The town of Collie (population ≈ 7500) is located 160 km south-southeast of the city of Perth and is the centre of coal mining in the Australian state of Western Australia (Fig. 1). Major land uses in the basin are coal mining, timber production, power generation, and agriculture. The Collie coal basin covers an area of ≈ 225 km2 and consists of two sub-basins, the Cardiff sub-basin (151 km2) to the west and the Premier sub-basin (74 km2) to the east, separated partially by a faulted basement high known as the Stockton Ridge (Moncrieff 1993). There are an estimated 6400 Mt of coal resource in the basin, of which economic-demonstrated resources account for 930 Mt (Moody 2017). ‘Collie’ coal is a sub-bituminous coal with low sulfur content (0.3–0.9%), and low caking and low ash (4–9%) properties (Stedman 1988).
Collie is situated in a Mediterranean climate (Koppen-Geiger climate type Csa; Perera et al. 2015), with hot, dry summers (range 11.7–30.5 °C) and cool, wet winters (range 4.2–16.3 °C) (Commonwealth of Australia, Bureau of Meteorology (BOM) 18/5/2018). 75% of the annual rainfall occurs from May to September, and outside these months, evaporation rates exceed rainfall. The 100-year mean annual rainfall for the Collie Basin was 933 mm (BOM 18/5/2018), although this has decreased to an average of 731 mm over the last 15 years. Total annual rainfall in Collie between 2004 and 2018 ranged between a minimum of 390 mm in 2010 and a maximum of 902 mm in 2005.
The Collie River has two branches, the main ‘eastern’ branch and smaller ‘southern’ branch (hereafter referred to as the ‘Collie River South Branch’ or CRSB). The CRSB is an intermittently flowing river due to seasonal rain falls, with total annual flows from 2010 to 2019 averaging 10.7 × 106 m3 (Dept. of Water and Environmental Regulation, unpublished gauging data). Land clearing for agriculture has caused secondary salinization in the catchment (Cramer and Hobbs 2002; Lymbery et al. 2003), particularly in the Eastern branch, which has degraded overall catchment water quality (Platt et al. 2012). The CRSB has a catchment area of 66,047 ha, which was primarily native jarrah (Eucalyptus marginata) forest with 24% cleared for farmland and < 5% disturbed by mining (Harper et al. 2005). Under mining agreements with the state government, companies are mandated to return mined land to jarrah forest, which included reforesting pit lake catchments.
Open-cut coal mining has created a ‘district’ of 10 pit lakes in the Collie region (Fig. 1). Previous research by the authors demonstrated that the water quality in these pit lakes varied spatially and temporally despite their close proximities (Lund and McCullough 2008; Lund et al. 2012, 2014; Phillips et al. 2000; Zhao et al. 2009). In general, low levels of pyrite oxidation, ferrolysis, and secondary mineralization combined with the low buffering capacity of surrounding rock resulted in the Collie pit lakes tending to have a low pH and low acidity, with some high metal and metalloid concentrations (Lund et al. 2012; Zhao et al. 2009).
This study sampled five pit lakes: Stockton, WON9, Kepwari, WO5H, and Black Diamond A (hereafter referred to as ‘Black Diamond’) (Fig. 1). Historical water quality parameters can be found in supplemental Table 1. All these pit lakes were warm monomictic (mixing typically in June and July) and at the time of the study all had oxic hypolimnions (authors’ unpublished data). Additionally, all the lakes consisted of a single basin, with steep sides and limited shallow areas. The largest pit lake sampled was Kepwari (maximum volume is 24 GL; maximum depth is 65 m), which had the Collie River South Branch (CRSB) diverted around it during mining. The final lake water level was established in 2008 under a rapid-fill program whereby river water was diverted into the mine void during periods of high flow with no discharge from the lake (Salmon et al. 2008). Rapid fill was used to limit acidification of the lake by preventing further oxidation of sulfides in exposed coal seams. Following an accidental river inflow (see McCullough et al. 2012), the state government approved a 3-year trial of connecting the river to the lake as a closure strategy (Lund and Blanchette 2018). During this trial period (2012–2018), water was diverted into the lake via a gate-controlled inlet. The final closure and relinquishment plans involve backfilling the diversion channel and constructing permanent uncontrolled inlet and outlet structures to allow the river to flow unimpeded through the lake. The riparian and catchment areas around Kepwari that were disturbed during mining were contoured for safety and stability and revegetated with upland vegetation in 2002, with some areas revegetated earlier (bathymetry in Müller et al. 2010).
