Techniques for Targeting Erosion Control and Riparian Protection in the Goulburn and Broken Catchments, Victoria

Transcription

1 Techniques for Targeting Erosion Control and Riparian Protection in the Goulburn and Broken Catchments, Victoria Report to Land & Water Australia Scott Wilkinson, Amy Jansen, Robyn Watts, Yun Chen, Arthur Read Logo of partner. Delete box if not required. CSIRO Land and Water Client Report August 2005

2 Copyright and Disclaimer 2005 CSIRO To the extent permitted by law, all rights are reserved and no part of this publication covered by copyright may be reproduced or copied in any form or by any means except with the written permission of CSIRO Land and Water. Important Disclaimer: CSIRO advises that the information contained in this publication comprises general statements based on scientific research. The reader is advised and needs to be aware that such information may be incomplete or unable to be used in any specific situation. No reliance or actions must therefore be made on that information without seeking prior expert professional, scientific and technical advice. To the extent permitted by law, CSIRO (including its employees and consultants) excludes all liability to any person for any consequences, including but not limited to all losses, damages, costs, expenses and any other compensation, arising directly or indirectly from using this publication (in part or in whole) and any information or material contained in it.

3 Techniques for Targeting Erosion Control and Riparian Protection in the Goulburn and Broken Catchments, Victoria Scott Wilkinson 1, Amy Jansen 2, Robyn Watts 2, Yun Chen 1, Arthur Read 1 1 CSIRO Land and Water, GPO Box 1666 Canberra ACT Johnstone Centre, School of Science and Technology, Charles Sturt University, LMB 588, Wagga Wagga, NSW 2678 CSIRO Land and Water Client Report August 2005 CSIRO Land and Water Page i

4 Acknowledgements This work was funded by the National Rivers Consortium through Land & Water Australia, as part of the project Catchment Assessment Techniques to help determine priorities in river restoration. The authors thank Ian Prosser and Alistar Robertson for their leadership in the formative stages of the project. We also acknowledge Wayne Tennant (Goulburn-Broken Catchment Management Authority) for providing access to data and helping to guide this study, Albert van Dijk (CSIRO Land and Water) for useful suggestions in reviewing this report, and Heinz Buettikofer (CSIRO Land and Water) for assistance with publishing the report. CSIRO Land and Water Page ii

5 Executive Summary This report presents assessment techniques for setting catchment-scale priorities for improving water quality and riparian condition in the Goulburn and Broken catchments. Separate priorities are provided for erosion control and protection, and for rehabilitation of riparian vegetation. Erosion sources and sediment transport are assessed using the SedNet model, and the condition of existing riparian vegetation is assessed using tree cover data. The SedNet assessment indicates that erosion control should target riverbank and gully erosion as the primary sediment sources. Priorities for spatially targeting erosion control to address three different objectives are demonstrated; the first objective to reduce total suspended sediment supply to the river network, second to reduce suspended sediment export to the Murray River, and the third to reduce suspended sediment supply to reaches identified in the Goulburn-Broken Regional River Health Strategy as having high priority for improving water quality. Priorities for improving riparian condition are based on the principle that it is more efficient to protect riparian areas in good condition before rehabilitating riparian areas in poor condition. The priorities are thus to protect the least degraded sections first, then work from least to most degraded. Spatial priorities for riparian protection and rehabilitation are demonstrated that meet two different objectives; to improve riparian condition across the catchment, and to improve riparian condition in priority reaches for riparian improvement specified in the Goulburn-Broken Regional River Health Strategy. The assessments are tested using independent data. The benefits of the techniques in reducing suspended sediment loads and improving riparian condition are evaluated by estimating the magnitude of improvement from implementing the priorities. Priorities for research to further improve model performance are also recommended. CSIRO Land and Water Page iii

6 Table of Contents Copyright and Disclaimer i Acknowledgements ii Executive Summary iii 1 Introduction 2 2 SedNet assessment of sediment sources and transport Method for constructing SedNet sediment budgets Defining the river network and associated catchments Hillslope erosion Gully erosion Riverbank erosion Hydrology Floodplain deposition Deposition in reservoirs Contribution of suspended sediment to catchment export Summary of model improvements SedNet results Evaluating model capacity for targeting erosion control Evaluation method Results and discussion Conclusions Recommended priorities for further data collection and research Priorities for reducing suspended sediment yield Introduction Scenarios to reduce suspended sediment supply Scenarios to reduce suspended sediment export 26 3 Assessment of riparian condition Background on RARC Extending the RARC to enable assessment of riparian condition at catchment scales Assessment of riparian zone condition across the Goulburn-Broken catchment Defining the riparian zone Classifying the overall riparian condition of each link Catchment-scale priorities for protecting and restoring riparian condition 37 4 Priorities for implementing the Regional River Health Strategy Introduction Water quality Riparian condition 43 5 Conclusions 47 6 References 48 CSIRO Land and Water Page iv

7 Appendix 1: Priorities for reducing suspended sediment supply to priority reaches in the Goulburn-Broken Regional River Health Strategy 50 CSIRO Land and Water Page v

