Buffalo Pound Lake is a shallow, eutrophic reservoir located near the headwaters of the Qu'Appelle River drainage system in southern Saskatchewan, Canada (50.65127° N, 105.508225° W; Fig. 1). Buffalo Pound Lake is the drinking water source for ∼25 % of the province's residents, servicing the cities of Regina (population of 249 000) and Moose Jaw (population of 34 000). The lake is the site of a provincial park and a popular recreation spot for activities that include fishing, swimming, and camping; however, agriculture is the dominant land use in the region. The climate in this region is semi-arid with long, cold winters and short, warm summers. The mean annual precipitation is 320 mm (McGowan et al., 2005), although it varies between dry and wet climate phases, and the average annual temperature is ∼3 °C (Haig et al., 2021). Buffalo Pound Lake was created by damming a natural lake's outflow to the Moose Jaw River to raise water levels (beginning in 1939). The lake is long and narrow (∼29 km × 1 km) with an average depth of 3 m (max depth of 5.6 m) and storage capacity of 9×107 m³ (BPWTP, 2020), with a short residence time (8 months to 2.5 years; BPWTP, 2016; Vogt et al., 2018). Its shallow depth and exposure to regular wind-induced mixing result in a polymictic system that only periodically establishes thermal stratification (Finlay et al., 2019).

Buffalo Pound Lake has two major water sources: the local catchment and upstream diversions from Lake Diefenbaker (Fig. 1), a reservoir with a volume of 9×109 m³, approximately ∼100 times greater than that of Buffalo Pound Lake. Importantly, the local catchment water quality and inflows from Lake Diefenbaker differ with respect to multiple attributes, including the DOC and sulfate (SO42-) concentrations. For example, measured local catchment DOC and SO42- concentrations are on the order of 20 mg L⁻¹ and > 1000 mg L⁻¹, respectively, whereas the respective concentrations in Lake Diefenbaker are typically < 5 mg L⁻¹ (Saskatchewan Water Security Agency, unpublished data) and < 100 mg L⁻¹ (John-Mark Davies, unpublished data).

Lake Diefenbaker lies outside of the natural catchment boundary for Buffalo Pound (BPWTP, 2020) and receives a continuous flow sourced from the South Saskatchewan River, originating in the Rocky Mountains. Water releases from Lake Diefenbaker maintain Buffalo Pound lake levels within an established operating range (30 cm) which has some seasonal variability. During the summer months, due to the importance of the water supply and recreational interests, levels are typically managed within the upper 10 cm of the operating range, but they are reduced by about 15–20 cm in winter. The second major water source to Buffalo Pound is the local catchment (3310 km², of which only 38% is effective), which is dominated by agriculture with nutrient-rich runoff that contributes to eutrophic conditions (Finlay et al., 2019; Hammer, 1971). These inflows are generally intermittent, occurring primarily during spring snowmelt and (infrequent) extreme summer precipitation. During wet years, which lead to higher catchment flows to Buffalo Pound, managed releases from Lake Diefenbaker are much lower than in dry years. In contrast, dry years, with little or no catchment flow, require larger releases from this upstream reservoir to maintain lake levels within the operating range. Groundwater flows are not thought to be important in terms of volume or DOC contributions to the lake.

Conventional hydrological modelling is not an easy task in the Buffalo Pound Lake catchment due to the poorly defined drainage system distributed over a relatively flat terrain (e.g., Mekonnen et al., 2015; Pomeroy et al., 2005). This forms complicated hydrological conditions, under which the hydrologic responses heavily depend on landscape feature detail (Fang et al., 2007; Van der Kamp et al., 2003; Spence, 2010) and are subject to nonlinear hysteresis (Shook and Pomeroy, 2011). With few exceptions (Clark and Shook, 2022), current hydrological models do not represent dominant hydrological processes in prairie landscapes and fail to reproduce the observed runoff. Here, we attempted to maximize the use of observed gauge information (Table S1) by implementing a data-driven reconstruction scheme. This approach allows us to consider three water sources in our analysis: diversions from Lake Diefenbaker (QLD); upstream inflows to Buffalo Pound Lake, representing a mix of local catchment flows and QLD (QBP); and flows originating from the local catchment (QLC).

Considering the mass conservation and ignoring other potential losses, the upstream inflow to Buffalo Pound Lake (QBP) can be described as follows: 1QBP=QLD+QRC+QIC+QU-ΔSΔt.

