Environmental Modelling & Software 26 (2011) 913e928 Contents lists available at ScienceDirect Environmental Modelling & Software journal homepage: www.elsevier.com/locate/envsoft Estimating catchment-scale impacts of wildfire on sediment and nutrient loads using the E2 catchment modelling framework Paul M. Feikema a, *, Gary J. Sheridan a, c, Robert M. Argent b, c, Patrick N.J. Lane a, c, Rodger B. Grayson b, c a Department of Forest and Ecosystem Science, The University of Melbourne, 221 Bouverie Street, Carlton, VIC 3053, Australia Department of Civil and Environmental Engineering, The University of Melbourne, Parkville, Victoria, Australia c eWater Cooperative Research Centre, Canberra, Australian Capital Territory, Australia b a r t i c l e i n f o a b s t r a c t Article history: Received 24 December 2009 Received in revised form 21 January 2011 Accepted 1 February 2011 Available online 3 March 2011 Approximately 1.3 million ha of forested and agricultural land in south-eastern Australia was burnt by wildfires in early 2003. This paper describes a modelling process to assess the impacts of the fires on the quality of receiving waters and river systems in the fire-affected catchments. First, we construct and parameterise the E2 catchment modelling framework to represent the flow, and sediment and nutrient loads for the water storages and river systems in fire-affected sub-catchments and second, we assess likely impacts of the fires on loads of total suspended sediments (TSS), total nitrogen (TN) and total phosphorus (TP). Very good calibration (with coefficient of efficiency values generally greater than 0.8) of the rainfallrunoff models to observed flow at several gauging stations within each catchment was achieved. Digitised land use layers were reclassified to form functional units representing unburnt and burnt land uses. Pre- and post-fire loads of TSS, TN and TP predicted by the model at end of catchment outlets and water storages were then compared relative to pre-fire loads. Compared to pre-fire conditions, the models predicted that the Ovens, Kiewa, Upper Murray and Snowy catchments would deliver, on average, approximately 30 times greater TSS, 5 times greater TN, and 8 times the amount of TP. Proportional increases in predicted loads at the catchment outlets were generally smaller than increases observed at water quality monitoring sites. These differences reflect the proximity of the monitoring stations to the burnt areas, the total percentage of catchment burnt, and the amount of rainfall. The predictions of load increases carry important assumptions and limitations, and such an approach can only be used for making relative assessments of the impacts of the fires on average suspended sediments and nutrient loads. Ó 2011 Elsevier Ltd. All rights reserved. Keywords: Catchment modelling E2 Water quality Wildfire Sediment loads Nutrient loads 1. Introduction During January and February 2003, approximately 1.3 million hectares of forested and agricultural land in south-eastern Australia was burnt by wildfire. Elevated sediment and nutrients loads are widely recognised as a consequence of severe wildfire, and scale and severity of the 2003 fires were suggestive of substantial changes in the quality and quantity of water available for downstream users. The fire-affected region included large areas within the Upper Murray, Kiewa, Ovens and Snowy river basins that drain into several key water storages, including the Hume Reservoir and Lake Dartmouth. The impact of these fires on * Corresponding author. Tel.: þ61 3 8344 0715. E-mail address: pfeikema@unimelb.edu.au (P.M. Feikema). 1364-8152/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.envsoft.2011.02.002 the water quality of water courses and receiving bodies has the potential to be significant. Following the fires, a program of water quality monitoring was initiated by the Victorian Department of Sustainability and Environment (DSE) in several of the Victorian catchments affected by the fires. The objective of the monitoring program was to enable post-fire changes in loads of total suspended solids (TSS), total phosphorus (TP) and total nitrogen (TN) to be estimated. Sheridan et al. (2007a) analysed water quality data collected from numerous hydrologic stations downstream of the fire-affected area. These data were collected from automated pumping samplers installed following the fire, and from routine periodic samples collected before and after the fire. Large impacts were recorded for all constituents, however, the levels of change were not consistent across the different catchments (Sheridan et al., 2007a). Despite uncertainty surrounding many of the load estimates, it was clear 914 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 that the fires had had a significant impact on water quality in most of the sampled catchments. Forests most often occur in upland areas, and yet it is the downstream effect of forest fires, often some distance away, that is of most interest. The changes in constituent export at larger scales are of particular importance to land and catchment managers where the potential downstream impacts may be severe, both in terms of damage to rural landscapes but also to property and infrastructure in urban areas. While observed increases after the 2003 fires in sediment and nutrient loads in the upper sub-catchments at the water quality monitoring stations were high, the loads delivered to water impoundments in the lower parts of catchments or at the catchment outlet are likely to be substantially lower. This is because sediment and adsorbed nutrients are stored within stream channels in the lower reaches of the stream network as the slope of the stream channel is reduced. Some of this material may, however, be remobilised during subsequent periods of high flow. Studies reporting concentrations of N and P in streams after wildfire are more extensive than those estimating exports (Smith et al., 2011). Studies into wildfire responses are generally opportunistic (e.g. Lane et al., 2006) which are often constrained by experimental design, and often lack good pre-fire data, making it difficult to infer post-fire responses (Shakesby and Doerr, 2006). Relatively few studies have been carried out in Australia of impacts at catchment scales (e.g. Brown, 1972; Chessman, 1986; Prosser and Williams, 1998; Lane et al., 2006, 2008; Blake et al., 2006; Wilkinson et al., 2009a, b) and these have generally been small catchments, less than 50 km2 in the case of the studies by Lane et al. (2006, 2008) and Prosser and Williams (1998). Research into the effects of fire on water quality has tended to be conducted on relatively small catchments that have been completely or almost completely burnt. The response of larger catchments (>50 km2) to wildfire has been less studied, partially due to the difficulty in establishing an maintaining monitoring equipment, the relatively long recovery periods at larger scales, and the greater spatial heterogeneity in catchment characteristics including climate, topography, geology, soils, vegetation, and fire severity. The effects of wildfires on erosion processes are spatially variable (Sheridan et al., 2007b), and together with spatially and temporally variable post-fire rainfall, lead to punctuated deliveries of constituents greater than the expected pre-fire deliveries (Moody and Martin, 2009). While at the plot scale, fire has been shown to increase sediment transport primarily through reduced ground cover, but this tends to be very localised, and much lower increases in sediment yield may be observed after fire at the catchment scale (Prosser and Williams, 1998; Scott, 1993; Moody and Martin, 2001). Within the fire area, there was one reported case where postfire runoff-generated debris flows occurred (Nyman et al., 2011). This was from a cluster of small catchments (<200 ha) in the upper reaches of the Ovens catchment in a month after the fires, after a very high intensity, short duration storm event (estimated at 150 mm in 1 h) (Smith et al., 2011). A pulse of highly turbid water travelled into the Ovens River, and turbidity levels in the Ovens River remained elevated at Wangaratta (over 150 km downstream) twelve days after the event (Lyon and O’Connor, 2008). There is a need to develop approaches that allow estimates of post-fire impacts at broader scales as a function of the main catchment characteristics. We are not aware of any studies that have attempted to scale changes in constituent loads in fire-affected subcatchments to larger, mixed land use catchments. Furthermore, scaling or extrapolating changes in sediment or nutrient loads is problematic because the nature of wildfires on water and constituent dynamics is diverse across both temporal and spatial scales. Prosser et al. (2001) identified this as the main reason why prediction of sediment export at large scales has been based on empirical relationships, and encouraged the use of simple physically based approaches that account for spatial patterns of delivery. Previous limits to computing capacity have made catchment-scale analysis particularly difficult for larger catchments. Modelling applications on large catchments were often conceptual and did not incorporate spatial data (e.g. Sivapalan et al.,1996; Zammit et al., 2005) and the accuracy of those that do may decrease for large (>100 