Temporal evolution of spatial water quality heterogeneity in Oman’s largest water supply reservoir

Abstract

Maintaining optimal water quality in reservoirs is a global challenge, particularly in arid or semi-arid regions where these water bodies are critical yet vulnerable ecosystems. A prerequisite for effective environmental preservation is the precise delineation of spatiotemporal and vertical variability in water-quality variables, especially across a complete seasonal cycle. This study was therefore designed to systematically investigate and document the detailed spatial and seasonal variation in the water quality of the Wadi Dayqah Dam (Oman). This characterization was based on a robust, year-long dataset spanning 19 separate sampling occasions and incorporating high-resolution depth profile measurements of temperature, dissolved oxygen (), oxidation-reduction potential (), pH, turbidity and chlorophyll-a from 20 distinct depth-profiling stations. Self-organizing map ) was used to identify spatial zones, while thermal stratification dynamics were quantified using the Schmidt Stability Index () and the Thermocline Strength Index (). This integrated approach combines high-resolution spatial clustering with complementary stratification metrics to reveal zone-specific water quality patterns typically obscured by single-point monitoring. differentiated distinctly phased lacustrine, transitional, and riverine regions with spatial heterogeneity in water quality, with pronounced longitudinal gradients in temperature, , turbidity, and chlorophyll-a across zones, an aspect traditionally overlooked by localized single-point, as opposed to distributed, observation. This spatial heterogeneity highlights the importance of zone-specific reservoir management strategies. On-site, observations revealed pronounced yearly patterns of thermal stratification and mixing, with the most intense and persistent stratification occurring in the lacustrine zone, where Schmidt stability reached a maximum of approximately 1144 . Notably, chemical stratification closely followed the temporal dynamics of thermal stratification, resulting in pronounced hypolimnetic anoxia in the deeper lacustrine waters, typically persisting from May to October. In contrast, suspended particulate matter and organic substances exerted a stronger control on turbidity and chlorophyll-a concentrations, with the latter reaching peak values of approximately 27 . Recurrent severe hypolimnetic anoxia was observed throughout the stratified period, indicating persistent water quality impairment in the deeper zones of this arid-region reservoir.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-54947-6.

Introduction

Reservoirs constitute indispensable infrastructure for global water security, supplying domestic, agricultural, and industrial demands, while also generating hydropower^1–3. Yet their long-term viability hinges on water quality management, a challenge compounded by these systems’ inherent dynamics. Unlike static pools, reservoirs function as complex, spatially heterogeneous ecosystems where water quality fluctuates across seasons, depths, and longitudinal gradients^4–6. Capturing this multidimensional variability remains pivotal for safeguarding their ecological and operational integrity.

Central to this complexity is stratification: the formation of density-separated layers that govern biogeochemical trajectories^7–9. Thermal stratification separates the water column into a warm, oxygenated surface layer (epilimnion) and a cold, isolated deep layer (hypolimnion), with a sharp temperature gradient (thermocline or metalimnion) between them. Once stratified, vertical mixing is suppressed, cutting off the hypolimnion from atmospheric oxygen supply. Without replenishment, decomposition of organic matter consumes the remaining oxygen until anoxic conditions develop^7,10,11. Critically, this physical barrier stifles reaeration, enabling organic matter decomposition that leads hypolimnetic dissolved oxygen () to anoxic () conditions^12–14. Such conditions trigger cascading deterioration: sediment-bound phosphorus and nitrogen mobilize under reducing conditions, fueling algal blooms^15,16, while redox-sensitive metals (e.g., , ) and toxic gases like (hydrogen sulfide) leach into the water column, compromising drinking water safety through taste, odor, and corrosion issues^17–19. Though thermal structure dominates these processes, nutrient influxes^20–22, light availability²³, and inflow/outflow regimes^22,24 further modulate biological responses such as chlorophyll-a proliferation.

Quantifying stratification thus demands integrated stability metrics. While indices like the Brunt-Väisälä frequency or Thermocline Strength Index () capture the intensity of thermal gradients^25,26, others, such as the Wedderburn Number (which characterizes resistance to wind-driven mixing)²⁷ or Schmidt Stability Index (; representing the total energy required to fully homogenize the water column)^28–30, quantify the resilience of the stratified system. Equally vital is chemical stratification, which may decouple from thermal patterns (e.g., Sabalan Reservoir¹³ yet critically influences water quality^14,31. The dimensionless Index of Chemical Stratification () introduced by Yu et al. (2010) was employed to assess chemocline strength: higher values, reflecting greater epilimnetic than hypolimnetic solute concentrations, were used to distinguish systems dominated by biogeochemical reactions from those controlled by physical mixing^25,32,33.

A critical methodological gap persists, however. Most stratification studies extrapolate from single-point measurements at deep stations^13,14,34, ignoring reservoirs’ longitudinal compartmentalization into riverine, transitional, and lacustrine zones, each with distinct hydrodynamics and biogeochemistry^35,36. This spatial oversimplification not only biases stability indices but also obscures lateral and longitudinal water quality gradients, leading to flawed management decisions. For instance, undetected anoxic conditions in shallow riverine or transitional zones may result in poorly sited or undersized aeration systems, reducing their effectiveness. Similarly, nutrient control strategies based solely on deep-station data may misallocate resources if pollution hotspots arise in under-sampled littoral or inflow-affected regions. Consequently, spatially blind monitoring undermines both diagnostic accuracy and the efficiency of remediation efforts. To resolve such complexity, multivariate techniques offer promise: Cluster Analysis () delineates spatial heterogeneity (e.g., Turkish³⁷, Chinese³⁸, Korean³⁹, Indian⁴⁰, and Ethiopian reservoirs^41,42, Principal Component Analysis () identifies drivers like natural vs. anthropogenic solute sources^43–45, and Discriminant Analysis () validates zonal classifications while pinpointing key monitoring variables^46,47. Together, they enable source apportionment and targeted intervention design.

