Emeritus Professor George Kuczera

Traditionally, reservoir management has been synonymous with the operation of engineering infrastructure systems, with the majority of literature on the topic focusing on strategies that optimize their operation and control. This is despite the fact that reservoirs have major impacts on society and the environment, and the mechanics of how to best manage a reservoir are often overshadowed by both environmental changes and higher-order questions associated with societal values, risk appetite and politics, which are highly uncertain and to which there are no "correct" answers. As a result, reservoirs have attracted more controversy than any other type of water infrastructure. In this paper, we address these often-ignored issues by providing a review of reservoir management through the lens of wickedness, competing objectives and uncertainty. We highlight the challenges associated with reservoir management and identify research efforts required to ensure these systems best serve society and the environment into the future.

The calibration of highly parameterized hydrological models is a major computational challenge, especially for models with long run times. This challenge motivates the reconsideration of gradient-based algorithms often overlooked for their perceived lack of robustness. Our study evaluates two Gauss-Newton algorithms, robust Gauss-Newton (RGN), and Levenberg-Marquardt (PEST), and two stochastic algorithms, Shuffled Complex Evolution (SCE), and Dynamically Dimensioned Search (DDS), on a 38-parameter SWAT model calibration problem. Algorithm performance is comprehensively assessed using trajectory plots from 100 invocations and by analyzing the distribution of estimated optima at fixed budgets of 200, 500, 1,000, 2,000, 3,000, and 5,000 objective function evaluations (model runs). Empirical results indicate that: (a) Gauss-Newton algorithms are more likely than stochastic algorithms to locate good solutions for the budgets considered in this work, and more likely to locate satisfactory solutions when budget is tight (200¿500 model runs) and (b) RGN shows the fastest initial convergence amongst the algorithms under consideration and has the highest chance of finding satisfactory solutions when the budget is tight. The results indicate that Gauss-Newton algorithms offer an attractive choice for the calibration of highly parameterized hydrological models.

Estimates of forest stocking density per hectare (NHa) are important in characterising ecological conditions and assessing changes in forest dynamics after disturbances due to pyrogenic, anthropogenic and biotic factors. We use Unmanned Aircraft Systems (UAS) LiDAR with mean point density of 1485 points m-2 across 39 flight sites to develop a bottom-up approach for individual tree and crown delineation (ITCD). The ITCD algorithm was evaluated across mixed species eucalypt forests (MSEF) using 2790 field measured stem locations across a broad range of dominant eucalypt species with randomly leaning trunks and highly irregular intertwined canopy structure. Two top performing ITCD algorithms in benchmarking studies resulted in poor performance when optimised to our plot data (mean Fscore: 0.61 and 0.62), which emphasises the challenge posed for ITCD in the structurally complex conditions of MSEF. To address this, our novel bottom-up ITCD algorithm uses kernel densities to stratify the vegetation profile and differentiate understorey from the rest of the vegetation. For vegetation above understorey, the ITCD algorithm adopted a novel watershed clustering procedure on point density measures within horizontal slices. A Principal Component Analysis (PCA) procedure was then applied to merge the slice-specific clusters into trunks, branches, and canopy clumps, before a voxel connectivity procedure clustered these biomass segments into overstorey trees. The segmentation process only requires two parameters to be calibrated to site-specific conditions across 39 MSEF sites using a Shuffled Complex Evolution (SCE) optimiser. Across the 39 field sites, the ITCD algorithm had mean Fscore of 0.91, True Positive (TP) trees represented 85% of measured trees and predicted plot-level stocking (NP) averaged 94% of actual stocking (NOb). As a representation of plot-level basal area (BA), TP trees represented 87% of BA, omitted trees represented slightly smaller trees and made up 8% of BA, and a further 5% of BA had commission error. Spatial maps of NHa using 0.5 m grid-cells showed that omitted trees were more prevalent in high density forest stands, and that 63% of grid-cells had a perfect estimate of NHa, whereas a further 31% of the grid-cells overestimate or underestimate one tree within the search window. The parsimonious modelling framework allows for the two calibrated site-specific parameters to be predicted (R2: 0.87 and 0.66) using structural characteristics of vegetation clusters within sites. Using predictions of these two site-specific parameters across all sites results in mean FScore of 0.86 and mean TP of 0.77, under circumstances where no ground observations were required for calibration. This approach generalises the algorithm across new UAS LiDAR data without undertaking time-consuming ground measurements within tall eucalypt forests with complex vegetation structure.

Growing agricultural water demand is dramatically affecting the implementation of, and compliance with, water sharing plans in regions such as the Murray-Darling Basin (MDB). Problems can arise from water theft, poor resourcing or questionable actions from stakeholders. Recent actions from MDB governments have resulted in improved regulation, although more is required in a technical, governance and cultural space to create a comprehensive and transparent management framework. This is pivotal in improving overall trust in water regulators. We discuss an integrated water resource management approach for improved water regulation, involving the implementation of remote sensing technologies to complement metering, coupled with a focus on a stronger compliance culture in a range of stakeholder groups and regulatory changes that allow quicker adoption of unbiased best practice science and technology.

