## wpiea2024141-print-pdf

## Source details

**Canonical URL:** [wpiea2024141-print-pdf](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024141-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024141-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024141-print-pdf.pdf.json)

---

### Purpose and scope
- Aim: bridge economic and climate literature with a simple framework to analyze forward-looking losses from floods and tropical cyclones (TCs).
- Coverage: forward-looking losses estimated for 183 countries for floods and 89 countries for TCs under three IPCC scenarios (SSP1-2.6, SSP2-4.5, SSP5-8.5).
- Hazards considered: riverine and coastal floods (excluding rainfall-induced flooding) and tropical cyclones.
- Data sources: hazards from Jupiter Intelligence; exposure proxied by spatially disaggregated GDP with global coverage; vulnerability from published damage functions (Huizinga and others (2017) for floods; Eberenz and others (2021) for TCs).

### Conceptual framework and components
- Damages arise from interaction of three components:
  - Hazard severity: projections under IPCC scenarios using Jupiter Intelligence (leveraging GCMs and proprietary models).
  - Exposure: spatially disaggregated GDP consistent with future socio-economic projections, available decennially to 2100.
  - Vulnerability: damage functions translating hazard magnitude into damages—Huizinga et al. (2017) for floods; Eberenz et al. (2021) for TCs.
- Outputs: damage rates (loss of value of assets, expressed in percent of asset value before event) computed at location level (latitude/longitude) and aggregated to country-level expected annual damage rates.

### Data: hazards, vulnerability, exposures
- Flood hazard data (Jupiter Intelligence):
  - River and coastal flood depth projections from GCM inputs.
  - Unified estimate of expected water levels combining sea-level rise, storm surge and tides, long-term lake levels.
  - Data provided for mean flood depth and fraction of land flooded in 3 arcseconds grids for return periods: 10, 20, 50, 100, 200, and 500-year.
  - Baseline: climate for 10-year period centered on 1995. Projections: 2020-2100 in 5-year increments under SSP1-2.6, SSP2-4.5, SSP5-8.5.
  - Rainfall-induced flooding not available.
  - Jupiter’s dataset includes global data for 100,000 locations selected to maximize global and country-level GDP coverage (coverage of 97 percent of global GDP in year 2040).
- Tropical cyclone (TC) hazard data:
  - Maximum 1-minute sustained wind speed at 10 meters above ground level.
  - Produced using GCMs and a synthetic TCs model; used only where TCs have been observed historically.
  - Data include atmospheric variables and use multiple GCMs from CMIP5 and CMIP6.
- Vulnerability — damage functions:
  - Floods: Huizinga et al. (2017) piecewise linear damage functions by continent and asset type (buildings and contents, transport, infrastructure, agriculture).
  - TCs: Eberenz et al. (2021) regionally calibrated sigmoidal function:
    - f = v_n^3 / (1 + v_n^3), where v_n = MAX[(V − V_thresh), 0] / (V_half − V_thresh).
    - V is 1 min sustained wind speed; V_thresh is threshold below which damage is zero; V_half is wind speed at which impact reaches 50 percent.
    - Calibration note: red dotted line denotes windspeed of 25.7 m/s, or 92.5 km/h, threshold below which damage rate is assumed zero.
- Exposures:
  - Downscaled GDP from Murakami and others (2021) gridded GDP at 30 arcminutes available 1980–2100 by 10-year intervals.
  - 1980–2010: downscaled actual GDPs by country. 2020–2100: downscaled projected GDPs under SSP1-2.6; SSP2-4.5; SSP5-8.5.
  - Gridded GDP centroids correspond to grid cell latitude/longitude.