This study confronts these gaps at Oman’s Wadi Dayqah Dam, the nation’s largest reservoir, by leveraging a high-resolution spatiotemporal dataset (20 depth-profiling stations across 19 sampling campaigns) and novel analytical integration. The study’s main contributions are threefold: (1) High-resolution spatial zoning via Self-Organizing Map (), capturing holistic water quality patterns across longitudinal gradients; (2) Cross-zonal stratification diagnostics through site-specific and reservoir-wide stability indices (, , ); (3) Actionable insights for management by explicitly linking spatial heterogeneity to water quality outcomes.

Study area

Wadi Dayqah Dam (23°5′N, 58°49′E), located approximately 83 km southwest of Muscat, Oman (Fig. 1a, b), is the nation’s largest reservoir with a storage capacity of 100 million cubic meters (). It serves as the primary source of potable water for the Muscat and Qurrayat regions and provides flood mitigation for downstream communities⁴⁸. The reservoir lies in an arid to semi‑arid environment with low, highly variable rainfall and high evaporative demand, conditions that promote strong and persistent thermal stratification^48,49. Such stratification, together with long residence times, can favor oxygen depletion in deep waters and redox‑driven changes in nutrient and metal cycling, as shown in stratified reservoirs under similar climates⁵⁰. Based on the reservoir’s hypsometric relationship (See Fig. S3), the water surface area at full supply level is approximately 400 ha (4.0), with a maximum depth of ~ 71 m near the dam wall. The bathymetry exhibits relatively steep-sided morphology in the lacustrine zone, transitioning to shallower, gently sloping margins in the riverine and transitional zones. The flow of water into the reservoir is irregular and occurs in short, intense events. These events are caused mainly by spring rainfall in April and occasional summer storms in the Eastern Hajar Mountains. During these times, flash floods move through the wadi channels, carrying large amounts of sediment, nutrients, and organic matter into the reservoir. This creates strong variations in water quality along the length of the reservoir^51,52. Water is withdrawn via an intake tower equipped with six gated outlets at elevations of 124, 128, 137, 146, 155, and 164 m above sea level. During the 2023 study period, active withdrawals occurred primarily from the 128, 146, and 155 m gates to supply: (1) drinking water (~ 1 month⁻¹) to Muscat and Qurrayat; and (2) biannual agricultural and environmental releases (~ 20 total), mainly to supply the environmental rights of farmers located in the downstream region of the dam⁵⁰. These coupled hydrometeorological and geomorphological controls make Wadi Dayqah an ideal natural laboratory for examining the interactions among climate forcing, reservoir operations, and biogeochemical processes that determine water quality in arid-region reservoirs^49,50.

Fig. 1.

(a) The geographical location of the study area. (b) The catchment area of the dam (≈ 1000 km²). (c) The in-situ water quality sampling points. This map was generated using ArcGIS v10.1 (https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview).

Materials and methods

The analytical framework for this study is depicted in Fig. 2. High-resolution field monitoring was first integrated with rigorous quality control procedures, followed by feature engineering to prepare the dataset for clustering. The analysis partitioned the reservoir into three hydrologically coherent zones (lacustrine, transitional, and riverine)³⁶, which then served as the spatial framework for parallel analysis. In the thermal stratification branch, reservoir-wide and local-scale stratification dynamics were quantified using and , respectively. A k-means clustering approach was subsequently applied to derive an objective, site-specific threshold for stratification onset, following the data-driven methodology established by Castrillo et al. (2024)⁵³. Simultaneously, water quality variables were analyzed within each -derived zone to assess spatial heterogeneity, temporal patterns, and cause-and-effect relationships between stratification phases and water quality impairment. Chemical stratification dynamics were then evaluated using the ²⁵, integrated with multivariate diagnostics to dissect the spatiotemporal dynamics of Wadi Dayqah. This pathway was designed to facilitate direct interrogation of how thermal and chemical stratification govern critical water quality variables within each reservoir zone. The approach thereby transcends limitations inherent in conventional single-point assessment methodologies.

Fig. 2.

An illustration of the proposed analytical framework for this study.

In situ water quality sampling and data preparation

Twenty sampling stations were selected from candidate locations based on two criteria applied jointly: (1) spatial representativeness across the full longitudinal gradient from the main inflow to the dam wall, ensuring coverage of all three reservoir zones; and (2) depth distribution, prioritizing stations that capture the full water column depth profile, particularly in the lacustrine zone. Logistical constraints (boat access and equipment deployment within a single campaign day) also constrained the final number to 20. These stations, dispersed throughout the Wadi Dayqah reservoir (Fig. 1c), were then utilized to collect in-situ water quality data. Using an AAQ-RINKO multi-variable water quality profiler (JFE Advantech Co., Ltd.), sampling was conducted over a one-year period comprising 19 campaigns starting from January 15th to December 6th. While monthly coverage was maintained (minimum once per month), frequency varied (1–3 samples/month) to capture seasonal transitions and specific hydrological events. All profiles were collected within a consistent daytime sampling window during each field campaign to maintain comparability among stations and minimize the potential influence of diel variability. In addition, operational data (inflow and withdrawal rates) were recorded to contextualize the hydraulic conditions (see Fig. S2 in supplementary materials). Vertical hydrochemical profiles were acquired from surface to reservoir bottom at all 20 sampling stations by profiler. Continuous measurements were recorded at 2–3 cm depth intervals, generating high-resolution records of temperature (), turbidity, chlorophyll-a, pH, oxidation-reduction potential (), and . All raw records underwent quality control: readings affected by bottom disturbance (agitated sludge or sediment) were identified and removed, and turbidity data were filtered to retain only values within the acceptable range of 0.3–6 ⁵⁴. The resulting dataset captures variations in the water quality variables along the length, depth, and width, as well as over time, providing a basis to investigate the reservoir’s spatiotemporal water quality dynamics.

