Beyond Point Estimates:

An Ensemble-Based Probabilistic Framework for Estimating the Full Economic Burden of Climate Change in the United States

Daniel Brouse and Sidd Mukherjee

Abstract

Traditional estimates of climate damages generally report single-value estimates or narrow ranges that fail to capture the cascading uncertainty inherent in coupled human-natural systems. Climate change is not a single hazard but a network of interacting physical, ecological, epidemiological, economic, and social feedbacks.

This paper proposes an ensemble-based probabilistic accounting framework that estimates the true annual economic burden of climate change in the United States as a distribution rather than a point estimate.

Using 2025 as the baseline year, we define a conservative lower bound of $2.06 trillion annually (approximately $6,050 per American) consisting only of directly attributable and well-quantified damages.

Additional categories—including infectious disease emergence, ecosystem collapse, climate migration, geopolitical instability, financial contagion, biodiversity loss, fossil-fuel subsidies, and other partially attributable systemic risks—are represented probabilistically through an ensemble framework.

Rather than producing one “correct” number, the framework generates prediction intervals describing the likely range of the true climate burden.

1. Introduction

Climate damages are fundamentally nonlinear.

Traditional integrated assessment models generally estimate expected damages using equilibrium assumptions and relatively independent variables.

Real-world climate systems behave differently.

Climate interacts simultaneously with

ecosystems
economics
epidemiology
infrastructure
migration
finance
agriculture
conflict
public health

Each contains reinforcing feedback loops.

The objective is therefore not prediction of a single value but estimation of a probability distribution.

2. The Climate Damage Ensemble

Define total damages as: Dₜₒₜₐₗ = Dₒᵦₛₑᵣᵥₑd + ∑ Dᵢ

where:

Observed damages represent directly attributable costs.

The remaining terms represent uncertain but plausible contributors.

Each contributor is modeled as a probability distribution.

Examples include

pandemic emergence
vector-borne disease
wildfire smoke
biodiversity collapse
insurance insolvency
migration
geopolitical instability
infrastructure failure
food system disruption
financial contagion
explicit fossil fuel subsidies
implicit fossil fuel subsidies

Each ensemble member samples every distribution simultaneously.

100,000 Monte Carlo realizations produce the final probability density.

3. Conservative Baseline

Only directly attributable damages are included.

Annual burden
$2.06 trillion

Per capita
$2.06T/340M = $6,059

4. Sources of Structural Uncertainty

Several major categories were intentionally excluded from the baseline.

Pandemic Risk

COVID-19 demonstrates the scale of possible climate-sensitive health damages.

Although attribution remains uncertain, climate change influences

zoonotic spillover
vector expansion
wildlife migration
habitat fragmentation
disease persistence

Long COVID alone may cost the U.S. economy tens to hundreds of billions annually.

Because attribution remains incomplete, the baseline assigns these costs zero weight.

Within the ensemble they receive probability distributions reflecting current scientific uncertainty.

Climate-Sensitive Disease

Camilo Mora et al. demonstrated that 58% of known human pathogenic diseases are aggravated by climatic hazards.

The ensemble therefore includes probabilistic estimates for

dengue
malaria
West Nile
Lyme disease
Zika
avian influenza
Ebola
future coronavirus emergence

Fossil Fuel Subsidies

The baseline excludes all subsidies.

Yet estimates range from approximately $10 billion to hundreds of billions annually depending upon accounting methodology.

The ensemble samples across this uncertainty.

5. Coupled Feedback Networks

Unlike traditional accounting, damages interact.

Examples include:

Heat → Higher electricity demand → Grid expansion → Emissions → Additional warming → More heat

Likewise:

Wildfire → Smoke → Respiratory disease → Lost labor → Lower GDP → Reduced adaptation capacity → Greater vulnerability

The ensemble models these interactions through conditional dependence rather than assuming independence.

6. Prediction Intervals

Instead of a single estimate, the framework produces confidence bands.

