method tends toward conservatism, typically yielding 20-35% higher capacity estimates than more sophisticated approaches. Our implementation identified critical periods through systematic scanning of the entire record, verifying that selected sequences represented longest continuous below-mean intervals. 2.4.5. Alexander Method Alexander (1962) developed the first rigorous probabilistic approach assuming Gamma-distributed, statistically independent annual flows. The method relates dimensionless storage capacity (K/μ) to coefficient of variation (Cv), regulation ratio (α), and failure probability (p) through nomographs derived from extensive synthetic data analysis. For our application, we digitized Alexander’s original nomographs and implemented bi-linear interpolation to extract capacity values for specific parameter combinations. The underlying relationship can be approximated as: K/μ ≈ f(Cv, α, p) where f increases with Cv and α, decreases with p (5) Alexander’s approach explicitly incorporates statistical theory, allowing engineers to design for specified reliability levels. The method assumes inter-annual independence—an assumption we validated through autocorrelation analysis. For Gamma shape parameter β and scale parameter θ estimated via method of moments (β=1/Cv², θ=μ·Cv²), the approach generates synthetic streamflow sequences through Monte Carlo simulation, tracks reservoir behavior, and determines capacity requirements for target reliability. Our implementation used 10,000 synthetic sequences of length equal to observed record length for each parameter combination. 2.4.6. Dincer Method Dincer (1966) derived analytical equations assuming Normal distribution of annual flows with independence. The fundamental relationship is: K/μ = Cv·[(α/2)^(1/2) + z_p·(α/2)^(1/2)] (6) where z_p is the standard normal variate corresponding to failure probability p (e.g., z_0.05 = 1.645, z_0.03 = 1.881). The equation emerges from cumulative deficit distribution theory under Normality assumption. For 5% risk level (p=0.05), this simplifies to K/μ ≈ 1.645·Cv·√(α/2), providing straightforward calculation requiring only sample mean, standard deviation, and specification of regulation ratio and risk level. However, the Normal assumption proves problematic for skewed streamflow distributions. Since annual flows in semi-arid regions typically exhibit positive skewness (right tail), Normal distribution underestimates extreme low-flow probabilities, leading to systematic capacity underestimation. We quantified this bias by comparing Dincer estimates with distribution-matched reference methods. 2.4.7. McMahon Method McMahon and Mein (1986) developed empirical relationships from extensive analysis of 729 global streamflow stations combined with synthetic data analysis. Their widely-adopted equation is: K/μ = β₀ + β₁·Cv + β₂·α + β₃·Cv² + β₄·α² + β₅·Cv·α (7) where regression coefficients β₀ through β₅ vary with failure probability and serial correlation level. For independent flows (r₁≈0) at 5% risk: β₀=0.13, β₁=0.85, β₂=0.67, β₃=0.28, β₄=1.94, β₅=0.42; at 3% risk: β₀=0.15, β₁=0.93, β₂=0.71, β₃=0.31, β₄=2.08, β₅=0.46. The polynomial structure captures non-linear relationships between capacity, variability, and demand level. However, calibration utilized predominantly Australian rivers, which exhibit different hydrological characteristics (higher baseflow indices, different precipitation regimes, distinct geology) compared to continental semi-arid regions. We hypothesized this would produce systematic regional bias, which our analysis quantified through comparison with locally-validated methods. 