Water quality zoning in the reservoir

Water quality zoning was implemented to delineate reservoir regions with coherent chemical and thermal signatures using an , an unsupervised neural network projecting high-dimensional data onto a low-dimensional grid⁵⁵. The is particularly well-suited to this task because it detects complex, nonlinear patterns and relationships in multivariate datasets without requiring preassigned class labels. s have been successfully deployed in prior water quality studies for spatial classification and pollution source apportionment^56,57, a capability extended herein through explicit integration of temporal dimensions into the clustering framework. To encapsulate seasonal cycles and hydroclimatic forcing events, the input matrix was compiled from spatiotemporal water quality records across all 20 sampling stations. Clustering features for each station were computed as the minimum, maximum, and mean values of six hydrochemical discriminants: temperature, turbidity, pH, , , and chlorophyll-a. This preprocessing step enabled the to delineate zones that concurrently represent spatial heterogeneity and temporal trajectories, capturing reservoir dynamics beyond static spatial snapshots.

Variables exhibiting minimal temporal variance were excluded to prevent clustering bias toward non-discriminatory dimensions, enhancing the interpretability of resultant hydrological zones. Specifically, temperature minima and maxima were omitted due to negligible inter-station variability, which was less than 0.5 , leaving only mean temperature as a representative thermal metric. The resulting feature set comprises 14 informative variables that provide a robust basis for the analysis; the complete input matrix and the retained variables are reported in Table S1 in Supplementary materials. All 14 hydrochemical variables were subsequently scaled to the range [0, 1] using Min-Max normalization prior to training, ensuring equal contribution of each feature to the distance calculations underlying the ‘s competitive learning process. The was implemented using the MiniSom library⁵⁸ in Python, configured as a 5 × 5 rectangular grid with a Gaussian neighborhood function, an initial neighborhood radius () of 1.5, and a learning rate of 0.5. The network was trained for 5,000 iterations using random weight initialization with a fixed random seed of 42 to ensure full reproducibility. The choice of a 5 × 5 topology yielding 25 neurons was guided by the commonly applied heuristic of N ≈ , where n is the number of observations^59,60; for n = 20 stations this yields N ≈ 22, making the 25-neuron grid the nearest standard rectangular topology that satisfies the criterion.

Following training, each station was assigned to its Best Matching Unit (), the neuron whose weight vector most closely matched the station’s normalized feature vector in Euclidean distance. The trained codebook, comprising the weight vectors of all 25 neurons, was subsequently passed to the k-means algorithm, which partitioned the nodes into k discrete clusters. Each station then inherited the cluster label of its , completing the two-stage , k-means pipeline⁶⁰. Cluster validity was quantified using the Silhouette Score⁶¹ and the Calinski–Harabász () Index⁶², both evaluated on the station-level normalized feature matrix using the propagated cluster labels.

The final zonation was consolidated into three clusters (k = 3) to reflect the canonical reservoir compartmentalization into lacustrine, transitional, and riverine zones, a framework widely adopted in reservoir limnology and management^35,36. The sensitivity of this choice was assessed by comparing solutions across k = 3, 4, 5, and 6, as well as across three normalization strategies: Min-Max scaling, Z-score standardization, and log-score transformation, to confirm the robustness of the selected configuration (see Supplementary Material, Fig. S4). Both validity indices declined consistently as k increased beyond 3 regardless of the normalization method applied, indicating diminishing cluster compactness and separation with finer partitioning. Solutions with k > 3 yielded additional clusters that could not be attributed to any recognisable hydrodynamic or limnological archetype within the reservoir, while k = 2 collapsed the transitional zone into one of the two end-member archetypes, obscuring ecologically meaningful longitudinal gradients along the reservoir axis. Taken together, the quantitative indices and the limnological interpretability of the resulting spatial patterns converge on k = 3 as the optimal zonation for this system.

Thermal stratification strength

Thermal stratification in Wadi Dayqah Reservoir was quantified using two complementary metrics:^63,64 and ²⁵. Both indices were computed from depth-resolved temperature profiles for each sampling date and bathymetric layer to characterize the energetic resistance to mixing () and the sharpness of the vertical temperature gradient (). was calculated as shown Eq. 1 using the standard discrete summation approximation of the continuous integral^63,64, which yields the energy per unit surface area () required to mix the stratified column to an isothermal condition:

1

2

3

Where is the cross-sectional area at depth (), denotes the reservoir surface area (), is depth (), is maximum depth (), is water density at depth (), which calculation formula is shown in Eq. 3 denotes mean density at depth (), which calculation formula is shown in Eq. 2, is depth to the center of gravity of stratified reservoir (), \:dz corresponds to the thickness of the bathymetric layer, and represents the layer volume (). SSI is a reservoir-wide metric, computed by integrating temperature–density profiles from all 20 stations across the full bathymetric volume using hypsometric data. To obtain continuous density fields across the reservoir, temperature and density measurements from the 20 stations were first vertically interpolated (linear interpolation) to standard depth intervals at each station. These vertically consistent station profiles were then spatially interpolated across the reservoir area using inverse distance weighting (), assigning density values to unsampled locations based on a weighted average of neighboring stations. The water surface elevation across the reservoir was treated as a spatially constant mean value for each sampling date, which is a standard simplification for compact reservoirs (surface area ≈ 4 at full capacity); the longitudinal gradient in surface elevation is small relative to the total depth range and does not affect the calculation. , in contrast, is a station-specific metric, calculated independently at P5 (lacustrine), P31 (transitional), and P6 (riverine) to capture zone-specific thermocline sharpness. The onset and termination of stratification were quantified by comparing the to a defined threshold. Although thresholds reported in the literature vary widely, from 30 ⁶⁵ to 800 ⁶⁶, a site-specific onset threshold was derived objectively via k-means clustering (k = 3) applied to the daily time series of , , and at the lacustrine station P5. This threshold represents the 25th percentile (Q1) of the Moderate stratification cluster. This value represents a site‑specific stratification onset threshold for Wadi Dayqah Reservoir and illustrates a methodological framework that may be applied to reservoirs with comparable morphometry and climatic forcing.

