Effects of drought on the physicochemical, nutrient, and carbon metrics of flows in the Savannah River, Georgia, USA

Hydrological drought has wide‐ranging impacts on water quality, nutrient and carbon metrics, and given the uncertainty of climate change and the predicted increased frequency and intensity of drought in the future, investigations into changes induced by drought become increasingly important. This study compared physicochemical parameters (temperature, conductivity, pH and DO), nutrients (TN, NOX [NO2 + NO3], NH3 and TP) and carbon (TOC and DOC) between hydrological drought conditions (2006–2008) and hydrological normal conditions (2016–2019) at five sites along the lower Savannah River (Georgia, USA). Although we had predicted that water temperatures would increase from drought, we instead found temperature was significantly lower during drought conditions. Levels of pH and DO were significantly higher during drought. Further, TN, TOC and DOC concentrations were significantly lower during drought, but NOX concentrations were significantly higher during drought. Conductivity varied at the lower river sites, being significantly higher during drought at sites located below the city of Augusta, GA. These complex changes could be attributed to volume reductions coupled with an increase in the percentage of total flow originating from groundwater as well as limnetic reservoir inputs, persistent point source pollution, reduced natural catchment inputs and/or reduced floodplain interactions. The changes that occurred during drought may be disruptive to aquatic life, not only from reduced water quantity but also due to a scarcity of some biologically essential materials and lower food resources, combined with artificially high levels of some other potentially stressful materials.

have been observed in temperature, dissolved oxygen and salinity (Whitehead et al., 2009).Longer residence times and reduced water volumes typically cause increased water temperatures, lowered dissolved oxygen concentrations, increased salinity and changes in other water quality measures (Baurès et al., 2013;Sprague, 2005).Dissolved oxygen is inversely related to temperature, and thus, responses of dissolved oxygen have ranged widely following changes in temperature (Ylla et al., 2010;Zieli nski et al., 2009).Water temperature increases from 1 to 2 C during drought have been observed (Hrdinka et al., 2012;Zieli nski et al., 2009), and extreme temperature increases (7 C) have also been noted (Ha et al., 1999).Further, major ions have been shown to increase during drought and these findings were attributed to reduced dilution of more saline groundwater inputs and increased evapotranspiration (Hrdinka et al., 2012).Although some physicochemical responses have been frequently documented during drought, efforts to assess changes in nutrient concentrations have been less extensive.
Predictions of rising global air temperatures and increased drought forecast an increase in nutrient loads (Whitehead et al., 2009).However, the response of nutrients to drought has been mixed and varies among river systems.Reduced water volume and reduced dilution of point source pollution have been shown to increase nutrient concentrations (Van Vliet & Zwolsman, 2008), especially where wastewater or industrial effluent is present (Andersen et al., 2004;Battaglin & Chapin, 2022).Increased nutrients have been observed with reduced groundwater dilution or reduced river connection to the floodplains (Golladay & Battle, 2002).Although, mixed responses of different nutrients to drought within the same system have also been observed (Sprague, 2005;Wilbers et al., 2009;Zieli nski et al., 2009).Low flows and longer residence times may facilitate an increased internal recycling of nutrients and primary production, which could account for decreased water column nutrients during drought (Hosen et al., 2019).Further, reduced catchment inputs and low turbidity may result in greater primary production, especially where algae are light limited (Andersen et al., 2004;Baurès et al., 2013).
Carbon dynamics in larger river systems during drought also have varied considerably.Much of our understanding of carbon dynamics is associated with flooding and the resultant mobilization of carbon stores (Whitworth et al., 2012), and few studies have emphasized carbon dynamics during drought.After floods recede, carbon and nutrients return to the main river channel and thus have been found in higher concentrations during low flows (Baurès et al., 2013).In contrast, some studies of drought noted decreased carbon concentrations because of reduced catchment inputs and reduced transportation of organic matter downstream (Ylla et al., 2010;Zieli nski et al., 2009).A system's response to drought ultimately depends on the local environment and becomes increasingly complex where industrial, municipal, and agricultural water usage, as well as hydroelectric power generation, is important.
This study compares physicochemical attributes (temperature, conductivity, pH and dissolved oxygen), nitrogen (TN, NO X , NH 3 ), total phosphorus (TP) and carbon (TOC and DOC) levels between hydrological drought conditions (defined here as 2006-2008) and hydrological normal conditions (defined here as 2016-2019) at five sites on the Savannah River, Georgia, USA.We hypothesized that (1) physicochemical metrics during drought, especially water temperature and dissolved oxygen concentrations, would be impacted by flow reductions because drier periods are typically associated with hotter air temperatures.We predicted that increased air temperatures, increased evapotranspiration rates and reduced water volumes would result in an increase in water temperature and a decrease in dissolved oxygen.Further, persistent point source pollution and reduced water volume would increase ionic concentrations; and (2) dissolved nutrients and carbon would increase with reduced flows because levels of nutrients and carbon will not change, but a reduced volume would result in higher nutrient concentrations.We predicted nitrogen and phosphorus concentrations would increase during drought from point source pollution.While concentrations of dissolved chemicals may increase, we predicted that material fluxes might decrease due to decreased flow volumes, or at least remain the same, if inputs did not change.
The Savannah River is a large river system supporting many industrial, municipal and recreational water uses and has experienced variability in drought conditions throughout the course of this study.
Changes in nutrient load and physical parameters on large river systems like the Savannah River such as, increased water temperatures (Bonacina et al., 2023), nitrogen levels (Beketov, 2004) and water acidification (Ganong et al., 2021) have been shown to negatively impact aquatic life.Likewise, these changes can largely affect the people and industries that rely on these systems to maintain a certain standard of water quality.Given this, an understanding of the impact that drought conditions have on these parameters is important not just from an ecological standpoint, but a social and economic one as well.