km2) catchments (e.g. Bhuyan et al., 2003). This may result, in part, from the lack of spatially distributed input data and the difficulty in parameterising large mixed land use catchments (Letcher et al., 2002). When compared to European and North American catchments, Australian hydrology typically has peakier flows, lower proportion of base flows, smaller runoff coefficients, and longer and more variable dry periods (Peel, 1999), and this warrants modelling at temporal scales sufficiently fine to capture these processes. This requires adequate temporal and spatial resolution rainfall data to address a range of questions. Technological advances over the past decade have facilitated the development of modeling platforms that combine flexible conceptualisation of catchment dynamic processes (Argent et al., 2009) and allow for extrapolation of changes in constituents based on catchment and water dynamics. This paper describes a modelling process to estimate the impacts of the fires on water quality of receiving waters and river systems in large, mixed land use catchments. First, we construct and parameterise models using the E2 catchment modelling framework (Argent et al., 2009) to represent the flow, and sediment and nutrient loads for the water storages and river systems in fireaffected sub-catchments, and second, we assess likely impacts of the fires on loads of total suspended sediments (TSS), total nitrogen (TN) and total phosphorus (TP) in the water storages and rivers of the entire catchments. 2. Catchment descriptions The four fire-affected catchments included in this study, Ovens, Kiewa, Upper Murray and Snowy, are shown collectively in Fig. 1. For all catchments, model predictions were made at the gauging station that included as much of the burnt areas as possible. The catchments vary in size (1714e13,485 km2), in mean annual runoff (68e394 mm) and in the percentage area burnt (21e80%). Elevation, slope and aspect information for each catchment are provided in Table 1. Individual catchment land use and fire extent information is shown in Fig. 2 and is described in more detail in the following sub-sections. 2.1. Ovens catchment The Ovens catchment, showing the water gauging stations and sub-catchment boundaries, is provided in Fig. 3. The gauging station at the Ovens River at Rocky Point was selected as the catchment outlet, and we did not set this further down the catchment (closer to the River Murray) at Ovens River at Peechelba due to the amount of uncertainty in representing diversions and the influences of more intensive agricultural land uses on water flows and quality beyond Ovens River at Rocky Point, and limited stream flow data for calibration. Predictions were made at Ovens River at Peechelba by extrapolation from predictions made upstream at Ovens River at Rocky Point. Pre-fire water quality data for the Ovens River at Peechelba was obtained from the Victorian Water Resources Data Warehouse (http://www.vicwaterdata.net/vicwaterdata/home.aspx) and comprised fortnightly samples between 1999e2002 for TSS, and 1981e2002 for TN and TP. A constant factor for each constituent was used to relate mean loads of these constituents to those in the P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 915 Fig. 1. Map showing fire-affected areas, catchment areas and catchment outlets included in this study. Rocky Point. Increases in post-fire constituent loads predicted at Rocky Point were related to the estimated pre-fire loads at Peechelba to derive estimates of factor increases in loads. As the prefire loads at Peechelba were higher than those upstream at Rocky Point (with lower concentrations but higher flows) factor changes at Peechelba were lower than those predicted at Rocky Point. The majority of the Ovens catchment (65%) is forested (with the southern half of the catchment nearly completely forested). The fireaffected areas were restricted to the southern portion, and virtually all the fire-affected areas were forested. The fire burnt about 16% of the forested area within the catchment (Table 2; Fig. 2a). The data used to calibrate constituent loads comprised a relatively large pre-fire dataset of 189 samples, resulting in a fairly low level of Table 1 Catchment characteristics. Average elevation and slope are shown (with range in brackets). Percentages of the catchment area with north, east, south and westerly aspects are also shown. Elevation (mASL) Ovens Kiewa Upper Murray Snowy 670 (195e1870) 13 (0e42) 25/26/19/30 625 (160e1960) 9.6 (0e36) 26/29/16/29 905 (225e1955) 10 (0e36) 26/27/22/25 Ovens at Bright Kiewa at Bandiana Mitta at Tallandoon Snowy at Jarrahmond 870 (10e2180) Slope ( ) % Aspect N/E/S/W Catchment Gauging station/ location 6.2 (0e38) 21/29/23/27 uncertainty in pre-fire estimates of constituents (Sheridan et al., 2007a). Similarly for the post-fire load increases which were based on 105 samples from 30% of the storm events from the following year. While there is relatively higher statistical certainty for this dataset, the sampling frequency was very low 3 years after the fire, with high variability and uncertainty in the data. (Sheridan et al., 2007a). Geology of the catchment comprises Ordovician sandstones and mudstones of marine origin (Hotham Group) predominantly in the upper catchment, with minor occurrences of Devonian granites (Mt Buffalo). Streams throughout the catchment are characterised by Quaternary sediments (lacustrine clays and sands; Coonambidgal formation) with more widespread occurrence of Quaternary sediments (silts and sands; Shepparton formation) lower in the catchment (DPI and DSE, 2004). The upper catchment is dominated by gradational friable earths (Dermosol; Isbell, 2002) and lower catchment is dominated by red duplex soils (Kandosol; Isbell, 2002). 2.2. Kiewa catchment For the Kiewa catchment, the station at Kiewa River at Bandiana represented the closest point to the River Murray system, and was chosen as the catchment outlet (Fig. 4). Modelling objectives were different in the Kiewa catchment than for the other catchments because the water quality data was already available for the station 916 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 Fig. 2. Maps of the (a) Ovens, (b) Kiewa, (c) Upper Murray and (d) Snowy catchments showing spatial distribution of unburnt and burnt functional units (land use). (Kiewa River at Bandiana) at the end of catchment. The model was still constructed to allow estimates of variability in loads resulting from changes in streamflow to be made. The catchment is primarily forested (58%), with the elevated areas close to the catchment boundaries completely forested. The fire-affected areas were restricted to the southern portion of the catchment, and virtually all the fire-affected areas were forested (Table 2; Fig. 2b). The fire burnt about 35% of the forested area. Both pre- and post-fire TSS calibration data for the Kiewa catchment are relatively poor, and with only 19 pre-fire fixedinterval samples, the pre-fire estimates are likely to be strongly negatively biased (Sheridan et al., 2007a). Furthermore, no samples collected in the year after the fire were from storm events, indicating that post-fire EMC values are likely to be underestimated. (Sheridan et al., 2007a). However, the large and continuous fixedinterval dataset (with greater confidence) for TN and TP from the Kiewa River suggested that the loads of these nutrients were only slightly changed by the fire. The geology is dominated by Ordivician-Silurian metamorphics, with Quaternary sediments (lacustrine clays and sands; Coonambidgal formation) along the water course (DPI and DSE, 2004). The upper catchment is dominated by gradational friable earths (Dermosol; Isbell, 2002) and lower catchment is dominated by duplex soils (Chromosol; Isbell, 2002). 2.3. Upper Murray catchment For the Upper Murray catchment, the closest gauging station to the Hume Reservoir, and the point of model predictions, was the gauging station at Mitta Mitta River at Tallandoon (Fig. 5). Estimates of changes in sediment and nutrient loads were made at the Dartmouth Dam and at the catchment outlet (the closet station above the Hume Reservoir) at Mitta Mitta River at Tallandoon. Predictions made at this station were used directly to infer changes in sediment and nutrient loads entering the Hume Reservoir. 917 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 Fig. 3. Ovens catchment showing water monitoring stations and catchment boundaries. The Upper Murray catchment is primarily forested (85%), with pockets of agricultural land uses in the north and south representing 15% of the catchment. The fires burnt approximately 85% of the forested areas and 47% of the agricultural areas (80% of the total area) within the catchment (Table 2; Fig. 2c). Pre-fire data for calibration of E2 for the Mitta Mitta River had a high degree of confidence, whereas the post-fire data has a high degree of uncertainty due to the presence of a few extreme values (Sheridan et al., 2007a). The upper catchment is predominantly Ordivician-Silurian metamorphics with Ordivician marine sediments in the north part of the catchment (DPI and DSE, 2004). Soils in the upper catchment are generally gradational earths (Kandosol; Isbell, 2002) while gradational friable earths (Dermosol; Isbell, 2002) occupy most of the lower catchment. 