The () quantifies thermocline strength as shown in Eq. 4²⁵:

4

Where is the temperature difference recorded across the thermocline boundaries (), and is the vertical depth at which this temperature change occurs (). Thermocline presence was assessed in previous studies using a gradient threshold of 0.1 , consistent with Koue et al. (2018)⁶⁷ and Duka et al. (2021)⁶⁸; the substantially higher threshold of 1 suggested by Horne and Goldman (1994)²⁶ was judged excessive for the subtle gradients observed in this case study. Because raw temperature profiles often exhibit small, noisy variations that can obscure the location and magnitude of the steepest gradient, estimation was improved using the empirical function proposed by Rimmer et al. (2005) and corrected by Cook and Rimmer (2010)^53,69,70, which describes the temperature–depth relationship as shown in Eq. 5⁷⁰, where and \:n are shape parameters related to metalimnion height and thickness, m is calculated as shown in Eq. 6, is epilimnion temperature, and is hypolimnion temperature.

5

6

Parameters were obtained by non-linear least squares fitting with physically constrained bounds (; n > 1.01; max 20,000 iterations). The thermocline depth Z was calculated from the inflection point as shown in Eq. 7.

7

Where Rimmer fitting failed to converge, a linear regression fallback was applied and the fit type was recorded. This method produces a more robust and reproducible estimate of thermocline strength.

Chemical stratification strength

Chemical stratification strength is determined by the difference in concentration of a water quality variable between the surface (epilimnion) and deep (hypolimnion) layers. This strength is influenced by both the variable’s overall concentration and the reservoir’s eutrophication state. To account for these factors, Yu et al. (2010) proposed a dimensionless chemical stratification strength index, denoted as , which is calculated as shown in Eq. 8²⁵:

8

Where is the constituent concentration in the surface layer and is the constituent concentration in the deep reservoir layer. According to Shi et al. (2022), the value ranges from − 2 to + 2, with 0 representing perfect mixing¹⁴. When thermal stratification is weak (i.e., when thermal indices like and are low), an value close to zero indicates that physical mixing effects are more dominant than biogeochemical reactions. Conversely, an value consistent with the dominance of biogeochemical reactions over physical mixing under low thermal stratification may reflect external loading events (e.g., inflow-driven turbidity pulses) in addition to internal stratification effects, and should be interpreted accordingly. Note that can also be negative: negative values indicate higher deep-layer concentrations relative to the surface (e.g., accumulation of reduced substances such as dissolved metals or high-turbidity sediment layers at depth), while positive values indicate higher surface concentrations (e.g., and pH during stratification).

Results and discussion

clustering, applied to 14 informative variables including spatiotemporal water-quality data and partitioned the sampling stations into three hydrologically distinct zones: lacustrine, transitional, and riverine. The resulting 5 × 5 grid is provided in Fig. S5 of the Supplementary Materials, and its spatial configuration is presented in Fig. 3. The lacustrine zone, characterized by greater depth, stronger thermal stratification and minimal direct influence from riverine inflow, was composed of stations P4, P5, P7, P8, P9, P10, P16, P23, and P29. In contrast, the riverine zone, characterized by shallow depths, well-mixed conditions and dominant hydrological forcing from the primary inflow, included stations P6, P24, P27, and P34. Stations P14, P15, P17, P22, P31, P32, and P33 were assigned to the transitional zone, which exhibits hybrid hydrodynamic and biogeochemical traits intermediate between lacustrine and riverine archetypes.

Fig. 3.

Spatial clustering of the 20 sampling points into three hydrodynamic zones: lacustrine, transitional, and riverine. This map was generated using ArcGIS v10.1 (https://www.esri.com/en-us/arcgis/products/arcgis-desktop/overview).

Cluster validity was evaluated using the Silhouette Score⁶¹ and CH Index⁶², yielding values of 0.323 and 13.42, respectively. Among the alternatives tested, k = 3 in combination with Min–Max normalization produced the most favorable results; comparative evaluations across other k values and normalization schemes are provided in Fig. S4 of the Supplementary Materials. While indicative of only moderate inter-cluster separation, this metric was deemed acceptable in light of the high dimensionality of the input feature space and the constrained sample size inherent to reservoir-scale monitoring⁷¹. As corroborated by Dai et al. (2013), the resulting zonation is thus regarded as a robust and reproducible framework for subsequent interrogation of spatiotemporal water quality dynamics within the reservoir³⁶.

Thermal stratification analysis

Thermal stratification dynamics in Wadi Dayqah Reservoir were assessed through a two-step analytical workflow: (1) quantification of zone-specific thermocline strength via curve-fitted TSI, and (2) derivation of an objective, site-specific stratification onset threshold via unsupervised clustering. Station selection was guided by both representativeness and data availability. P5 (lacustrine zone) was chosen because it is located near the dam wall, is one of the deepest stations, and consistently exhibits strong thermal stratification with minimal riverine influence. P31 (transitional zone) was selected as it falls within the mean range of both geographic location and maximum depth for its zone, offering a balanced representation of transitional hydrodynamics. For the riverine zone, P34 was initially considered due to its position near the primary inflow, but it was excluded due to missing data on several sampling dates. P6, which belongs to the same -derived cluster as P34, was therefore selected as the most reliable alternative. Although P6 is located further downstream, its shallow depth and the influence of surrounding mountain inflows result in biogeochemical conditions closely resembling those at P34, making it a suitable representative of riverine dynamics. A summary of each station’s -assigned zone, is provided in Fig. S5 of Supplementary Materials.

