Case study: Drought forecasting & water allocation markets — a city or utility pilot and the results so far

Australia's Murray-Darling Basin generates approximately AUD 24 billion in agricultural output annually and supplies drinking water to more than 3 million people across four states and the Australian Capital Territory. When the Millennium Drought (1997 to 2009) reduced inflows to the basin's river system by 60% relative to the long-term average, the resulting water allocation failures exposed fundamental weaknesses in static, rule-based allocation systems. Since then, Australian water authorities have invested heavily in drought forecasting and market-based allocation mechanisms. This case study examines a pilot program launched in 2022 by the Murray-Darling Basin Authority (MDBA) in partnership with the Bureau of Meteorology (BOM), the New South Wales Department of Planning and Environment, and a private analytics firm to integrate seasonal ensemble forecasting with real-time water market pricing in the southern connected Murray system.

Why It Matters

The Asia-Pacific region faces the most acute water scarcity pressures of any global region. The Asian Development Bank estimates that by 2050, water demand in the region will exceed supply by 40% under business-as-usual scenarios. Australia, while not the most water-scarce country in absolute terms, operates the most mature water trading system in the world, making it a natural laboratory for innovations that combine predictive hydrology with market mechanisms.

The Murray-Darling Basin Plan, enacted in 2012, established a framework for sustainable diversion limits and environmental water recovery. However, the plan's implementation has repeatedly collided with the reality of extreme climate variability. The 2019 to 2020 bushfire season and associated drought reduced general security allocations in the NSW Murray to zero for extended periods, triggering significant economic hardship in irrigation-dependent communities. Annual water market turnover in the Murray-Darling Basin exceeded AUD 2.8 billion in 2024, with temporary water allocation trades accounting for approximately 70% of total volume.

The Australian Competition and Consumer Commission's 2021 Murray-Darling Basin Water Markets Inquiry identified information asymmetry as a critical market failure. Large corporate irrigators and water investors with access to sophisticated forecasting tools consistently outperformed smaller family farms in timing their market participation. The ACCC recommended measures to improve forecast accessibility and market transparency, creating the policy foundation for the pilot examined here.

Climate change projections from CSIRO indicate that rainfall in the southern Murray-Darling Basin will decline by 5 to 15% by 2050, with autumn and winter rainfall (the critical recharge season) experiencing the most pronounced reductions. Simultaneously, higher temperatures will increase evapotranspiration, amplifying the effective deficit. These projections make improved drought forecasting not merely operationally useful but existentially important for communities that depend on irrigated agriculture.

Key Concepts

Water Allocation Announcements in the Murray-Darling system determine the percentage of entitlement volume that licence holders can access in a given season. State water authorities issue allocation announcements monthly during the water year (July to June), progressively increasing allocations as inflows materialize and dam storage levels change. General security entitlements in the NSW Murray receive highly variable allocations, ranging from 0% in severe drought to 100% in wet years. High security entitlements receive priority allocations but are more expensive to acquire.

Seasonal Streamflow Forecasting produces probabilistic projections of river flows over the coming 1 to 12 months. The Bureau of Meteorology's Seasonal Streamflow Forecasting Service uses a statistical approach called the Bayesian Joint Probability (BJP) modelling framework, which combines antecedent catchment conditions (soil moisture, groundwater levels, and dam storage) with seasonal climate outliers from dynamical models. Forecasts are issued as tercile probabilities: the likelihood that flows will fall in the below-normal, near-normal, or above-normal range.

Water Market Spot and Forward Pricing reflects the intersection of supply expectations, demand patterns, and speculative positioning. Spot allocation prices in the southern Murray fluctuate between AUD 20 per megalitre in wet periods and AUD 800 or more per megalitre during severe drought. Forward contracts, traded on platforms such as Waterflow and H2OX, allow irrigators to lock in prices for future delivery, but market liquidity for forward contracts remains thin compared to spot markets.

Ensemble Hydrological Prediction Systems generate hundreds of plausible future streamflow scenarios by driving rainfall-runoff models with perturbed initial conditions and multiple climate forecast inputs. The resulting ensemble spread quantifies forecast uncertainty, enabling risk-based decision-making. The pilot's innovation was coupling ensemble hydrological forecasts directly with agent-based models of water market behaviour to produce integrated supply-price projections.