WO5H (volume 11.1 GL; maximum depth 81 m) was the second-largest Collie pit lake sampled in the study, and bathymetry details can be found in McCullough et al. (2009). WO5H was rapid-filled between 1997 and 2004 with a natural stream composed of mine water surface drainage, dewatering water, and rainfall (Zhao et al. 2009). WO5H is connected to a small stream that is used to manage water runoff from the mine site, including water pumped from operational areas. There is a potential for discharge from WO5H into a natural waterway (linked to EW1); however, the continuous extraction of lake water for power station cooling during the study period prevented discharge and altered water levels by several meters over relatively short periods (weeks). The riparian and adjacent terrestrial areas of WO5H have also been contoured for safety and stability with revegetation using upland taxa during the late 1990s. Planting extended to 2 m below the final water level to introduce organic matter to the lake (see Blanchette et al. (2019) for effect of planting on ecology of Lake WO5H).
The Stockton pit was abandoned by the mining company and ownership returned to the state of Western Australia in the 1960s without any form of rehabilitation at the time of abandonment. Stockton was managed by the state and used for public camping and water recreation, particularly water skiing. Lake Stockton has had some terrestrial remediation works to reduce risks from spontaneous combustion of coal deposits and highwall collapse. EW1 Creek (Fig. 1) is a natural stream supplemented on occasion by water from Ewington Mine that flows through Lake Stockton. The EW1 creek drains from Stockton into an adjacent swamp and then onwards to eventually join the CRSB (Fig. 1).
The Black Diamond pit was also abandoned and returned to the state of Western Australia in the 1960s without any form of rehabilitation, although, like Stockton, it has spontaneously but slowly regenerated with both native and exotic trees. Black Diamond was owned both by the state and privately, but was not actively managed, although it too has been used for water recreational activities. Black Diamond discharges into the Collie River at high water levels. At the time of the study, the catchment of Black Diamond had revegetated naturally and there were steep (1–4 m) cliffs around much of the shoreline.
Mining in WON9 ceased in 1991 and the lake has not yet reached predicted final water levels (and may never), resulting in 1–2 m high cliffs along the lake’s edge and few littoral areas. Due to its morphology, WON9 cannot discharge surface waters. Like WO5H and Kepwari, the catchment of WON9 has been contoured for safety and stability and upland vegetation was established throughout the catchment, including riparian areas.
Broad Experimental Design and Rationale
Stage one of the broad experimental design evaluated how pit lake catchment characteristics (age, degree of rehabilitation) correlated with potential nutrient sources, and the relationship between those sources and lake sediment nutrient concentrations (see Table 1). Four of the five catchments were sampled in July 2012—WO5H was not evaluated as part of stage one. Determination of catchment nutrient sources was assessed through measurement of soil concentrations of C, P, and N and litter biomass along transects. Age and measures of rehabilitation such as vegetation communities and tree biomass (density and basal area), litter, and shrub cover were recorded and compared to literature-derived data for co-occurring natural forest.
Stage two sampled Lakes Kepwari and WO5H during 2016, 2017, and 2018 (Table 1). The purpose of stage two was to measure sedimentation rates and the rate of C accumulation in sediment in pit lakes connected to watercourses. During the period of inflow into the lakes (September and October), sedimentation rates (total and C) were measured in the middle of each lake at multiple depths. Additionally, in 2018, sedimentation rates (total and C) close to the edge of these lakes was measured to evaluate the input of allochthonous material from the adjacent catchment. Benthic sediment data was collected from both lakes from different water depths in 2016 and 2018 to measure C accumulation in the sediments.
Stage three modelled the magnitude of nutrient inflows and losses for Lake Kepwari over the period 2013–2016. The simple model used river flow and nutrient concentration data (groundwater and river) collected directly from the Lake Kepwari closure trial as well as from the literature. The purpose of stage three was to determine the potential accumulation of nutrients from the Collie River South Branch and local catchment runoff to drive ecosystem development in Kepwari.