Temperature profiles at the zone-representative stations P5 (lacustrine), P31 (transitional), and P6 (riverine) were fitted using the Cook & Rimmer (2010) function (Eq. 5) to estimate thermocline depth, epilimnetic/hypolimnetic temperatures (see Fig. S9), and (). Curve-fitting results, alongside Rimmer’s equation parameters, for all three zones are presented in Figs. S6–S8 of Supplementary Materials, demonstrating robust convergence for > 90% of profiles; linear regression fallbacks were applied where necessary and flagged for quality control. This approach aligns with established practices for deriving robust thermocline metrics from high-frequency vertical profiles^70,72.

To identify the most representative station for reservoir-scale stratification assessment, we performed Pearson correlation analysis between station-specific , surface–bottom temperature difference (), and the reservoir-wide . Results (See Figs. S10, S11 in Supplementary Materials) indicate that P5 (lacustrine) exhibits the strongest and most significant correlations with for both (r = 0.952, ρ = 0.934) and (r = 0.897, ρ = 0.89), confirming that deep-zone thermal dynamics best capture whole-reservoir stratification behavior, a pattern widely observed in deep reservoirs where lacustrine stations serve as effective proxies for system-wide stability^73–75. Consequently, P5 thermal metrics were selected as the basis for threshold derivation.

An objective stratification onset threshold was derived by applying k-means clustering (k = 3) to the standardized time series of three features: P5 , P5 , and reservoir-wide . The three resulting clusters were interpreted as Weak, Moderate, and Strong stratification regimes based on their centroid values and temporal coherence. The transition from Weak to Moderate stratification, defined as the onset threshold, was conservatively set at the 25th percentile (Q1) of the Moderate cluster to minimize false-negative detection in management applications. The derived threshold values are: = 261.9 , = 0.1576 , and = 3.75 . Descriptive statistics for each cluster are reported in Table S2 of Supplementary Materials.

Using this data-driven framework, the annual stratification cycle was partitioned into five temporal phases that map onto the three clustering regimes (Fig. S12) consistent with multi-stage stratification classifications reported in other monomictic reservoirs^76,77. The Weak stratification cluster encompassed Phases 1 and 5, characterized by isothermal or weakly stratified conditions dominated by wind-driven mixing and seasonal cooling^74,78: Phase 1 (January to mid March; < 261.9 ) coincided with negligible inflow (Fig. S2), confirming atmospheric forcing as the primary destratification mechanism, while Phase 5 (late October to December; < 261.9 ) reflected stratification collapse driven by surface cooling and elevated withdrawal rates. The Moderate cluster comprised Phases 2 and 4, representing transitional stratification with developing (Phase 2: mid March to mid May; 261.9 < < 744 ) or weakening (Phase 4: late September to late October; 261.9 < < 744 ) thermal gradients; Phase 2 exhibited a delayed stratification onset consistent with the seasonal rise in air temperature (Fig. 4), attributable to inflow-driven mixing during spring discharge peaks (total March–May discharge: 21.1 ; peak: 39.7 on 28 April 2023; Fig. S2), whereas Phase 4 showed a gradual SSI decline as the surface–bottom weakened. Finally, Phase 3 (mid May to late September; > 744 ), assigned to the Strong cluster, exhibited persistent, high-amplitude stratification with maximum (1144 ), surface–bottom (~ 8.26 °C) and (0.4 ), corresponding to minimal inflows, maximal solar heating, and a stable water column structure that created conditions conducive to hypolimnetic oxygen depletion, a well-documented consequence of prolonged strong stratification in deep reservoirs^66,79,80.

Fig. 4.

Temporal variations in temperature profiles within three distinct zones: (a) lacustrine, (b) riverine, and (c) transitional.

Zone-specific comparisons reveal marked spatial heterogeneity in stratification behavior. While the lacustrine zone (P5) exhibited the full five-phase cycle with pronounced thermal gradients (Fig. 5a), the transitional (P31) and riverine (P6) zones showed only ephemeral, shallow stratification even during Phase 3. Although TSI time series (Fig. 5b and c) illustrate that localized thermocline intensification occasionally occurred at P31 and P6, peaking at 0.355 and 0.42 respectively, these instances generally did not coincide with and values exceeding the derived stratification onset thresholds. Consequently, despite these localized gradients, the majority of these periods were classified within the Weak stratification cluster rather than the Moderate or Strong regimes⁷⁹. The sole exceptions occurred at P31 on June 9th and July 4th, when conditions met the stratification threshold criteria. Atmospheric forcing and inflow turbulence dominated mixing in these shallower regions, limiting the development of stable density gradients required to satisfy the multi-metric classification. This underscores the importance of multi-station monitoring for capturing longitudinal variability.

Fig. 5.

(a) Time series of SSI; . (b–d) TSI; for the reservoir zones: (b) lacustrine, (c) transitional, and (d) riverine, panels (b–d) show the zonal temporal variability in thermocline strength.

Changes in water quality variables with reservoir depth

The vertical distribution of measured water quality variables: , , chlorophyll-a, , and turbidity, was analyzed across the three distinct zones of the reservoir. Consistent with the temperature analysis, sampling points P5, P31 and P6 were chosen to represent the lacustrine, transitional, and riverine zones, respectively. The corresponding results for each zone are presented in Figs. 6 and 7, and 8.

Fig. 6.

Temporal variations of water-quality variables profiles in the lacustrine zone: (a) DO, (b) ORP, (c) Chl-a, (d) pH, and (e) turbidity.

Fig. 7.

Temporal variations of water-quality variables profiles in the riverine zone: (a) DO, (b) ORP, (c) Chl-a, (d) pH, and (e) turbidity.

Fig. 8.

Temporal variations of water-quality variables profiles in the transitional zone: (a) DO, (b) ORP, (c) Chl-a, (d) pH, and (e) turbidity.