Sub-seasonal streamflow forecasts are important for a range of water resource management applications, with a distinct practical interest in forecasts of high flows (e.g., for managing flood events) and low flows (e.g., for managing environmental flows). Despite this interest, differences in forecast performance for high and low flow events are not routinely investigated. Our study reveals that while forecasts evaluated over the full flow range can appear reliable, stratification into high/low flow ranges highlights significant under/over-estimation of forecast uncertainty, respectively. We overcome this challenge by introducing a flow-dependent (FD) nonparametric component into a post-processing model of hydrological forecasting errors, the Multi-Temporal Hydrological Residual Error (MuTHRE) model, yielding the MuTHRE-FD model. The MuTHRE and MuTHRE-FD models are compared in a case study with 11 Australian catchments, the GR4J rainfall-runoff model and post-processed rainfall forecasts from ACCESS-S. Through its improved treatment of flow-dependence, the MuTHRE-FD model achieves practically significant improvements over the original MuTHRE model in the reliability of forecasted cumulative volumes for: (a) high flows out to 7¿days; (b) low flows out to 2¿days; and (c) mid flows for majority of lead times. The new MuTHRE-FD model provides seamless sub-seasonal forecasts with high quality performance for both high and low flows over a range of lead times. This improvement provides forecast users with increased confidence in using sub-seasonal forecasts across a wide range of applications.

Subseasonal streamflow forecasts, with lead times of 1¿30¿days, provide valuable information for operational water resource management. This paper introduces the multi-temporal hydrological residual error (MuTHRE) model to address the challenge of obtaining "seamless" subseasonal forecasts ¿ that is, daily forecasts with consistent high-quality performance over multiple lead times (1¿30¿days) and aggregation scales (daily to monthly). The key advance of the MuTHRE model is combining the representation of three temporal characteristics of hydrological residual errors: seasonality, dynamic biases, and non-Gaussian errors. The MuTHRE model is applied in 11 Australian catchments using the hydrological model GR4J and post processed rainfall forecasts from the numerical weather prediction model ACCESS-S, and is evaluated against a baseline model that does not model these error characteristics. The MuTHRE model provides "high" improvements (practically significant in the majority of performance stratifications) in terms of reliability: (i) at short lead times (up to 10¿days), due to representing non-Gaussian errors, (ii) stratified by month, due to representing seasonality in hydrological errors, and (iii) in dry years, due to representing dynamic biases in hydrological errors. Forecast performance also improves in terms of sharpness, volumetric bias, and CRPS skill score; these improvements are statistically but not practically significant in the majority of stratifications. Importantly, improvements are consistent across multiple time scales (daily and monthly). This study highlights the benefits of modeling multiple temporal characteristics of hydrological errors and demonstrates the power of the MuTHRE model for producing seamless subseasonal streamflow forecasts that can be utilized for a wide range of applications.

A major challenge in surface hydrology involves predicting streamflow in ungauged catchments with heterogeneous vegetation and spatiotemporally varying evapotranspiration (ET) rates. We present a top-down approach for quantifying the influence of broad-scale changes in forest structure on ET and hence streamflow. Across three catchments between 18 and 100 km2 in size and with regenerating Eucalyptus regnans and E. delegatensis forest, we demonstrate how variation in ET can be mapped in space and over time using LiDAR data and commonly available forest inventory data. The model scales plot-level sapwood area (SA) to the catchment-level using basal area (BA) and tree stocking density (N) estimates in forest growth models. The SA estimates over a 69 year regeneration period are used in a relationship between SA and vegetation induced streamflow loss (L) to predict annual streamflow (Q) with annual rainfall (P) estimates. Without calibrating P, BA, N, SA, and L to Q data, we predict annual Q with R2 between 0.68 and 0.75 and Nash Sutcliffe efficiency (NSE) between 0.44 and 0.48. To remove bias, the model was extended to allow for runoff carry-over into the following year as well as minor correction to rainfall bias, which produced R2 values between 0.72 and 0.79, and NSE between 0.70 and 0.79. The model under-predicts streamflow during drought periods as it lacks representation of ecohydrological processes that reduce L with either reduced growth rates or rainfall interception during drought. Refining the relationship between sapwood thickness and forest inventory variables is likely to further improve results.

Optimisation can assist in the management of riverine ecosystems through the exploration of multiple alternative management strategies, and the evaluation of trade-offs between conflicting objectives. In addition, it can facilitate communication and learning about the system. However, the effectiveness of optimisation in aiding decision making for ecological management is currently limited by four major challenges: identification and quantification of ecosystem objectives; representation of ecosystems in predictive simulation models; specification of objectives and management alternatives in an optimisation framework; and evaluation of model results against actual ecological outcomes. This study evaluates previous literature in ecology, optimisation and decision science, and provides a strategy for addressing the challenges identified. It highlights the need for better recognition and analysis of assumptions in optimisation modelling as part of a process that generates and shares knowledge.

Water distribution systems throughout the world are deteriorating in part due to corrosion of cast iron pipes. Developing a deeper understanding of the operational association with failure may lead to operational improvements extending asset life. Operations engineers can be expected to develop credible theories of pipe failure based on their experience dealing with failures in the field. However, these failure theories can be more rigorously tested by a systemwide investigation of a large number of pipe failures. A forensic framework was developed to investigate how system operations are associated with large-diameter trunk-main failures. The framework utilizes a pipe failure database and historical supervisory control and data acquisition (SCADA) data to analyze trunk-main failures. Using the framework, about 141 largediameter (=300 mm) trunk-main failures were investigated in the Lake Zone in Newcastle, Australia. A predominant system failure mode called pump offpeak (POP) was identified that accounted for approximately 38% of the failures. POP was characterized by an increasing static pressure in low demand periods with limited pressure relief from downstream reservoirs.