Stage 1: Catchment Sampling
Catchment and lake areas were estimated using Google Earth Pro (Google, USA) in 2012 and revised in 2019 using the polygon tool to trace visible contours and provide an area estimate. The catchment areas were also ground truthed during sampling using visual landmarks.
Pit lake catchments (excluding WO5H) were sampled in July 2012 for vegetation characteristics. The catchments were predominantly covered in jarrah forest, although the unrehabilitated catchments also contained some exotic tree species. Three transects were placed within each catchment perpendicular to the shore and heading upslope at three different localities around the lake to capture as much of any visually-assessed natural variability in vegetation as possible (Kent 2011). Transects were 5 m wide belts, starting 10 m into the lake and extending to the top of the catchment to a maximum length of 110 m.
For each transect, each tree rooted within the transect was measured, and terrestrial soil, litter, and sediment samples were collected (n = two samples of each per transect). Trees were identified to species, and their diameter at breast height (DBH) and crown width were measured with a tape measure. The basal area of each tree was calculated for each stem or trunk using the formula πr2, where applicable (where ‘r’ is half the DBH). A clinometer was used to measure tree height, and the slope (in degrees) of the ground was measured between trees. A section of the transect 1 m above and below each tree was visually assessed to determine the percentage cover of ground plants (< 0.3 m high) and shrubs. Two sites (2 m in length) per transect were established: 5 m back from the shoreline (hereafter referred to as ‘edge’) and approximately halfway along the transect (‘upslope’). At each edge or upslope site, the representative leaf litter depth (to the soil) was measured with a ruler and the percent of litter cover was visually estimated. Additionally, a 0.25 m × 0.25 m quadrat was randomly established within each site and all the litter and soil (to 50 mm) was collected and stored in clean cotton bags.
In the aquatic sections of each transect, sediments were collected 5 and 10 m out from the shore (n = four benthic sediment samples per transect). SCUBA was used to collect the benthic sediment samples using an acrylic tube, (0.6 (l) × 0.12 (dia.) m) to a sediment depth of at least 100 mm. Samples were removed by sealing the tubes top and bottom with rubber bungs and residual water was decanted. The top 10 mm of sediments (representing the most recent deposits) and then 10-50 mm (longer-term accumulations) were each sliced off from the collected sediment and stored separately in plastic bags. All samples (leaves, soil, sediments) were stored at ≈ 4 °C using freezer blocks in a cooler (or refrigerator, if overnight) for transport to the laboratory.
On returning to the laboratory, litter, soil, and sediment samples were manually homogenised and then stored at 4 °C before being oven-dried at 80 °C to constant weights. Loss on ignition (LOI) was used to assess the percent organic matter in leaf litter and sediments, following burning at 550 °C for 1 h to constant weight (Benfield 2007). All soil samples were analysed for organic C, nitrate-N, ammonium-N, P (total and Colwell), sulphur, electrical conductivity (EC), and pH (CaCl2) at CSBP Soil Analysis Laboratories (Perth, Western Australia) as per Rayment and Higginson (1992). Sediment samples were also analysed for total P concentrations at the Edith Cowan University Analytical Facility, as per APHA (2017). Where soil or sediment analysis resulted in values that were below detection, these values were assigned a value of half the detection limit for data analysis.
Tree density, stand basal area, and terrestrial surface litter cover data for co-occurring native forest unimpacted by mining was sourced from Abbott and Loneragan (1986), McCaw (2011), and Stoneman et al. (1997). These values were compared to data collected as part of this study to determine whether there were differences in pit lake vegetation communities compared to natural forest that might affect nutrient availability. The literature values were divided into ‘old growth and long unburnt forests’, representing high biomass and litter levels and ‘regrowth and regularly burnt’ (previously logged forests, regenerated, and regularly burnt through a prescribed hazard-reduction burning program) representing most of the jarrah forest in the Collie region.