We used available flow information at Ridge Creek (QRC) to estimate missing flows at Iskwao Creek (QIC∗). Likewise, we used Ridge Creek to estimate flows for the ungauged portion of the catchment upstream of the Buffalo Pound Lake inflow ((QU-ΔSΔt)∗), an area that includes Eyebrow Lake (Fig. 1), which periodically attenuates flow. While records for QLD and QRC are complete, we used a set of linear and nonlinear transfer functions, including polynomial, exponential, and inverse exponential functions, along with power and tangential functions and their linear combinations (resulting in a total of 420 functional mappings) to link QIC∗ and (QU-ΔSΔt)∗ to QRC. We note that the nature of landscape hydrologic response in the lake catchment is significantly different between frozen (December–March) and ice-free (April–November) months; therefore, we considered separate transfer functions for warm and cold seasons. The 1972–1992 data were used for model development, and the models were compared and falsified based on the Bayesian information criterion, which penalizes the number of model parameters. For Iskwao Creek, the following transfer functions performed well during the training period: 2QIC∗=0.3112QRC0.4537if cold season,0.4957QRC+0.1185if warm season and QRC>0,0if warm season and QRC=0. Here, QIC∗ and QRC are in cubic metres per second. Similarly, the total upstream ungauged flux, in cubic metres per second, was estimated based on QRC as follows: 3QU-ΔSΔt∗=0.01292QRC3-0.303QRC2if cold season, +1.249QRC-0.0956-0.05958QRC2if warm season.+2.77QRC-0.1463 Here, the storage term ΔSΔt accounts for changes in landscape storage, including at Eyebrow Lake. The efficiency of estimated QIC∗ and QU-ΔSΔt∗ was investigated by calculating the coefficient of determination (R2) for observed flows at Iskwao Creek and upstream of Buffalo Pound Lake during separate training (1972–1992) and testing periods. The R2 measures the percentage of described variance and was 0.71 during the training period and 0.68 during the testing period of 2008–2011 for QIC∗. For the total inflow to the lake (QBP), R2 was similarly high: 0.87 during training and 0.78 during testing (1993–1995).

Finally, because QBP is dominated by QLD, we used a third metric as a means of assessing the role of flows from the local catchment (QLC). In this instance, we approximate QLC, including runoff from the catchment area downstream of the gauge at the Buffalo Pound Lake inflow, according to the following: 4QLC≅QDCifQRC+QIC+QU-ΔSΔt∗<0,QRC+QICifQRC+QIC+QU-ΔSΔt∗+QU-ΔSΔt∗≥0.+QDC Here, we scale combined flows from the Ridge and Iskwao creeks to estimate flows from the effective catchment area that lies downstream of the upstream inflow to Buffalo Pound Lake (QDC), including areas along either side of the lake (Fig. 1). While this approach can only serve as an estimate, as any attenuation of flows in the ungauged portion of the Qu'Appelle River can be a mix of source waters from Lake Diefenbaker and the local catchment, it nonetheless provides a proxy for the flow contribution from the local catchment.

There were two major goals of our statistical analysis. First, we aimed to understand frequency-dependent relationships between the DOC concentration and a suite of environmental predictors that include upstream flows (QLD, QBP, and QLC) and in-lake water chemistry (SO42-, TP, SRP, Chl a, NO3-, and NH4+) using wavelet coherence analysis (Grinsted et al., 2004; Sheppard et al., 2016, 2017; Walter et al., 2021). Wavelet-based methods can be used to measure synchrony and coherence between environmental and biological variables and to investigate relationships that are not readily detected by conventional correlation methods (Reuman et al., 2021). Specifically, wavelet coherence analysis can reveal the degree of association between two variables, while the phase relationships provide information on the patterns of behaviour between these two variables. Variables that are in phase or antiphase are analogous to a positive or negative correlation, respectively, while the phase relationships also provide information on where there is a time lag between the variables (e.g., Walter et al., 2021). Importantly, this approach allows for investigation across different timescales, as it is possible that these relationships manifest in different ways at different timescales. We also note that these phase relationships do not identify dependence among variables. Variables identified as significantly coherent with DOC (α= 0.05) by wavelet coherence analysis were subsequently used to achieve our second objective, wherein generalized additive modelling (GAM) was used to investigate predictor–response relationships between the DOC concentration, flows, and in-lake water chemistry. Predictor–response relationships were assessed using GAM because we anticipated complex nonlinear trends between DOC and predictor variables that would be masked by traditional (e.g., Mann–Kendall) methods or would require some degree of a priori trend selection or functional form (e.g., linear and generalized additive models). This approach allows for investigation of the relationships between each predictor and the response variable, but we note that we use this technique as a tool to analyze complex relationships, not as a forecasting tool or to infer that these predictors are drivers of DOC. All statistical analyses were completed using monthly averages (n= 360; see Sect. 2.2.1).