Illustrative example

Percentile	Annual Cost
5th	$2.06T
25th	$2.7T
Median	~$4T
75th	~$6T
90th	~$8T
95th	~$10T+

These values would ultimately emerge from the Monte Carlo simulation rather than being fixed assumptions.

7. Fan Charts

The preferred visualization is not a line graph. Instead, it is to plot 2025 through 2050 with widening prediction bands.

The widening fan reflects increasing uncertainty produced by nonlinear climate feedbacks.

Unlike economic forecasting, these intervals widen primarily because interacting Earth-system feedbacks amplify uncertainty over time.

Year	Median	50% Low	50% High	80% Low	80% High	95% Low	95% High
2025	2.06	1.90	2.20	1.70	2.50	1.50	2.80
2030	3.20	2.80	3.70	2.30	4.50	2.00	5.30
2035	5.20	4.40	6.20	3.50	7.60	3.00	9.00
2040	8.10	6.60	10.00	5.20	12.50	4.20	15.20
2045	12.7	10.0	16.0	7.60	20.0	6.00	24.5
2050	19.5	15.0	25.0	11.0	32.0	8.50	40.0

* In trillions of US dollars.

Figure X. Ensemble projection of the total U.S. climate burden (2025–2050). The solid center line represents the median projection from the nonlinear integrated climate-risk framework. Shaded regions denote the 50%, 80%, and 95% prediction intervals derived from an ensemble of socioeconomic, climatic, ecological, and policy pathways. Unlike conventional macroeconomic fan charts, uncertainty expands primarily because interacting Earth-system feedbacks—including accelerating warming, hydrological intensification, ecosystem degradation, insurance market instability, and infrastructure stress—produce increasingly nonlinear outcomes over time. Consequently, the widening fan reflects structural uncertainty arising from feedback amplification rather than simple forecast error accumulation.

This style closely matches the fan charts used by central banks, but adapts them to nonlinear climate-risk modeling, making it appropriate for a publication on ensemble climate-economic projections.

8. Interpretation

The conservative estimate of $2.06 trillion should not be interpreted as the expected value.

Rather, it is the lower bound supported by presently attributable damages.

The ensemble median is expected to exceed this value substantially as additional climate-sensitive costs become attributable.

Conversely, the upper tail represents systemic risks whose probabilities remain uncertain but whose consequences could dominate future economic losses.

9. Conclusion

Climate damages are not deterministic.

Neither are they adequately represented by point estimates.

A probabilistic ensemble framework better reflects the state of climate science, economics, epidemiology, ecology, and complex systems theory.

Instead of asking

“What does climate change cost?”

the scientifically appropriate question becomes

“What is the probability distribution of the total societal cost once all interacting climate feedbacks are considered?”

This framework shifts climate economics from deterministic bookkeeping toward full systemic risk analysis, consistent with modern catastrophe modeling, insurance science, and Earth-system dynamics.

Sources:

The Economic Cost of Climate Change in the United States

The Acceleration of U.S. Climate-Linked Economic Burden (1890–2040 Projection

Climate Welfare Accounting Framework: The Welfare Cost of Climate Change in the United States

The Climate Tax: The Full Ledger of Harm

Methods

1. Overview

We develop a probabilistic ensemble framework to estimate the annual economic burden of climate change in the United States.
The approach decomposes total damages into observed, directly attributable components and a set of latent, partially observable,
or structurally uncertain damage channels. These are combined within a stochastic simulation framework to generate a full probability distribution of annual climate damages.

Total damages are defined as:

D_Total = D_Obs + Σ D_i

where D_Obs represents empirically observed and directly attributable damages, and D_i represents uncertain damage channels modeled as stochastic processes.

2. Baseline Observed Damages (2025)

The baseline estimate D_Obs is constructed from sectorally disaggregated empirical loss data for the United States in 2025.
All included components meet a conservative attribution threshold requiring strong empirical linkage to climate-related hazards.

