Treatment of Uncertainty in Hydro-economic Modeling of the Murray–Darling Basin, Australia

Presentation 1 Feb 2022

Jeff Connor

Centre for Market, Value and Inclusion

UniSA Business School

​

Objective

Present on the treatment of Uncertainty in Hydro-economic models applied to the Murray-Darling Basins

The Murray-Darling Basin

• 1/7^th of Australian land area
• 2 million irrigated hectares
• Extensive dams and regulation
• Iconic wetlands
• Flows, diversions, dams a source of conflict between States for more than 100 years

MDB has highly variable water supply & severe drought

Murray-Darling Basin Inflows

Historic infrastructure & institutional drought response

Federation Drought 1895-1902:

• Cooperative States agreement to ensure navigable river
• major storages and weirs

​

Mid-century Drought 1938-45:

• Inter-basin transfer/hydro-power

1980’s-90’s water institution developments – response to growing allocation

• salinity threatened end of system urban water supply

​

• Low flow Environmental threat to River Murray Delta

​

National Water Reforms 1994; National Water Initiative 2004

​

​

Capped diversions at 1993/4 levels

• Individual water property rights
• Volumetric & metered
• Automatically adjusted “allocation” in scarce supply (shares of storage)
• Tradable Independent of land
• Deliver critical human water & required conveyance first (municipal, industrial + stock)
• Aspiration for cost effectiveness in infrastructure investment
• Provide information and plan for risks of climate change, other threats to water supply
• Incentive payments for State to comply

(Connor and Kaczan, 2013)

​

Context: “Millennium Drought -2000-09” worst in recorded weather history, tested water reforms introduced in prior decade

Worst drought

in recorded

history led to

empty storages

and emergency

planning

In 2007

water trade provided irrigators an adaptation mechanism

1/3 of all irrigation traded at height of drought 2008/9

​

improved farm income in drought by $0.5-1 billion (10-20%)

​

​

​

Water traded from low value annual to high value perennial irrigation

​

- allowed sustaining high value perennials

​

- provided higher income to annual crop farmers than irrigation would have

To Fruit & wine

​

​

​

From rice, pasture, grain

(Connor et al, 2014)

(Connor and

Kaczan, 2013)

MDB Hydro-economic models reviewed

66 studies most >= 2000

ag water most common focus

Growing environmental focus

Simulations, optimisations

Partial and general equilibrium

Dynamic and static

Catchment to basin scale

​

​

​

Types of uncertainty: Stochastic v espistemic

unreducible

reducible

uncertainty arises from many sources:

1) underlying framing of the question and context that dictates scope of benefits/costs of water use considered, market, and institutional representations, value metrics, spatial and temporal boundaries and granularity of modelling;

2) the selection of model structure: algorithms, and interactions;

3) parameterization of uncertainty including choices of ranges and functional forms of uncertainties represented

Approaches to treat uncertainty in MDB ad other HEMs

​

Statistical confidence intervals quantify uncertainty response to water scarcity statistically

Climate variability and change (natural stochasticity) treatment in MDB HEMs

• State contingent (also know as two stage recursive) optimisation models most common climate treatment (14 reviewed models)
• e.g. Quiggin et al. 2010; Crean et al. 2013; Adamson and Loch 2014; Loch and Adamson 2015; Connor et al, 2009, 2012
• Econometric simulation is the 2^nd most common approach

State contingent (two stage recursive) optimisation climate treatment in MDB HEMs

• Danzig two-step programming with recourse (see, e.g., Danzig, 1955; McCarl et al., 1999).
• 1st stage = long-run capital investment decisions: irrigation equipment, permanent plantings
• conditional on probabilistic expectations of stochastic annual water allocation and water price.
• 2^nd stage = variable input levels (water application; land with irrigation capital to actually irrigate or leave fallow); made when stochastic water allocations & price levels for a given year are revealed

​

��State contingent optimisation climate treatment preoptimization steps: compute model water scarcity and price “states” and probabilities

Typical state contingent climate model response to water scarcity

• Connor et al, 2009 produced prior to millennium drought modelled very similar to observed responses to zero and near zero water allocations for Lower Murray
• E.g. buy water from other regions at high price, fallow nearly all annual crops, sacrifice 20% of perennial plantings

​

Most common treatment of parameter uncertain – one at a time (OAT) uncertainty analysis

• OAT uncertainty analysis involves varying key uncertain parameters alone or in combinations
• Example from …

​

Monte Carlo methods for uncertainty analysis

• Define uncertain parameter ranges and distribution type
• Repeated random draws from specified ranges
• Repeated recalculation of outcomes of interest
• Provides confidence interval estimates around expected value of outcomes
• Rowan et al. applied the approach in a dynamic profit maximization where losses and profits are carried over year to year

​

​

The frontier in treating uncertainty in HEMs:�Global sensitivity analysis for HEM

Monte Carlo simulation with draws from specified distributions of multiple uncertain model parameters simultaneous

​

Allows quantifying the variability of outputs from direct and interactive effects of multiple uncertain inputs.

​

Also allows “apportioning” - determining relative importance multiple input uncertainties on outcome uncertainty

​

Generally hasn’t been applied to HEMs to date, possible because of computational challenge

(requires 1000’s to 10,000’s of model runs),

​

though it is increasing common in Ecosystem services literature

Global sensitivity analysis water quality and carbon PES example (in submission

• First step is to specify ranges and functional forms of uncertain model inputs

​

​

Global sensitivity analysis water quality and carbon PES example �output 1�outcome�distributions

Global sensitivity analysis water quality and carbon PES example output 1�attribution of input uncertainty influence on outcomes