All statistical analyses were carried out in R (version 4.1.3; R Core Team, 2022), including “Wavelet approaches to synchrony” (wsyn) for wavelet analysis (version 1.0.4; Reuman et al., 2021), base (R Core Team, 2022), and spatstat.core (version 2.4.2; Baddeley et al., 2015) for wavelet plotting. Generalized additive models were fit using the R package mgcv (version 1.8.40; Wood, 2017), and graphics were plotted with ggplot2 (version 3.3.6, Wickham, 2016), gratia (version 0.7.3; Simpson, 2021), and patchwork (version 1.1.1; Pedersen, 2020). The graphics in Fig. 1 were plotted using the R package mapcan (McCormack and Erlich, 2019). The gross and effective drainage area shapefiles used in Fig. 1 were obtained from the Prairie Farm Rehabilitation Administration, and lake and river geospatial data were retrieved through the Government of Canada's Open Government License (Statistics Canada, 2006, 2011). Further details on wavelet analyses and GAMs are provided below.

Flow sources varied over the 30-year period with no distinct long-term or multiyear patterns (Fig. 2a–c). Buffalo Pound Lake inflows (QBP) and QLD each showed seasonal patterns, peaking around late spring to early summer and decreasing during winter months when water demands were lower and the lake was covered with ice (Fig. S2). Control structures from Lake Diefenbaker to Buffalo Pound Lake permit flows to continue during winter months when the lakes are ice-covered. Local catchment flows (QLC) typically peaked in late spring to early summer and did not contribute water to Buffalo Pound Lake in winter months or dry years. Peaks in QLD, QBP, and QLC are typically associated with spring freshet, but large rainfall events can also contribute considerable flow within the region. Streamflow averaged 3.2±2.6 (range: 0–12.4) m³ s⁻¹ at QLD and 3.6±2.8 (range: 0.1–18.9) m³ s⁻¹ at QBP (Table S3), highlighting much lower catchment flows which averaged 0.7±2.3 (range: 0–25.4) m³ s⁻¹. In years where QLC was low or absent, particularly in the late 1990s and early 2000s, QBP generally followed patterns of water release from Lake Diefenbaker, and fluctuations in QBP were comparable to those of QLD. Several years showed evidence of wetter-than-normal conditions (e.g., 1997 and 2014) with short-term episodes of considerable discharge from the local catchment. These episodes correspond to peaks in QBP and were generally associated with low QLD flows (Fig. 2).

The DOC concentrations fluctuated considerably over the 30-year observation period, ranging from 3.3 to 12.4 mg L⁻¹ (Fig. 3, Table S3). The mean (± standard deviation) DOC concentration was 6.8±1.8 mg L⁻¹. Several notable fluctuations occurred over multiyear periods. For example, after sustained high concentrations from 1997 to 2000 (mean 9.5 mg L⁻¹), DOC fell to 4.4 mg L⁻¹ by 2004 and remained near or below average until the mid-2010s. Between 1990 and 2000, there was a gradual increase in the DOC concentration where, alongside intra-annual increases and decreases, DOC concentrations increased from 3.9 mg L⁻¹ in 1990 to 9.3 mg L⁻¹ in 2000. Other notable changes occurred at shorter timescales: in 1991, DOC concentrations were ≤5.0 mg L⁻¹ from January to July, spiked to 10.5 mg L⁻¹ in August, and fell to 5.5 mg L⁻¹ by December. The mid-2010s also saw a sharp increase in DOC concentration followed by below-average levels within a 4-year span.

Water chemistry was also highly variable over the 30-year period (Fig. 2d–i, Table S3), with evidence of dramatic changes in the chemistry of Buffalo Pound Lake over time. For example, SO42- concentrations ranged from 56.8 to 340 mg L⁻¹ and were relatively stable from 1990 to 2009 before rapidly rising up to 250 mg L⁻¹ between 2011 and 2017 (Fig. 2d). By the end of 2019, SO42- returned to levels observed during 1990–2010. Nitrate also showed two distinct patterns. After averaging 0.20±0.28 mg L⁻¹ from 1990 to 1999 and peaking at 1.5 mg L⁻¹ in 1997, NO3- concentrations were nearly an order of magnitude lower from 2000 to 2019, averaging 0.05±0.08 mg L⁻¹ (Fig. 2h). Total phosphorus, SRP, Chl a, and NH4+ concentrations rose rapidly in 1991 (Fig. 2e–g, i), concomitant with the rise in DOC in that year, but this was transient. In the few years after 1991, chemical concentrations were average or below average, followed by increases in concentrations into the early 2000s. The mid-2010s also showed elevated levels of SO42-, TP, Chl a, and NH4+, concurrent with elevated DOC concentrations and QLC during this time.