2.4. Snowy catchment The end of catchment outlet for the Snowy catchment was the station closest to the ocean (Snowy River at Jarrahmond), where tidal influences did not affect monitoring (Fig. 6). Estimates closer to the mouth of the Snowy River (at Orbost) are difficult to derive due to the tidal nature of the river. Predictions of increases in nutrient and sediment loads made at Jarrahmond were used to make direct estimates of increases on loads at Orbost. Approximately 54% of the Snowy catchment is forested, with forests predominantly in the southern half and western fringes of the catchment. The fire-affected areas were restricted to the western portion of the catchment, leading to approximately 47% of the forested area, and 7% of the agricultural area being burnt (29% in total) (Table 2; Fig. 2d). The pre-fire dataset used for calibration of the model for the Snowy River had a relatively low level of uncertainty, yet the postfire data contained much larger uncertainty, with only 13% of the storm events sampled (Sheridan et al., 2007a). The estimates of load increases within the Snowy catchment should therefore be interpreted with caution. Geology of the catchment is dominated by Silurian granites, with minor occurrences of Ordivician marine sediments in the lower parts of the catchment (DPI and DSE, 2004). Kandosol soils Table 2 Pre-fire constituent loads and percentage land use and fire-affected land use for all catchments. Note that the total percentage of a particular land use is the sum of not fire- and fire-affected percentages. Fire severity classes (Department of Sustainability and Environment, 2003) are from most severe; 1, crown burnt; 2, severe crown scorch; 3, moderate crown scorch; and 4, light crown scorch. Pre-fire loada (T yr 1) Not fire-affected (%) Catchment Gauging station Ovens Ovens at Bright Above Lake Buffalo Ovens at Rocky Pt Ovens at Peechelba Kiewa Kiewa R West Brancha Kiewa at Bandiana 13,324 249 28 Upper Murray Mitta at Hinommunjie Above Lake Dartmouth Mitta at Tallandoon 3034 111 Snowy Snowy at McKillops Snowy at Jarrahmond 1959 75 TSS a From Sheridan et al. (2007a). 1773 TN 43 TP Fire-affected (%) Forest Agric. Other Forest Other 1 40.4 62.6 46.4 44.4 4.3 2.2 10.7 34.4 0.3 0.2 0.1 0.3 54.3 34.9 42.2 20.4 0.7 0.2 0.6 0.5 0 0 0 0 8 7 1 5 0.5 37.9 e 38.8 e 0.5 95.5 20.6 e 2.2 e 0 1 19 0 2 14 8.6 6.5 12.6 8.6 7.4 7.8 e e e 70.7 77.5 72.7 12.1 8.7 7.0 e 0 e 4 0 90 5 5 20.8 28.5 47.5 39.0 5.0 3.9 23.5 25.4 3.0 2.9 0.2 0.2 e e e 6 Agric. Fire severity (% of catchment) 2 e 3 4 918 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 Fig. 4. Kiewa catchment showing water monitoring stations and sub-catchment boundaries. (Isbell, 2002) are found throughout the catchment, with generally gradational soils in the upper part and duplex or uniform soils in the lower part of the catchment. 3. Model construction, development and calibration 3.1. Modelling framework We adopted a modelling approach similar to the Environmental Management Support System (EMSS) described in the comprehensive review of erosion and sediment transport models by Merritt et al. (2003). This approach has relatively low input requirements and complexity while providing for temporally and spatially dynamic simulation of runoff and constituent generation and transport. This is suited for our application to relatively large catchments (from 500 to over 13,000 km2 in our study) where the complexity of input data at these spatial scales is relatively low. The specific modelling framework applied in this study is known as E2, and provides a flexible approach to whole-of-catchment modelling (Argent et al., 2009; Murray et al., 2005) and is available from the Catchment Modelling Toolkit (www.toolkit.net. au). Models created using E2 can represent the hydrologic Fig. 5. Upper Murray catchment showing water monitoring stations and sub-catchment boundaries. P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 919 Fig. 6. Snowy catchment showing water monitoring stations and sub-catchment boundaries. behaviour of any sized catchment with tens to hundreds of subcatchments and it can be conFig.d to predict flow and constituent (e.g. sediments and nutrient) loads at any point in a river network over time. The main model structure is “node-link” where subcatchments feed water and material fluxes into nodes, from where they are routed through links. Spatial information including elevation, land use, management, climate and soil may be used in modelling with the sub-catchment node-link structure. We used the E2 framework for several reasons. It has a flexible representation of space and time that permits its application to a large range of spatial scales and to use the most appropriate temporal scale for the processes to be simulated. It allows for a choice of alternate models, methods and systems (e.g. runoff model, constituent generation model) and has the capacity to simulate more than one water constituent as was the case in our study. We used E2 as a daily time step model, and results were considered at monthly and annual time scales. We required that the constituent generation model should be able to represent DWC (dry weather concentration) and EMC (event mean concentration) so that it could be used with appropriate event-based water quality data. The inclusion of DWC and EMC is important because large runoff events can carry a large proportion of the load (Baginska et al., 2003). We also needed the ability to modify flow and constituent transport in relation to water storages and regulated flows, which is vitally important when applying models in management. In short, the framework provides sufficient flexibility to adjust the nature and complexity of the system model, and is inclusive of the dominant catchment processes. It is therefore a desirable solution for the application of land and water science to management issues. interest. The fire-affected areas and the catchment areas are shown in Fig. 1, and represent the majority (65%) of the total fire-affected area in Victoria and in New South Wales. The DEM was resampled at a resolution of 250 m, and converted to a text grid file. The DEM was pit-filled and then imported into E2 where stream network and sub-catchment boundaries were delineated. The stream threshold was set so that sub-catchment size was no greater than 20e50 km. The resulting node-link network and sub-catchment maps are shown in Fig. 7. 3.3. Climate Daily rainfall data were obtained from the SILO Data Drill. The SILO climate surfaces have been derived from spatial interpolation of ground-based observation data onto a 5 km  5 km grid on a daily time basis (Jeffrey et al., 2001). The rainfall data was preprocessed to provide a weighted average daily rainfall sequence for each sub-catchment before being imported into E2. This process resulted in each sub-catchment being assigned its own unique rainfall sequence. The spatial representation of rainfall is important, because the response of sediment and nutrient loads to forest fires is highly dependent on subsequent rainfall amount and distribution. Data for areal potential evapotranspiration (PET) were obtained from the Bureau of Meteorology. These data represents the long term mean monthly PET (mm month 1) on a 10  10 km grid resolution. Values of areal PET were converted to average total daily PET for model simulations. Average sub-catchment size was kept relatively small to allow an adequate spatial representation of the rainfall and PET distribution across each catchment. 3.2. Spatial 3.4. Hydrology A 20 m digital elevation model (DEM) of Victoria (DPI and DSE, 2004) was used to delineate upstream sub-catchments associated with water quality monitoring stations (for water quality calibration), gauging stations (for flow calibration) and other points of A ‘bucket-style’ rainfall-runoff model based on Denmead and Shaw (1962) and developed further by Kandel and Argent (2005) was used in the E2 models (Fig. 8). This model has two main 920 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 Fig. 7. Node-link network and sub-catchment map in E2 for the (a) Ovens, (b) Kiewa, (c) Upper Murray and (d) Snowy catchments (Not to scale). parameters; the size of the store (Smax), and the leakage rate (B). Water enters the store via rainfall and is removed from the store via evapotranspiration, surface runoff (saturation-excess) and leakage. Baseflow (or slow flow) on a given day is calculated as a product of the leakage rate (B) and the soil moisture store on that day. Quickflow occurs when the soil moisture store exceeds the soil moisture store capacity (Smax). These become inputs to the downstream area. While other rainfall-runoff models are available in the E2 framework, the additional complexity was not warranted. Potter et al. (2005) suggests that infiltration-excess runoff might be an important process especially for the summer-dominant rainfall catchments, but rainfall intensities are unlikely to be high enough to generate it, even after fire (Lane et al., 2004; Sheridan et al., 2007b). Furthermore, infiltration-excess runoff is more difficult to simulate than the storage-excess runoff considered in the model used here because it depends also on rainfall intensity as well as soil permeability and would require an additional unknown variate in the rainfall model. Evapotranspiration (ET) is estimated from the monthly areal potential and allows for some control on actual evapotranspiration (ETa) due to soil water (S), with the maximum ET (ETmax) defined as a function of vegetation (Fig. 9). The model uses the value of ETa if cell water is at Sa, or ETmax, if soil water is at or above 0.7  Smax (Sb in Fig. 9). In winter, the actual rate tends to be controlled by the potential value whereas in summer, soil water or vegetation controls can limit the rate. Parameters for the rainfall-runoff model were determined by calibration against observed flows at stream gauging stations. Parameters for catchment areas above water storages were obtained by calibrating to the next gauging station above the respective water storage. Releases from water storages were calibrated to observed flows at gauges below the lake outlet. Flow was calibrated over a 20 year period between 1980 and 1999 (inclusive). This allowed the optimum calibration for the available data. This period was chosen to contain series of relatively dry and wet years, and as such represented the variation in rainfall observed over the region. The model was run for one year prior to 921 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 categories were included to represent the ‘burnt’ portions of the original FUs (i.e. ‘burnt forest’, ‘burnt agriculture’ and ‘burnt other’). The resulting land use layer was used for the post-fire scenario model simulations. 