Drought Forecasting and Water Market KPIs: Benchmark Ranges

Metric	Below Average	Average	Above Average	Top Quartile
Seasonal Inflow Forecast Skill (CRPSS)	<0.1	0.1-0.25	0.25-0.4	>0.4
Allocation Announcement Prediction Accuracy	<55%	55-70%	70-85%	>85%
Water Price Forecast Error (30-day)	>30%	20-30%	10-20%	<10%
Drought Onset Detection Lead Time	<6 weeks	6-12 weeks	12-20 weeks	>20 weeks
Market Participation Rate (Pilot Users)	<15%	15-30%	30-50%	>50%
Farm-Level Water Productivity (AUD/ML)	<AUD 400	AUD 400-800	AUD 800-1,400	>AUD 1,400
Decision Lead Time Improvement	<1.5x	1.5-2.5x	2.5-4x	>4x

Pilot Design and Implementation

The pilot launched in July 2022 with 187 participating irrigators, 3 water utilities, and 2 environmental water holders across the NSW Murray and Murrumbidgee valleys. The program received AUD 4.2 million in funding through the Commonwealth National Water Grid Fund and AUD 1.8 million in co-investment from participating water utilities, Murray Irrigation Limited, and the private analytics partner.

The platform integrated three data streams. First, the Bureau of Meteorology's operational seasonal streamflow forecasts provided baseline probabilistic inflow projections for 14 key gauging stations. Second, the analytics firm developed a proprietary machine learning overlay using gradient-boosted decision trees trained on 120 years of streamflow records, satellite-derived soil moisture from NASA's SMAP mission, snow depth observations from Snowy Hydro's monitoring network, and real-time dam storage data from WaterNSW. Third, the platform ingested water market transaction data from the Bureau of Meteorology's Water Markets Dashboard, the Murray-Darling Basin Exchange, and Waterflow's trading platform.

The system produced three integrated outputs delivered through a web-based dashboard and weekly email briefings. The first output was a probabilistic allocation pathway showing the range of likely allocation percentages for general and high security water at monthly intervals through the water year. The second was a water price range forecast showing expected spot allocation prices 30, 60, and 90 days forward, calibrated against the allocation pathway projections. The third was a set of decision triggers tied to individual farm plans, indicating when purchasing or selling water, adjusting crop mix, or activating on-farm storage would be economically optimal given the forecast distribution.

A critical design decision was the development of a "traffic light" interface layer that translated complex probabilistic information into intuitive visual signals. Red indicated high drought risk with recommended defensive actions (reduce water-intensive plantings, secure forward purchases). Amber indicated moderate uncertainty requiring monitoring. Green indicated comfortable supply outlooks where expansion or spot market selling could be considered.

Measured Outcomes

The pilot ran for three complete water years (2022-23, 2023-24, and 2024-25), spanning conditions that ranged from the La Nina-influenced wet conditions of 2022-23 to the developing El Nino drought stress of 2023-24 and the return to near-normal conditions in 2024-25. This variability provided a rigorous test of the system across multiple climate regimes.

Forecast Accuracy

The machine learning overlay improved seasonal inflow forecast skill by 18 to 32% compared to the Bureau of Meteorology's baseline statistical forecasts alone, as measured by the Continuous Ranked Probability Skill Score (CRPSS). The improvement was most pronounced during transitional seasons (autumn and spring) when antecedent catchment conditions provided stronger predictability than atmospheric forcing alone. Winter forecasts showed more modest improvement because the skill of seasonal climate models in predicting Indian Ocean Dipole and ENSO influences was already relatively high during this period.

Allocation announcement predictions achieved 78% accuracy at the 60-day horizon, meaning the system correctly predicted the subsequent allocation percentage within plus or minus 5 percentage points for 78% of monthly announcements. At the 90-day horizon, accuracy declined to 64%, though the directional signal (whether allocations would increase, decrease, or remain stable) was correct 82% of the time.

Water price forecasts at the 30-day horizon achieved a mean absolute percentage error of 14%, outperforming the naive random walk benchmark (which assumes tomorrow's price equals today's price) by 25%. During the 2023-24 El Nino period, when allocation prices spiked from AUD 85 to AUD 420 per megalitre in the NSW Murray, the platform's 60-day forecasts correctly signalled the price trajectory six weeks before the spike materialized in spot markets.