If risk-based criteria are used in the design of high hazard structures (such as dam spillways and nuclear power stations), then it is necessary to estimate the annual exceedance probability (AEP) of extreme rainfalls up to and including the Probable Maximum Precipitation (PMP). This paper describes the development and application of two largely independent methods to estimate the frequencies of such extreme rainfalls. One method is based on stochastic storm transposition (SST), which combines the "arrival" and "transposition" probabilities of an extreme storm using the total probability theorem. The second method, based on "stochastic storm regression" (SSR), combines frequency curves of point rainfalls with regression estimates of local and transposed areal rainfalls; rainfall maxima are generated by stochastically sampling the independent variates, where the required exceedance probabilities are obtained using the total probability theorem. The methods are applied to two large catchments (with areas of 3550 km2 and 15,280 km2) located in inland southern Australia. Both methods were found to provide similar estimates of the frequency of extreme areal rainfalls for the two study catchments. The best estimates of the AEP of the PMP for the smaller and larger of the catchments were found to be 10-7 and 10-6, respectively, but the uncertainty of these estimates spans one to two orders of magnitude. Additionally, the SST method was applied to a range of locations within a meteorologically homogenous region to investigate the nature of the relationship between the AEP of PMP and catchment area.

One of the fundamental steps in regional rainfall frequency analysis is deciding the method by which rainfall stations are to be grouped together to form regions. This paper compares three methods of forming regions for use in estimating design rainfalls: a fixed region approach where all the available sites are included in a single region, a Region of Influence (ROI) approach based on geographical proximity and a hybrid approach where sites with similar topographic orientations are grouped together.The three region types were implemented in a Bayesian Generalized Least Squares Regression (BGLSR) framework which leads to regionalized regression equations that can be used to predict rainfall L-moments at ungauged sites. A leave-one-out cross validation approach was used to compare the relative accuracy, reliability and uncertainty of the derived rainfall statistics and resulting estimates of the rainfall quantiles. The study used data from two areas of Australia chosen for their highly varied topography and different climatic influences.It was found that all three methods provided good estimates of the L-moment statistics and the rainfall quantiles. The hybrid approach produced the smallest errors in the South-East Queensland region whilst for the Tasmanian region the fixed region approach was best. The results from this study show that although there is a slight benefit in using the proposed hybrid approach for BGLSR, these benefits were minor compared to maximizing the number of stations used to calibrate the BGLSR equations. This conclusion regarding the number of stations could be tested in future work by repeating the analyses in areas with sparser station density. Another test could be to simulate reduced station coverage in the current study areas by leaving stations out of the analyses. Finally it would be interesting to see if similar results are obtained by expanding the study area so that different climatological regimes are assessed.

Multi-objective optimization methods require many thousands of objective function evaluations. For urban water resource problems such evaluations can be computationally very expensive. The question as to which optimization method is the best choice for a given function evaluations budget in urban water resource problems remains unexplored. The main objective of this paper is to address this question. The second objective is to develop a new optimization algorithm, efficient multi-objective ant colony optimization-I (EMOACO-I), which exploits the good performance of ant colony optimization enhanced using ideas borrowed from evolutionary optimization. Its performance was compared against three established methods (NSGA-II, SMPSO, eMOEA) using two case studies based on the urban water resource systems serving two major Australian cities. The case study problems involved two or three objectives and 10 or 13 decision variables affecting infrastructure investment and system operation. The results show that NSGA-II was the worst performing method. However, none of the remaining methods was unambiguously superior. For example, while EMOACO-I converged more rapidly, its diversity was comparable but not superior to the other methods. Greater differences in performance were found as the number of objectives and case study complexity increased. This suggests that pooling the results from a number of methods could help guard against the vagaries in performance of individual methods.

Managers of forested water supply catchments require efficient and accurate methods to quantify changes in forest water use due to changes in forest structure and density after disturbance. Using Light Detection and Ranging (LiDAR) data with as few as 0.9 pulses m-2, we applied a local maximum filtering (LMF) method and normalised cut (NCut) algorithm to predict stocking density (SDen) of a 69-year-old Eucalyptus regnans forest comprising 251 plots with resolution of the order of 0.04 ha. Using the NCut method we predicted basal area (BAHa) per hectare and sapwood area (SAHa) per hectare, a well-established proxy for transpiration. Sapwood area was also indirectly estimated with allometric relationships dependent on LiDAR derived SDen and BAHa using a computationally efficient procedure. The individual tree detection (ITD) rates for the LMF and NCut methods respectively had 72% and 68% of stems correctly identified, 25% and 20% of stems missed, and 2% and 12% of stems over-segmented. The significantly higher computational requirement of the NCut algorithm makes the LMF method more suitable for predicting SDen across large forested areas. Using NCut derived ITD segments, observed versus predicted stand BAHa had R2 ranging from 0.70 to 0.98 across six catchments, whereas a generalised parsimonious model applied to all sites used the portion of hits greater than 37 m in height (PH37) to explain 68% of BAHa. For extrapolating one ha resolution SAHa estimates across large forested catchments, we found that directly relating SAHa to NCut derived LiDAR indices (R2 = 0.56) was slightly more accurate but computationally more demanding than indirect estimates of SAHa using allometric relationships consisting of BAHa (R2 = 0.50) or a sapwood perimeter index, defined as (BAHaSDen)1/2 (R2 = 0.48).

Mean sapwood thickness, measured in fifteen 73 year old Eucalyptus regnans and E. delegatensis stands, correlated strongly with forest overstorey stocking density (R² 0.72). This curvilinear relationship was used with routine forest stocking density and basal area measurements to estimate sapwood area of the forest overstorey at various times in 15 research catchments in undisturbed and disturbed forests located in the Great Dividing Range, Victoria, Australia. Up to 45 years of annual precipitation and streamflow data available from the 15 catchments were used to examine relationships between mean annual loss (evapotranspiration estimated as mean annual precipitation minus mean annual streamflow), and sapwood area. Catchment mean sapwood area correlated strongly (R² 0.88) with catchment mean annual loss. Variation in sapwood area accounted for 68% more variation in mean annual streamflow than precipitation alone (R² 0.90 compared with R² 0.22). Changes in sapwood area accounted for 96% of the changes in mean annual loss observed after forest thinning or clear-cutting and regeneration. We conclude that forest inventory data can be used reliably to predict spatial and temporal variation in catchment annual losses and streamflow in response to natural and imposed disturbances in even-aged forests. Consequently, recent advances in mapping of sapwood area using airborne light detection and ranging will enable high resolution spatial and temporal mapping of mean annual loss and mean annual streamflow over large areas of forested catchment. This will be particularly beneficial in management of water resources from forested catchments subject to disturbance but lacking reliable long-term (years to decades) streamflow records. Key Points: Correlations between mean annual streamflow and forest sapwood area are examined Annual streamflow can be predicted from forest inventory data and rainfall Changes in sapwood area account for changes in annual streamflow