Coherences between DOC and nine environmental predictors at short (≤ 18-month) and long (> 18-month) timescales were highly variable (Table 1, Fig. S3). At short timescales, coherence magnitudes ranged from 0.08 to 0.50 (median of 0.15), whereas at long timescales, they ranged from 0.05 to 0.74 (median of 0.54); this is consistent with potential bias in wavelet coherence, where greater coherence magnitudes tend to be returned at long timescales (Walter et al., 2021). Four DOC–predictor pairs were significant at short timescales (α=0.05 significance level): SO42-, TP, Chl a, and NH4+. At long timescales, SO42-, TP, and QLD were significantly coherent with DOC (α= 0.05 significance level). Significant coherence relationships at these timescales indicate that the magnitude of changes that occur in DOC concentrations and the environmental predictors are closely related. Phase relationships, expounded upon next, describe the time lags in these changes. Pairs of variables that are in phase will increase (and decrease) together, and a time lag can tell us if changes in one variable precede the other. Antiphase behaviour indicates that one variable is increasing while another decreases.

Phase relationships spanned the range of possible values (0 to ±π) (Fig. S4). For DOC–predictor pairs with significant coherence, most relationships were approximately (-π/4<ϕ<π/4) at both short and long timescales (Table 1). At short timescales, DOC was strongly in phase with TP, Chl a, and NH4+ (0<ϕ<π/4), whereas it tended to lag behind fluctuations in SO42- (-3π/4<ϕ<-π/2), indicating that DOC increases (or decreases) when these predictors increase (or decrease). Interestingly, the phase relationship between DOC and SO42- shifted to strongly in phase (-π/4<ϕ<0) at long timescales, while TP remained strongly in phase with DOC (0<ϕ<π/4). Unlike SO42- and TP, fluctuations in DOC were antiphase with and tended to lag behind fluctuations in QLD (-3π/4<ϕ<-π/2), indicating that DOC decreases follow increases in QLD. Significant coherence and phase relationships described here provide insight into the timescale and time-lag relationships between DOC and the environmental predictors. Next, analogies to traditional positive and negative correlation are discussed, using GAM.

Five environmental predictors (SO42-, TP, Chl a, NH4+, and QLD) found to be significantly coherent with the DOC concentration, as identified by wavelet coherence analyses at either short (≤18-month) or long (>18-month) timescales, were included in a simple GAM. A thin plate regression spline for each covariate explained 56 % of the deviance in the DOC concentration over the 30-year period, with an adjusted R2 of 0.50 (effective degrees of freedom = 16.9). Sulfate, TP, NH4+, and QLD all explained the significant variation in the DOC concentration (p<0.001 for NH4+ and QLD; p<0.0001 for SO42- and TP). The Chl a smooth term was the only nonsignificant predictor (p= 0.15; Fig. 4a). Total phosphorus and QLD showed linear positive and negative relationships with the DOC concentration, respectively (Fig. 4b, e). At NH4+ concentrations <∼0.1 mg L⁻¹ (61 % of the observed NH4+ data), NH4+ had an approximately linear positive effect on the DOC concentration but did not have much predictive power at concentrations >0.1 mg L⁻¹, as shown by the increasingly large confidence interval (CI; shaded region) at higher NH4+ levels (Fig. 4c). The relationship between DOC and SO42- (Fig. 4d) was generally positive but was more variable than the other environmental predictors. For example, at an SO42- concentration of ∼150–200 mg L⁻¹, increases in SO42- were concurrent with decreases in DOC; however, at SO42- concentrations <150 mg L⁻¹ (71 % of the observed SO42- data), DOC tended to increase with increases in SO42-. At higher concentrations of SO42-, TP, and NH4+ and high QLD, the GAM lost predictive power (large CI for the right tails of each smooth), but it performed well in the lower to moderate ranges (Fig. 4). Visually, the GAM provided a good fit between observed and fitted DOC at a concentration <∼ 7 mg L⁻¹, but considerable variation was apparent in the model fit at concentrations above a threshold of 8 mg L⁻¹ (Fig. S5). Fitting the GAM to the DOC time series (Fig. 5) revealed that using these predictors to estimate the DOC concentration was useful for identifying multiyear trends (e.g., in the 1990s and mid-2010s), but the model suffered from over-fitting intra-annual fluctuations, such that DOC minima and maxima are not always captured (Fig. 5).