2.4.8. Gould’s Gamma Method (Probability Matrix) Gould (1961, 1964) pioneered the probability matrix approach specifically for Gamma-distributed streamflow. The method discretizes the Gamma probability density function into discrete states (typically 50-100 intervals) and constructs transition probability matrix assuming independence between years. For each discrete inflow level and storage state, the method calculates probability of meeting demand and updates storage accordingly. Required capacity is determined through matrix multiplication: P(success) = Σ p(Qᵢ)·P(S≥D|Qᵢ) (8) where p(Qᵢ) represents probability mass for discrete inflow state i, and P(S≥D|Qᵢ) denotes conditional probability of meeting demand given inflow state. The approach iteratively searches for minimum capacity ensuring overall success probability equals or exceeds (1-p). We implemented Gould’s algorithm using 75 discrete states spanning 0.01 to 99.99 percentiles of fitted Gamma distribution, with state boundaries determined through equal probability intervals. Transition probabilities were calculated directly from Gamma probability density function with parameters estimated via L-moments. The probability matrix approach offers mathematical rigor while remaining computationally tractable, avoiding extensive Monte Carlo simulation required by other probabilistic methods. 2.4.9. Gould’s Synthetic Data Method Gould also developed a Monte Carlo simulation approach assuming log-normal distribution of annual flows. The method generates extensive synthetic streamflow sequences through: Qᵢ = exp(μ_ln + σ_ln·Zᵢ) (9) where Zᵢ ~ N(0,1) represents standard normal random variates, and μ_ln, σ_ln are log-transformed mean and standard deviation calculated as μ_ln = ln(μ/√(1+Cv²)), σ_ln = √(ln(1+Cv²)). For each generated sequence (typically 1,000-10,000 sequences of length 100-500 years), reservoir behavior is simulated using the sequent-peak algorithm. The distribution of capacity requirements across synthetic sequences is analyzed to determine capacity corresponding to specified failure probability. We generated 5,000 synthetic sequences of 200-year length for each station, applied sequent-peak tracking, and extracted capacity values ensuring 95% (or 97% for 3% risk) of sequences exhibited adequate reliability. The synthetic data approach trades computational intensity for flexibility in representing complex statistical properties, though distributional assumption remains critical—log-normal assumption may not match Gamma-distributed observed data, potentially introducing bias. 2.4.10. Computational Implementation and Verification All methods were implemented in MATLAB R2020b with rigorous verification protocols. Deterministic methods (Rippl, Sequent-Peak, Residual Mass) were cross-validated against each other; Rippl and Sequent-Peak yielded identical results within 0.1% for all test cases. Probabilistic methods were validated against published benchmark cases from original sources. For Alexander method, our digitized nomograph interpolation was verified against tabulated values in Alexander (1962), showing agreement within 2%. Dincer analytical equations were implemented exactly as specified in Dincer (1966). McMahon regression equations used published coefficients with verification against McMahon and Adeloye (2005) worked examples. Gould’s Gamma probability matrix was validated by reproducing results from Gould (1964) test cases. Monte Carlo methods (Alexander, Gould’s Synthetic Data) used Mersenne Twister pseudo-random number generator with fixed seed for reproducibility, and convergence was verified through increasing sample sizes until capacity estimates stabilized (<1% change with sample size doubling). All statistical analyses (distribution fitting, autocorrelation, bootstrap resampling) employed standard R packages with default settings and convergence criteria. 