We introduce a novel methodology for measuring stand sapwood area (SA), which provides a useful indicator of evapotranspiration from forest stands. The method is demonstrated in a 73-year-old Eucalyptus regnans forest comprising 784 stems over a 5ha area. We used photos of stump cross-sections to differentiate sapwood from heartwood and found 90% of stump segments to have a visible transition boundary. The digital images were corrected for lens distortion and scaled to an actual cross-sectional area, with resulting stump perimeters corresponding well with field-measured perimeters traced using string (root-mean-square error=4cm, R2=0.99). Calculated SA and basal area (BA) at stump height were coupled with tree and sapwood taper data to predict the SA:BA ratio at 1.3m height (R^1.3). Tree taper data were coupled with stump dimensions data in a mixed-effects model to predict each stump's BA at 1.3m, and sapwood taper data from buttress logs were used to improve each stems R^1.3. Using this procedure, we found our study site to have an R^1.3 of 0.21 and total stand SA at 1.3m height of 9.3m2ha-1. We quantified the bias in traditional R^1.3 estimates that use cores to measure sapwood thickness and diameter tape to calculate SA. We show traditional methods underestimate R^1.3 that increases with tree diameter and decreases with stem circularity, whereas our methodology gave more accurate measures of SA in large buttressing trees. Our methodology also provides a more efficient way of generating maps of SA variation across large forested catchments.

Significant population increase in urban areas is likely to result in a deterioration of drought security and level of service provided by urban water resource systems. One way to cope with this is to optimally schedule the expansion of system resources. However, the high capital costs and environmental impacts associated with expanding or building major water infrastructure warrant the investigation of scheduling system operational options such as reservoir operating rules, demand reduction policies, and drought contingency plans, as a way of delaying or avoiding the expansion of water supply infrastructure. Traditionally, minimizing cost has been considered the primary objective in scheduling capacity expansion problems. In this paper, we consider some of the drawbacks of this approach. It is shown that there is no guarantee that the social burden of coping with drought emergencies is shared equitably across planning stages. In addition, it is shown that previous approaches do not adequately exploit the benefits of joint optimization of operational and infrastructure options and do not adequately address the need for the high level of drought security expected for urban systems. To address these shortcomings, a new multiobjective optimization approach to scheduling capacity expansion in an urban water resource system is presented and illustrated in a case study involving the bulk water supply system for Canberra. The results show that the multiobjective approach can address the temporal equity issue of sharing the burden of drought emergencies and that joint optimization of operational and infrastructure options can provide solutions superior to those just involving infrastructure options. © 2014. American Geophysical Union. All Rights Reserved.

The development and application of evolutionary algorithms (EAs) and other metaheuristics for the optimisation of water resources systems has been an active research field for over two decades. Research to date has emphasized algorithmic improvements and individual applications in specific areas (e.g. model calibration, water distribution systems, groundwater management, river-basin planning and management, etc.). However, there has been limited synthesis between shared problem traits, common EA challenges, and needed advances across major applications. This paper clarifies the current status and future research directions for better solving key water resources problems using EAs. Advances in understanding fitness landscape properties and their effects on algorithm performance are critical. Future EA-based applications to real-world problems require a fundamental shift of focus towards improving problem formulations, understanding general theoretic frameworks for problem decompositions, major advances in EA computational efficiency, and most importantly aiding real decision-making in complex, uncertain application contexts.

Urban bulk water systems supply water with high reliability and, in the event of extreme drought, must avoid catastrophic economic and social collapse. In view of the deep uncertainty about future climate change, it is vital that robust solutions be found that secure urban bulk water systems against extreme drought. To tackle this challenge an approach was developed integrating: 1) a stochastic model of multi-site streamflow conditioned on future climate change scenarios; 2) Monte Carlo simulation of the urban bulk water system incorporated into a robust optimization framework and solved using a multi-objective evolutionary algorithm; and 3) a comprehensive decision space including operating rules, investment in new sources and source substitution and a drought contingency plan with multiple actions with increasingly severe economic and social impact. A case study demonstrated the feasibility of this approach for a complex urban bulk water supply system. The primary objective was to minimize the expected present worth cost arising from infrastructure investment, system operation and the social cost of "normal" and emergency restrictions. By introducing a second objective which minimizes either the difference in present worth cost between the driest and wettest future climate change scenarios or the present worth cost for driest climate scenario, the trade-off between efficiency and robustness was identified. The results show that a significant change in investment and operating strategy can occur when the decision maker expresses a stronger preference for robustness and that this depends on the adopted robustness measure. Moreover, solutions are not only impacted by the degree of uncertainty about future climate change but also by the stress imposed on the system and the range of available options.

