Skip to main content

Climate & Earth Sciences

Operations Research in Environmental Monitoring & Management

From IPCC assessment reports selecting representative climate model ensembles, to Argo float networks maximising ocean observation coverage, to US Forest Service wildfire dispatch under uncertainty — three operations research problems that shape how we understand and respond to the planet's most pressing environmental challenges.

Science Context

The Coupled Model Intercomparison Project Phase 6 (CMIP6) produces output from roughly 100 Global Climate Models (GCMs) spanning 50+ modelling centres. Using all models is computationally infeasible for regional impact studies. A randomly chosen subset risks being unrepresentative — missing key uncertainty dimensions like climate sensitivity or precipitation patterns. The challenge is to select k models that balance individual performance, inter-model independence, and diversity across the projection space.

Problem type: p-dispersion / maximum diversity selection. Select k models from a pool maximising the minimum pairwise distance in a multi-dimensional climate projection space, subject to a performance threshold ensuring each selected model meets minimum observational fidelity.

Mathematical Formulation max min{i,j in S} d(m_i, m_j)
s.t. |S| = k // select exactly k models
     perf(m_i) ≥ θ   ∀ m_iS // performance threshold
// d() = Euclidean distance in PCA projection space

Ensemble Solver

4
30%
Random k
Greedy Diversity
p-Dispersion
Adjust parameters and click Solve.
Evidence Base
  • Herger, N., et al. (2018). Selecting a climate model subset to optimise key ensemble properties. Earth System Dynamics, 9, 135-151. Published
  • Merrifield, A., et al. (2023). A weighting scheme applied to the CMIP6 ensemble. Geoscientific Model Development, 16, 4715-4747. Published
  • Knutti, R., Furrer, R., Tebaldi, C., Cermak, J., & Meehl, G. (2010). Challenges in combining projections from multiple climate models. IPCC Expert Meeting on Assessing and Combining Multi Model Climate Projections. Published
Science Context

The Argo programme maintains ~4,000 autonomous profiling floats across the global ocean, measuring temperature and salinity. Weather station networks face similar tradeoffs: each sensor costs money to deploy and maintain. The goal is to maximise information gain about the full spatial field given a limited budget of k sensors. Because mutual information is submodular, a greedy algorithm provides a provable (1 - 1/e) ≈ 63% approximation guarantee (Krause et al., 2008).

Problem type: Submodular maximisation / maximum coverage. Place k sensors on a discrete grid to maximise the conditional mutual information between observed and unobserved locations, subject to a cardinality constraint.

Mathematical Formulation max f(S) = H(Y_V) - H(Y_V | Y_S)
s.t. |S| ≤ k // budget constraint
// H() = differential entropy
// V = all grid cells, S = sensor set
// Greedy: (1-1/e) approximation guarantee

Sensor Solver

5
Random
Greedy (Info Gain)
Grid Uniform
Adjust parameters and click Solve.
Evidence Base
  • Krause, A., Singh, A., & Guestrin, C. (2008). Near-optimal sensor placements in Gaussian processes: Theory, efficient algorithms and empirical studies. Journal of Machine Learning Research, 9, 235-284. Published
  • Argo Science Team. Argo: Global ocean observing with profiling floats. Operational programme maintained by JCOMMOPS with ~4,000 active floats. Operational
Science Context

Western US wildfire seasons now routinely involve dozens of simultaneous fires competing for finite suppression resources: hotshot crews, air tankers, engines, and dozers. Fire growth is stochastic — driven by wind, terrain, and fuel moisture — and new ignitions arrive as random events. The US Forest Service and CAL FIRE must decide, in real-time, how to allocate resources across fires to minimise total expected damage. Integer programming formulations by Donovan & Rideout (2003) demonstrated 15-20% cost savings over priority-based heuristics.

Problem type: Stochastic multi-resource assignment. Assign heterogeneous suppression resources to concurrent fires to minimise total containment plus damage costs, subject to resource capacity, stochastic fire growth, and random new ignitions.

Mathematical Formulation min Σ (containment_cost + damage_cost)
s.t. Σfires x_ijcapacity_j   ∀ resource j
     growth(fire_i) ~ Stochastic(wind, fuel, terrain)
     x_ij ∈ {0,1} // binary assignment

Wildfire Dispatch Solver

Air Tankers (3)
Crews (4)
Priority (Greedy)
ILP (Optimal)
Predictive (Lookahead)
Click Run Simulation to dispatch resources.
Evidence Base
  • Donovan, G. & Rideout, D. (2003). An integer programming model to optimize resource allocation for wildfire containment. Forest Science, 49(2), 331-335. Operational
  • MacLellan, J. & Martell, D. (1996). Basing airtankers for forest fire control in Ontario. Operations Research, 44(5), 677-686. Published

Explore More Applications

See how the same mathematical families — facility location, combinatorial optimization, stochastic programming — apply across agriculture, healthcare, logistics, and energy.

Portfolio
ESC