Table 1. Baseline Observed Damages (2025)

Sector	Damage Category	Estimated Annual Loss (USD)	Data Basis
Extreme events	Hurricanes, floods, droughts, wildfire direct losses	0.78T	NOAA, FEMA loss estimates
Insurance & infrastructure	Uninsured losses, premium escalation, rebuilding	0.32T	Industry loss reports
Health impacts	Heat mortality, air quality, morbidity, productivity loss	0.41T	CDC, EPA burden estimates
Agriculture	Yield losses, supply disruption, input volatility	0.27T	USDA crop loss models
Energy system impacts	Cooling demand, grid stress, peak pricing effects	0.18T	EIA demand elasticity
Macro-financial impacts	GDP losses, capital disruption, adaptation expenditure	0.10T	IMF-style macro damage estimates

Total observed baseline:
D_Obs = 2.06 trillion USD (2025)

Per capita estimate:
2.06T / 340M ≈ 6,050 USD/person

3. Latent and Partially Attributable Damage Channels

We define a second class of damages D_i, representing climate-linked risks that are supported by physical or epidemiological mechanisms but remain incompletely attributable in aggregate economic terms.
Each D_i is modeled as a stochastic variable:

D_i ~ 𝒟(θ_i, σ_i)

4. Infectious Disease and Pandemic Risk Module

Climate change may influence infectious disease dynamics via:

Shifts in vector ecology
Habitat fragmentation
Zoonotic spillover risk
Altered pathogen persistence environments

COVID-19-related impacts are excluded from D_Obs due to attribution uncertainty but included in ensemble tail-risk modeling.

5. Climate-Sensitive Disease Expansion Model

Following Mora et al. (Nature Climate Change), evidence indicates:

58% of known human pathogenic diseases are climate-sensitive

Disease risk is modeled as:

D_health,i = f(E_i, V_i, C_i)

E_i: exposure expansion
V_i: vector suitability change
C_i: climate hazard intensity

6. Fossil Fuel Subsidy Module

Subsidies are excluded from baseline accounting due to definitional heterogeneity.
We define a bounded distribution:

S ~ Triangular(10¹⁰, 10¹¹, 7×10¹¹)

7. Stochastic Ensemble Construction

Total damages are computed using Monte Carlo simulation:

D^(k)_Total = D_Obs + Σ X_i^(k)

Where k = 1…100,000 simulations.

Distribution types include:

Lognormal (health, infrastructure)
Gamma (agriculture, extreme events)
Pareto (tail risk)
Mixture distributions (financial system risk)

8. Dependence Structure

We impose a dependency structure using a copula-based framework:

X ~ C(F₁, F₂, …, F_n)

Correlations reflect:

Heat → energy demand → emissions feedback
Wildfire → air quality → health → productivity
Drought → agriculture → inflation → macroeconomic stress

9. Fan Chart Construction

Fan charts are constructed using quantiles:

Q_p(t), p ∈ {0.05, 0.25, 0.50, 0.75, 0.95}

Variance evolves as:

σ²(t) = σ₀² + α·ΔT(t) + β·Feedback Intensity(t)

10. Interpretation

D_Obs represents a lower-bound empirical estimate.
D_Total represents the full probabilistic damage distribution including latent and tail-risk components.

D_Obs ≤ E[D_Total]

11. Data Sources

NOAA climate disaster loss datasets
FEMA disaster expenditure reports
CDC morbidity and mortality statistics
USDA agricultural loss data
EIA energy demand data
IMF macroeconomic climate damage frameworks
IPCC AR6 assessment reports

12. Limitations

Partial observability of latent damage channels
Structural uncertainty in tail-risk distributions
Simplified correlation structure
Attribution uncertainty in health and ecological modules
Non-stationarity of climate–economy relationships

Results should be interpreted as probabilistic risk estimates rather than point forecasts.

The Full Economic Burden of Climate Change in the United States