| Study sites
The Savannah River is a major river in the southeastern United States and forms the border between South Carolina and Georgia.The drainage basin extends from the southeastern portion of the Appalachian Mountains, where the Tugaloo and Chattooga Rivers converge to form the headwaters of the Savannah River.The Savannah River is approximately 484 km long, drains an area of 27,390 km 2 and supports two major cities in Georgia: Augusta located mid-watershed near the Fall line (transition between the Piedmont and Coastal Plain) and Savannah located at the mouth of the river and the Atlantic Ocean, where the Savannah River drains.The natural hydrology of the Savannah River has been greatly altered by dams, with the largest of the dams, J. Strom Thurmond Dam, located near the Fall line and upstream from our study area, and ultimately affecting all flows in the mid and lower river to some extent.The Savannah River watershed is altered by agriculture, silviculture, municipalities, industries and energy generation.Silviculture is an especially significant industry in the Savannah River basin, with approximately 890,000 hectares of commercial forested land.As of 2018, there were 177 industries and municipalities authorized to discharge wastewater in the basin (Georgia River Networks, 2018).
Five study sites were selected for this study, located in the middle and lower portions of the Savannah River (Figure 1, Table A1).Site 1 (325 river kilometers; hereafter RKM) was the most upstream  affected 49.9% of Georgia land (Akyuz, 2017).Therefore, the 2006-2008 study period encompassed significant drought conditions (i.e., D1-4 were p < 0.01), while the 2016-2019 study period had near normal rainfall.Hereafter, "drought" conditions refer to the 2006-2008 period and "normal" conditions refer to the 2016-2019 period.