Uncertainty in the rainfall inputs, which constitute a primary forcing of hydrological systems, considerably affects the calibration and predictive use of hydrological models. In a recent paper, Ajami et al. [2007] proposed the Integrated Bayesian Uncertainty Estimator (IBUNE) to quantify input, parameter and model uncertainties. This comment analyzes two interpretations of the IBUNE method and compares them to the Bayesian Total Error Analysis (BATEA) method [Kavetski et al., 2002, 2006a]. It is shown that BATEA and IBUNE are based on the same hierarchical conceptualization of the input uncertainty. However, in interpretation A of IBUNE, the likelihood function, and hence the posterior distribution, are random functions of the inferred variables, which violates a standard requirement for probability density functions (pdf). A synthetic study shows that IBUNE-A inferences are inconsistent with the correct parameter values and model predictions. In the second interpretation, IBUNE-B, it is shown that a specific implementation of IBUNE is equivalent to a special Metropolis-Hastings sampler for the full Bayesian posterior, directly including the rainfall multipliers as latent variables (but not necessarily storing their samples). Consequently, IBUNE-B does not reduce the dimensionality of the sampling problem. Moreover, the jump distribution for the latent variables embedded in IBUNE-B is computationally inefficient and leads to prohibitively slow convergence. Modifications of these jump rules can cause convergence to incorrect posterior distributions. The primary conclusion of this comment is that, unless the hydrological model and the structure of data uncertainty allow specialized treatment, Bayesian hierarchical models invariably lead to high-dimensional computational problems, whether working with the full posterior (high-dimensional sampling problem) or with the marginal posterior (high-dimensional integration problem each time the marginal posterior is evaluated). © 2009 by American Geophysical Union.

Hidden Markov models (HMMs) offer a plausible representation of long-term hydroclimatic persistence in rainfall and streamflow observations. Persistent climate processes influence hydrological observations at various time scales. This paper develops the stochastic framework of two-state HMMs to better represent climate-rainfall interactions at both monthly and annual levels. Two new models, a hierarchical HMM and a non-homogeneous HMM are introduced, and fitted to monthly rainfall and streamflow observations from Australia. The value of these models to identify two-state persistence is compared to that of existing two-state HMMs.

Sub-seasonal streamflow forecasts are important for a range of water resource management applications, with a distinct practical interest in forecasts of high flows (e.g. for managing flood events) and low flows (e.g. for managing environmental flows). Despite this interest, differences in forecast performance for high and low flow events are not routinely investigated. Our study reveals that while forecasts evaluated over the full flow range can appear reliable, stratification into high/low flow ranges highlights significant under/over-estimation of forecast uncertainty, respectively. This study introduces a flow-dependent (FD) non-parametric component into a post-processing model of hydrological forecasting errors, the Multi-Temporal Hydrological Residual Error (MuTHRE) model, yielding the MuTHRE-FD model. We use a case study with 11 catchments in the Murray Darling Basin, the GR4J rainfall-runoff model and post-processed rainfall forecasts from ACCESS-S, to compare the MuTHRE and MuTHRE-FD models. Through its improved treatment of flow-dependence, the MuTHRE-FD model achieves practically significant improvements over the original MuTHRE model in the reliability of forecasted cumulative volumes for: (i) high flows out to 7 days; (ii) low flows out to 2 days; and (iii) mid flows for majority of lead times. Example cumulative flow time series are provided in Figure 1. The new MUTHRE-FD model provides sub-seasonal forecasts with high quality performance for both high and low flows over a range of lead times. This improvement provides forecast users with increased confidence in using sub-seasonal forecasts across a wide range of applications.

Calibration and prediction in conceptual rainfall-runoff (CRR) modelling is affected by input, model and response error (Figure 1a). This study works towards the goal of developing a robust framework for dealing with these sources of error and focuses on model error. The characterization of model error in CRR modelling has been thwarted by poor conceptualizations of error propagation (Figure 1b) and the convenient but indefensible treatment of CRR models as deterministic descriptions of catchment dynamics. It is argued that CRR fluxes are fundamentally stochastic because they involve spatial and temporal averaging. Acceptance that CRR models are intrinsically stochastic paves the way for a more rational characterization of model error. The hypothesis advanced in this paper is that CRR model error can be characterized by storm-dependent random variation of one or more CRR model parameters that affect fluxes. A simple sensitivity analysis is developed to assist in identifying the parameters most likely to behave stochastically. A Bayesian hierarchical model is formulated to explicitly differentiate between input, response and model error - this provides a very general framework for calibration and prediction, as well as the testing of hypotheses regarding model structure and data uncertainty. A case study using daily data from the Abercrombie catchment (Australia) and employing a 6-parameter CRR model demonstrates the considerable potential of this approach. Figure 2 illustrates the excellent fit to the observed data. Of particular significance is the use of posterior diagnostics to test key assumptions about errors. The assumption that the storm-dependent parameters are log-normally distributed is only partially supported by the data, which suggests that the parameter hyperdistributions have thicker tails. Further research is aiming to refine this approach to characterizing model error.

© 2015, Engineers Australia. All rights reserved.East Coast Lows (ECLs) are intense low-pressure systems which occur over the subtropical east coasts of Southern Hemisphere continents, including Australia. These systems are typically associated with gale-force winds, large seas, storm surges, heavy rainfall and flooding. While these ECL impacts are mostly negative the rainfall associated with ECLs is also very important for urban water security within the heavily populated eastern seaboard of Australian (ESA). This study examines historical ECLs to gain insights into the timing, frequency and location of ECL occurrence as well as the magnitude and spatial extent of ECL impacts on rainfall. The different characteristics and impacts associated with the different ECL sub-types are highlighted and it is demonstrated how this information can be used to stochastically simulate daily rainfall such that the statistics important for catchment scale hydrology (e.g. clustering of extreme events, long-term persistence, frequency/duration/magnitude of wet and dry spells etc.) are realistically preserved. These simulated rainfall sequences, that incorporate the spatial and temporal hydroclimatic variability caused by ECLs and other climate phenomena, are important inputs into hydrological models used to determine current and future urban water security.