Stage 2: Sedimentation, Sediment Sampling and Laboratory Processing
In the centres of WO5H and Kepwari, a single set of sediment traps made of plastic cylinders were deployed for 30 days in Sept. 2016, 42 days in Sept./Oct. 2017, and 23 days in Sept./Oct. 2018 (Table 1). A ‘set’ of sediment traps (Fig. 2) consisted of groups of 3 cylinders (0.98 m long, 80 mm dia.) sealed at the bottom and attached to a 50 m rope deployed at depths of 10, 25, and 40 m representing the top, middle, and bottom of the water column, respectively (Bloesch and Burns 1980). The vertical cylinders were attached ≈ 0.2 m apart, minimising contamination from any algal growth that might occur on the rope. The 50 m rope was kept vertical, with a 15 kg weight at one end, and tensioned by a buoy mounted 2 m below the water surface, with a second small free-floating buoy used to locate the traps (Fig. 2). Additionally, in 2018, three sets of sediment traps were deployed at equidistant sites around the edges of both lakes (Table 1). However, each set consisted of only a single group of three cylinders mounted at 5 m depth on a 10 m rope—otherwise, they were designed and deployed as per the central sediment traps.
At the end of each deployment, all sediment traps were removed from the lakes and each cylinder was immediately capped after collection. Each capped cylinder was manually shaken for at least 20 s to homogenise any captured sediment, and then a 4 L aliquot was decanted into storage containers for transport at ambient temperatures to the laboratory. Samples were stored at 4 °C for less than 48 h prior to processing. In the laboratory, duplicates of each sample were filtered under vacuum pressure through 0.5 µm (47 mm) pre-weighed glass fibre filter paper (Pall Metrigard, USA) until it clogged (85–2455 mL) for immediate determination of dry weight (DW) following drying at 80 °C to constant weight. Dried filters were then burnt for 1 h at 550 °C to allow for determination of carbon content via loss on ignition (LOI). Carbon content of the sediment collected from the sediment traps was estimated as half of the LOI value, as per Pribyl (2010).
Five equidistant sites were randomly selected in both Kepwari and WO5H for benthic sediment collection (Table 1). At each site, one sample was collected from each of the following water depths: < 10 m (‘shallow’), 10-20 m (‘intermediate’), and > 20 m (‘deep’). Sediment was collected in 2016 using either an Ekman dredge or Peterson sampler (Wildco, Fl, USA) and in 2018 using a Standard Ponar sampler (Wildco, Fl, USA). Sediment samples consisted of ≈ 0.5 L from the top 100 mm of sediment, regardless of collection device. In the laboratory, each sediment sample was manually homogenised, and then triplicate subsamples were dried to a constant weight at 60 °C and then burnt for 1 h at 550 °C for determination of LOI.
Stage 3: A Simple Nutrient Budget for Lake Kepwari
The water budget for Kepwari reported in Lund and Blanchette (2018) and Lund et al. (2018) was simplified into a mean annual model for 2013 to 2016. Key components of the water budget were the inflows: surface runoff, direct rainfall into the lake, river, and groundwater, and the outflows: river, evaporation, and groundwater. Other components of the water and nutrient budget such as denitrification and interflow (see Grigg 2017), were considered to have a relatively small impact on the nutrient budget and were not investigated further. Data for each water budget component was averaged (n = 4) over the time period.
Nutrient data was obtained from Lund and Blanchette (2018) and monitoring data collected by Premier Coal (unpublished) as part of its requirements under the river flow through trial. Nutrient concentrations in surface runoff from the catchment, prior to river connection, at Kepwari were taken from Salmon et al. (2008) and were based on an opportunistic single sample recorded in 2004. Mean nutrient concentration (from all available data n = 18–43) over 2013–2016 for river inflows and outflows was determined.
Groundwater inflow nutrient concentrations were estimated from two nested groundwater bores on the upstream western side of Kepwari sampled by Premier Coal (unpublished data) in Oct. 2015. Mean nutrient concentrations were determined across both bores and screened depths (n = 6). Groundwater outflow out of Kepwari was not measured; however, nutrient concentrations should reflect those of the lake’s hypolimnion. Therefore, a mean (n = 11–17) of hypolimnetic water samples collected by (Lund and Blanchette 2018) were used to represent groundwater outflow concentrations. No ammonia was measured in the groundwater inflow, but it is typically low relative to NO3/NO2 (Wetzel 2003).