3.6. Constituent generation Fig. 8. Conceptual basis of the bucket model. the calibration period, so that the soil water store was representative of the water storage at the commencement of the calibration period. Calibration was based on the Nash and Sutcliffe (1970) coefficient of efficiency (E) for monthly flow records, and on changes in flow mean, standard deviation and coefficient of variation between observed and predicted values of monthly flow. Emphasis was given to calibrating streamflow to observed monthly and annual flow, so that the predicted average long term streamflow was the similar to the observed long term average flow. 3.5. Land use A digitised land use layer containing polygon features for Victoria was obtained from the Corporate Geospatial Data Library (DPI and DSE, 2004). A digitised land use layer containing polygon features for the NSW portion of the Snowy catchment (Bordas and Lesslie, 2002) was obtained from the Bureau of Rural Sciences, Canberra. The files were converted to gridded datasets using standard GIS processing software (ESRI/ArcInfo). The lack of available data to assign constituent generation rates led us to aggregate land use categories into three major categories, or functional units (FUs): ‘forest’ (native forest, remnant vegetation, hardwood and softwood plantations), ‘agriculture’ (pasture, cropping and horticulture) and ‘other’ (mainly townships and other human habitation). This land use layer was used to represent the pre-fire scenario simulations in E2. A digitised layer of the fire-affected areas of Victoria and New South Wales was obtained from DPI and DSE (2004). This layer was overlain onto the original land use layer. New land use or FU Sheridan et al. (2007a) applied a range of load estimation methods were applied to the water quality dataset, including interpolation, regression, averaging and Monte Carlo simulation as well as several data stratifications including base flow separation, and separation of the hydrograph into rising limb and falling limbs. The most valid method for the data was the averaging method using the arithmetic mean of the concentration data, stratified into base flow and event flow concentrations. An event mean concentration (EMC) and dry weather concentration (DWC) component model was used for each FU within each catchment, with user-specified values (EMCg and DWCg) defined for each of total suspended solids (TSS), nitrogen (N) and phosphorus (P). The simulated values for EMC and DWC are a scaled derivation of the DWCg and EMCg (input) data, where flow is scaled by concentration to give output load. Calibration of these parameters is described in the following section. Constituent generation is spatially variable, and in the first year in many cases can be attributed to channel erosion and incision of drainage lines in response to flash flooding. Erosion is not uniformly distributed across the landscape, and calibrated values for DWCg and EMCg represent average upstream landscape values. 3.7. Parameterisation and calibration The water quality data encompasses values for DWC and EMC appropriate for a particular land use at a particular spatial scale. However, it is not possible to derive DWC and EMC values for individual land uses in mixed land use catchments as they represent a net DWC-EMC from more than one land use. Different land uses are characterised by different generation rates, and so should be represented by different DWC and EMC values, yet data for these are scarce. We applied a weighting based on data from Chiew and Scanlon (2002) and Grayson and Argent (2002) to each land use to reflect these differences in constituent generation. Values for DWCg and EMCg were ‘linked’ together by applying constant multiplication factors to the forest generation concentrations using relative differences calculated from data from landscapes in south-eastern Queensland by Chiew and Scanlon (2002) and in similar Victorian landscapes by Grayson and Argent (2002) (Table 3). An implicit assumption in this approach is that the relative differences in generation rates from the broad classes of forest and agriculture land uses are similar across different regions. Given that relative differences in data from the two land uses in both Table 3 Multiplication factors used to calculate constituent concentrations for functional units. Factors were derived from data by Chiew and Scanlon (2002) and Grayson and Argent (2002) and are presented relative to the functional unit for forest DWC (assigned a value of 1 here) for each constituent. Functional Unit TSS DWCga EMCgb DWCga EMCgb DWCga EMCgb Forest Agriculture Other (urban) 1.0 1.4 1.0 4.6 5.3 4.1 1.0 2.3 3.7 3.3 4.1 2.8 1.0 1.4 3.0 1.6 4.2 2.0 a Fig. 9. Conceptual approach to representing evapotranspiration. b TP Dry weather concentration. Event mean concentration. TN 922 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 studies were similar, we feel this assumption is justified in this application. Water quality monitoring stations with pre- and post-fire sediment and nutrient loads calculated by Sheridan et al. (2007a) were used to calibrate the generation of constituents in the models. Pairs of values for DWCg and EMCg for each functional unit were increased and/or decreased iteratively until matches with observed pre-fire loads for TSS, TN and TP were reached. Once this had been achieved, a new digitised land use layer incorporating burnt areas of forest, agriculture and other, was imported into the model to create a ‘burnt’ scenario. Values of DWCg and EMCg for each functional unit were adjusted iteratively until the observed post-fire loads for TSS, TN and TP (as calculated by Sheridan et al., 2007a) were obtained. Pre- and post-fire loads of TSS, TN and TP predicted by the model at points of interest (catchment outlets and water storages) were then compared in terms of the relative changes (rather than absolute changes) in loads. 3.8. Comparison of modelled DWC and EMC values with observed values There is a significant lack of data available to test whether values for DWC and EMC for the different functional units (land uses). However, we were able to examine whether values for DWC and EMC (as parameterised by DWCg and EMCg) generated by the model were similar to those calculated from a limited set of measured water quality data. The only data available for this test was that from the monitoring station at Kiewa River West Branch at upstream of Takeoff. This sub-catchment (101 km2) is in the upper reaches of the Kiewa catchment and is unique because it is completely forested, and nearly all of it (95%) was mapped as being fire-affected (Fig. 2b). There was a variation in fire intensity (Table 2) with significant areas of low intensity and as such it is characteristic of a mosaic of fire severity often observed in eucalypt forests. The water quality data were sufficient in quality and quantity to allow estimates for TSS of DWC and EMC to be calculated for the pre-fire conditions, and for values of EMC only to be estimated for 2003 (one year post-fire) and 2004 (two years postfire) (Table 4). The value for DWC predicted from model runs for forest pre-fire for all catchments ranged between 0.4 and 5.0 mg l 1 across the four catchments and was very similar to the mean value of 2.3 mg l 1 calculated from water quality data in the Kiewa River West Branch (Table 4). Values for EMC used in pre-fire model simulations ranged from 2 to 23 mg l 1 across the catchments, while that calculated for the Kiewa River West Branch was 2.8 mg l 1. While the model values are higher than that observed, there is an acceptable level of agreement given the variable nature of the catchments modelled and the inherent variability in fire and recovery processes (Sheridan et al., 2007a). The calibrated model value for burnt forest EMC TSS in the Kiewa catchment one year after the fire was 38 mg l 1, and compared reasonably well to the value of 108 mg l 1 calculated Table 4 Values of TSS DWC and EMC (mg l 1) for forest measured at