The Bayesian total error analysis (BATEA) framework allows model calibration and prediction informed by estimates of data and model uncertainty. However, full BATEA applications are currently limited to studies with relatively short record lengths which do not require real-time updating of model predictions. This is due to the use of batch calibration strategies, which rapidly become computationally expensive when input and/or model errors are inferred directly. This study seeks to develop a recursive implementation of the BATEA framework based on particle filters that efficiently manage time invariant parameters and stochastic state variables. For real-time updating, recursive estimation techniques can be considerably faster than batch methods, facilitating the application of BATEA to applications such as forecasting. It is shown how particle filtering techniques, traditionally used in automatic control and signal processing applications, can be adapted to provide a robust recursive implementation of BATEA. This study assesses the performance of the resample-move particle filter using noise models that preserve physical constraints when applied to the calibration of a conceptual rainfall-runoff model with time-invariant parameters and time-varying model states.

The aim of this paper is to share the experience gained during the development of an optimization model for Sydney's water supply system for the Metropolitan Water Plan (MWP) 2015. The objective is to identify the optimal portfolio for water supply and demand management measures to secure water for drought and for growth. In 2013, an optimisation model for Sydney's supply system called MetroNet based on Wathnet5 was developed for screening supply measures and optimization operation. The key objective is to minimise the total present worth cost of operating and augmenting the system while maximimising the demand that could be met without breaching the design criteria that constrain security and reliability. Using MetroNet, the current system called 'Reference' was optimised modifying its operating triggers without the introduction of any new supply measures. New supply measures were considered subsequently. The decision variables were selected based on previous knowledge gained during supply system yield estimation. The first two criteria were used as constraints in optimization. The outcomes of an optimisation are presented as a 'Pareto' optimal set of portfolios for which it is impossible to improve the performance on one objective without reducing the performance on another. The results of the optimization runs were analysed and trade-offs were explained to decision makers. This paper discusses the processe by which objectives were formulated and decisions selected as well as approximations to make optimization run times feasible.

A new regional flood frequency estimation (RFFE) model for Australia has been developed as a part of 4th edition of Australian Rainfall and Runoff, which is referred to as 'RFFE Model 2015'. To develop and test this model, flood data from 853 gauged catchments have been utilised, which includes 798 gauged catchments from the humid coastal areas and 55 catchments from the arid and semi-arid areas. The model allows the derivation of design flood estimates for annual exceedance probabilities (AEP) of 50% to 1% for small to medium catchments anywhere in Australia. In the RFFE Model 2015, the humid coastal and arid/semi-arid areas of Australia have been divided into five and two different regions, respectively. The boundaries between the arid and humid coastal regions have been drawn approximately based on the 500 mm mean annual rainfall contour line. To reduce the effects of sharp variation in flood estimates for the ungauged catchments located close to these regional boundaries, seven fringe zones have been delineated. In the humid coastal regions, a region-of-influence approach has been adopted to derive design flood estimates for ungauged catchments. In developing the prediction model for the regionalised Log Pearson Type 3 distribution, a Bayesian generalised least squares regression technique has been applied, which considers the inter-station correlation and variation in record lengths of the annual maximum flood series across different sites. For the arid/semiarid regions, a simple index type regional method has been adopted. For easy application by the industry, an application tool has been developed, which automates the application of the RFFE Model 2015. The user is required to provide simple input data (e.g. catchment area and catchment location) to obtain design flood quantiles and associated uncertainty. This paper provides essential technical information, which will assist the user to apply the RFFE Model 2015 in practice with confidence. Further details can be found in the ARR draft chapter on regional flood methods and technical reports.

This study investigated the potential to simulate a river basin with significant wetland hydrodynamics within a network flow programming (NFP) framework. The Macquarie River Catchment, which encompasses the Macquarie Marshes, was selected as a case study to test the approach. The Marshes support water-dependent ecological communities which require flooding events, triggered through natural and environmental flows, to inundate wetland areas for ecologically sufficient periods of time. Ecological responses of the Macquarie Marshes with respect to inundation are currently simulated using the Macquarie Valley Integrated Quantity and Quality Model (IQQM). This model represents the Marshes hydraulics using a rules-based algorithm informed by detailed hydrodynamic modelling of the Marshes. However, the model does not have the potential to optimise water resource and ecological management. The primary objective of this study was to use NFP to emulate the Macquarie Marshes hydrodynamics in a manner consistent with the Macquarie Valley IQQM model. This paper presents and evaluates several NFP techniques to emulate the Marshes hydraulics. It was found that iterative use of NFP with side constraints almost exactly matched the emulator rules and implemented them more accurately than the IQQM model. This successful outcome will enable multi-criterion optimisation of the basin operation to deal directly with the ecological responses of water sensitive communities.

Hydrological model calibration has benefited from rapid advances in optimization algorithms and computing power in the last few decades. Stochastic evolutionary optimization methods have received particular attention because, compared to classical gradient-based algorithms, they are generally less sensitive to the irregularity, multi-optimality and non-smoothness of objective function surfaces commonly reported in calibration of hydrological models. In particular, the Shuffled Complex Evolution (SCE) search is currently one of the most widely used stochastic evolutionary optimization methods in hydrology. However, the improved robustness of evolutionary optimization usually comes at considerable computational cost. For example, the SCE search has been reported to become very expensive as the number of calibrated parameters increases. This work revisits the use of modern gradient-based algorithms. We investigate the performance of quasi-Newton and Gauss-Newton optimization algorithms using the hydrological models SIMHYD (7 parameters) and FUSE-536 (14 parameters) calibrated to three Australian catchments. These Newton-type algorithms are compared to the SCE search. Analysis of the objective function surfaces found micro-scale roughness and parameter insensitivity in both SIMHYD and FUSE-536. SCE search was the most robust algorithm for calibrating SIMHYD, but, somewhat surprisingly, struggled in the case of FUSE-536. In terms of computational costs, the Newton-type algorithms required about 20 times fewer objective function evaluations than SCE search for SIMHYD and 50 times fewer evaluations for FUSE-536. Taking into account the chance of converging to the global optimum and the computational cost, we suggest that modern Newton-type algorithms may be competitive with, or even outperform, the SCE search. Any improvements to the robustness of Newton-type algorithms will further strengthen this result.