### Methodology (calculation of damages)
- Focus: direct damages only (exclude indirect damages such as business interruption, spillovers, second round effects).
- Damage rates: loss of value of assets expressed in percent of value before hazard.
- Notation (as provided):
  - depth_{c,i,RP,t,s}, frac_{c,i,RP,t,s}, wind_{c,i,RP,t,s} denote flood depth, fraction of land flooded, maximum sustained windspeed respectively in country c, location i, return period RP, projection year t and scenario s.
  - Damage function for floods: df_floods(depth).
  - Damage function for storms: df_storms(wind).
- Location-level damage rate formulas:
  - Floods: d_{c,i,RP_j,t,s} = frac_{c,i,RP_j,t,s} * df_floods(depth_{c,i,RP_j,t,s}).
  - TCs: d_{c,i,RP_j,t,s} = df_storms(wind_{c,i,RP_j,t,s}).
- Expected annual damage rate (location level):
  - E[d_{c,i,t,s}] = Σ_{j ∈ RP̅} (1/RP_j − 1/RP_{j+1}) * (d_{c,i,RP_j,t,s} + d_{c,i,RP_{j+1},t,s}) / 2  + d_{c,i,RP_500,t,s} / RP_500,
    where RP̅ = {10,20,50,100,200,500}.
- Country-level aggregation:
  - GDP_{c,i,t,s} is gridded GDP at location i; GDP_{c,t,s} total GDP.
  - E[D_{c,t,s}] = Σ_{i=1}^{n} E[d_{c,i,t,s}] * (GDP_{c,i,t,s} / GDP_{c,t,s}).
- Baseline defined: hazard for the 10-year period centered around 1995 and GDP in 2000.
- Flood exposures split into built-up and non-built-up using Copernicus BuiltUp-CoverFraction-layer (2019) downsampled to 5 arcminutes; built-up assets combine residential, commercial, industrial damage functions equally; non-built-up use agriculture and infrastructure functions.

### Key findings — Floods (high-level)
- General:
  - Most countries experience increased flood damages by mid-century; damages higher under more severe climate scenarios.
  - Flood and TC risks are unevenly distributed globally; highest flood damages concentrated in tropical and subtropical areas and in small land area countries.
- Country counts and medians (Table 3 — flood damages):
  - 2050 SSP1-2.6: 107 countries increasing (0.35), 60 countries decreasing (0.22).
  - 2050 SSP2-4.5: 115 countries increasing (0.28), 52 countries decreasing (0.32).
  - 2050 SSP5-8.5: 123 countries increasing (0.32), 45 countries decreasing (0.32).
  - 2100 SSP1-2.6: 121 countries increasing (0.36), 47 countries decreasing (0.26).
  - 2100 SSP2-4.5: 117 countries increasing (0.36), 51 countries decreasing (0.27).
  - 2100 SSP5-8.5: 124 countries increasing (0.41), 44 countries decreasing (0.25).
- Aggregate shares:
  - Increasing flood risks for countries representing 74-80 percent of global GDP in 2020.
  - Depending on scenario, 58-67 percent of the countries (representing 74-80 percent of global GDP in 2020) display an increase in expected annual country-level damage rate as of 2050 relative to the baseline.
  - 25-33 percent of countries (representing 18-25 percent of global GDP in 2020) display a decrease in flood risks; the rest display no change.
- Median country-level damage rates (countries with increasing flood risks):
  - 2050: 0.28-0.35 percent (depending on scenario).
  - 2100: 0.36-0.41 percent (depending on scenario).
- Median country-level damage rates (countries with decreasing flood risks):
  - 2050: 0.22-0.32 percent (depending on scenario).
  - 2100: 0.25-0.27 percent (depending on scenario).
- Distributional and geographic patterns:
  - Largest relative changes in damage rates: South America, Africa, Southeast Asia.
  - Median relative change across countries: 8.46 percent by 2050; 17.37 percent by 2100.
  - Median levels change: 0.02 percentage points in 2050; 0.03 percentage points in 2100.
- Drivers:
  - Country-level damage rate changes mainly driven by changes in hazard severity (flood depth).
  - Some high-latitude regions (e.g., Canada and some parts of Europe) may see decreased riverine flood risks due to drier climates.

### Key findings — Tropical cyclones (Section 5.2)
- Sample and aggregate shares:
  - Projections for 89 countries with historical TC records.
  - Depending on scenario, 66-67 percent of the exposed countries (exposed areas representing 40-42 percent of global GDP in 2020) display an increase in country-level expected damage rate in 2050 relative to baseline.
  - 16-17 percent (exposed areas representing 4-5 percent of global GDP in 2020) display a decrease in TC risks.
  - Number of countries with increasing damage rates highest for SSP5-8.5.
- Key statistics (Table 4 — TC damages):
  - 2050 SSP1-2.6: 59 countries increasing (median 0.11 percent); 15 decreasing (median 0.0019 percent).
  - 2050 SSP2-4.5: 60 countries increasing (median 0.11 percent); 14 decreasing (median 0.0006 percent).
  - 2050 SSP5-8.5: 59 countries increasing (median 0.12 percent); 15 decreasing (median 0.0025 percent).
  - 2100 SSP1-2.6: 57 countries increasing (median 0.12 percent); 17 decreasing (median 0.0015 percent).
  - 2100 SSP2-4.5: 60 countries increasing (median 0.13 percent); 14 decreasing (median 0.0012 percent).
  - 2100 SSP5-8.5: 63 countries increasing (median 0.16 percent); 11 decreasing (median 0.0007 percent).
- Distributional patterns and extremes:
  - Variability of damage rates increases for more severe scenarios at the end of the century for countries with intensifying risk.
  - Number of extreme observations is overall stable for countries with increasing and decreasing damage rates; all extreme values shown are island countries.
  - Small island countries are among the top five countries by damage rate for 2050 and 2100 across scenarios.
- Geographic concentration:
  - Highest TC damage rates concentrated in the Caribbean, South and Southeast Asia, Eastern Africa, and Oceania.
  - Median relative change: 6.95 percent by 2050; 14.20 by end of century.
  - Largest changes across North and Central and South America, and coastal countries from Eastern Africa to Oceania.
- Notes:
  - When only parts of a country are exposed to TCs, only the GDP within those grids is counted.