8 BARMAH LAKE NATHALIA Broken River KATAMATITE LOCH GARRY GREEN LAKE LAKE COOPER WARANGA BASIN TATURA MURCHISON SHEPPARTON LAKE MOKOAN BENALLA REEDY LAKE LAKE NAGAMBIE EUROA LAKE NILLAHCOOTIE SEYMOUR Goulburn River MANSFIELD KILMORE YEA ALEXANDRA BUXTON LAKE EILDON EILDON JAMIESON Legend Locations Roads Streams Reservoirs and Lakes Goulburn-Broken Catchment Kilometres Figure 1: The Goulburn-Broken catchment, showing major rivers and reservoirs, towns and roads. CSIRO Land and Water Page 1

9 1 Introduction There is a growing interest in restoring rivers. In many cases the restoration work required to fully restore river systems exceeds the resources available. Consequently, management actions need to be targeted to achieve the greatest environmental benefit. Prerequisites for targeting management actions include assessing spatial variation in condition, and identifying assets and problems (Rutherfurd et al., 2000). Resource allocation decisions by governments and agencies are often made at river basin scales ( km 2 ) and historically there has been a paucity of data on catchment and stream condition on which to base decisions at this scale. Assessing river condition at a whole of catchment scale also provides the potential to link upstream sources with downstream impacts. This report describes the development and application of two catchment-scale assessment techniques in the Goulburn-Broken catchment (Figure 1). The purpose of the techniques is to identify those river reaches in the catchment where rehabilitation actions are most likely to result in significant improvement in river health, and to identify the magnitude of expected improvement. The first technique is the SedNet model (sediment budgets for river networks). SedNet is used to determine erosion problems causing elevated suspended sediment loads, and priorities for reducing these loads. The report updates previous SedNet modelling of waterborne sediment sources in the Goulburn-Broken catchment (DeRose et al., 2003a). The representations of erosion processes in the SedNet model have been improved, and several datasets have also been improved (Section 2). We also demonstrate several ways in which the SedNet results can be used to determine priorities for erosion control, and the predicted response following implementation of those priorities. The second technique is a catchment-scale assessment of riparian condition, which is based on the Rapid Appraisal of Riparian Condition (RARC). The RARC is a site-based assessment of the condition of riparian zones (Jansen et al., 2004) and provides a measure of riparian degradation caused by vegetation clearing and grazing by domestic livestock. Here we use remotely sensed vegetation data to assess riparian condition across the Goulburn-Broken catchment, validating this technique against RARC site assessments (Section 3). This assessment is used to recommend priorities for protecting and restoring riparian vegetation to improve riparian condition. As well as using the assessments to determine priorities for the spatial location of management actions, the report also describes how the assessments can assist planning management actions when priorities are set by another process, the Goulburn-Broken Regional River Health Strategy (GBCMA, 2004). The purpose of this report is to present priorities at a regional scale. There is insufficient detail in the priorities for designing protection and restoration measures at the site-scale; field surveys and established design procedures should also be employed in site-scale design of works. CSIRO Land and Water Page 2

10 2 SedNet assessment of sediment sources and transport The SedNet model was developed for the National Land and Water Resources Audit (NLWRA; Prosser et al., 2001) and is a conceptual process model that identifies the major sources, sinks and loads of sediments. These are used to construct sediment budgets that, in this application, are designed to represent the average condition over the past 30 to 50 years. Because of the limited data available over large regions, the model is relatively simple. 2.1 Method for constructing SedNet sediment budgets This application of SedNet to the Goulburn-Broken catchment uses a similar method to that outlined in Wilkinson et al. (2004a). The important features of the method are outlined below, including the model developments made in this application. In the model the river network is divided into a series of links, which are the basic unit of calculation for the sediment budgets. A link is the stretch of river between adjacent stream junctions. Each link has an internal sub-catchment, which is the catchment area added to the link between its upper and lower nodes (Figure 2). Sediment inputs to each link come from local hillslope erosion, gully erosion and riverbank erosion, and from upstream tributaries (Figure 3). Sediment is then either deposited within the link or is transported downstream. This calculation is carried out in each river link working from the top of the catchment to the bottom, so that by the final link the mean annual export is calculated. An advantage of this spatial budget approach to estimating sediment exports is that the exported sediment can be tracked back upstream to identify its origins. Separate budgets of bedload and suspended sediment load are constructed for each link. Gully and riverbank erosion supply 60% of their sediment to the bedload budget, and 40% to the suspended load budget, with the proportions chosen so that model outputs reflected the observed balance between suspended and bedload sediment fluxes (DeRose et al., 2003b). Hillslope erosion only supplies sediment to the suspended load budget. This report focuses on suspended sediment budgets, due to the focus of the Goulburn-Broken Catchment Regional River Health Strategy on improving water quality in the Goulburn-Broken catchment (GBCMA, 2004) Figure 2: Link node network as used in SedNet, showing link Shreve order (Shreve, 1966), and sub-catchments for first and fourth order links. CSIRO Land and Water Page 3