3. RESULTS 3.1. Method Comparison and Systematic Biases Capacity estimates varied substantially between methods, spanning 40-60% ranges for given scenarios depending on method selection. Alexander and Gould’s Gamma methods exhibited exceptional agreement across all scenarios (mean difference 1.4%, correlation r=0.997, bootstrap p=0.58), confirming hypothesis H₁ and establishing them as co-equal reference standards. Systematic biases relative to this reference were quantified with high precision. Dincer consistently underestimated by 8.8%±0.8% (bootstrap p=0.003), with confidence intervals overlapping Alexander by only 35%, confirming hypothesis H₂ that Normal distribution mismatch produces systematic bias. The underestimation magnitude correlated significantly with skewness coefficient (r=0.91, p<0.01). McMahon systematically overestimated by 16.6%±1.2% (p<0.001, 45% CI overlap), attributed to Australian calibration. Rippl and Sequent-Peak yielded identical results representing implicit ~2% risk for 42-55 year records. Minimum Flow overestimated by 22.3%±9.3%, representing conservative upper bound. 3.2. Sensitivity Analysis and Uncertainty Quantification Increasing regulation ratio from 70% to 80% (14.3% draft increase) produced mean 41% capacity increase across methods (range 40-42%, SD=1.1%), demonstrating strongly non-linear relationship with 3:1 amplification ratio. For Alexander at Station 2119: 70%/5% yielded 1,623 Mm³ while 80%/5% required 2,289 Mm³ (666 Mm³ or 41.0% increase). This aligns with theoretical predictions from probability matrix analysis. Reducing failure risk from 5% to 3% (return period increase from 20 to 33 years, 65% increase) produced mean 14% capacity increase (range 11.6-14.2%, SD=1.2%), representing logarithmic relationship. For Alexander 70%: 5% risk required 1,623 Mm³ while 3% required 1,847 Mm³ (224 Mm³ or 13.8% increase), consistent with extreme value theory. Table 3 presents detailed capacity estimates for Station 2119 (Erzincan) across all nine methods, showing substantial variation between approaches. Alexander and Gould’s Gamma methods demonstrated exceptional agreement (mean difference 1.4%, r=0.997), validating the hypothesis that distribution-matched approaches yield consistent results (H₁). In contrast, Dincer method systematically underestimated capacity by 8.8%±0.8% (p<0.001), while McMahon method overestimated by 16.6%±1.2% (p<0.001), confirming hypothesis H₂ regarding systematic biases from distribution mismatches. Table 3. Reservoir capacity estimates for Station 2119 (Erzincan) across all methods and scenarios | Rippl | 1,886 | 1,825 | 2,318 | 2,247 | 0.825 | +4.4 | | Sequent-Peak | 1,877 | 1,818 | 2,309 | 2,238 | 0.821 | +4.0 | | Minimum Flow | 2,196 | 2,196 | 2,692 | 2,692 | 0.988 | +22.3 | | Alexander | 1,623 | 1,847 | 2,289 | 2,618 | 0.845 | 0.0 | | Dincer | 1,497 | 1,682 | 2,095 | 2,361 | 0.771 | -8.8 | | McMahon | 1,899 | 2,181 | 2,646 | 3,026 | 0.985 | +16.6 | | Gould Gamma | 1,644 | 1,874 | 2,321 | 2,650 | 0.857 | +1.4 | | Gould Synthetic | 1,590 | 1,696 | 2,257 | 2,434 | 0.804 | -4.7 | Note: Shaded methods serve as references (mean diff 1.4%, r=0.997). Bootstrap confidence interval analysis (1,000 resamples) revealed mean ±9.8% uncertainty for 50-year records (range 8.2-11.6%), reflecting combined parameter estimation and sampling variability. Deterministic methods exhibited slightly wider intervals (10.8%) compared to probabilistic methods (9.3%). Among probabilistic approaches, Alexander and Gould Gamma showed narrowest CIs (9.1%), McMahon widest (10.2%). Coefficient of variation uncertainty contributed most substantially; Cv standard error of 0.023 propagates to ~10% capacity uncertainty given K/μ ∝ Cv^1.5 to Cv^2 relationship. These uncertainty levels provide adequate precision for engineering design when combined with safety factors. 