Total inflow and outflow loads were determined by summing individual component loads of total P, N, and organic C. However, as total N and P are commonly not measured for groundwater, ammonia and NO3/NO2 loads and groundwater outflow were used for groundwater inflow instead.
Surface runoff export rates were determined for the Kepwari catchment prior to connection to the CRSB and for the CRSB catchment. Export rates were calculated by dividing the relevant nutrient load by the catchment area.
The broad approach to data analysis involved use of descriptive statistics, correlations (Pearson), ANOVA, and multivariate analysis using ordination, followed by hypothesis testing using PERMANOVA. Multivariate analysis was undertaken using PRIMER v7 and PERMANOVA + (PRIMER-e, Quest Research Ltd., Auckland, New Zealand) and univariate analysis using SPSS v24 (IBM Corporation, Armonk, NY). Univariate data were transformed (log10 or log10 + 1) as required to achieve homoscedasticity as determined by the Levene’s Test. Areas where homoscedasticity was not demonstrated after transformation has been indicated in the results and caution was used in interpretation. Multivariate data was also log10 transformed and normalised prior to determination of resemblance based on Euclidean distance (Clarke and Gorley 2015).
Counts of trees and basal area were scaled to a hectare basis based on the number of trees and total basal area per transect, and the area (length × 5 m) covered by transects above the water line. Slope, height, shrub, litter, ground cover, and litter depth were averaged across all measurements per transect. Vegetation characteristics were ordinated using principal components analysis (PCO) to ascertain differences among catchments. All vegetation characteristics were then correlated to the ordination space using Pearson’s r and shown (where r > 0.6) as vectors on the PCO plot. In multiparameter analysis, if there were only one or two missing values for the parameter (e.g. tree height), then the parameter was retained and the ‘Missing’ routine in PRIMER (Clarke and Gorley 2015) was used to estimate the missing variable(s); otherwise, the parameter was removed from the analysis. The null hypothesis of no significant difference (p > 0.05) among vegetation characteristics from catchments (fixed factor) was tested using a one-way PERMANOVA (pseudo-F is reported as Fp; 99999 permutations). Significant differences (p < 0.05) for each vegetation characteristic between catchments were then tested using one-way ANOVA, followed by pairwise comparisons using the Tukey B test. One-way ANOVA was also used to determine significant differences in catchment litter loads across all transects between upslope and edges.
Physico-chemical data (soil concentrations of total P, ammonium, NOx, Colwell P, sulphur, organic C, measures of EC and pH (CaCl2)) were log10 transformed and ordinated in a PCO. All physico-chemical data were then correlated to the ordination space using Pearson’s r and shown as vectors on the PCO plot. The null hypothesis of no significant (p > 0.05) difference in physico-chemical data among catchments and location (upslope and edge) was tested using a two-way PERMANOVA with the lake as a fixed factor and location as a random factor in the analysis. Relationships between individual soil parameters were examined by Pearson’s r (α = 0.05).
The null hypotheses of no significant difference in % organic matter or total P in sediments between catchments, sediment depth, and distance from the shore (fixed factors) were individually tested in three-way ANOVAs between catchments. In addition, the null hypothesis of no significant difference in % organic matter in sediments between depths, years, and lakes (Kepwari and WO5H) was tested using three-way ANOVA (fixed factors). Homoscedasticity was not demonstrated, but a log10 transformation reduced heteroscedasticity; however, caution should be used during results interpretation, especially in light of the small sample size (Table 1).
Data (% organic matter, C, and total dry weight sedimentation rates) from the three cylinders per depth were averaged to determine the characteristics of pelagic sediment at different depths from the sediment traps. Occasionally one cylinder out of the three for a depth would have substantially different values from the other two (> 50% different), mainly due the cylinder not hanging vertically in the water column or collecting a clump of material; these results were excluded from the average (Lund et al. 2019). There was no replication of the central sediment traps within lakes (Hulbert 1984), so results at the lake-scale were descriptive only; however, sediment traps around the edge were replicated within lakes (n = 3 per lake) and differences in C sedimentation rate and dry weight of sedimentation were tested individually using one-way ANOVA.