Semi-arid regions are subject to high inter-annual variation in moisture conditions. Buffalo Pound Lake is a shallow reservoir in a DOC-rich landscape (Labaugh et al., 1987; Waiser, 2006) with high climatic variability (Pham et al., 2009; Pomeroy et al., 2007; Vogt et al., 2018) where prolonged periods of drought are often followed by multiple high-precipitation years. This leads to very high variability in DOC, with a range of ∼10 mg L⁻¹ over the 30-year observation period (Fig. 3). Rapid changes in DOC were sometimes observed, likely related to the lake's short residence time (8 months to 2.5 years; BPWTP, 2016; Vogt et al., 2018) that makes the system highly responsive to changing inputs. In regard to our first hypothesis, the flow source appears to have an important influence on DOC. Wavelet analyses demonstrated that flow from the upstream reservoir (QLD) was coherent with DOC concentrations in Buffalo Pound Lake at long timescales, with changes in the DOC concentration lagging behind changes in flows. This suggests that “managed” flows from Lake Diefenbaker have a flushing (diluting) effect, which is consistent with the generally lower DOC concentrations of this lake compared with local catchment runoff (Saskatchewan Water Security Agency, unpublished data). Thus, we fail to reject the null hypothesis (that QLD is not related to DOC).

We expected that QLC would be strongly related to lake DOC for the following reasons: prairie catchment runoff tends to be high in solutes, including DOC, salinity, and nutrients (e.g., Casson et al., 2019; Fasching et al., 2019), and prairie wetlands become important flow sources during periods of high hydrologic connectivity (Ali and English, 2019; Nachshon et al., 2014) and can have extremely high DOC concentrations (sometimes in excess of 100 mg L⁻¹) (Arts et al., 2000; Waiser, 2006). Instead, coherence of QLC with DOC was low (0.10) at short timescales and moderate (0.51) at long timescales, with neither being significant (Table 1). Statistically, the smaller and intermittent contribution of the local catchment to flow makes signals challenging to ascertain without much longer time series. Specifically, QLC, although sometimes large, is infrequent (Fig. 2) due to the ephemeral nature of these sources and their tendency to have no or minimal flow in drier years. There is also some potential for the attenuation of upstream catchment flows and for unknown lags in these flows reaching the lake, which may impair statistical detection. In contrast, QLD is large, even in years with high QLC. In addition, QLC is also estimated from smaller areas of gauged flow and, therefore, has greater uncertainty than QLD. In addition, total flows into Buffalo Pound Lake are not coherent with DOC (at short or long timescales; Table 1). Because QBP combines flows from Lake Diefenbaker and parts of the local catchment, this indicator may obscure relationships by integrating different mechanisms by which the two flow sources impact chemistry (e.g., greater allochthonous inputs from the local catchment and more dilute waters from Lake Diefenbaker).

Ultimately, the lack of statistical significance between QLC and lake DOC does not negate the likely importance of catchment DOC inputs to Buffalo Pound Lake. Due to the fact that there are only two flow sources to the lake and our analysis revealed a flushing effect of QLD, we would not expect this flushing to occur without the input of water of a different chemical character from the catchment. The wettest multiyear period in the long-term record (2010–2015) coincided with a prolonged period of DOC increase in Buffalo Pound Lake (Fig. 3), suggesting that a pulse of inflows from the local catchment influenced DOC and that the local catchment is likely an important source of allochthonous DOC when the QLC magnitude is large. Existing streamflow data from the local catchment indicate much higher DOC concentrations than in Buffalo Pound Lake. The Ridge Creek and Iskwao Creek DOC concentrations are each on the order of 20 mg L⁻¹ (Saskatchewan Water Security Agency, unpublished data), while water diverted from Lake Diefenbaker is similar to the minimum observed concentrations in Buffalo Pound Lake (<5 mg L⁻¹; Saskatchewan Water Security Agency, unpublished data). Further, observed increases in lake DOC due to autochthony and other in-lake processes are modest (Baron, 2023): DOC concentrations measured at eight locations along the length of Buffalo Pound Lake from 2016 to 2019 showed a 1.2 mg L⁻¹ increase from near the lake's inflow to near its outflow. In contrast, DOC concentrations from all lake locations over this period ranged from 3.7 to 8.1 mg L⁻¹ (Baron, 2023), As such, increases in DOC from autochthonous sources are unlikely to be the driving factor in the observed across-year increases in DOC. It is important to emphasize that QLD is reduced when QLC is high. This is a management-driven relationship that stems from the need to maintain stable water levels and limit downstream water transfers during wet periods (due to flooding effects). However, reducing the inputs from Lake Diefenbaker (QLD) when the inputs from the local catchment are high reduces the likelihood that higher-DOC waters will be flushed.