Economic Impact on Participants

An independent evaluation conducted by ABARES (Australian Bureau of Agricultural and Resource Economics and Sciences) in mid-2025 estimated that pilot participants achieved average water procurement cost savings of AUD 38,000 per farm per year compared to matched non-participants in the same irrigation districts. The savings arose from three mechanisms: better timing of spot purchases (buying earlier in the season when forecasts indicated future scarcity), more effective use of forward contracts (locking in prices before anticipated spikes), and improved crop planning that aligned water-intensive plantings with favourable supply outlooks.

Water utilities reported that improved allocation pathway forecasts enabled more efficient urban water supply planning. South East Water, serving communities along the Murray downstream of the Hume Dam, estimated AUD 1.2 million in avoided costs over the pilot period from deferred activation of contingency supply measures that would have proved unnecessary.

Environmental water holders, particularly the Commonwealth Environmental Water Holder, used the forecasts to optimise the timing of environmental flow deliveries. By scheduling releases during periods when forecast inflows would supplement environmental allocations, the CEWH achieved an estimated 15% increase in effective environmental watering volume without acquiring additional water entitlements.

Participation and Adoption Patterns

Of the 187 initially enrolled irrigators, 163 (87%) remained active participants through the full three-year pilot. The 24 who disengaged cited three primary reasons: insufficient digital literacy to interpret dashboard outputs (9 users), dissatisfaction with forecast accuracy during a specific period when the platform underperformed (8 users), and change in farming operations unrelated to the platform (7 users).

Engagement data revealed that irrigators with annual water entitlements exceeding 500 megalitres were 3.2 times more likely to actively use the decision trigger features than smaller entitlement holders. This pattern mirrored the ACCC's earlier finding about information asymmetry and suggested that the platform, despite its accessibility design, had not fully closed the sophistication gap between large and small operators.

What Worked and What Did Not

Effective Strategies

The traffic light interface was consistently cited in user surveys as the single most valuable design feature. Irrigators with limited technical background could engage with the system without understanding ensemble statistics or CRPSS scores. The interface abstracted complexity without sacrificing decision utility.

Co-locating the analytics team with Murray Irrigation Limited's extension officers for the first six months of the pilot enabled rapid iteration on outputs based on direct farmer feedback. Features that the development team considered important (such as detailed ensemble spread visualisations) were deprioritised in favour of features farmers requested (such as SMS alerts when allocation pathways crossed critical thresholds).

The integration of market pricing with hydrological forecasts created a product that neither component could deliver alone. Pure hydrological forecasts tell irrigators about water availability but not about cost. Pure price forecasts tell irrigators about cost but not about the physical supply drivers creating price movements. The combined product enabled strategic decisions that accounted for both dimensions simultaneously.

Persistent Challenges

Data latency from WaterNSW's dam storage and river flow monitoring systems introduced 24 to 48 hour delays that degraded forecast responsiveness during rapidly changing conditions. During the October 2023 inflow event on the Murrumbidgee, the platform's price forecast lagged the actual market response by three days because the underlying inflow data arrived too slowly to update the ensemble in real time.

The platform struggled with "regime shift" events where market behaviour departed from historical patterns. In March 2024, the announcement of a major almond plantation sale triggered a sudden release of 12,000 megalitres of water entitlements onto the market, causing prices to drop 35% in a week. No amount of hydrological forecasting could have predicted this market-driven event, highlighting the fundamental limitation of supply-side forecasting in markets influenced by speculative and structural factors.

Scaling the pilot beyond its initial geographic footprint faced institutional barriers. Each state water authority operates under different allocation frameworks, data systems, and regulatory requirements. Adapting the platform for Victorian Murray allocations required substantial re-engineering because Victoria's resource manager, Goulburn-Murray Water, uses different allocation determination methodologies and data formats than WaterNSW.

Action Checklist

• Assess current drought management decision processes and identify specific points where probabilistic forecasts could improve lead times
• Evaluate existing hydrological monitoring infrastructure and data accessibility, including real-time telemetry coverage and data latency
• Engage with Bureau of Meteorology seasonal forecasting services to understand baseline forecast skill for your catchment or region
• Pilot integrated forecast-market tools with a subset of water users, ensuring representation across different scales of operation and technical sophistication
• Design user interfaces that translate probabilistic information into decision-relevant signals calibrated to users' operational language
• Establish independent evaluation frameworks before pilot launch, including matched control groups and pre-defined outcome metrics
• Address digital literacy barriers through extension partnerships and face-to-face training alongside digital platform delivery
• Coordinate with neighbouring jurisdictions early to anticipate data format and regulatory compatibility challenges for future scaling