the Kiewa River West Branch and values predicted for the modelled catchments. Catchment Kiewa River West Ovens Kiewa Upper Murray Snowy Pre-fire 1 year post-fire 2 years post-fire from water quality data at the Kiewa River West Branch (Table 4). More generally, model values for burnt forest (post-fire) TSS EMC in 2003 for all catchments ranged between 38 and 3264 mg l 1. The large range can be explained in part by the higher variability, and greater uncertainty, of post-fire loads used to calibrate the models. Model values for burnt forest (post-fire) TSS EMC two years after the fires ranged between 146 and 1225 mg l 1, as compared to a calculated value of 92 mg l 1for the Kiewa River. Note that there is no TSS DWC data for the Kiewa catchment in 2004 or 2005 for comparison. Given the large variation and uncertainty in the water quality data, model values of DWC and EMC for the forest and burnt forest functional units agreed fairly well with observed values calculated from water quality data. 3.9. Water quality and water storages There is a lack of data to demonstrate the effect of water storages on differences between inflow and outflow water quality. It is therefore difficult to make estimates of post-fire constituent concentrations of storage outflows. We assigned values of constituent mean concentrations of water released from water storages for pre-fire conditions, and a different set of values for post-fire conditions (Table 5) based on limited observed data. The values were first calculated for the Dartmouth Reservoir (Upper Murray catchment), using changes in pre- and post-fire water quality data from a monitoring station downstream of the reservoir (Mitta Mitta River at Colemans) together with post-fire water quality data taken by Alexander (2004) from within the reservoir at the lake wall. Prefire water quality data for Lake Buffalo (Ovens catchment) were scaled using the factor changes for the Dartmouth Reservoir after allowance was made for differences in the proportion of the catchment area burnt (86% for the Dartmouth Reservoir and 35% for Lake Buffalo). 3.10. Variability in constituent loads Values of baseflow and event flow concentrations corresponding to the lower and upper 95% confidence intervals of the water quality data from Sheridan et al. (2007a) were used to estimate the likely minimum and maximum proportional increases in loads at the water quality stations. The likely minimum increase was derived by using the upper 95% confidence interval values of prefire concentrations and the lower 95% confidence interval values of post-fire concentrations (the ‘low’ scenario). Similarly, estimates of the likely maximum increase was derived by relating the lower 95% confidence interval values of pre-fire concentrations to the upper 95% confidence interval values of post-fire concentrations (the ‘high’ scenario). 3.11. Assumptions and limitations The simulations and resulting predictions of load increases carry the following assumptions and limitations.  The model outputs are based on long term flow and rainfall distribution (between 1980 and 1999). Table 5 Mean constituent concentrations (mg l DWC EMC DWC EMC DWC EMC 2.3 2.4 5.0 3.8 0.4 2.8 11 23 17 2.0 e 96 8.2 714 268 108 439 38 3264 1225 e 32 e 128 4.0 92 146 e 583 18 Constituent TSS TN TP 1 ) of water released from water storages. Lake Buffalo Dartmouth Dam Pre-fire Post-fire Pre-fire Post-fire 2.0 0.21 0.015 2.8 0.24 0.016 1.4 0.27 0.017 2.9 0.35 0.020 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928  No allowance has been made for changes in runoff generation and stream flow in response to the fires.  No allowance is made for the storage, deposition or remobilisation of sediments and nutrients within catchments (except for Lake Buffalo and Lake Dartmouth) or within streams, and therefore actual increases in loads will be lower than those predicted.  No consideration is given to differences in fire severity.  The fate of increased sediment and nutrient loads to the water storages in the study area (Lake Buffalo and Lake Dartmouth) is not known. Concentrations of water released from the water storages was set as a constant for pre- and post-fire conditions, representing the average pre-and post-fire concentrations observed at water quality monitoring stations below the water storages.  There are uncertainties associated with calculations of changes in loads from the water quality data used to calibrate the model.  Calculated increases in loads are independent of flow, and are based only on changes in DWC and EMC. Annual flow of the respective post-fire year was also applied to the pre-fire data for the purposes of estimating loads, e.g. annual flow in 2004 was used for both pre-fire and post-fire 2004 load calculations.  Water quality data represents a catchment with several land uses. It is not possible, from the data available, to calculate constituent generation parameters for individual functional units. Therefore, a weighting was applied to the functional units, to represent the relative differences in these generation parameters observed in other studies.  The same parameter defining constituent generation from a particular functional unit (or land use) was assigned across the whole of each catchment (from the calibrated sub-catchment). At smaller scales, these parameters are likely to vary with many factors, including vegetation type, fire severity, soil type and topography. A number of the assumptions listed above will have had systematic effects on the predictions made by the models. While most of the processes could have been included in the E2 modelling framework, there was no data available to either parameterise or calibrate them. The assumptions made in this modelling study are crucial to the magnitude of the predictions made, and required careful thought and justification during the modelling process. Processes not represented in the model if included, may reduce the predicted increases in loads, including the process of storage of sediment and adsorbed nutrients within stream channels in the lower reaches of the stream network. Therefore, observed increases in loads are likely to be less than the values reported for the various scenarios. 4. Simulation results and discussion 4.1. Calibration of streamflow Observed and predicted data for the calibration periods are provided in Fig. 10. Calibration of streamflow was first performed on the highest gauging stations within each respective catchment. Sub-catchments progressively closer to the end of catchment outlet were then calibrated. Results for the flow calibrations for each subcatchment, including monthly summary statistics, are presented in Table 6. The calibration above the Lake Dartmouth resulted in a relatively poor E (coefficient of efficiency) value of 0.38 (Table 6). However, because the lake is highly regulated, flows from the lake in the model were represented by observed flows from the lake 923 between 1980 and 1999, and this led to a relatively higher E value of 0.96 downstream of the lake at the catchment outlet at Tallandoon. The models developed in E2, together with gridded rainfall data from SILO (Jeffrey et al., 2001) led to excellent calibration results (where E > 0.80 and differences between observed and predicted mean streamflow <1.0%) for all catchments and for most subcatchments. 4.2. Potential impact of fire on constituent loads Once models had been calibrated to represent the observed proportional increases in loads at the respective water quality monitoring station, model predictions of loads pre- and post-fire at end of catchment outlets and at the water storages provided estimates of impacts of the fires relative to pre-fire conditions. Sediment and nutrient loads predicted by the model in the ‘burnt’ scenarios were compared to predictions ‘unburnt’ scenario to derive factor changes in post-fire loads. Post-fire changes in loads of TSS, TN and TP one year, two years and three years after the fire for the water quality monitoring sites (used to calibrate the models), water storages, and catchment outlets, together with location information, are presented in Table 7. An important driver of post-fire erosion and sediment delivery in burned catchments is the magnitude, intensity and frequency of post-fire rainfall and associated flow events (Moody and Martin, 2009). A review by Smith et al. (2011) highlights the effect of rainfall patterns on the large variation in post-fire suspended sediment yields, where large sediment yield (560 times pre-fire loads) for the first year after fire can largely be attributed to channel erosion and incision of drainage lines in response to flash floods generated by periods of short duration, high intensity summer rainfall. In south-eastern Australia, a single summer storm accounted for 45% and 47% of the total suspended sediment yield in the first year after fire from two small wet Eucalyptus forest mountain catchments (Lane et al., 2006). While we report relative changes in metric tonnes per hectare on an annual basis, this may be misleading in the sense that erosion is not uniformly distributed across the landscape. We account for this in part by calibrating to measurements of in-stream water quality (rather than to plot based generation rates) and representing the spatial and temporal variation in rainfall. The average change in constituent loads for the large catchments