Dissolved organic carbon was not the only chemical variable that underwent dramatic change within Buffalo Pound over the study period. Indeed, 6-fold variation in the concentration of SO42- was observed (Fig. 2d). At short timescales (≤18 months), DOC lagged behind SO42- on average, but this relationship was strongly in phase at long timescales (>18 months), suggesting that within-year or seasonal DOC and SO42- dynamics may differ; however, the patterns in these chemical variables become more synchronous at longer timescales. The rapid rise and fall in DOC and SO42- concentrations in the 2010s may be linked to both internal and external processes. The majority of SO42- fluxes to lakes come from soils and wetlands in the catchment, often in mineral form or bound in allochthonous organic matter containing oxygen-bound S (e.g., as ester-SO₄ or other organo-SO₄ species) (Couture et al., 2016), helping to explain coherence between SO42- and DOC. Much like prairie catchments have high DOC, they can also have high SO42-, again, due to soil and wetland pools. Pothole ponds have a wide range of SO42- concentrations (median levels of 163 (mean: 519) mg L⁻¹), with some ponds reaching concentrations of 5500 mg L⁻¹ (Miranda and Whitfield, 2024), typically much higher than in many other regions, owing to sulfur-rich geologic sources and the potential for evapoconcentration in surface waters. Prairie soils can be similarly high in SO42- (Fennell and Bentley, 1998). In contrast, annually measured SO42- samples at Lake Diefenbaker from 1990 to 2019 showed median concentrations of only 58 mg L⁻¹ and maxima of <90 mg L⁻¹ (John-Mark Davies, unpublished data), which are values near the low range of SO42- seen in Buffalo Pound Lake. The much higher median Ridge Creek SO42- concentrations (estimated at ∼1094 mg L⁻¹; unpublished data) combined with below-average QLD (and higher QLC) in the early 2010s under more “natural” flow conditions mean that catchment SO42- sources must have contributed to the higher Buffalo Pound Lake SO42- concentrations during the latter part of the observation record. Extreme flooding across the prairies in 2011 could have created an exceptional fill-and-spill (Shaw et al., 2012) scenario (see Nachshon et al., 2014; Shook and Pomeroy, 2012), during which water containing elevated levels of SO42- and DOC was transported via runoff from spilling pothole ponds to Buffalo Pound Lake. Similar mechanisms were likely important in 2014 and 2015, 2 years with periods of high precipitation in the region (WSA, 2018) and above-average catchment inflows, when SO42- and DOC concentrations further increased. There is also the potential that groundwater may be infusing minerals, such as SO42-, into the lake (BPWTP, 2022), and analogous patterns have been observed during multiyear wet periods that have led to the activation of subsurface flow pathways that transport solutes from the soil to surface water (Draude, 2025). Prairie soils and lake sediments rich in gypsum (CaSO₄) and pyrite (FeS₂) deposited during glaciation 10 000 years ago may also play a role. The reoxidation of lake sediments rich in H₂S, CaSO₄, FeS₂, and other sulfides can also increase surface water SO42- concentrations (Holmer and Storkholm, 2001); moveover, in freshwater sediments, high rates of oxidation of reduced S compounds can shift the sediments from net sinks to sources of SO42- (Holmer and Storkholm, 2001), particularly if groundwater SO42- sources to the lake are not high.

Lake DOC concentrations in cold regions tend to vary both seasonally and inter-annually. Our results highlight in-phase relationships of DOC and TP, Chl a, and NH4+ (Table 1). Seasonal cycles alter the rates of biological DOC production and hydrological export (Clark et al., 2010), and DOC influxes often peak during spring snowmelt, whereas biological DOC production peaks in summer (Buffam et al., 2007; Clark et al., 2010; Laudon et al., 2004). In cold regions where lakes are ice-covered for a portion of the year, DOC exclusion from ice can be important for driving winter metabolism and DOC concentrations (Guo et al., 2012; Kurek et al., 2022), whereas photo- and biodegradation processes (Hansen et al., 2016) and autochthonous DOC production (Baron, 2023) may dominate during the open-water season. In many lakes, the seasonal variation in the DOC concentration can be 1–2 orders of magnitude greater than the long-term rate of change over decadal timescales (Clark et al., 2010). Within Buffalo Pound Lake, we observed both high inter- and intra-annual variability in the DOC concentration, with fluctuation as high as 7 mg L⁻¹ within a single year.