3.3. Cross-Station Validation Inter-station coefficient of variation for dimensionless capacity (K/μ) ranged 0.06-0.14 across methods, meeting established consistency criteria (CV<0.15). Distribution-matched methods exhibited higher consistency than mismatched approaches. Gamma prevalence at all four stations, combined with similar findings from previous Turkish studies, suggests Gamma may be normative for Turkish semi-arid basins generally. Pattern consistency across stations with different drainage areas (3,005-9,310 km²), record lengths (42-65 years), and elevations (1,156-1,538 m) demonstrates robustness within hydrologically similar regions and supports extrapolation to ungauged sites with comparable physiographic characteristics. Cross-station validation results are presented in Table 4, demonstrating the generalizability of method performance across different hydrological regimes. The low coefficient of variation (CV<0.15) for dimensionless capacity ratios indicates that relative method performance remains consistent despite substantial differences in basin characteristics and mean flows. Table 4. Cross-station capacity estimates (70% regulation, 5% risk) | Rippl | 2,189 | 678 | 531 | 0.819 | 0.073 | 0.089 | | Sequent-Peak | 2,180 | 674 | 527 | 0.815 | 0.072 | 0.088 | | Alexander | 1,934 | 572 | 448 | 0.705 | 0.048 | 0.068 | | Dincer | 1,798 | 524 | 410 | 0.650 | 0.041 | 0.063 | | McMahon | 2,245 | 673 | 527 | 0.825 | 0.062 | 0.075 | | Gould Gamma | 1,951 | 581 | 455 | 0.714 | 0.052 | 0.073 | Note: Low CV (<0.15) validates generalizability. Statistical significance testing of systematic biases is summarized in Table 5, using Alexander method as the reference standard. Bootstrap resampling (1,000 iterations) revealed that Dincer and McMahon methods exhibit highly significant systematic biases (p<0.001), while Gould’s Gamma method shows no significant deviation (p=0.476), further supporting its equivalence to the Alexander approach. Table 5. Systematic bias relative to Alexander method | Rippl | +16.2% | -1.2% | +1.3% | -14.2% | +4.4±8.2% | 0.303 | NS | | Sequent-Peak | +15.6% | -1.6% | +0.9% | -14.5% | +4.0±8.0% | 0.330 | NS | | Minimum Flow | +35.3% | +18.9% | +17.6% | +2.8% | +22.3±9.3% | 0.011 | * | | Dincer | -7.8% | -8.9% | -8.5% | -9.8% | -8.8±0.8% | <0.001 | *** | | McMahon | +17.0% | +18.1% | +15.6% | +15.6% | +16.6±1.2% | <0.001 | *** | | Gould Gamma | +1.3% | +1.5% | +1.4% | +1.2% | +1.4±0.1% | 0.476 | NS | | Gould Synthetic | -2.0% | -8.2% | -1.4% | -7.0% | -4.7±3.4% | 0.050 | * | Note: *p<0.05, ***p<0.001. Pink=systematic bias. Bootstrap n=1,000. 4. DISCUSSION 4.1. Physical Mechanisms Underlying Negligible Autocorrelation Consistent negligible autocorrelation across all stations reflects fundamental hydro-climatological mechanisms specific to continental highland semi-arid regions. Annual snowpack reset effectively decouples consecutive years’ runoff—each year’s snowmelt depends on current-year winter precipitation rather than previous conditions, creating hydrological ”reset” breaking inter-annual memory. Limited groundwater storage in fractured crystalline bedrock (specific yield inter-annual carryover through subsurface pathways. Quantitative assessment reveals total groundwater storage of ~31,500 Mm³ represents only 1.6 years equivalent runoff, with long residence times (8-20 years) limiting active annual exchange to relief ratio 0.008) and rapid drainage (time-of-concentration 18-24 hours) minimize surface storage. High runoff coefficient (0.63) indicates limited infiltration opportunity. Monthly flow analysis shows 55-60% of annual runoff occurs in April-May snowmelt peak, with 95-98% evacuated by September, leaving <2-5% carryover insufficient for autocorrelation. These mechanisms validate independence assumptions fundamental to Alexander and Gould Gamma methods. 4.2. Economic Implications and Method Selection

selection differences translate directly to construction costs and failure risks. For typical 2,000 Mm³ reservoir at $30-50/m³ unit cost, Dincer’s 8.8% underestimation (176 Mm³ deficit) saves $35M construction but increases actual failure risk from 5.0% to 50-year life exceeds construction savings, yielding net economic loss. McMahon’s 16.6% overestimation (332 Mm³ excess) costs additional $66M with reliability improvement from 5.0% to ~3.8% risk, but cost-benefit analysis reveals net loss as construction expense exceeds failure cost reduction. Alexander and Gould Gamma provide optimal cost-reliability balance, with their exceptional agreement (1.4%) providing confidence interval substantially smaller than systematic biases from distribution-mismatched methods. For Gamma-distributed data, we recommend applying both methods as mutual verification; agreement <5% indicates high confidence, while larger discrepancies warrant data reexamination. 