are summarised by relationships between annual predicted constituent loads and streamflow (Fig. 11). Proportional increases in loads at the catchment outlets predicted using E2 were generally smaller than increases observed at the water quality monitoring sites. These differences reflect the proximity of the monitoring stations to the burnt areas, the total percentage of catchment burnt, the amount of rainfall and the location and size of water storages. The simulations indicate that the magnitude of impacts decreases further down the catchment as the distance from the burnt areas increases, and the proportion of the catchment burnt decreases. While we are unable to validate the predicted loads for whole catchments with observed data, the predictions made appear sensible and consistent with expected changes sediment and nutrient loads in relation to the proportion of the catchments affected. Predicted increases of sediment and nutrient loads were greater that those estimated from water quality data measured below the water storages. Post-fire N and P concentration changes in lakes and reservoirs have not been widely reported, with published work largely confined to natural lakes in forest regions of North America (see Smith et al., 2011). However, in north-east Minnesota, Wright (1976) reported an increase in post-fire TP inputs from tributary streams but no increase in TP concentrations in lakes, supporting the observations of our study. 924 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 Fig. 10. Annual observed and predicted streamflow for all catchments. Very high increases in loads were predicted at the Lake Dartmouth within the Upper Murray catchment, with loads entering the lake of 170 (TSS), 20 (TN) and 40 (TP) times higher post-fire (Table 7). These predictions are the highest of the four catchments modelled, and reflect the large increase in loads calculated at the station at Mitta Mitta River at Hinommunjie and the high proportion (80%) of the catchment burnt. Increases of this order are likely to have serious effects on water quality and health of the dam, however, no significant impact of these increased nutrient and sediment loads on water within the lake was detected one year after the fires (Alexander, 2004). Furthermore, measured increases in concentrations of these constituents in water released from the lake were much less (2.1 times for TSS, 1.3 times for TN and 1.2 times for TP) suggesting that mixing and/or deposition of these constituents is occurring and/or the additional sediments and nutrients have not yet made their way to the lake outlet. Consequently, increases in loads predicted at the outlet of the Upper Murray catchment were substantially smaller than those predicted at the entry of the lake. Estimates of proportional increase in loads below the Dartmouth Dam, at Mitta Mitta River at Tallandoon were lower than those made above the dam (Table 7). Analysis of limited pre-and 925 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 Table 6 Calibration results for monthly flows for the 20 year climate period between 1980 and 1999. E is Nash and Sutcliffe (1970) coefficient of efficiency, and D% are percentage difference in mean (Mean), standard deviation (St Dev) and coefficient of variation (Cv). Area (km2) Flow (GL yr 1) D% Mean D% St Dev D% Cv Catchment Gauging station/location Ovens Ovens at Bright Above Lake Buffalo Ovens at Rocky Pt 493 1142 2969 218 443b 1116 0.80 0.77 0.85 0.13 1.06 0.32 10.6 2.75 0.37 10.7 3.85 0.05 Kiewa Kiewa R West Brancha Kiewa at Bandiana 101 1714 122 675 0.57 0.64 1.29 0.13 43.9 7.74 43.2 7.60 Upper Murray Mitta at Hinommunjie Above Lake Dartmouth Mitta at Tallandoon 1525 3568 4774 407 749b 1160 0.37 0.38 0.96 0.04 0.53 0.70 24.5 6.4 6.8 24.4 5.9 7.4 Snowy Snowy at McKillops Snowy at Jarrahmond 613 910 0.84 0.83 0.65 0.12 a b 10,654 13,485 E 1.96 0.71 2.60 0.60 Based on 8 years of streamflow record. Predicted flow from E2 model. post-fire water quality data from a monitoring station below the dam (Table 5) suggested that increases of approximately 2.1 times (TSS), 1.3 times (TN) and 1.2 times (TP) were observed. It is possible that future increases in sediment and nutrient loads may be observed below the dam, as loads from higher in the catchment and already within the dam reach the dam outlet, and if deposited constituents are remobilised within the dam. The Dartmouth Dam acted as a buffer in dampening the increases in loads observed higher in the catchment. Predicted increases in loads at the catchment outlet, and at the Hume Reservoir, were approximately 72 times (TSS), 5.0 times (TN) and 11 times (TP) greater than before the fires. While increases in loads below the Dartmouth Dam were between 25 and 40% lower than those increases above the dam, the increases in themselves are still large, and may have serious consequences on the water quality and health of the Upper Murray catchment in the longer term. Predictions suggest that the Ovens, Kiewa, Upper Murray and Snowy catchments would receive, on average, approximately 30 times more TSS, 5 times more TN, and 8 times the amount of TP one year post-fire. These values are similar to those from previous work in south-eastern Australia by Wilkinson et al. (2009b), where loads of TSS and P after wildfire that burnt 69% of a 629 km2 catchment dominated by Eucalyptus spp. were higher (109e250 times higher for TSS and 4e10 times higher for P) in the year immediately following fire. The Ovens, Kiewa and Upper Murray catchments all drain into the River Murray where issues relating to river health and associated problems of algal blooms are more critical. Model simulations suggest that collectively, increases of 15 times for TSS, 3 times for TN, and 3 times for TP may have occurred across these three inland-flowing catchments in the first year after the fires. The increase in loads declined after the first year, being approximately 3 times greater TSS, and no increase in loads for both TN and TP three years after the fires for all catchments combined. Post-fire increases in sediment (TSS) loads were generally greater than for nutrient (TN and TP) loads, and may take longer to recover to pre-fire levels. The Snowy was the only catchment in which the predicted increase in loads was greater at the catchment outlet, than at the water quality monitoring station higher in the catchment. There were burnt areas below the monitoring station, resulting in a greater proportion of the catchment burnt (above the catchment outlet) and the proximity of the station to the outlet. The increases in loads were very high given that the proportion or catchment burnt was relatively low (29%). This has arisen, in part due to altered flow conditions higher in the catchment, and by the close proximity of the monitoring station (used to calibrate the model) to the burnt areas. The Lakes Eucumbene and Jindabyne within the Snowy catchment form part of the Snowy Mountains Hydro-electric Scheme. The Scheme collects and stores the water that would normally flow east to the coast and diverts it through trans-mountain tunnels and Table 7 Predicted factor changes in loads of TSS, TN and TP after the wildfires. Catchment Gauging station or water storage Area (km2) Mean flow (GL yr 1) MARd (mm yr % Burnt 1 Year post-fire 2 Years post-fire 3 Years post-fire TSS 1 ) TSS TN TP TN TP TSS TN TP 4 3 3 1 3 2 3 1 2 2 2 2 2 2 2 2 1 1 1 2 Factor increase in load Ovens Ovens at Brighta Lake Buffalob Ovens at Rocky Ptb Ovens at Peechelbac 493 1142 2969 6360 218 443e 1116 1750 441 388 376 275 55 35 43 21 Kiewa Kiewa at Bandianaa 1714 675 394 23 1 1 1 1 1 1 1 Upper Murray Mitta at Hinommunjiea Lake Dartmouthb Mitta at Tallandoonb 1525 3568 4774 407 749e 1160 267 210 243 83 86 80 170 170 70 20 20 5 40 40 10 30 30 15 6 6 2 6 6 3 30 35 15 5 5 2 7 8 3 57 68 27 29 43 36 110 140 15 30 30 30 3 5 110 160 3 8 3 3 4 4 4 5 1 2 2 3 1 1 3 3 3 3 2 2 1 1 2 3 1 1 Snowy at McKillopsa 10,654 613 13,485 910 Snowy at Jarrahmondb Total for inland-flowing catchments (Ovens, Kiewa & Upper Murray) Total for all four catchments (Ovens, Kiewa, Upper Murray & Snowy) Snowy a b c d e Water quality monitoring stations. Predicted using spatial E2 model. Extrapolated from E2 prediction. Mean annual runoff. Predicted flow from E2. 