11 Hillslope erosion (t/y) Tributary supply (t/y) HSDR Riverbank erosion (t/y) Gully erosion (t/y) Floodplain area Downstream yield (t/y) Figure 3: Conceptual diagram of the SedNet suspended sediment budget for one river link. HSDR is hillslope sediment delivery ratio Defining the river network and associated catchments The river network and associated internal catchment areas were defined from a 20m Digital Elevation Model (DEM). Each first-order link in the river network begins at a point with a catchment area of 20 km 2. This area threshold was selected to limit the number of links across the assessment area, while providing a good representation of the channel network. First order links shorter than 1km long are removed to simplify the stream network Hillslope erosion The input from hillslope erosion is estimated using the Revised Soil Loss Equation (RUSLE; Renard et al., 1997). The RUSLE is an empirical model relating hillslope characteristics of soil erodability, rainfall erosivity, slope and vegetation cover to observed erosion rates. The same method used in the National Land and Water Resources Audit was applied here (NLWRA; Lu et al., 2001), apart from cover factor. Landuse data from Department of Natural Resources and Environment (DNRE) were converted to grid format at a resolution of 20 m and incorporated in the calculation of cover factor. The cover factor was computed at the same resolution of the landuse grid, accounting for differences in cover factor between landuses and using satellite observations of vegetation greenness to indicate variation in the cover across the catchment for each landuse. A detailed description of this is contained in Wilkinson et al. (2005). Very little soil eroded from the hillslope is delivered to the stream; most of the sediment is trapped within the hillslope. The extent of this trapping is represented by the hillslope sediment delivery ratio (HSDR). A HSDR of 5% was applied in this study; observed yields from catchments dominated by hillslope erosion indicate that this is an appropriate HSDR for southern Australia (Prosser et al., 2001) Gully erosion Gully erosion is defined as incision and erosion of drainage lines since European settlement. It is assumed in the model that gully erosion represents incised, eroding drainage lines that are not represented on the modelled stream network due to their small catchment areas. CSIRO Land and Water Page 4

12 This means that all modelled channel erosion occurs either along the defined stream network, or along mapped gullies. Sediment supply from gullies is calculated as the product of gully length, average cross-sectional area (10 m 2 ) and average dry bulk soil density (1.5 t/m 3 ), divided by the time over which gullies have developed (100 years). Prosser et al., (2001) describes the basis for these values. This algorithm provides an average sediment supply over the life of the gullies. The spatial pattern of gullies in the Goulburn-Broken catchment was derived directly from a gully map of Victoria produced by Lindsay Milton and others in the 1960 s (Ford et al., 1993). The proportion of gully and riverbank sediment contributing to the suspended sediment budget was set at 0.4 (DeRose et al., 2003b). The rate of supply of suspended sediment from gullies was reduced to 0.5 times the long-term average, to represent decay in gully activity following natural gully recovery, based on recent measurements (Caitcheon, 2004) Riverbank erosion The mean annual bank erosion rate (BE x, m/y) along a river link x is predicted as a function of stream power, and the proportion of the link length having erodable soil along the banks: BE = ρgQ x bf S x E x where ρ is the density of water (kg/m 3 ), g is the acceleration due to gravity (m/s 2 ), Q bf is bankfull discharge (m 3 /s), S x is the river bed slope, E x is the proportion of the river link with erodable soil. River bed slope is measured by determining link length from the stream network and node elevations by interpolation between contour lines. This method was found to avoid errors associated with noise in DEM elevations along streamlines. The coefficient ( ) was calibrated to produce bank erosion rates that match observed reach average rates of 0.3 m/y for the Goulburn River upstream of Seymour and 0.1 m/y downstream of Seymour (Erskine et al., 2004). Bankfull discharge is assumed to be equivalent to a 2.5 year recurrence interval discharge on the annual maximum series (Wilkinson et al., 2004a). Erodable soil in the riparian zone is identified using the Multi Resolution Valley Bottom Flatness (MRVBF) terrain analysis technique, which predicts the presence of deposition soil at the same resolution as the DEM (Gallant and Dowling, 2003). A minimum cell value of 2.5 was used to represent depositional soil; Gallant and Dowling (2003) recommend 1.5 as an appropriate threshold but there were minimal areas with MRVBF between 1.5 and 2.5. The absence of depositional soils (MRVBF < 2.5) is assumed to indicate rocky terrain, and the link-average bank erosion rate is calculated assuming zero bank erosion in those areas. Clearing of riparian vegetation makes the channel banks more susceptible to mass failure (Abernethy and Rutherfurd, 2000), fluvial scour (Hickin, 1984; Prosser et al., 2000), or catastrophic erosion during floods (Brooks and Brierley, 1997; Brooks, 1999), although ascertaining the relative rates of bank erosion with and without vegetation is difficult. Within the proportion of the link having erodable soils, we apply the full bank erosion rate where the riparian zone has no riparian vegetation, and reduce the bank erosion rate by 95% in areas with intact riparian vegetation. Riparian vegetation tree cover is determined using the Victorian TREEDENS25 dataset; this data is superior to national vegetation datasets. The polygon dataset was converted to a grid of resolution 25 m, with 1 for all densities of tree cover and 0 for areas with no tree cover. The riparian zone is defined as a 40m strip either side of the channel margins. Both the grid resolution and riparian buffer width were selected to ensure appropriate clipping of the vegetation data given the registration uncertainty between the DEM-derived stream network and the TREEDEN25 data. The channel margins are defined by buffering the DEM-derived stream network by half the channel width, where width is predicted using a relationship with catchment area. The location of the DEM-derived network can deviate considerably from its true topographic position, particularly in flatter floodplain reaches. Where the riparian zone CSIRO Land and Water Page 5