| Savannah River hydrology
Discharge data were gathered from the U.S. Geological Survey (USGS) gage at Augusta (#02197000) for Site 1 (USGS, 2006).Site 1 was located approximately 25 RKM upstream of the USGS gage at Augusta.Discharge from the Horse Creek gage (#02196690), a major tributary approximately 8 RKM downstream from Site 1 but upstream from the Augusta gage, was subtracted to estimate discharge at Site 1. Discharge from the Augusta gage (#02197000) was used to estimate discharge at Sites 2 and 3; no significant tributaries flow into the Savannah between that gage and these sampling sites.Discharge from the Waynesboro gage (#021973269) was used to estimate discharge at Site 4 and discharge from the Clyo gage (#021973269) was used to estimate discharge at Site 5. We calculated average daily discharge yield (m 3 km À2 s À1 ) by obtaining drainage area (km 2 ) from the associated USGS gage and dividing it from daily discharge (m 3 s À1 ).Average discharge during the drought period was found to be significantly lower than the normal period across sites (F 1,220 = 6.22,p = 0.01; difference at the uppermost and lowest gage between drought and normal periods.The current Savannah River Drought Plan establishes a minimum daily average release from Thurmond dam of 107.6 m 3 s À1 (USACE, 2012).This minimum was established so that municipalities and industries downstream of the dam would be in compliance with state permitting requirements for the dilution of wastewater and to ensure adequate river flows for industrial water supply.

| Field methods
We collected monthly field measurements and discrete water samples from February 2006 to January 2008 and from May 2016 to December 2019.Field measurements of temperature ( C), dissolved oxygen (mg L À1 and % saturation), pH and specific conductance (μS cm À1 ) were collected using a YSI multiparameter sonde (Yellow Springs Instruments, Yellow Springs, OH).Following USGS protocol (USGS, 2006), samples were collected in the field using non-isokinetic pump sampling methods.A portable pump was used to collect a depth-integrated sample by continuously lowering and raising the pump vertically in the water column at a constant rate.One vertical sample was obtained in well-mixed areas or the thalweg of the river for each sampling event.Samples were collected in acid-rinsed polypropylene bottles, stored on ice, transported to the laboratory and processed within 48 h.
Extensive quality control measures were used to maintain data consistency (USGS, 2006).Field blank and replicate samples were collected for each sample batch, which was every five samples or at least once per month.Results from blank samples were used to indicate contamination across samples and replicate samples were used to indicate variability from the collection, processing, and laboratory analyses.
In the laboratory, samples were transferred to a churn splitter (Phinizy Center for Water Sciences; Augusta, Georgia) and homogenized to ensure the particulate organic material was evenly distributed.Aliquots were separated for each analyte and preserved with sulfuric acid to a pH ≤2 for subsequent analyses.For dissolved organic carbon (DOC), samples were filtered through a 0.45 μm glass fiber filter prior to analysis.Samples were stored between 4 C and 6 C until the time of analysis.

| Analytical methods
Samples were analyzed within 28 days for total nitrogen (TN), ammonia (NH 3 ), nitrate + nitrite (NO X ), total phosphorus (TP), total organic carbon (TOC) and dissolved organic carbon (DOC).Environmental Protection Agency (EPA) method 350.1 for the determination of NH 3 (USEPA, 1993a); method 353.2 for the determination of NO X (USEPA, 1993b), method 365.1 for the determination of TP (USEPA, 1993c), and method 415.1 for the determination of TOC and DOC (USEPA, 1974).Prior to 2018, EPA method 351.2 was used to determine TN (USEPA, 1993d), and from 2018 to 2019, a comparable method, American Standard Methods D8083 was used because of safety concerns working with mercury (ASTM, 2016).In each analysis batch quality control consisted of a calibration control blank and a calibration control verification to check for instrument drift, a laboratoryfortified matrix to determine instrument repeatability and accuracy, a laboratory fortified duplicate to determine the precision of the instrument and certified reference material to determine the analysis accuracy (Eaton et al., 1998).During 2016-2019, samples were externally verified by Pace Analytical at least once per year.