25 10 15 15 9 5 5 2 9 4 6 2 8 4 6 6 926 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 Fig. 11. Relationships between annual post-fire constituent loads and streamflow. power stations. The water is then released into the Murray and Murrumbidgee Rivers for irrigation. Records from the Dalgety gauging station (below Jindabyne) show that all aspects of the flow regime have been modified since the commissioning of the Snowy Mountains Scheme (Rose and Bevitt, 2003). Significant reductions have occurred in flow volume, magnitude and frequency of floods, and a complete loss of seasonal flow variability particularly the spring snowmelt. Median annual flows before operation of Lake Jindabyne were approximately 1,100,000 ML, and were reduced to 13,500 ML, or 1.2% of pre-operation flows. The relatively low mean annual flow and relatively low concentrations for the Snowy catchment results in relatively low pre-fire sediment and nutrient loads and it has by far the lowest constituent loads when expressed on an areal basis (T km 2 yr 1). In absolute terms, sediment and nutrient load increases observed and predicted in the Snowy catchment are lower than those in the Upper Murray and Ovens catchments. Therefore, changes in land use within the Snowy catchment that result in increases in constituent generation, may be exaggerated by the low pre-change loads, resulting from low flows due to upstream dam regulation. Changes that occur in the higher rainfall areas within the catchment (such as the bushfires in this case) will also lead to proportionally higher loads generated. considerable variability. To provide some indication of this variability, two additional scenarios were established to represent a minimum likely change in constituent loads (‘low’ scenario), and the maximum likely change in loads (‘high’ scenario). It is highly likely that the proportional increase in post-fire loads lie between the ‘low’ and ‘high’ proportional increases listed in Table 8. The predicted increases in loads in this study are expressed as a function of long term (20 year) streamflow. Variability is also introduced into these predictions from variation in (annual) streamflow experienced during the simulation period. The range in predicted annual constituent loads for 1, 2 and 3 years post-fire is shown in Fig. 11. Observations of the variability of annual streamflow during the 20 year simulation period were compared to the variability in constituent loads. Uncertainty from constituent generation was greater than that from variability in annual flows, and supports general observations that uncertainty in predictions Table 8 Predicted factor changes in loads of TSS, TN and TP one year (in 2003) after the wildfires for low and high impact scenarios. Catchment Gauging station or water storage Low scenario TSS TN TP TSS TN Ovens Ovens at Brighta Lake Buffalob Ovens at Rocky Ptb Ovens at Peechelbac 10 5 8 8 7 3 4 2 4 2 3 1 80 40 50 50 10 7 8 3 20 10 10 4 Kiewa Kiewa at Bandianaa 1 1 1 3 2 1 30 30 10 9 9 3 10 10 4 630 660 270 40 40 10 120 130 40 30 40 9 10 50 70 670 830 50 60 230 320 4.3. Variability of potential fire impacts on constituent loads The variability in sediment and nutrient loads following fire are expected to be large, with generation rates reflecting the interrelated processes of generation, movement, and deposition at a catchments scale. The errors of measurement are generally unspecified and assigning generation rates inherent in the approach described here requires subjective interpretation and assessment (Baginska et al., 2003). Estimates of constituent loads calculated by Sheridan et al. (2007a) from water quality represent means of a dataset with a Upper Murray Mitta at Hinommunjie Lake Dartmouthb Mitta at Tallandoonb Snowy Snowy at McKillopsa Snowy at Jarrahmondb a b c Water quality monitoring stations. Predicted using spatial E2 model. Extrapolated from E2 prediction. High scenario TP P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 of water quality is larger than in water quantity (Croke and Jakeman, 2001). For example, predicted increases in TSS loads at Snowy at Jarrahmond (with average 140 times) ranged from 40 to 830 times in the ‘low’ and ‘high’ scenarios (Table 8), whereas this range was negative 3 to positive 280 times based on flow variability. Similarly, predictions made of load increases in TSS from the Ovens catchment at Ovens at Peechelba (15 times on average) ranged from 8 to 50 times in the ‘low’ and ‘high’ scenarios based on variability in constituent concentrations, compared to a range between 10 and 20 times based on flow variability alone. Our study supports conclusions made by Baginska et al. (2003) that the allocation of land use and corresponding generation rates remains critical in estimating diffuse catchment exports. While available data may be limited, care should be taken when assigning generation rates because they may be atypical or not representative of the catchments. Generation rates assigned to land uses that dominate the catchment (e.g. if the majority of a catchment is fireaffected) will be highly influential in downstream predictions. High generation rates after wildfire are generally associated with relatively small areas (Moody and Martin, 2009). Scaling up to larger catchments remains difficult due to the difficulty in obtaining and validating information on sediment sources, paths, and rates of transport and delivery (Merritt et al., 2003). Merritt et al. (2003) also concede that a complete understanding of the processes and their interactions is unlikely to occur in the near future, and that uncertainties associated with model structures and processes must be considered explicitly. 4.4. Recovery While observed and predicted increases in loads of TSS in the year after the fires in most cases was high (between 15 and 170 times), these increases were lower in the subsequent two years (between 2 and 30 times; Table 7, Fig. 11). In a study on experimental catchments, Lane et al. (2006) reported a 10 fold increase in total sediment loads in the year following wildfire, and 2e4 fold in the second year, and this appeared to be related to surface vegetation cover. Similar trends in reduced rates of erosion were shown by Wallbrink and Croke (2002) in forested areas subjected to timber harvesting and low severity fire. This is characteristic of wet Eucalyptus forests of south-eastern Australia or in drier forest types where fire severity is not extreme. This is in contrast to other forest types where recovery can take much longer (Shakesby et al., 2007). As for TSS, relatively high increase in loads of TN in the year after the fire (2e30 times) declined over the following two years to a point where the increase in loads of TN across all catchments was marginal (1.3 times pre-fire loads for all catchments). Increases in loads of TP in the year after the fires in one case was high (160 times for the Snowy catchment), they were generally in the order of 10 times the pre-fire levels. These increases in loads declined over the following two years where the increase in loads of TP across all catchments was marginal (1.4 times pre-fire loads for all catchments). Lane et al. (2008) reported increases in loads of TN and TP by 5e6 times following wildfire, and that there was a rapid decline in nutrient concentration in the first 6 months. For the Ovens, Kiewa and Upper Murray catchments, which all lead inland towards the Murray River, issues relating to river health and the associated problems of algal blooms are more critical. Model simulations suggest that increases of 15 times for TSS, 2.7 times for TN, and 2.8 times for TP may have occurred across these catchments in the first year after the fires. However, these increases were predicted to be much lower in the following two years, as a result of the recovery of fire-affected catchments after the fires. The predictions suggest that the catchments will recover relatively quickly in relatively few years after the fire, and that relative 927 increases in nutrients (TN and TP) in 2005 (up to three years after the fires) may be in the order of 1.3e1.4 times the pre-fire loads. Post-fire increases in sediment (TSS) loads were generally greater than for nutrient (TN and TP) loads, and may take longer to recover to pre-fire levels. Nonetheless, the simulations show the recovery process to be highly variable between catchments. 5. Conclusions Proportional increases in loads at the catchment outlets modelled using E2 were generally smaller than increases observed at the water quality monitoring sites. These differences reflect the greater proportion of the catchment burnt above the water quality monitoring stations, the presence and location of water storages, and the effect of dilution further down the catchment. Despite the uncertainty surrounding many of the load estimations used to calibrate the models and the assumptions in the modelling process itself, it is clear that the 2003 fires have had a significant impact on water quality in many of the catchments. The variability of impacts is most likely to be a function of the severity of the burn and speed of subsequent vegetation recovery, the intensity and volume of rainfall following the fires, and interaction of these factors. Many of these processes are not captured in this modelling study and this must be remembered when interpreting the results. Although uncertainties associated with the load estimates used to calibrate these models are very high in some cases, the estimated values represent the most likely impacts based on the experimental data set. This study highlights the difficulty in scaling the effects of mosaic changes in the landscape and the land use attributes on hydrologic response. While one of the main limiting factors has been the accuracy of precipitation estimates in many Australian catchments, inclusion of SILO Data Drill data led to good stream flow calibrations, and prediction of water quality effects had a much higher uncertainty than that of water yield. Our study supports a more widely held view that the quality of model predictions largely depends on the data used to support the model, and that improvements in model structures must be undertaken with efforts to improve data quality and monitoring (Merritt et al., 2003). Confidence in predictions of the effect of land use change on water quality attributes can only be achieved if there are reliable historic observations of water quality for the specified change in land cover. This would include better coordination of monitoring activities, to collect data over longer periods and over a broad range of event magnitudes, and to gather water quality information at nested sites within catchments to allow investigation of instream processes and increase understanding of key processes and model parameterisation at different scales (Letcher et al., 2002). Acknowledgements This project was funded by the Victorian Department of Sustainability and Environment as part of the Bushfire Recovery Program. We thank DD Kandel, Jean-Michel Perraud, and Joel Rahman for assistance with model construction and development. Jodie Smith from the Bureau of Rural Sciences kindly provided land use data for the Snowy River Catchment, and Keith Moodie (NRM) provided SILO climate data. We thank the anonymous reviewers for their constructive comments which greatly improved the manuscript. References Alexander, S.J., 2004. Effect of wildfires on receiving waters, Eastern Victoria. Interim Report, Water Studies Centre, Monash University, May 2004, unpublished. 