13 contains a narrow strip of trees, the error in position is likely to result in under-estimation of the extent of riparian vegetation. To summarise, the riparian zone for each link is divided cell-by-cell into three proportions, with bank erosion being calculated differently in each: Erodable soils and non-vegetated; full bank erosion rate applied Erodable soils and vegetated; 5% of full bank erosion rate applied Non-erodable soils; zero bank erosion applied This approach is an improvement on previous studies, where the effects of erodable soil and riparian vegetation were both applied as reduction factors when determining the link-average bank erosion rate. The new approach recognises that there is often a spatial correlation between depositional soils and vegetation clearance for agriculture, with less erodable steeper slopes often remaining vegetated (see Figure 4). The spatial correlation can considerably affect the link-average bank erosion rate for links with erodable soil along only part of their length. For a link with half its length in non-erodable rocky terrain, and half in depositional, erodable soils; if riparian vegetation has been removed in the half having erodable soils, the previous method would calculate the active proportion of the link as 0.5 * 0.5 = The new method calculates the active proportion as 0.5; recognising that all of the erodable part of the link is non-vegetated. CSIRO Land and Water Page 6

14 Yellow denotes nonvegetated areas of the riparian zone with erodable soil Dark areas of the riparian zone have non-erodable soil, where the channel impinges on the bedrock valley margin Light green areas of the riparian zone have erodable soil, but are vegetated Figure 4: Erodable soil and vegetation in the riparian zone, illustrated on a threedimensional visualisation of the Digital Elevation Model (with vertical exaggeration) The potential bank erosion rate, also termed the bank erosion hazard (m/y), is defined as the erosion rate that occurs in the absence of riparian vegetation (tree cover). This measure is useful for prioritising riparian revegetation because it identifies the erosion rate predicted to be occurring in the area of each link having erodable soil, and degraded riparian vegetation. This measure also predicts the bank erosion rate that would occur if existing riparian vegetation is removed, and is therefore useful for prioritising protection of existing riparian vegetation. For sediment budgets, the erosion rate is expressed in units of tonnes per year of sediment eroded along the length of each river link. This is determined as the bank erosion rate multiplied by the length of the river link, the height of the bank (based on a catchment area regionalisation), and the dry bulk density of sediment (1.5 t/m 3 ). The sediment from bank erosion is apportioned 60% to the bedload budget and 40% to the suspended sediment budget (cf Section 2.1). CSIRO Land and Water Page 7

15 2.1.5 Hydrology Mean-annual runoff from each sub-catchment is predicted using a physically-constrained regionalisation relationship between observed runoff coefficient and the ratio of mean annual potential evapotranspiration to rainfall. Derivation of the regionalisation relationship is described in Wilkinson et al. (In Press), and follows the method of Zhang et al. (2004). Observed runoff from 27 gauged catchments throughout the Goulburn-Broken was used in the regionalisation. The estimated runoff volume from each sub-catchment is accumulated through the river network to determine mean annual runoff at any point in the network. Mean annual runoff is used in the prediction of reservoir deposition. As well as mean annual runoff, three characteristic values of daily flow are regionalised from the mean annual runoff regionalisation. A measure of daily flow variability is regionalised to predict bedload transport; daily flow of recurrence interval 2.5 years on the annual maximum series is regionalised as a representation of bankfull flow for predicting bank erosion; and the median value of daily overbank flows is regionalised to predict floodplain deposition (Wilkinson et al., 2004a; Wilkinson et al,. In press). For links downstream of reservoirs, the regionalisations are modified to reflect contemporary changes in flow volume and variability Floodplain deposition Deposition of suspended sediment becomes significant when flows spread onto floodplains, or enter reservoirs, because flow velocity is greatly reduced in these environments. The equation used to estimate floodplain deposition is given in Prosser et al. (2001). A new method for defining floodplain area has been applied in the equation for the present application. In the NLWRA floodplains were defined by hydraulic modelling of the 250 m DEM, and significantly over predicted floodplain extent, particularly in upland streams. The method developed for the present application uses MRVBF (Gallant and Dowling, 2003). By identifying the flattest valley bottoms at a range of scales it is possible to define areas of floodplain deposition. Over the long term, the amount of fine sediment in transient storage in river channels will not change, and can be neglected in a mean-annual sediment budget Deposition in reservoirs Sediment deposition in reservoirs is a function of an empirical rule based upon the mean annual inflow into the reservoir and its total storage capacity (Heinemann, 1981). Reservoir capacities and construction dates are given in Table 1. The suspended sediment budgets are valid for the period since reservoir construction. Table 1: Reservoir construction dates and capacity Reservoir Construction date Capacity (gigalitres) Lake Eildon Waranga Basin Lake Mokoan Lake Nillahcootie Lake Cooper, Green Lake & Horse Shoe Lake Goulburn Weir & Lake Nagambie CSIRO Land and Water Page 8