Time series simulation of reservoir behaviour using synthetically generated streamflow and rainfall data is an integral part of calculating drought risks for many water utilities in Australia. Two key parameters that are required for the generation of the synthetic climate data are the mean and standard deviation of the annual data for each site, these being estimated from historic data. Stedinger and Taylor (1982) explored the impact of uncertainty in these parameters on the simulation of reservoir behaviour and the size of reservoir that would be required in order to maintain a specified target release for a 50 year sequence of inflow. The contemporary focus of drought risk assessments is different and the decision making criteria now require substantially more computational effort. Today, the focus of the assessments is generally on how much water can be supplied from a given set of infrastructure for a given set of reliability targets, rather than on how large the infrastructure needs to be to meet the reliability target. Further, the reliability targets are commonly expressed in terms of low frequency events, which means that many replicates are required per assessment to estimate these low risks. In this paper the impact of uncertainty in the mean and standard deviation of the historic climate data is explored in terms of its impact on a contemporary calculation of system yield. It is found that the uncertainty in these parameters has an appreciable impact on uncertainty associated with the estimated system yield.

The configuration of the Macquarie Marshes is a mosaic-like collection of swamps, marshes and lagoons. The Macquarie Marshes is also one of the most ecologically important wetland systems in Australia. It contains unique plant communities that serve as a sanctuary for many species of waterbirds and other fauna such as frogs and mammals. A significant deterioration of the ecological features of the Macquarie Marshes has been recorded in the past decades. This fact is mostly attributed to reductions of the input discharges to the marshes due to water allocations for industrial, agricultural and domestic usage. Reduction of water supply translates into changes of the hydraulic regime which has a direct impact on the flood dependent vegetation species of the marshes. The complexity of the system and its ecological significance requires the use of an adequate computational tool that would allow for a realistic assessment of the site. In this paper we present initial work regarding the development of a vegetation dynamics model that can integrate vegetation establishment with time aggregated characteristics of the flow. We simulate floods on a fictional wetland by implementing a quasi-2D hydrodynamic model (VHHMM 1.0) over a rectangular cell grid. This same grid constitutes the basis for a cellular vegetation model that can calculate changes in the vegetation for each element inside the domain. The work presented here for a fictional site was developed in order to test the capability of our model to recreate consistent vegetation gradients by using deterministic transitional rules. These rules relate time aggregated characteristics of the flow such as flood period and depth of water to water requirements of different vegetation communities. We found that a well calibrated set of deterministic transitional rules based on water preferences can recreate consistent vegetation distributions; however, succession and critical conditions for succession rules will have to be defined for a specific site application. Further development of this model will result in a strategic tool for managing environmental water allocations and water sharing plans in the Macquarie Marshes.

The Bureau of Meteorology recently released a new streamflow forecast product based on a dynamic approach where the forecasts are generated with a lumped hydrological model (GR4J) that is forced by statistically downscaled rainfall forecasts obtained from the Bureau's Predictive Ocean Atmosphere Model for Australia (POAMA). This registered user service provides ensemble forecasts of the 1 month and 3 month streamflow volume at 100 locations across Australia. The forecast system is composed of three different components: (1) downscaling of gridded outputs from POAMA version M2.4 to catchment scale rainfall forecasts; (2) hydrological model calibrated with a statistical tool that accounts for predictive uncertainty (BATEA) and forced with downscaled rainfall forecasts; and (3) post processed streamflow forecasts to correct for residual biases and adjust the ensemble spread. In this paper, we describe the complexities involved and challenges faced in operationalising the dynamic approach. Performance of this system is reviewed by computing performance metrics for historical reforecast in a cross-validation framework. The forecast performance exceeds the one obtained with a climatological forecast for a vast majority of sites, for all metrics, and for both monthly and seasonal forecasts. In addition, it reaches a comparable level of performance with those derived from the existing statistical model (BJP) currently used by the Bureau to issue seasonal forecasts. These results demonstrate a major achievement considering the number of modelling components involved, their respective complexity, the number of forecast sites covered, the range of climate conditions encountered and the constraints of running such system in an operational setting Regardless of completion of this important milestone, the forecast skill could be improved under certain conditions, particularly in dry catchments and several forecast sites in Queensland and Tasmania.

This paper describes the development and application of two largely independent methods to estimate the annual exceedance probability (AEP) of Probable Maximum Precipitation (PMP). One method is based on the Stochastic Storm Transposition (SST) approach, which combines the "arrival" and "transposition" probabilities of an extreme storm using the total probability theorem. The second method - termed "Stochastic Storm Regression" - combines frequency curves of point rainfalls with regression estimates of areal rainfalls; the regression relationship is derived using local and transposed storms, and the final exceedance probabilities are derived using the total probability theorem. The methods are applied to two large catchments (with areas of 3550 km2 and 15280 km2) located in inland southern Australia. In addition, the SST approach is used to derive regional estimates for standardised catchments within the Inland GSAM region. Careful attention is given to the uncertainty and sensitivity of the estimates to underlying assumptions, and the results are used to help formulate draft ARR recommendations.