Total phosphorus, Chl a, and NH4+ concentrations were strongly in phase with DOC at short timescales, indicating that increases in these three variables are accompanied by increases in DOC. Total phosphorus remained strongly in phase at long timescales (Table 1). Possible mechanisms for these relationships include (1) internal production and transformations and (2) synchronous inputs of nutrients and allochthonous DOC from the local catchment, as expected from SO42- behaviour. We anticipate that the latter plays a more prominent role, given the modest DOC increases as water moves through the lake (above). Using SO42-, TP, NH4+, Chl a, and QLD as predictors in a GAM describing DOC explained 56 % of the deviance in the DOC concentrations observed over this 30-year data set, where all predictors were significant except for Chl a. This result presents evidence that the DOC concentration in Buffalo Pound Lake is strongly linked to in-lake nutrient and solute chemistry in addition to its major flow source, Lake Diefenbaker. We attempted to incorporate time-lagged relationships of DOC with SO42- and QLD into the GAM approach; however, a feature of wavelet coherence analysis is that phase relationships can only be determined on average over the timescale of interest (J. Walter personal communication, 2022), and thus specific time lags (e.g., 6 months or 24 months) cannot be accurately measured with this approach. The true time-lag relationship between QLD and DOC also likely varies depending on flow magnitude. To assess concerns over potential collinearity between DOC and TP, we ran a GAM with SRP in place of TP but found no discernable differences in model results.

Strong in-phase coherence between DOC and TP at both short and long timescales and the strong linear effect of TP on DOC (Fig. 4e) suggest that there is synchrony in the processes influencing DOC and TP in Buffalo Pound Lake. As with SO42-, prairie ponds can also be important phosphorus (P) sources to downstream lakes during periods of elevated catchment flow, particularly because ponds high in SO42- tend to have greater sediment P release rates (Jensen et al., 2009). Spring TP concentrations measured at ∼150 ponds across the prairie region in 2019 revealed mean TP concentrations in excess of 500 µg P L⁻¹ (McFarlan, 2021); when water is mobilized from the catchment through wetland fill-and-spill behaviour (Shaw et al., 2012), QLC is elevated and increased nutrient export from the catchment to the lake is expected. Periods of water flow from the catchment also prompt lower QLD, as less water is diverted from Lake Diefenbaker, during the natural flow events. This would reduce the dilution effects of Lake Diefenbaker inputs for both DOC and TP. Sediment processes may also support this coherence. Larger pulses of internal P released from sediments during transient stratification events could be linked to mid-summer algal blooms (D'Silva, 2017; Painter et al., 2023), potentially supporting pulses of greater autochthonous DOC production. In addition, anoxia and organic P are important to internal P loading in lakes further downstream (Wauchope-Thompson et al., 2024) which could similarly support in-phase DOC and TP relationships at shorter timescales.

The Chl a concentration, a measure of algal abundance, was not significant in the GAM but showed in-phase coherence with DOC concentrations at short timescales, suggesting a seasonal effect associated with cyanobacterial/algal biomass in the lake. Inefficient conversion of light energy to organic molecules during photosynthesis means that a portion of the organic materials produced by algal cells are released as DOM/DOC to ambient water over the lifetime of the algae (Fogg, 1966; Myklestad, 1995). Extracellular DOC release, via the overflow model (Fogg, 1966, 1983; Williams, 1990) or the passive diffusion model (Bjørnsen, 1988; Fogg, 1966), has been shown to be an important source of autochthonous DOC, where up to 5 %–35 % of fixed organic carbon may be release immediately as DOM, often with high proportions of DOC (Carlson and Hansell, 2015). Other mechanisms for the release of autochthonous carbon, such as release by grazers and following sedimentation, are also likely important (Keller and Hood, 2011). Additionally, DOC concentrations can increase by 1–2 mg L⁻¹ between the Buffalo Pound Lake inflow and outflow in years dominated by flow from Lake Diefenbaker (Baron, 2023), consistent with autochthonous production and the potential for extracellular release.

Observed in-phase coherence between DOC and NH4+ at short timescales and the significance of NH4+ as a predictor of DOC in the GAM may be related to several mechanisms. The lake is sometimes N-limited (Swarbrick et al., 2019), suggesting that increased NH4+ could support increased productivity as well as autochthonous production. This temperate lake is also highly seasonal, with major summer blooms (Painter et al., 2022, 2023), followed by long, cold winter periods when the lake is ice-covered and NH4+ dynamics undergo rapid change (Cavaliere and Baulch, 2020). The early-winter phase is marked by increasing NH4+ concentrations associated with decreased autotrophy, oxygen depletion, and organic matter mineralization. We observed several years during which DOC concentrations decreased in the early-winter period. The late-winter phase shows a decline in NH4+ likely associated with biotic uptake. Seasonal variation in the DOC : DON ratios are also apparent (Fig. S6). Higher C : N ratios occur under ice cover and then decrease beginning around spring ice-off (∼100th day of year), reaching a minima in late summer, concurrent with seasonal maximum Chl a concentrations.