| Data analyses
To assess the effects of discharge on nutrient concentrations, mass flux (kg day À1 ) was used to account for reduced flow rates during drought, integrating concentration and daily discharge averages.This was achieved by gathering average daily discharge for each sampling date from the nearest USGS gage (Table A1) and then calculating mass flux for each nutrient (Aulenbach et al., 2007).Mass flux (Φ) was calculated as the product of constituent concentration (C) and discharge (Q) integrated over time (t): The following equation was used to convert nutrient concentrations into kilograms per day (Goolsby et al., 1999) We calculated means, medians, minimums, maximums, standard deviations (SD), standard errors (SE) and coefficients of variation (CV) to summarize each metric.Data were then compared using a 2-way Analysis of Variance (ANOVA), with time period (i.e., drought, normal) and sampling sites (i.e., 1-5) as the factors (a priori α = 0.05).Goodness of Fit Tests were used to ensure normal distributions of data, and Levene Tests were used to ensure equal variances.If metrics were found to have significant interactions between the time periods and the sites, a series of 1-way ANOVAs were used to assess the effect of time period within each of the 5 sites.When a significant effect of site was indicated by the 2-way ANOVAs, we used Tukey-HSD tests to separate site means.R (Version 3.6; R Core Team, 2020) was used to run 1-way and 2-way ANOVAs, Goodness of Fit Tests, Levene Tests and Tukey tests.

| Drought impacts on physicochemical parameters
Mean water temperature ( C) was found to be significantly lower during the drought period than the normal period (F 1,220 = 4.27, p = 0.04; Figure 4a).Variation in water temperature, as reflected by coefficient of variation (CV), was highest at site 3, at 74.1% during drought and 103.9% during normal conditions and decreased steadily downstream with Site 5 exhibiting the lowest CV of 17.7% during drought and 50.7% during normal conditions (Table A3).Dissolved oxygen levels were higher during drought for both concentration (mg L À1 ) (F 1,220 = 11.74,p < 0.01; Figure 4b) and saturation (%) (F 1,220 = 6.72, p = 0.01, Figure 4c).Levels of pH were significantly higher at all sites during drought (F 1,220 = 11.99,p < 0.01; Figure 5a).