The Integrated Water Supply Scheme (IWSS) provides a secure water supply to the Perth region. This study reports on the findings of a multi-objective optimization analysis to derive insights on the best ways to manage the current IWSS. With precedents based urban systems dominated by surface water, it was realized that the drying climate experienced by the Perth region in recent decades would make formulating the optimization problem itself a challenging task. The optimization had to be embedded in a discovery/learning process characterized by iterative review of results and reformulation to develop a deeper understanding of system behaviour and an adequate articulation of stakeholder needs. The key insights to emerge from this process were recognition of the major changes on system behaviour brought about by the drying trend in Perth's climate. With reduced dependence on surface water, reservoirs were found to operate in a narrow range, rarely spilling. Traditional drought contingency measures based on storage-triggered restrictions and a variable groundwater abstraction rule proved to be ineffective. The availability of surplus desalination water offers the prospect of banking water in underutilized reservoirs to provide a contingency source of water. However, a major cost trade-off is required to sustain high levels of banking.

© 2014 The Authors. Water distribution networks throughout the world are ageing, which increasingly leads to sudden pipe failure. About 108 trunk-main pipe failures in an urban sub-network were investigated using a pipe failure data base and SCADA (Supervisory Control and Data Acquisition) data to understand the contribution of hydraulic pressure to pipe failure using multiple lines of evidence. The forensic investigation revealed a dominant system-wide failure mode which was characterized by predominately off-peak high speed pumping with limited pressure relief from downstream reservoirs. A frequency analysis was conducted for greater understanding of the dominant failure mode.

Increasingly, water management models are being used in decision making contexts that involve selecting a "preferred" course of action by weighing performance against competing objectives. The role of models within the decision process is often poorly articulated, uncertainty is not well accounted, and risk only evaluated after the selection has been made. For example, the technical performance of alternative integrated urban water management options are evaluated using various models, and the results presented in a technical report. The decision makers base their selection on the predicted performance, and the preferences of the stakeholders around the table. Risk assessment is then undertaken to identify and control any areas of high risk, which might be costly or even unachievable. A well structured decision process might have resulted in a different choice. eWater CRC is delivering a range of new tools to support decision-making in the water industry, ranging from selecting water sensitive design of new urban allotments to exploring policy options for Australia's large regulated rivers. Central to this effort is a user requirement to incorporate uncertainty analysis, risk analysis, optimisation and prioritisation into the tools. This paper describes a decision making framework that places models, and other sources of knowledge, into a decision making context. The framework articulates the role of optimisation, risk analysis and prioritisation in the decision making process and clarifies the pervasive role of uncertainty. The framework provides a guide for the inclusion of these decision elements into modelling products, either as generic software elements that can be applied to multiple models, or as supporting material such as documentation or training. Using the framework, water-management stakeholders articulate problems by iterating around a cycle that defines objectives based on an initial problem statement, and determines what metrics will be used to ascertain that the objective has been achieved within the context of a well-defined system. Different proposed solutions are then evaluated in terms of the agreed metrics, and the outcomes are compared to select the "best" solution. Selection of "best" option is achieved by including considerations beyond the direct outputs of performance prediction models. By tracking uncertainties and providing assessment methods for risk and optimisation in an environment of compound considerations, a rational and scientifically justified suite of preferred options can be generated. These preferred solutions in turn inform a multi-objective decision process, which allows stakeholders to express preferences and assign weightings to make their final choice, while making full use of the outcomes of detailed scientific analysis. An understanding of the quality of the evidence used to support each step of the process enhances the value of the decision support. The decision framework complements a common model structure that is used to integrate the various component models developed by eWater CRC. Together the decision framework and the common model structure form the conceptual architecture of the eWater product offering.

River managers and modellers use long term planning models to inform river operators and planners on how to best operate regulated river systems. Long term planning models simulate the regulated river system using either rules based approaches or linear optimisation techniques. This paper compares these two approaches and examines the potential for objective driven solutions to be used to generate better rules in the Lachlan River System in NSW, Australia. Multiple supply path problems occur when water can be sourced from storages in parallel, storages in series or delivered by parallel distribution paths. Multiple supply path problems are typically complex and difficult to solve. Both rules based models and objective driven (optimisation) models are used to solve multiple supply path problems for long term planning in Australia's rural catchments. Rules bases models such as the Integrated Quantity and Quality Model (IQQM) and MSM-BigMod have been used to model rural catchments in Queensland (IQQM), NSW (IQQM) and the Murray River (MSM-BidMod). Optimisation models have been used extensively in Australia for urban water supply modelling and in Victoria for rural and urban systems (REALM, (Diment, 1991), and WATHNET). Specific system information on tradeoffs between the two modelling methods (e.g. efficiency, accuracy of complex processes, and runtime) could be obtained relatively easily by modellers if the software allows a choice of approaches or a combination of approaches. Currently rules based long term planning models are typically run on a daily time step while the optimisation models are run on a monthly time step. It is considered important to be able to run models on either a monthly, daily or sub daily time step. There are run time implications for use of optimisation on daily to sub daily time steps. It may be preferable to model only part of the system with optimisation and the rest with a rules based model which is an option for the Lachlan example. This paper focuses on a case study for supply through multiple paths on the Lachlan River System in NSW that is traditionally modelled using a rules based model (IQQM). Implementing an objective driven model decreased the volumes ordered from the multiple supply paths by 55% and reduced shortfalls by 7% of total demand relative to the rules based model. Using the NetLP solution to generate new distribution rules for orders in IQQM reduced the volumes ordered by 13% and reduced shortfalls by 5.4% of total demand relative to the original IQQM. This illustrates the benefit to river operators and planners of having NetLPs in software packages for long term planning models. Objective driven solutions can be used to generate more efficient rules where a rules based model is preferred, however there will still be tradeoffs in efficiency, modelling accuracy of complex processes, and runtime.