Flood risk and drought underpin water management concerns in the Prairie region; cyclic wet–dry phases mean that these systems frequently alternate between years of having too much or too little water associated with elevated flood and drought risk, respectively. Examples include extensive flooding in 2011 and 2014, a major drought from 1999 to 2004, and extreme drought conditions in recent years (2021–2023). Natural and managed flow scenarios associated with wet–dry cycles each present challenges for Buffalo Pound Lake water quality and for the Buffalo Pound Water Treatment Plant (BPWTP). Our analysis highlights the importance of flows from Lake Diefenbaker as a means of “flushing” the lake and contributing to lower DOC concentrations. While our analysis did not yield significant relationships between QLC and DOC at long timescales, this may be partly attributed to the intermittent nature of this flow source. Given that there are only two flow sources in this system, we argue that the flushing mechanism identified for QLD would not likely exist if years with high local catchment flows (natural flow scenario) were not contributing to high DOC and poor water quality. These conditions can take years to dissipate. For example, during a prolonged drought from 1999 to 2005, specific ultraviolet absorbance at 254 nm normalized to DOC concentration (SUVA254) values were very low (∼0.9–1.6 L mg C⁻¹ m⁻¹ over this period), but SUVA₂₅₄ exceeded 2.5 L mg C⁻¹ m⁻¹ after a very wet 2011 that resulted in extensive flooding and then remained above 1.5 L mg C⁻¹ m⁻¹ into 2019 (Baron, 2023; Fig. S7).

There are important trade-offs here that include balancing the provision of water from Buffalo Pound Lake to meet the needs of ∼300000 people. While QLD has a flushing effect, diverting additional water during wetter periods when more source water originates from the local catchment, to manage DOC concentrations (and support drinking water treatment), can be expected to exacerbate downstream flooding and is currently avoided. It should be noted that dry conditions that prompt elevated flows from Lake Diefenbaker to Buffalo Pound Lake also represent a trade-off, as this diversion of water means that less flow is available for hydroelectric power generation via water exiting Lake Diefenbaker to the South Saskatchewan River at the Gardiner Dam. Diversions to maintain Buffalo Pound Lake levels during dry phases have also been associated with worsening algal blooms in Buffalo Pound Lake, in terms of magnitude (Painter et al., 2022), and coincide with autochthonous DOC production in the lake (e.g., Baron, 2023). Periods of elevated DOC under either flow scenario can create added challenges and costs for the BPWTP to maintain safe drinking water standards. Disinfection byproduct formation is a major concern for the BPWTP (Williams et al., 2019), particularly when the DOC concentration is high. Under these circumstances, the BPWTP has used a pre-chlorination step to reduce algal growth and prevent rising floc (organic materials that coalesce and resist coagulation) that can impact treatment capacity by accumulating on filtration media (Painter et al., 2022). Pre-chlorination has typically been used in summer months when algae are abundant in Buffalo Pound Lake (Painter et al., 2022); however, this can make it difficult for the BPWTP to meet regulatory limits for trihalomethanes in years in which DOC is also elevated (BPWTP, 2016). Untreated DOC can also contribute to poor taste and odour problems (Matilainen et al., 2011).

Environmental flows represent a framework that could be used to better manage the competing needs for drinking water and the hydrological functions of the reservoir, as well as its critical fish and wildlife habitat. The variable magnitude, timing, and water quality of flows from two water sources and the impacts on DOC underscore some of the complexity that would accompany the development of environmental flow rules for Buffalo Pound Lake. Maintaining stable water levels within the lake to limit valley slumping and erosion is important (SERM, 2001), and competing regional priorities for freshwater, including hydropower production, flood control, and irrigation, are also key factors in water management (WSA, 2013; Wheater and Gober, 2013). Water quality concerns are broad and include increasing nutrient inputs to agricultural lands; agricultural drainage of wetlands, which can have major impacts on catchment streamflow (Spence et al., 2022); effluent disposal practices on agricultural land; leaking septic fields; and pollution from tourism and recreational activities. Ultimately, concurrent efforts to consider environmental flow rules and advance source-water protection could help ensure that the maximum benefit is attained from current investments in upgraded water treatment.