| Drought impacts on nutrient and carbon parameters
Nitrogen concentrations, including total nitrogen (TN), ammonia (NH 3 ), and nitrate + nitrite (NO X ), were found to vary between drought and normal hydrological conditions (Figure 6, Table A4).TN concentration (mg-N L À1 ) was significantly lower during drought (F 1,220 = 5.23, p = 0.02; Figure 6a), as was TN flux (kg-N day À1 ) (F 1,220 = 11.29,p < 0.01; Figure 6b).NO X concentration (mg-N L À1 ) was significantly higher during drought (F 1,219 = 4.04, p = 0.05; We had hypothesized that water temperatures would increase and dissolved oxygen levels would decrease during drought due to flow reductions.However, water temperatures instead decreased and dissolved oxygen levels increased during the drought.Maximum air temperatures in our study area were similar between the time periods, although minimum air temperatures were significantly lower during drought.For reference, we assessed water temperatures in the Ogeechee River, an adjacent, largely unregulated, free-flowing river in Georgia, and found no differences in water temperature between drought and normal hydrological periods (Table A3).Thus, the observed differences in the Savannah River were likely attributable to the regulation of the upstream large reservoir (Thurmond Dam).A higher percentage of total flows in the river likely originated from hypolimnetic discharges through the dam during drought (Sprague, 2005).To corroborate this, we assessed temperature variation (as reflected by coefficients of variation) and found the least variability at the most upstream site, closest to the dam, and a steady increase in the temperature variability longitudinally during drought (i.e., sites became more variable as the regulatory impacts of the dam waned) (Table A5).Controlled flows from the dam dominated total flows near the dam, which caused decreased water temperature levels and variability, a common observation for dams (Ruhi et al., 2018).We predicted that DO levels would be related to water temperature, and this was true, although, like for temperature, inversely to our prediction.We found that DO increased during drought conditions.We did not find any similar instances in the literature, but we found several studies reported no changes in DO levels (Caruso, 2002;Hrdinka et al., 2012).They attributed the lack of temperature increases to similar air temperature and dilution from cooler reservoir inputs, point sources, or groundwater downstream.
In addition, we predicted drought would cause increased pH and conductivity with reduced water volumes and inputs of point source pollutants.We found that pH and conductivity, at some locations, were higher during the drought, similar to findings by Sprague (2005) and and flushing from precipitation (Mosley, 2015;Mosley et al., 2012), and in some cases to point source pollution (Van Vliet & Zwolsman, 2008).We found that conductivity was only elevated downstream from the city of Augusta (Sites 3-5), which suggested that anthropogenic inputs of ions from Augusta might be involved.
This pattern was observed in both time periods but was exacerbated during drought.
We also hypothesized that reduced flows would increase the concentrations of dissolved nutrients and carbon, but that flux would decrease with reduced water volumes.This hypothesis was observed for nitrate + nitrite but was not observed for any of the other nutrients.Increased nutrients during drought commonly occurs in rivers where point sources (e.g., industrial, domestic, agricultural, or wastewater discharge) of nutrients predominate (Baurès et al., 2013;Sprague, 2005).Higher concentrations of nitrate and nitrite during drought probably indicates a decreased dilution of effluent from municipal and industrial sources.This was corroborated by flux calculations, as they were similar between drought and normal conditions, suggesting unchanging effluent discharge.However, ammonia and phosphorus did not follow this trend.Ammonia is the most bioavailable form of nitrogen, and phosphorus is a limiting nutrient in freshwater systems.Therefore, they may have been utilized by primary producers and assimilated into biomass before detection, which has been reported by others (Caramujo et al., 2008;Caruso, 2002;Nguyen et al., 2020).
Contrary to our original hypothesis, we found that total nitrogen and carbon levels were reduced during drought.In low gradient streams, nitrogen and carbon that originates from the terrestrial landscape can accumulate in floodplains and wetlands (Batzer & Sharitz, 2014;Richardson et al., 2018).The flood-pulse concept (Junk et al., 1989) emphasized the importance of river-floodplain exchanges.
During drought, rivers receive reduced overland flows and catchment inputs, and floodplain connection declines.This reduced connectivity with terrestrial landscapes and floodplains is considered a likely mechanism for decreased nutrients in river waters during drought (Golladay & Battle, 2002;Hunt et al., 2005).This has been corroborated by the mobilization of these stored nutrients and carbon during floods (Frazar et al., 2019;Kaushal et al., 2014).Nutrient dynamics are obviously complex during drought, and in our study were likely complicated by flow regulation, and by point source and non-point source pollution.The results of this study indicate that drought conditions affect water quality metrics, but not always in consistent or expected ways.
Our observations were likely related to several mechanisms, including volume reduction, persistent point source pollution, reduced natural catchment inputs, and reduced floodplain interactions.Although we were not aware of what management actions were implemented in response to drought, we acknowledge that management efforts to mitigate changes in regulated nutrients may have affected differences between our study time periods.Given the sensitivity of aquatic life, especially macroinvertebrates, to minute changes in water chemistry, drought was likely disruptive to aquatic biological communities, especially those which may be affected by changes in pH and total nitrogen, considering the differences seen between normal and drought conditions.Likewise, animals which feed on these more sensitive organisms may have been affected.That said, impacts during drought were dynamic and complex, and a more complete understanding of these interactions in large rivers will become increasingly important with the anticipated increases in intensity, frequency and duration of drought disturbances in the future.
T A B L E A 5 Coefficient of variation (CV) (%) of water temperature (ºC) for drought and normal hydrological conditions on the Savannah River for Sites 1, 2, 3, 4 and 5.
site and Site 5 (98 RKM) was the most downstream site.Site 1 was approximately 7 RKM below the Augusta Diversion Dam at the base of a rocky shoals.Site 2 was located 12 RKM downstream from the Horse Creek confluence, whose watershed drains the cities of North Augusta and Aiken, South Carolina.Site 3 was located 18 RKM downstream from New Savannah Bluff Lock and Dam, 12 RKM downstream from the Butler Creek confluence, whose watershed drains Augusta, GA, and 4.3 RKM downstream from a large pulp and paper mill.Site 4 was located 6.4 RKM downstream of Plant Vogtle, a nuclear electric generating plant.Site 5 was located in the lower Coastal Plain ecoregion near Clyo, Georgia.Our study sites spanned the majority of the lower river, from the Fall line across the Coastal Plain to the upper extent of any tidal influences.