16 2.1.8 Contribution of suspended sediment to catchment export Because of losses to floodplain and reservoir deposition, not all suspended sediment delivered to a river network is exported from the catchment outlet. If the objective is to reduce sediment export, we want to target erosion control to areas which supply sediment to the river network that is actually transported to the catchment outlet (Prosser et al., 2001). Therefore, we calculate the proportion of the suspended sediment supplied to each link and sub-catchment that reaches the catchment outlet. This represents a sediment delivery ratio for the river network (RSDR) (Equation 1), where n is the number of links between link x and the catchment outlet, Sup x = H x + Gx + Bx + Tx, the total suspended sediment supplied to a link from local hillslope ( H x ), gully ( G x ) and bank ( B x ) erosion and upstream tributaries ( T x ), and Y x is sediment yield from the downstream end of a link (all in units t/y). The contribution of a link/sub-catchment pair to suspended sediment export ( Cont x ; t ha -1 y - 1 ), is then given by Equation 2, where A x is the area of the sub-catchment. Y Y +... x x 1 RSDR x = Supx Supx+ 1 Cont H + G + B Yn Sup n Equation 1 x x x x = RSDRx Equation 2 Ax The contribution of each erosion process is determined by including only that process in Equation 2. A consequence of Equation 1 is that, all other factors being equal, the further a sub-catchment is from the catchment outlet(s), the lower the probability of sediment reaching the outlet. This behaviour is modified, however, by differences between links in erosion and deposition rates Summary of model improvements This report describes the third application of SedNet to the Goulburn-Broken catchment. The first application was as part of the National Land and Water Resources Audit (NLWRA; Prosser et al., 2001). The second application applied similar model algorithms, but used regional-scale datasets (De Rose et al., 2003), and this study for the NRC (National Rivers Consotrium) has added further improved algorithms and datasets (Table 2). These improvements are discussed in the evaluation of SedNet results (Section 2.3). CSIRO Land and Water Page 9

17 Table 2: Improvements in data and algorithms in Goulburn-Broken SedNet studies Datasets: Riparian vegetation Landuse Floodplain extent Algorithms: Bank erosion equation Channel slope Bank soil erodability Gully erosion rate Proportion of suspended sediment from bank and gully erosion Resolution of hillslope erosion cover factor algorithm Hydrology NLWRA NLWRA regional study This study 250 m 2 resolution BRS data (Barson et al., 2000) 1 km 2 resolution BRS data 250 m resolution hydraulic modelling Bankfull discharge rule Average value for DEM cells along river link Not considered Average since European settlement Interpolation of ISC (Ladson, 1999) surveys TREEDEN25 (20 m resolution) 20 m resolution landuse Updated 20 m resolution landuse FLOODWAY25 Bankfull stream power rule Average value for DEM cells along river link Erodable proportion of link length related to floodplain width; spatial correlation with riparian vegetation not considered Average since European settlement km 1 km 20 m Regionalisation against catchment area and rainfall Regionalisation against catchment area and rainfall, accounting for flow regulation Terrain analysis (MRVBF) Bankfull stream power rule, calibrated to observed rates Link rise/run, using contour line elevations and link length Terrain analysis (MRVBF); combined cell-by-cell with riparian vegetation Reduced 50% from the average since European settlement to reflect declining gully activity Physically constrained catchment water balance, accounting for flow regulation CSIRO Land and Water Page 10

18 2.2 SedNet results The stream network defined as described in Section in the Goulburn-Broken catchment has a total length of 4,768 km. The network comprises 826 separate river links, with and average length of 5.8 km. The total catchment area is 22,872 km 2, with each sub-catchment having an average area of 28 km 2. When each term in the bedload and suspended sediment budgets is totalled across the whole river network, the proportions of each source and loss term indicate the dominance of riverbank and gully erosion as sediment sources (Table 3). These two sediment sources can together be considered as channel erosion, with gullies representing small, incised streams that drain to the model river network. Table 3: Sediment sources and losses in the SedNet budget, totalled across the Goulburn-Broken catchment Sediment inputs kt/y Sediment outputs kt/y Suspended sediment from hillslopes 50 Floodplain suspended deposition 150 Suspended sediment from gullies 100 Channel bedload deposition 450 Bedload sediment from gullies 300 Reservoir suspended deposition 80 Suspended sediment from riverbanks 150 Reservoir bedload deposition 70 Bedload sediment from riverbanks 220 Export suspended sediment 70 Export bedload sediment 0 Total inputs 820 Total output 820 Bank erosion supplies 45% of the total sediment and 48% of the suspended sediment to the river network. A total of 3227 km of river network length (68% of the total network length) has degraded riparian zone (no tree cover). The low levels of riparian vegetation (Figure 5) are a major factor in the high rates of predicted bank erosion (Figure 6), particularly in steep, nonvegetated foothill areas. Gully erosion supplies 49% of the total sediment, and 33% of the suspended sediment that is delivered to the river network. The spatial pattern of gully density (Figure 7) represents the extent of gully erosion, triggered largely by degradation or removal of vegetation from convergent zones along valley floors in grazing and agricultural areas. Of the entire catchment area, 46% has an average gully density exceeding 0.1 km/km 2 ; within this, 36% has medium density ( km/km 2 ) and the other 10% has high density (> 0.51 km/km 2 ). Hillslope erosion is relatively low in the majority of the catchment (soil loss rate < 0.5 t ha -1 y - 1 ) (Figure 8). The areas with a soil loss rate > 0.5 t ha -1 y -1 relate primarily to areas of high slope on agricultural or grazing lands. It is predicted that about 1100 kt of soil is moved annually on hillslope. This equates to an average soil erosion rate of 0.48 t ha -1 y -1. The values of hillslope erosion represent local movement of soil on hillslope. Considering the 5% HSDR, the total annual input of sediment from sheet and rill erosion to streams is estimated to be about 50 kt, which is only about 18% of suspended sediment delivered to the river network. CSIRO Land and Water Page 11