Lacustrine zone

In January and February, concentrations near the bottom of the lacustrine zone ranged from 5 to 7 (Fig. 6a). During this period, no anoxic regions were detected in the reservoir. Elevated wind speeds, along with thorough mixing and reoxygenation between surface and deeper layers, led to consistently high levels throughout the water column. This pattern aligns with findings by Saber et al. (2019)⁸⁰. By March, anoxic conditions ( concentration < 2 ;⁸¹ began to develop at the bottom. Following the onset of stratification in April, a distinct vertical gradient in emerged, resulting in a completely anoxic hypolimnion ( concentration = 0 ). Surface oxygen concentrations remained relatively high ( concentration in 7–9 range) due to atmospheric reoxygenation and algal photosynthesis, as also reported by Shi et al. (2022)¹⁴. This trend of stratification was sustained through October when the thermally driven decomposition of the stratification allowed for the re-oxygenation of the hypolimnion. The primary cause of the oxygenation deficit in the reservoir was intense thermally driven stratification. In November and December, the conditions reverted to the start of the year. These trends align with those of Noori et al. (2021), Shi et al. (2022) and Ma et al. (2025)^13,14,82.

Even as concentrations dropped to 0 in the hypolimnion, remained relatively stable between April and June, fluctuating between + 75 mV and + 150 mV (Fig. 6b), suggesting the presence of other oxidizing agents such as nitrate or ferric iron¹⁴. This stability occurs because the reduction of and , oxidizing agents that succeed consumption, helps buffer the before it plunges to sharply negative values associated with sulfate or methane reduction⁸³. Nevertheless, from June to July, in the hypolimnion experienced a sharp decrease, from + 110 mV to -200 mV, pushing the environment toward more reducing conditions. This condition persisted until October, reaching its minimum in September ( = -366 mV), then the overall increased in the water column to + 168 mV in November and + 105 mV, due to mixing and reoxygenation of the reducing zone of the water column.

This zone also experienced a major chlorophyll-a event (Fig. 6c). While chlorophyll-a levels were nearly uniform across all depths during isothermal months, they hit an annual peak of 26.96 in the hypolimnion in September. While elevated chlorophyll-a below the thermocline may seem unexpected given light limitation, deep chlorophyll maxima are a documented feature of many stratified water bodies^10,84,85. Several non-exclusive mechanisms could contribute to this pattern: (i) phytoplankton may accumulate near the metalimnion where light and nutrients are both available; (ii) senescent biomass from surface blooms may settle downward; (iii) or episodic mixing may temporarily redistribute particles. Persistent low turbidity (0–3 ; Fig. 6e) suggests that sediment resuspension was unlikely to dominate under observed conditions. Additionally, fluorometer-based chlorophyll-a estimates can be influenced by water optical properties, though no site-specific calibration was available to assess this effect. The sharp decline in chlorophyll-a following a deep withdrawal event in October (total monthly outflow: 7.13 ; peak daily rate: 15.1 ; Fig. S2) is consistent with physical export influencing vertical distributions. While these observations are presented as a tentative management hypothesis, that strategically timed deep withdrawals could help modulate hypolimnetic biomass, confirmation would require targeted measurements or modeling beyond the scope of this study.

at P5 recorded a broader range of 6–8.5 (Fig. 6d). Notably, pH levels in the hypolimnion fell during periods of low oxygen. This is a common and expected observation, typically driven by the accumulation of carbon dioxide () produced from the decomposition of sinking organic matter, which forms carbonic acid () and lowers the water’s in the isolated hypolimnion⁸⁶. Throughout the year, turbidity in the lacustrine zone remained consistently low, typically in the 0–3 range across most of the water column (Fig. 6e). The low turbidity variation here is due to its limited influence from inflows, which are the primary drivers of turbidity through factors like suspended particles, total dissolved solids, and algae concentrations⁸⁷.

Riverine zone

The riverine zone exhibited significantly different dynamics compared to the lacustrine and transitional zones (see Fig. 7). No instances of anoxia were observed throughout the study period, with concentrations consistently exceeding 2 ( minimum = 2.4 on May 30th ) (Fig. 7a). This persistent oxygenation is consistent with enhanced vertical mixing driven by wind forcing and inflow-induced turbulence, which maintain water column homogeneity in shallow, hydrologically active regions. While slightly lower levels were observed in deeper portions of the riverine zone during May and June, these values remained well above anoxic thresholds and likely reflect seasonal reductions in wind-driven mixing rather than stratification-driven isolation.

measurements in the riverine zone exhibited modest variability, with values fluctuating between − 25.7 mV and + 22.1 mV during the June sampling campaign (Fig. 7b). This transient decrease may be attributed to elevated inflows in April and May (total March–May discharge: 21.1 ; peak: 39.7 on 28 April 2023; Fig. S2), which likely introduced pulses of organic matter and nutrients that temporarily increased oxygen demand^88–90. The absence of negative values later in the year indicates recovery to sufficiently oxidizing conditions, consistent with the zone’s high flushing rates and limited residence time. Nutrient concentrations are also typically elevated in shallow, near-inflow regions due to catchment loading⁹¹, which may further contribute to short-term redox fluctuations.

This zone did not experience eutrophication, with peak chlorophyll-a concentrations reaching only 4.2 in January (Fig. 7c). Turbidity was consistently higher than in the lacustrine zone (Fig. 7e), reflecting greater suspended sediment loads associated with inflow events and shallow-water resuspension. pH remained slightly alkaline throughout the year, fluctuating narrowly between 7.5 and 8.0 across most of the water column (Fig. 7d), well within typical ranges for drinking water sources. Collectively, these observations: persistent oxygenation, elevated turbidity, limited algal biomass, and stable pH, are consistent with the hydrodynamic and material-loading characteristics expected for riverine reservoir zones, mirroring patterns documented in other inflow-dominated systems^79,92.