2. 2
| Drought In Georgia, maximum daily air temperature ( C) during the 2006-2008 study period averaged 25.6 ± 0.8 C and minimum daily air temperature averaged 10.8 ± 0.2 C.During the 2016-2019 study period, maximum daily air temperature averaged 26.2 ± 0.2 C and minimum daily air temperature averaged 12.4 ± 0.2 C (NOAA Station USW00003820).Maximum daily air temperature was statistically similar between the two study periods, but average minimum temperature was significantly (F 1,2554 = 21.86,p < 0.01) lower during the 2006-2008 study period (Table A2).However, during the 2006-2008 study period, Georgia experienced considerable precipitation deficits.Total precipitation averaged 81.3 ± 5.1 millimeters (mm) per month for the 2006-2008 study period and 106.7 ± 7.6 mm per month for the 2016-2019 study period, a difference of 23.8% (Young et al., 2020).Drought data were obtained from the US Drought Monitor, which combines the Palmer Drought Severity Index, the Standardized Precipitation Index and other climatological inputs (Akyuz, 2017).The 2006-2008 study period was significantly drier for all drought metrics (Figure 2).A particularly intense period of drought occurred the week of December 11, 2007, when exceptional drought (D4) conditions F I G U R E 1 Study sites on the Savannah River included Site 1 (33.50277,À81.99067), Site 2 (33.38391,À81.93174), Site 3 (33.31791,À81.89093), Site 4 (33.11608,À81.69772) and Site 5 (32.52474,À81.26239).The Savannah River forms the border between Georgia and South Carolina, USA.EPA level III ecoregions are indicated with black lines including 45 (Piedmont), 63 (Middle Atlantic Coastal Plain), 65 (Southeastern Plains) and 75 (Southern Coastal Plain), and gray lines indicate other river systems.The transition between the Piedmont ecoregion and Southeastern Plains ecoregion is known as the Fall line.F I G U R E 2 Area (% area) of Georgia experiencing drought from 2006 to 2019.Bracket indicates study periods (drought [2006-2008] or normal [2016-2019]).Data were obtained from the US Drought Monitor.Drought was significantly higher during 2006-2008 for abnormally dry (D0), moderate drought (D1), severe drought (D2), extreme drought (D3) and exceptional drought (D4) and significantly lower during 2016-2019 for no instances of drought.

Figure 3 )
Figure 3).Discharge for the drought study period averaged 150.0 ± 8.9 m 3 s À1 at the uppermost gage and increased to 193.1 ± 10.3 m 3 s À1 at the lowest gage.In contrast, discharge for the normal study period averaged 206.3 ± 23.0 m 3 s À1 at the uppermost gage and 262.8 ± 25.7 m 3 s À1 at the lowest gage, an approximate 27% Samples from the 2006-2008 study period were analyzed by Pace Analytical (Columbia, SC, USA), and samples from the 2016-2019 study period were analyzed by Pace Analytical and Phinizy Center for Water Science (Augusta, GA, USA).NO X , NH 3 and TP were determined by colorimetric analysis with a discrete analyzer (Seal Analytical v. 4, Mequon, WI), and TOC and DOC were determined by combustion catalytic oxidation and non-dispersive infra-red detection by a total organic analyzer (TOC-L) with TN module with a chemiluminescence detector (Shimadzu v. 1.04, Columbia, MD).Methods used to analyze included 64 Â 10 À3 ¼ kilograms day Data were organized by Julian date and grouped by drought conditions (2006-2008) and normal conditions (2016-2019).Data were blocked by historical flow seasonality (Benke, 2001) to ensure equal number of samples were collected during both natural high flow and low flow periods.The natural high flow period was 1 December to 31 May, and the natural low flow period was 1 June to 30 November.