19 Bedload deposition in the channel network is the dominant sediment sink in the catchment. Reservoirs trap approximately 18% of total sediment supplied to the river network, and floodplain deposition of suspended sediment (19% of total supply) are also important sediment sinks. As a result of deposition, the amount of suspended sediment exported from the catchment outlet is less than 25% of the suspended sediment supplied to the river network. Suspended sediment load in the river network is highest along the Goulburn and Broken Rivers (Figure 9). CSIRO Land and Water Page 12

20 NATHALIA KATAMATITE TATURA MURCHISON SHEPPARTON BENALLA EUROA SEYMOUR MANSFIELD KILMORE YEA ALEXANDRA EILDON BUXTON JAMIESON Riparian Vegetation Proportion Kilometres Figure 5: Proportion of erodable riparian zone in each link having tree cover used in SedNet CSIRO Land and Water Page 13

21 NATHALIA KATAMATITE TATURA MURCHISON SHEPPARTON BENALLA EUROA SEYMOUR MANSFIELD KILMORE YEA ALEXANDRA EILDON BUXTON JAMIESON Bank Erosion (m/y) Kilometres Figure 6: Predicted mean annual bank erosion rate CSIRO Land and Water Page 14

22 NATHALIA KATAMATITE TATURA MURCHISON SHEPPARTON BENALLA EUROA SEYMOUR MANSFIELD KILMORE YEA ALEXANDRA EILDON BUXTON JAMIESON Gully Density (km/km2) Kilometres Figure 7: Gully density calculated from gully mapping. CSIRO Land and Water Page 15

23 NATHALIA KATAMATITE TATURA MURCHISON SHEPPARTON BENALLA EUROA SEYMOUR YEA ALEXANDRA EILDON KILMORE BUXTON MANSFIELD JAMIESON Hillslope Erosion (t/ha/y) > Kilometres Figure 8: Mean annual hillslope erosion rate used in SedNet CSIRO Land and Water Page 16

24 NATHALIA KATAMATITE TATURA MURCHISON SHEPPARTON BENALLA EUROA KILMORE SEYMOUR YEA BUXTON ALEXANDRA EILDON MANSFIELD JAMIESON Suspended Sediment Load (kt/y) obs_gauges Kilometres Figure 9: Predicted mean annual suspended sediment load. The stream gauges with observed water quality data are labelled by identification number. Coloured circles denote gauges where load is under-predicted by more than 20% (red) and overpredicted by more than 20% (green); discussed in Section 2.3. CSIRO Land and Water Page 17

25 2.3 Evaluating model capacity for targeting erosion control The preceding sections concentrate on SedNet modelled sediment budgets, with little discussion of their accuracy. At the scale of individual links, there are considerable uncertainties in each of the budget terms. The sources of uncertainty include assumptions in the erosion process algorithms, and uncertainties in the data used to parameterise the algorithms. Here we assess the model uncertainties and determine whether they are small enough that the model results are useful for targeting erosion control Evaluation method Model appropriateness for targeting erosion control to reduce suspended sediment yields can be determined by the capacity of the model to predict spatial variations in suspended sediment yield. This is because variations in yield are closely linked to variations in upstream sediment supply, and because reductions in yield are often what drive erosion control. We evaluate model capacity to predict spatial variations in suspended sediment yield by comparing predicted yields against yields estimated from stream flow and suspended solids concentration measurements using concentration rating curves. Because SedNet sediment budgets are constructed independently of the estimated yields this is an independent test of the model results. Previous SedNet studies have found that for catchments larger than 3,000 km 2 the discrepancy between sediment yield predicted by the model and that estimated is generally less than 30%, but for catchments smaller than 3,000 km 2, the relative error in predicted loads can be much larger (DeRose et al., 2004; Prosser et al., 2001). In other words, model uncertainty reduces as catchment area increases. This suggests that spatial averaging of errors occurs and that the sediment budget is reasonably well constrained. It also indicates that largest errors are likely in small sub-sets of a catchment, and so model evaluation should use estimated yields from small catchments. Another advantage of using small catchments is that deposition is generally less important, meaning suspended sediment yield is more closely related to sediment supply, and the evaluation can be considered to be of predicted erosion rates as well as sediment yield. We test the model performance using loads estimated from stream flow and suspended solids concentration measurements for eight small subsets of the total Goulburn-Broken catchment. The estimated yields were determined by DeRose et al. (2003a), using the method of constructing rating curves of suspended solids concentration against flow. The eight catchment gauges are shown in Figure 9, and have an average catchment area of 378 km 2 ; range km 2 (Table 4). Gauge is also included in Figure 9 and Table 4 as a separate evaluation of model performance at predicting end-of-valley sediment yield. Because sediment yield increases with catchment area as well as erosion intensity, model performance is evaluated using specific sediment yield; yield divided by upstream catchment area Results and discussion Table 4 and Figure 10 show the estimated and predicted suspended sediment yields for the eight small catchments. The downstream gauge was not included in the evaluation of model capacity to predict spatial variation in suspended sediment yield, but indicates that the model prediction of end-of-valley sediment yield for the present study is within 10% of the load estimated from monitoring data. CSIRO Land and Water Page 18