Transitional zone

For most of the year, the transitional zone exhibited dynamics similar to the riverine zone, with effective mixing (Fig. 8a). However, because this zone is somewhat deeper than P6, it is more susceptible to thermal stratification, leading to more pronounced stratification dynamics developed in the deeper layers of this zone over an approximately two-week period in late spring: concentrations began to decline on May 25th and reached a minimum of 0.035 on June 6th at depths below 20 m (Fig. 8a). This transient anoxic episode coincided with strong thermal stratification (SSI > 261.9 ) and was confined to the deeper portion of the transitional zone, where local depth exceeds 20 m. readings in this zone were never negative compared to P5 and P6, suggesting a lower rate of organic matter buildup (Fig. 8b). Chlorophyll-a concentrations did not indicate eutrophication, peaking at 4.68 in January (Fig. 8c). The was nearly identical to the riverine zone, fluctuating between 6.5 and 8.5 (Fig. 8d). Turbidity variations were primarily driven by river inflow peaking at 6.34 in April and were more pronounced in this zone than in others due to its location upstream of the reservoir (Fig. 8e). Overall, transitional zones are characterized by a mix of riverine and lacustrine characteristics; while the P31 sampling point analyzed here displayed trends closer to the riverine zone, other points within the transitional area could exhibit greater similarity to the deeper, stratified lacustrine zone.

Chemical stratification of the reservoir

Relationships between and for DO (), pH (), chlorophyll-a Chl-a), and turbidity () at three representative sampling points in Wadi Dayqah Dam are shown in Fig. 9. For P5 (lacustrine zone), increased with rising until the hypolimnion became fully anoxic, at which point thermal stratification stabilized and reached its theoretical maximum of 2 (Fig. 9a). This indicates that distribution at P5 was strongly controlled by thermal stratification. In contrast, values at P31 (transitional zone) and P6 (riverine zone) remained close to zero throughout most of the study period. A brief spike in at P31 coincided with a transient anoxic event during one sampling campaign under high , but stratification quickly dissipated, leading to rapid recovery of vertical homogeneity.

Fig. 9.

Relationships between SSI and for: (a) DO, (b) pH, (c) chlorophyll-a and (d) Turbidity.

values were consistently near zero across all zones, showing only slight increases at higher levels (Fig. 9b). This reflects the inherent buffering capacity of the reservoir water, which limits substantial pH shifts in response to either biogeochemical reactions or physical stratification processes. No clear relationship was observed between Chl-a and at any of the sites. At P6 and P31, Chl-a remained around − 1, while at P5 it exhibited greater variability, suggesting that chlorophyll-a dynamics, particularly in the lacustrine zone, are influenced more strongly by factors other than thermal stratification, such as nutrient availability or light regime (Fig. 9c). Similarly, showed patterns comparable to Chl-a (Fig. 9d), likely due to their close interdependence through algal biomass and suspended particulate matter.

To characterize chemical stratification dynamics across the reservoir, dimensionless values for DO, pH, chlorophyll-a, and turbidity were computed at representative stations P5 (lacustrine zone), P31 (transitional zone), and P6 (riverine zone) throughout the monitoring period (Supplementary Table S3). These stations were selected to capture the dominant zonal trends observed within their respective regions, although some intra‑zonal variability may exist.

Complete vertical mixing, defined as () ≈ 0, was rarely observed for and pH during the stratified period from April to October, particularly in the lacustrine zone where () consistently approached the theoretical maximum of 2.0 under persistent hypolimnetic anoxia. Near-complete mixing ( < 0.1) occurred only during the isothermal periods in January and November–December. In contrast, the riverine zone exhibited weak stratification, with () < 0.4 observed on 14 out of 19 sampling dates at P6, reflecting frequent vertical homogenization driven by wind forcing and inflow turbulence. The transitional zone displayed intermediate behavior, with transient anoxic events producing brief spikes ( > 0.9) during May and June that rapidly dissipated due to mixing. pH stratification remained weak across all zones, with () consistently below 0.3, which is consistent with the reservoir’s inherent buffering capacity.

For biologically mediated parameters, namely chlorophyll-a and turbidity, values displayed greater temporal variability with frequent sign reversals. These reversals indicate shifts between surface-enriched and bottom-enriched conditions depending on algal growth phases, inflow pulses, and sediment resuspension events. This decoupling from thermal stratification patterns confirms that non-thermal drivers dominate the vertical distribution of optical water quality parameters in this arid-zone reservoir.

Chemical stratification correlation analysis

Comprehensive correlation analyses of for , pH, chlorophyll-a, and turbidity across the three monitoring stations confirmed that chemical stratification patterns can be systematically quantified and compared along the reservoir’s longitudinal gradient, consistent with the original framework and its integration with thermal stability metrics (See Figs. S13 to S19).

Exceptionally strong / correlations in the lacustrine zone (r = 0.914) were interpreted as evidence of tightly coupled metabolic processes, aligning with established observations that DO and pH change together over time due to photosynthetic oxygen production and respiratory consumption^93–95. Positive DO–pH relationships, and positive correlations of both with Chl‑a under productive conditions, have been reported in reservoirs and other aquatic systems where algal photosynthesis elevates both oxygen and pH and respiration lowers them. Strong -a / associations (r = 0.833 at P5) supported the interpretation that algal biomass significantly contributes to turbidity in stratified zones, consistent with reservoir and estuarine studies documenting positive chlorophyll-a –turbidity relationships driven by phytoplankton optical properties^96–98.