FAQ

Q: How does the Murray-Darling water market work, and why does forecasting matter for market participants? A: The Murray-Darling Basin operates the world's most developed water market, with over AUD 2.8 billion in annual trade. Water entitlement holders receive seasonal allocations that vary based on dam storage and inflows. They can trade these allocations on spot or forward markets. Forecasting matters because allocation and price volatility create significant financial exposure. An irrigator who purchases water at AUD 400 per megalitre in December, only to see prices fall to AUD 150 by March as late-season rains arrive, faces material financial loss. Accurate forecasts enable better timing of purchases and sales, crop planning aligned with expected water availability, and strategic use of forward contracts to manage price risk.

Q: Can these approaches be applied in other Asia-Pacific water markets? A: The methodology is transferable, but market maturity varies dramatically across the region. China's Yellow River Basin has experimented with water rights trading since 2014, and India's groundwater markets operate informally in many states, but neither has the institutional infrastructure to support the kind of integrated forecast-market platform deployed in the Murray-Darling. Southeast Asian river basins managed under the Mekong River Commission face transboundary governance challenges that add complexity. The most promising near-term expansion opportunities are in Chile (where water rights are constitutionally protected and tradeable) and parts of Central Asia where the World Bank is supporting water allocation reform.

Q: What level of forecast accuracy is needed for water market participants to benefit? A: Research from the pilot indicates that forecasts need to outperform climatology (the historical average) by at least 10 to 15% on the Continuous Ranked Probability Skill Score to generate statistically significant economic benefits for users. Below this threshold, the noise in forecasts overwhelms the signal, and users following forecast guidance perform no better than those using historical averages. The pilot achieved 18 to 32% improvement over climatology, well above this utility threshold. Importantly, directional accuracy (correctly predicting whether conditions will be drier or wetter than average) delivers more value than precise quantitative accuracy for most farm-level decisions.

Q: What are the costs of implementing a drought forecasting and water market analytics platform? A: The pilot's total cost of AUD 6 million over three years included platform development, data integration, user engagement, and independent evaluation. For an operational deployment serving 200 to 500 users, annual operating costs are estimated at AUD 1.2 to 2 million, covering data acquisition, model maintenance, platform hosting, and user support. Per-user costs decline significantly with scale. The ABARES evaluation found average user savings of AUD 38,000 per farm per year, suggesting a strongly positive return on investment at platform costs of AUD 3,000 to 6,000 per user annually.

Q: How do environmental water managers benefit from drought forecasting? A: Environmental water holders manage allocated water entitlements to achieve ecological outcomes such as waterbird breeding events, fish spawning, and floodplain vegetation health. These outcomes are highly sensitive to flow timing and magnitude. Improved seasonal forecasts enable environmental water managers to schedule releases that coincide with natural inflow pulses, amplifying ecological impact without additional water purchases. The Commonwealth Environmental Water Holder's participation in the pilot demonstrated a 15% increase in effective environmental watering volume through optimised delivery timing, representing significant value given that environmental water entitlements in the Murray system are valued at several billion dollars.

Sources

• Murray-Darling Basin Authority. (2024). Basin Plan Annual Report 2023-24: Water Resource Assessments and Allocation Outcomes. Canberra: MDBA.
• Australian Competition and Consumer Commission. (2021). Murray-Darling Basin Water Markets Inquiry: Final Report. Canberra: ACCC.
• Bureau of Meteorology. (2025). Seasonal Streamflow Forecasting Service: Methodology and Verification Report. Melbourne: BOM.
• ABARES. (2025). Evaluation of Integrated Drought Forecasting and Water Market Analytics in the Southern Murray-Darling Basin. Canberra: Australian Bureau of Agricultural and Resource Economics and Sciences.
• CSIRO. (2023). Climate Change Projections for the Murray-Darling Basin: Rainfall, Temperature, and Water Availability. Canberra: CSIRO.
• Wheeler, S. A., Loch, A., Zuo, A., and Bjornlund, H. (2024). Water Market Efficiency and Information Asymmetry in the Murray-Darling Basin. Water Resources Research, 60(3).
• Commonwealth Environmental Water Office. (2025). Annual Report on Environmental Watering Outcomes 2024-25. Canberra: CEWO.