26 Table 4: Estimated and predicted suspended sediment yields. Gauge ID Estimated yield (kt/y) Catchment area (km 2 ) Predicted yield (this study) Predicted yield (regional NLWRA study; DeRose et al., 2003a) (DeRose et al., 2003a) Kt/y Relative error (%) Kt/y Relative error (%) Downstream gauge with estimated yield LOG (predicted suspended sediment yield (kt/y)) LOG (observed suspended sediment yield (kt/y)) This study Regional NLWRA study Figure 10: Average annual suspended sediment yields predicted from SedNet compared against estimated loads from eight gauging stations. CSIRO Land and Water Page 19

27 20 Predicted suspended sediment yield (t/km2/y) This study Regional NLWRA study Observed suspended sediment yield (t/km2/y) Figure 11: Average annual specific suspended sediment yield predicted from SedNet compared against estimated specific yields from eight gauging stations Average relative error Median relative error Relative error % NLWRA 1 NLWRA 2 This 3 Continental regional study Figure 12: Average and median relative errors in predicted specific suspended sediment yields, relative to that estimated from rating curves, for eight small catchments in the Goulburn-Broken catchment. CSIRO Land and Water Page 20

28 Figure 11 shows the model performance for specific suspended sediment yield for the two most recent studies. Figure 12 illustrates the average and median error in specific sediment yield relative to the estimated yields for the eight catchments, for the NLWRA continental study as well as the Regional NLWRA study and the present study. The mean error indicates no improvement in model performance for the present study relative to the Regional NLWRA study, while the median error does indicate improvement. The difference in the mean and median errors between this and the previous study occurs because the present study predicts suspended sediment yields closer to the estimated yields than the previous study for 5 of the 8 gauged catchments, but performs poorly for gauge ; the Acheron River at Taggerty, where yield is over-predicted almost 3-fold. The four catchments with model error greater than 20% are in the south of the catchment (see coloured circles in Figure 9). In considering this evaluation of model capacity to predict spatial variations in suspended sediment yield, it is important to remember that the model results represent best estimates of erosion and sediment transport given the catchment data available. The yields estimated from the water quality observations using rating curves also contain uncertainty; of the order of 30% Conclusions Four important conclusions can be made from the model evaluation over the eight small catchments: 1. The evaluation indicates that the model results are useful for targeting erosion control, at least down to the scale of the gauged catchments (average catchment area 378 km 2 ). Specific sediment yield varies by more than 300% over the gauged catchments (Figure 11), and by more than two orders of magnitude over the entire Goulburn-Broken catchment. This is a larger range of variation than the errors observed for the gauged catchments (median error of 20%, mean error of 51%, maximum error of 173%; Table 4). 2. Present model performance indicates that an uncertainty in predicted suspended sediment yield of 50% is typical for catchments around 400 km 2 in area, assuming similar attention to data preparation and model parameterisation as described here. 3. The improvements in model performance over previous SedNet studies in the Goulburn-Broken catchment demonstrate the value of improved regional data for sediment modelling, and can also be attributed to improvements in model algorithms (Figure 12; Table 2). Given the size of spatial variation in bank erosion, it is believed that the improvement between the present and previous study is due mainly to improved riparian vegetation data and representation of soil erodability in the bank erosion equation. The large error of the NLWRA continental SedNet study over the eight small catchments indicates that a continental study is not appropriate for regional-scale targeting of erosion control. However, the NLWRA had the objective to predict continental-scale patterns in sediment loads, and not regional-scale patterns. 4. Over-prediction of sediment yield in the Acheron River is due to over-prediction of bank erosion in that catchment. Preliminary investigation indicates that predicted bank erosion is about double what it should be, in catchments draining the wetter mountain parts of the south-east Goulburn River catchment, but only where erodable soil is present and riparian vegetation is degraded. This model error is not large enough to prevent model results being used for present targeting of erosion control, as discussed later. CSIRO Land and Water Page 21

29 2.3.4 Recommended priorities for further data collection and research Further improvements to SedNet modelling in the Goulburn-Broken catchment would further improve targeting of specific erosion processes and refine the spatial location of erosion hotspots. The modelling can be regarded as an ongoing experiment rather than a final solution; current modelling is an improvement over that of a couple of years ago and further improvements are to be expected in the future. Priority should be placed on bank and gully erosion given the dominance of bank and gully erosion as sediment sources (Table 3), and the less mature science used to predict these processes. In around two years, an updated modelling study could address sources of remaining model uncertainty including hydrology, riparian condition and bank resistance, and also gully density and gully sediment yield. Modelling of bedload sediment could also be incorporated to predict distribution of habitat sedimentation; and modelling of nutrient budgets could assist water quality improvement. Updated monitoring data would better define uncertainty in predicted spatial patterns. CSIRO Land and Water Page 22