Spatiotemporal distributions of water quality variables

To account for variations in space, time, and depth, boxplots were used to illustrate the distribution of each variable at five-meter depth intervals within each zone (depth below surface). Data were pooled across all stations within each zone and across all sampling dates within each stratification phase (stratified vs. non-stratified). The lacustrine zone includes 9 stations (n = 99 per depth bin during stratified conditions, 72 during non-stratified), the transitional zone 7 stations (n = 77 stratified, 56 non-stratified), and the riverine zone 4 stations (n = 44 stratified, 32 non-stratified). Representing the data at five-meter depth intervals was necessary due to challenges in precisely visualizing the data at finer scales. This section compares the general spatiotemporal and depth-related behavior of the variables. The distribution of variation at five-meter intervals is shown in Fig. 10a–c. Differences were observed in the maximum measured depth between stratified and non-stratified periods across both the lacustrine and riverine zones. During stratified periods, the deepest measurement was recorded at a lower depth in the lacustrine zone, remained unchanged in the transitional zone, and was deeper in the riverine zone. These differences are attributed to a combination of access limitations (some deeper stations could not be sampled) and quality control trimming (near-bottom noise or sediment disturbance led to profile truncation). Such variability is considered inherent to field sampling in a dynamic reservoir and is acknowledged here to aid interpretation. Initially, the study focuses on how stratification affects changes with depth. As expected, on stratified dates DO levels in the deeper portions of the reservoir were consistently lower than in surface waters across all zones, with the deficit being greatest in the lacustrine zone, intermediate in the transitional zone, and smallest in the riverine zone. Surface DO levels remained greater than at depth on all sampling dates. For , as shown in Fig. 10d–f, average values are relatively uniform among zones and sampling days. The riverine zone does experience a slight negative shift in , primarily due to the fact that there is greater nutrient input in this zone. Both riverine and lacustrine zones experience more negative values at greater depths. Generally, values are smaller in the stratified phase compared to non-stratified phases, reflecting the influence of stratification on the redox state of the reservoir. In the lacustrine and transitional zones, the average concentration of chlorophyll-a fluctuated between 1 and 2 throughout the stratification period, while in the riverine zone, it remained steady at around 2 (Fig. 10g-i). However, signs of eutrophication were observed in the bottom layers at certain locations in September. During the stratification-free period, chlorophyll-a concentrations were generally higher across the three zones, ranging from 2 to 3 , which is due to increased mixing and increased oxygen availability in the water column. As stratification progressed, levels invariably decreased with depth in all three zones, with the lowest of 6 being recorded at the deepest point in the lacustrine zone, indicating mildly acidic conditions (Fig. 10j-l). Despite this observation, values in all three zones continuously fell within the acceptable range of potable water⁹⁹, signaling no alarm. Additionally, turbidity followed an increasing pattern from the stratified to the non-stratified season for all zones, with the riverine zone recording high turbidity values compared to the transitional and lacustrine zones (Fig. 10m-o) .

Fig. 10.

The distribution of Water quality variables across different depth intervals within three distinct environments: (a–c) DO, (d–f) ORP, (g–i) Chlorophyll-a, (j–l) pH, and (m–o) Turbidity.

Study limitations

Certain limitations regarding monitoring scope should be acknowledged. The absence of concurrent nutrient measurements (nitrogen, phosphorus) precluded detailed analysis of nutrient-algal feedbacks, while limited carbonate system data (, alkalinity) constrained mechanistic interpretation of pH dynamics. Similarly, although high-frequency data were available, the lack of concurrent measurements of redox-sensitive metals (e.g., iron, manganese) precluded direct assessment of their biogeochemical coupling. Furthermore, while this study established a robust link between thermal stratification metrics and hypolimnetic anoxia, direct quantification of hypolimnetic oxygen demand () and external organic matter loading was not feasible. Notwithstanding these constraints, the methodological framework presented here remains transferable to other arid-zone reservoirs, and future work incorporating these additional variables would further refine predictive capacity for water quality management.

Concluding remarks

This study systematically characterized the spatiotemporal water quality dynamics of Wadi Dayqah Dam through an integrated framework combining -based spatial zoning, process-specific stratification metrics, and chemical stratification diagnostics. clustering delineated three hydrologically distinct zones: lacustrine, transitional, and riverine, revealing pronounced longitudinal heterogeneity in thermal structure and water quality responses. The lacustrine zone exhibited persistent thermal stratification ( max = 1144 ) and seasonal hypolimnetic anoxia ( = 0 , April–October), accompanied by strong coupling between and (r = 0.914) indicative of tightly linked metabolic processes. In contrast, riverine and transitional zones displayed ephemeral stratification and minimal chemical stratification indices, reflecting inflow-driven mixing that decouples local dynamics from reservoir-scale stability.

A data‑driven derivation of a site‑specific stratification onset threshold ( = 261.9 ; = 0.158 ; = 3.75 ) was achieved via k-means clustering, enabling robust classification of the annual cycle into five thermal phases mapped onto three stratification regimes (Weak/Moderate/Strong). at the deepest lacustrine station (P5) strongly correlated with reservoir-wide (r = 0.952), supporting its conditional use as a proxy for whole-system mixing potential in data-scarce contexts, a relationship that weakened substantially in upstream zones, underscoring the necessity of zone-specific validation.

Chemical stratification analysis revealed that biogeochemical and optical water quality drivers operate largely independently of thermal structure: while saturated near its theoretical maximum (≈ 2) under persistent anoxia in the lacustrine zone, stratification indices for pH, chlorophyll-a, and turbidity remained uncorrelated with SSI system-wide. Strong -a/ associations (r = 0.833 at P5) indicated algal contributions to light attenuation, whereas cross-zonal coherence between riverine and transitional stations (r = 0.845) reflected shared hydrological forcing. These patterns suggest that thermal metrics alone may be insufficient for predicting certain critical water quality outcomes in arid reservoirs.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

All authors contributed equally to the development of the methodology, analysis, preparation, and editing of the manuscript. During the preparation of this work the authors used Chat GPT in order to improve language and readability. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Funding

The authors thank Sultan Qaboos University (SQU) for financial support under His Majesty (HM) grant number SR/ENG/CAED/22/01.

Data availability

The data used in this study were provided by the Ministry of Agriculture, Fisheries, and Water Resources and are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.