F
I G U R E 4 (a) Water temperature ( C), (b) dissolved oxygen (DO) (mg L À1 ) and (c) dissolved oxygen (DO) (% saturation).Box and whisker plots in white represent the drought period and box and whisker plots in gray represent the normal period.The bottom and top of each box are the 25th and 75th percentiles of the samples, the line in the middle of each box is the median, whiskers extend above and below each box to 1.5 times the interquartile range and observations beyond the whisker length are marked as outliers with an individual symbol.Where site effects were significant ( p < 0.05), Tukey-HSD tests were used to separate means (indicated by small letters), and sites indicated by the same letter are not different.The most extreme values are not shown for figure clarity.
Zieli nski et al. (2009).Elsewhere, higher pH and dissolved ions during drought have been attributed to more saline and more alkaline (bicarbonate) groundwater inputs, coupled with decreased dilution F I G U R E 5 Physicochemical ionic metrics including (a) pH and (b) overall conductivity (μS cm À1 ).For conductivity, a significant interaction existed between sites and time periods, so a series of 1-way ANOVAs were used; Sites 1 and 2 were similar between time periods, but sites 3, 4 and 5 were significant (denoted as ***).Box and whisker plots in white represent the drought period and box and whisker plots in gray represent the normal period.The bottom and top of each box are the 25th and 75th percentiles of the samples, the line in the middle of each box is the median, whiskers extend above and below each box to 1.5 times the interquartile range and observations beyond the whisker length are marked as outliers with an individual symbol.Where site effects were significant ( p < 0.05), Tukey-HSD tests were used to separate means (indicated by small letters), and sites indicated by the same letter are not different.The most extreme values are not shown for figure clarity.

F
I G U R E 6 Nutrient metrics including (a) total nitrogen (TN) concentrations (mg-N L À1 ), (b) total nitrogen flux (kg-N day À1 ), (c) NO X (nitrate +nitrite) concentrations (mg-N L À1 ), (d) nitrate + nitrite flux (kg-N day À1 ), (e) ammonia (NH 3 ) concentration (mg-N L À1 ), (f) ammonia flux (kg-N day À1 ), (g) total phosphorus (TP) concentration (mg-P L À1 ) and (h) total phosphorus flux (kg-P day À1 ).Box and whisker plots in white represent drought conditions and box and whisker plots in gray represent normal conditions.The bottom and top of each box are the 25th and 75th percentiles of the samples, the line in the middle of each box is the median, whiskers extend above and below each box to 1.5 times the interquartile range, and observations beyond the whisker length are marked as outliers with an individual symbol.Where site effects were significant ( p < 0.05), Tukey-HSD tests were used to separate means (indicated by small letters), and sites indicated by the same letter are not different.The most extreme values are not shown for figure clarity.

F
I G U R E 7 (a) Total organic carbon (TOC) concentrations (mg-C L À1 ), (b) total organic carbon flux (kg-C day À1 ), (c) dissolved organic carbon (DOC) concentration (mg-C L À1 ) and (d) dissolved organic carbon flux (mg-C L À1 ).Box and whisker plots in white represent the drought conditions and box and whisker plots in gray represent the normal conditions.The bottom and top of each box are the 25th and 75th percentiles of the samples respectively, the line in the middle of each box is the median, whiskers extend above and below each box to 1.5 times the interquartile range, and observations beyond the whisker length are marked as outliers with an individual symbol.Where site effects were significant ( p < 0.05), Tukey-HSD tests were used to separate means (indicated by small letters), and sites indicated by the same letter are not different.The most extreme values are not shown for figure clarity.