30 2.4 Priorities for reducing suspended sediment yield Introduction SedNet provides a means for targeting erosion control to most effectively reduce suspended sediment yield. There are three main considerations when prioritising where in the catchment to do erosion control, and what type of management action is most effective (Wilkinson et al., 2004): 1. Is the objective to reduce suspended sediment or bedload sediment, or both? This will help determine the relative proportion of investment that should be directed to sources of suspended and bedload sediment. 2. What is the allocation of management effort between the erosion processes? If one process dominates, then a proportional reduction in this input may achieve the greatest impact for the least expense. Where more than one erosion process is important, the allocation should be in the same proportions as the proportion of sediment delivered from each erosion process. Separate spatial priorities should be determined for each erosion process, because each process has a different spatial pattern. 3. Is there a sediment control point in the river network, which is the focus for reducing sediment load? This may be the catchment outlet, or the delivery to a reservoir. If there is a control point, then target erosion control to upstream links in descending order of contribution, for each process. If the objective is to reduce sediment loads throughout the river network, target sub-catchments in descending order of specific erosion rate for each process (t/km 2 /y). For bank erosion, prioritise riparian revegetation in descending order of potential bank erosion rate, or the potential bank erosion contribution. This priority is appropriate for both revegetation, and for protection of existing vegetation. Another advantage of using quantitative spatial modelling to assess erosion and sediment movement is that different management options can be simulated, and the outcomes compared. However, given the model uncertainty (Section 2.3.1), site-scale design of rehabilitation should include field assessments, to identify erosion hotspots within each link and account for uncertainties in the erosion predictions for each link. We demonstrate priorities for erosion control across the catchment, using the predicted patterns of erosion and suspended sediment transport. Two objectives of reducing total supply and reducing export of suspended sediment were used (Sections and respectively). In Section 4, we determine priorities to assist implementation of the Regional River Health Strategy, which identifies specific reaches in which suspended sediment loads are to be reduced Scenarios to reduce suspended sediment supply In the Goulburn-Broken catchment, SedNet identifies bank and gully erosion as roughly equally important sources of suspended sediment on a whole of catchment basis. Reducing both bank and gully erosion will therefore provide the greatest reduction in sediment supply, and suspended sediment export from the catchment as well. We define separate spatial CSIRO Land and Water Page 23

31 priorities for bank and gully erosion, to achieve the largest reduction in both sediment supply and suspended sediment export from each of these erosion processes. Using reduction in suspended sediment supply as the objective for erosion control leads to targeting erosion hotspots (t/ha/y), since a percentage reduction in these will have the largest impact on total sediment supply (t/y). This objective implies that reductions in sediment load anywhere have equal value. When modelling the predicted sediment supply, we assume complete treatment of riverbanks and gullies in priority areas. In reality, practical limitations on access and within-link variability in erosion rates may mean that complete treatment is impossible or unnecessary. Revegetating a proportion of the non-vegetated length of each link, and a proportional reduction in sediment supply from gully erosion in a treated sub-catchment may be a more realistic scenario. Bank erosion priorities Priorities for protecting existing riparian vegetation, and revegetating degraded riparian vegetation to reduce bank erosion, were set in descending order of potential bank erosion rate; the rate which would occur everywhere if vegetation were removed, and which currently occurs on the non-vegetated portion of each link. Reserves could be excluded from the priorities, since river channels in reserves will generally not require protection or revegetation. However, given the small amount of reserves in the Goulburn-Broken catchment (total riparian length 260 km), we do not exclude them from the simulated scenarios. Figure 13 shows the spatial priorities of potential bank erosion rate. Highest priority areas are those with highest stream power. The effect of existing riparian vegetation in mitigating bank erosion can be determined by comparing Figure 13 with Figure 6, which shows present linkaveraged bank erosion rate. Figure 14 illustrates the response of suspended sediment supply to targeting bank erosion control according to Figure 13. The brown links in Figure 13 containing a total length of non-vegetated, erodable soils, of 250 km, as shown in Figure 14. The non-linear shape of the curve is the result of targeting highly eroding links first. In contrast, the response to riparian revegetation implemented randomly would be approximately linear (dashed line). Once 1000 km of revegetation has been implemented (right hand side of graph), there is little further reduction predicted in sediment supply from bank erosion (near-zero gradient of the response curve). The total sediment supply then remaining is the amount supplied from gully and hillslope erosion. Accounting for the over-prediction of bank erosion in the wetter, south-east of the Goulburn catchment, in links having erodable soil and poor riparian vegetation (Section 2.3.3) would result in reclassification of around half of the dark brown tributaries in that area to the light brown class, the rest remaining dark brown (Figure 13). In other words, the area remains a hotspot of potential bank erosion, but to a less extreme level than currently indicated. CSIRO Land and Water Page 24