4.3. Climate Change Implications Regional climate projections (10-20% precipitation reduction, 2-4°C warming by 2100) have implications extending beyond simple mean flow adjustment. Temperature increase advances snowmelt timing 2-4 weeks, potentially misaligning peak runoff with irrigation demand and increasing evapotranspiration 15-25%. Mean flow reduction of 10-20% effectively increases regulation ratio; reservoir designed for 70% may operate at 77-84% under reduced flows, entering non-linear capacity escalation zone where small demand increases require disproportionate capacity additions. Distribution shape changes complicate matters—warming may reduce snowpack contribution, potentially altering from current Gamma toward more variable or skewed distributions, affecting method applicability. We recommend 1.15-1.30 safety factors for 2050-2100 horizons, scenario analysis using downscaled projections, adaptive management frameworks enabling periodic reassessment, and flexible infrastructure allowing future expansion. Portfolio approaches diversifying water sources beyond single large reservoirs provide additional resilience. 4.4. Study Limitations and Future Research Several limitations warrant consideration. Annual time steps neglect within-year dynamics including demand-supply mismatches, seasonal evaporation variations, and flood control constraints; monthly simulation typically reveals 5-15% higher capacity requirements for systems with strong seasonal patterns. Independent station analysis ignores cascade reservoir interactions where upstream operations affect downstream inflows. Data uncertainties remain despite quality control, including rating curve errors, gap-filling estimation uncertainty, potential land use impacts on record homogeneity, and spatial variability. Stationarity assumptions may be violated under climate change. Future research priorities include monthly simulation analysis quantifying annual method bias, cascade system optimization for Keban-Karakaya-Atatürk sequence, comprehensive Turkish recalibration developing regionally-specific empirical equations, climate non-stationarity incorporation through time-varying parameter methods, economic optimization balancing construction costs against failure risks, and transferability analysis to other semi-arid regions. 5. CONCLUSIONS This systematic comparison of nine critical period methods using 42-65 year records from four Upper Euphrates stations provides evidence-based guidance for reservoir capacity estimation in semi-arid regions. Our principal findings include: All stations exhibited Gamma-distributed annual streamflow with negligible autocorrelation (|r₁|<0.11), validating distribution-based selection and independence assumptions. Physical mechanisms—annual snowpack reset, limited groundwater storage, steep topography, rapid drainage—explain this behavior. Alexander and Gould’s Gamma methods demonstrated exceptional agreement (mean difference 1.4%, r=0.997, p=0.58), establishing them as co-equal reference standards and providing mutual validation despite independent derivation. We quantified systematic biases with high precision: Dincer underestimated 8.8%±0.8% (p<0.001) due to Normal distribution mismatch correlating with skewness; McMahon overestimated 16.6%±1.2% (p<0.001) reflecting Australian calibration; Rippl implicitly represented ~2% risk for 40-60 year records. Regulation ratio sensitivity proved strongly non-linear (70%→80% yielded 41% capacity increase with 3:1 amplification); risk level sensitivity exhibited logarithmic relationship (5%→3% yielded 14% increase). Bootstrap analysis indicated ±9-11% uncertainty