928 P.M. Feikema et al. / Environmental Modelling & Software 26 (2011) 913e928 Argent, R.M., Perraud, J.-M., Rahman, J.M., Grayson, R.B., Podger, G.M., 2009. A new approach to water quality modelling and environmental decision support systems. Environmental Modelling and Software 24, 809e818. Baginska, B., Pritchard, T., Krogh, M., 2003. Roles of land use resolution and unitarea load rates in assessment of diffuse nutrient emissions. Journal of Environmental Management 69, 39e46. Bhuyan, S.J., Koelliker, J.K., Marzen, L.J., Harrington Jr., J.A., 2003. An integrated approach for water quality assessment of a Kansas watershed. Environmental Modelling and Software 18, 473e484. Blake, W.H., Wallbrink, P.J., Doerr, S.H., Shakesby, R.A., Humphreys, G.S., 2006. Magnetic enhancement in wildfire-affected soil and its potential for sediment-source ascription. Earth Surface Processes and Landforms 31, 249e264. Bordas, V., Lesslie, R., November 2002. Land Use Mapping for the Snowy River Catchment in NSW. Bureau of Rural Sciences, Canberra. Brown, J.A.H., 1972. Hydrologic effects of a bushfire in a catchment in South-Eastern New South Wales. Journal of Hydrology 15, 77e96. Chessman, B.C., 1986. Impact of the 1983 wildfires on river water quality in East Gippsland, Victoria. Australian Journal of Marine and Freshwater Research 37, 399e420. Chiew, F.H.S., Scanlon, P.J., February 2002. Estimation of Pollutant Concentrations for EMSS Modeling of The South-east Queensland Region. Technical report 02/2. Cooperative Research Centre for Catchment Hydrology. Croke, B.F.C., Jakeman, A.J., 2001. Predictions in catchment hydrology: an Australian perspective. Marine Freshwater Research 52, 65e79. Denmead, O.T., Shaw, R.H., 1962. Availability of soil water to plants as affected by soil moisture contents and meteorological conditions. Agronomy Journal 54, 385e389. Department of Primary Industries (DPI) and Department of Sustainability and Environment (DSE), 2004. In: O’Brien, M. (Ed.), Corporate Geospatial Data Library Catalogue, 17th ed. DPI/DSE. Department of Sustainability and Environment (DSE), 2003. Victorian Alpine Fire Maps, CD-Rom Grayson R.B., Argent R.M., 2002. A tool for investigating broad scale nutrient and sediment sources from the catchments of the Gippsland Lakes. A Report to the Department of Natural Resources and Environment Nutrient Management Program. Isbell, R.F., 2002. In: The Australian Soil Classification, revised ed. CSIRO Publishing, Melbourne. Jeffrey, S.J., Carter, J.O., Moodie, K.B., Beswick, A.R., 2001. Using spatial interpolation to construct a comprehensive archive of Australian climate data. Environmental Modelling and Software 16, 309e330. Kandel, D.D., Argent, R.M., 12e15 December 2005. Estimating Sediment and nutrient loads in Gippsland Lakes Catchments using E2 modelling framework. Paper presented at MODSIM05: 16th International Congress of the Modelling and Simulation Society of Australia and New Zealand, Melbourne, Australia. Lane, P.N.J., Croke, J.C., Dignan, P., 2004. Runoff generation from logged and burnt convergent hillslopes: rainfall simulation and modelling. Hydrological Processes 18, 879e982. Lane, P.N.J., Sheridan, G.J., Noske, P.J., 2006. Changes in sediment loads and discharge from small mountain catchments following wildfire in south eastern Australia. Journal of Hydrology 331, 495e510. Lane, P.N.J., Sheridan, G.J., Noske, P.J., Sherwin, C.B., 2008. Phosphorus and nitrogen exports from South Eastern Australian forests following wildfire. Journal of Hydrology 361, 186e198. Letcher, R.A., Jakeman, A.J., Calfas, M., 2002. A comparison of catchment water quality models and direct estimation techniques. Environmental Modelling and Software 17, 77e85. Merritt, W.S., Letcher, R.A., Jakeman, A.J., 2003. A review of erosion and sediment transport models. Environmental Modelling and Software 18, 761e799. Moody, J.A., Martin, D.A., 2001. Initial hydrologic and geomorphic response following a wildfire in the Colorado front range. Earth Surface Processes and Landforms 26, 1049e1070. Moody, J.A., Martin, D.A., 2009. Synthesis of sediment yields after wildland fire in different rainfall regimes in the western United States. International Journal of Wildland Fire 18, 96e115. Murray, N., Perraud, J.-M., Podger, G., Argent, R., 2005. E2 Catchment Modelling Software e Component Manual. Cooperative Research Centre for Catchment Hydrology. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models, 1. A discussion of principles. Journal of Hydrology 10, 282e290. Nyman, P., Sheridan, G.J., Smith, H.G., Lane, P.N.J., 2011. Evidence of debris flow occurrence after wildfire in upland catchments of south-east Australia. Geomorphology 125, 383e401. Peel, M.C., 1999. Annual runoff variability in a global context. Ph.D. Thesis, University of Melbourne. Potter, N.J., Zhang, L., Milly, P.C.D., McMahon, T.A., Jakeman, A.J., 2005. Effects of rainfall seasonality and soil moisture capacity on mean annual water balance for Australian catchments. Water Resources Research 41, W06007. doi:10.1029/ 2004WR003697. Prosser, I.P., Williams, L., 1998. Effect of wildfire on runoff and erosion in native Eucalyptus forest. Hydrological Processes 12, 251e264. Prosser, I.P., Rutherford, I.D., Olley, J.M., Young, W.J., Wallbrink, P.J., Moran, C.J., 2001. Large-scale patterns of erosion and sediment transport in river networks, with examples from Australia. Marine and Freshwater Research 52, 81e99. Rose, T.A., Bevitt, R., 2003. Snowy River Benchmarking and Environmental Flow Response Monitoring Project: Summary Progress Report on Available Data from 1999e2001, for Environment Australia. DIPNR, Cooma NSW. Scott, D.F., 1993. The hydrological effects of fire in South African mountain catchments. Journal of Hydrology 150, 409e432. Shakesby, R.A., Doerr, S.H., 2006. Wildfire as a hydrological and geomorphological agent. Earth-Science Reviews 74, 269e307. Shakesby, R.A., Wallbrink, P.J., Doerr, S.H., English, P.M., Chafer, C.J., Humphreys, G.S., Blake, W.H., Tomkins, K.M., 2007. Distinctiveness of wildfire effects on soil erosion in south-east Australian eucalypt forests assessed in a global context. Forest Ecology and Management 238, 347e364. Sheridan, G.J., Lane, P.N.J., Grayson, R.B., Noske, P.J., Feikema, P.M., Sherwin, C., 2007a. Impact of the 2003 Alpine Bushfires on Streamflow e Estimated Changes in Stream Exports of Sediment, Phosphorus and Nitrogen Following The 2003 Bushfires in Eastern Victoria. School of Forest and Ecosystem Science, University of Melbourne, Murray-Darling Basin Commission. MDBC Publication No. 22/08. ISBN 978 1 921257 62 9. Sheridan, G.J., Lane, P.N.J., Noske, P.J., 2007b. Quantification of hillslope runoff and erosion processes before and after wildfire in a wet Eucalyptus forest. Journal of Hydrology 343, 12e28. Sivapalan, M., Viney, N.R., Jeevaraj, C.G., 1996. Water and salt balance modelling to predict the effects of land-use changes in forested catchments. 3. The large catchment model. Hydrologiccal Processes 10, 429e446. Smith, H.G., Sheridan, G., Lane, P.N.J., Nyman, P., Haydon, S., 2011. Wildfire effects on water quality in forest catchments: a review with implications for water supply. Journal of Hydrology 396, 170e192. Wallbrink, P.J., Croke, J., 2002. A combined rainfall simulator and tracer approach to assess the role of best management practices in minimising sediment redistribution and loss in forests after harvesting. Forest Ecology and Management 170, 217e232. Wilkinson, S.N., Prosser, I.P., Rustomji, P., Read, A.M., 2009a. Modelling and testing spatially distributed sediment budgets to relate erosion processes to suspended sediment yields. Environmental Modelling and Software 24, 489e501. Wilkinson, S.N., Wallbrink, P.J., Hancock, G.J., Blake, W.H., Shakesby, R.A., Doerr, S.H., 2009b. Fallout radionuclide tracers identify a switch in sediment sources and transport-limited sediment yield following wildfire in a eucalypt forest. Geomorphology 110, 140e151. Wright, R.F., 1976. The impact of forest fire on the nutrient influxes to small lakes in northeastern Minnesota. Ecology 57, 649e663. Zammit, C., Sivapalana, M., Kelsey, P., Viney, N.R., 2005. Modelling the effects of land-use modifications to control nutrient loads from an agricultural catchment in Western Australia. Ecological Modelling 187, 60e70.