for 50-year records, adequate precision when combined with safety factors. Coefficient of variation uncertainty dominated, propagating to capacity estimates given K/μ ∝ Cv^1.5-2.0 relationship. Economic analysis demonstrated Alexander/Gould Gamma provide optimal cost-reliability balance. Distribution-mismatched methods incur either excess failure costs (Dincer) or unnecessary construction expenses (McMahon), both yielding net economic losses. Climate projections suggest 1.15-1.30 adjustment factors for 2050-2100 horizons. Regulation ratio drift from declining flows requires particular attention given strongly non-linear sensitivity. For Gamma-distributed streamflow with low autocorrelation—characteristic of semi-arid snowmelt-dominated basins—we unequivocally recommend Alexander and Gould’s Gamma methods, supplemented by 1.25-1.30 safety factors accommodating climate uncertainty. Apply both methods for mutual verification; agreement identified knowledge gaps through longest records used in Turkish reservoir studies, most comprehensive method comparison, explicit bias quantification enabling correction factors, multi-station validation strengthening generalizability, and climate adaptation guidance. Results transfer to global semi-arid regions exhibiting similar characteristics. As water systems globally face mounting pressures, robust design methodologies become increasingly critical. This research contributes evidence-based guidance ultimately supporting more reliable, economically efficient, and sustainable water infrastructure in semi-arid regions worldwide. ACKNOWLEDGMENTS The authors gratefully acknowledge the General Directorate of State Hydraulic Works (DSI) and former Electrical Power Resources Survey and Development Administration (EIE) for providing long-term streamflow data essential for this research. This study is based on the M.Sc. thesis completed by the second author (Sinan Bazancir) at the Department of Civil Engineering, Ataturk University, Erzurum, Turkey, under the supervision of the first author (Serkan Şenocak). Serkan Şenocak: Conceptualization, Methodology, Formal analysis, Investigation, Writing - original draft, Writing - review & editing, Supervision, Project administration. Sinan Bazancir: Investigation, Data curation, Formal analysis, Writing - review & editing. AUTHOR CONTRIBUTIONS This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. FUNDING The streamflow datasets analyzed during this study are available from the General Directorate of State Hydraulic Works (DSI), Turkey, and the former Electrical Power Resources Survey and Development Administration (EIE). Restrictions apply to the availability of these data, which were used under license for this study. Data are available from the authors upon reasonable request and with permission of DSI. DATA AVAILABILITY Conflict of Interest: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Ethical Approval: This article does not contain any studies with human participants or animals performed by any of the authors. Declaration of Generative AI and AI-Assisted Technologies in the Writing Process During the preparation of this manuscript, the authors used Claude 3.5 Sonnet (Anthropic) to improve language clarity, refine technical descriptions, and enhance the overall structure and readability of the text. The AI tool was employed specifically to: 1. Improve grammatical accuracy and sentence flow in the manuscript 2. Enhance clarity of complex methodological explanations 3. Suggest alternative phrasing for technical concepts to improve reader comprehension 4. Assist with structural organization of sections All AI-generated suggestions were critically reviewed, extensively edited, and verified by the authors. The authors take full responsibility for the accuracy, scientific validity, and integrity of the published content. No AI tool was used in data collection, analysis, interpretation of results, or generation of scientific conclusions.

1. Adeloye AJ, Montaseri M (2002) Preliminary streamflow data analyses prior to water resources planning study. Hydrol Sci J 47(5):679-692 Alexander GN (1962) The use of the gamma distribution in estimating regulated output from storages. Civ Eng Trans Inst Eng Aust 4(1):29-34 Bacanlı ÜG, Özkul S, Baran T (2003) Design of storage reservoirs by different methods in the Dicle Basin Botan River example (in Turkish). Proc 1st Natl Water Eng Symp, Gümüldür, İzmir, Turkey, pp 389-401 Bayazit M (1997) Biriktirme Haznelerinin Tasarımı ve İşletilmesi (Reservoir Design and Operation, in Turkish). Istanbul Technical University, Istanbul, 197 pp Bazancir S (2013) Comparison of critical period methods for reservoir design (in Turkish). M.Sc. Thesis, Department of Civil Engineering, Ataturk University, Erzurum, Turkey Bozkurt D, Sen OL (2013) Climate change impacts in the Euphrates-Tigris Basin based on different model and scenario simulations. J Hydrol 480:149-161 Burnham KP, Anderson DR (2002) Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach, 2nd edn. Springer, New York Dincer T (1966) The use of the gamma distribution in regulated reservoir studies. J Hydrol 4:39-50 DSI (2021) Water Resources of Turkey. General Directorate of State Hydraulic Works, Ankara, Turkey Freeze RA, Cherry JA (1979) Groundwater. Prentice-Hall, Englewood Cliffs, New Jersey Gould BW (1961) Statistical methods for estimating the design capacity of dams. J Inst Eng Aust 33(12):405-416 Gould BW (1964) Statistical methods for reservoir yield estimation. Water Research Foundation of Australia, Report No. 2, Melbourne Halden ŞM, Özkul S (2004) Comparison of Rippl method and Alexander method in reservoir design (in Turkish). İMO Tek Der 15(1):3051-3069 Hosking JRM (1990) L-moments: analysis and estimation of distributions using linear combinations of order statistics. J R Stat Soc Ser B 52(1):105-124 Kandemir ŞY (2009) Comparison of reservoir design methods in Gediz Basin (in Turkish). PhD Dissertation, Dokuz Eylül University, İzmir Loucks DP, van Beek E, Stedinger JR, Dijkman JPM, Villars MT (2005) Water Resources Systems Planning and Management. UNESCO Publishing, Paris McMahon TA, Adeloye AJ (2005) Water Resources Yield. Water Resources Publications, Littleton, Colorado McMahon TA, Mein RG (1986) River and Reservoir Yield. Water Resources Publications, Littleton, Colorado Rippl W (1883) The capacity of storage reservoirs for water supply. Minutes Proc Inst Civ Eng 71:270-278 Thomas HA, Burden RP (1963) Operations Research in Water Quality Management. Harvard Water Resources Group, Harvard University, Cambridge, MA Türkeş M, Erlat E (2003) Precipitation changes and variability in Turkey linked to the North Atlantic Oscillation during the period 1930-2000. Int J Climatol 23(14):1771-1796 Vogel RM, Bolognese RA (1995) Storage-reliability-resilience-yield relations for over-year water supply systems. Water Resour Res 31(3):645-654 Vörösmarty CJ, Green P, Salisbury J, Lammers RB (2000) Global water resources: vulnerability from climate change and population growth. Science 289(5477):284-288 WMO (2008) Guide to Hydrological Practices Volume I: Hydrology—From Measurement to Hydrological Information, 6th edn. WMO-No. 168, World Meteorological Organization, Geneva Data Availability Statement The monthly streamflow data from Turkish State Hydraulic Works (DSI) gauging stations analyzed in this study are available upon reasonable request from DSI (https://www.dsi.gov.tr). Data access requires formal request to DSI in accordance with Turkish government data sharing policies. Python code implementing all nine critical period methods, statistical preprocessing, and bootstrap uncertainty analysis is available from the corresponding author upon reasonable request. Contact: Dr. Serkan Şenocak, [email protected], Department of Civil Engineering, Ataturk University, 25240 Erzurum, Turkey, ORCID: 0000-0002-4962-5349 Information & Authors Information Version history Copyright This work is licensed under a Non Exclusive No Reuse License.

Authors Metrics & Citations Metrics Article Usage 188views 122downloads Citations Download citation Serkan Şenocak, Sinan Bazancir. Systematic Evaluation of Critical Period Methods for Reservoir Storage Estimation: Distribution-Based Framework for Semi-Arid Basins. Authorea. 16 December 2025. DOI: https://doi.org/10.22541/au.176590686.62006932/v1 DOI: https://doi.org/10.22541/au.176590686.62006932/v1 If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download. For more information or tips please see 'Downloading to a citation manager' in the Help menu.