### Practical applications and policy relevance
- Diagnostic tool: forward-looking, granular country-level indicators of flood and TC physical risks to complement backward-looking EM‑DAT–based indicators.
- Macro-financial analysis: damage estimates can be inputs to macro models to generate macro-financial scenarios incorporating climate physical risks (e.g., for banks’ stress tests).
- Micro- and sectoral applications: damage/loss components can be integrated with borrower-level data for micro approaches to climate risk analysis.
- Adaptation and mitigation appraisal: hazard damages can be used in macroeconomic models to evaluate macroeconomic and financial implications of alternative investment and financing programs.

### Limitations and methodological caveats (Annex VIII)
- Hazards data and model uncertainty:
  - Wide range of climate models yields sizable model uncertainty.
  - Jupiter Intelligence uses CMIP6 and CMIP5 models, selecting based on availability, ensemble members, and model quality.
  - Jupiter provides uncertainty bounds for hydrological model parameter uncertainty (flood depth) and GEV fitting uncertainty (wind); future work could incorporate these bounds.
  - Spatial resolution: floods at 3 arcseconds (90 meters by 90 meters at equatorial sea level); TC spatial resolution 15 arcminutes before bilinear interpolation.
  - 100,000 locations maximize economic coverage but geographical coverage is not always ideal; 100K locations give gridded GDP coverage above 95 percent.
- Assumptions affecting estimates:
  - Lowest return period available is 1-in-10 years; damage rates for return periods below 1-in-10 years set to zero, implying underestimation where low-severity recurrent events matter.
  - Correlation of hazard data across locations assumed zero, implying potential overestimation of damages.
  - Net effect of these assumptions on damages is unclear.
- Aggregation and compound events:
  - Study evaluates fluvial and coastal floods by taking maximum depth between river and coastal outputs; joint hazard distributions where hazards are coincident may be underestimated.
  - Floods and TC damages are calculated in isolation; caution when combining them.
  - Compound floods and cascading multi-hazard effects are not modeled here.
- Exposure and vulnerability limitations:
  - Downscaled GDP is a proxy for real economic exposures; granular geospatial asset data would be ideal.
  - TC damage functions rely on EM-DAT calibrated reported damages, inheriting EM-DAT biases and under-reporting; largest uncertainties for North-West Pacific regions.
  - Damage functions focus on direct damages; indirect damages excluded.
- Adaptation and mitigation:
  - Study does not explicitly account for future adaptation and mitigation measures, providing an upper-bound estimate of damages.
  - Exception: for floods in the USA, England, the Netherlands and Germany, hazard data assume current levee protections hold up to a 100-year flood in the future.

*Source: wpiea2024141-print-pdf*

### 1. Introduction ........................................................................................................

### 1. Introduction

### Purpose and scope
- Aim: bridge economic and climate literature with a simple framework to analyze forward-looking losses from floods and tropical cyclones (TCs).
- Coverage: forward-looking losses estimated for 183 countries for floods and 89 countries for TCs under three IPCC scenarios (SSP1-2.6, SSP2-4.5, SSP5-8.5).
- Hazards considered: riverine and coastal floods (excluding rainfall-induced flooding) and tropical cyclones.
- Data sources highlighted: hazards from Jupiter Intelligence; exposure proxied by spatially disaggregated GDP with global coverage; vulnerability from published damage functions (Huizinga and others (2017) for floods; Eberenz and others (2021) for TCs).

### Conceptual framework and components
- Damages arise from the interaction of three components:
  - Hazard severity: projections of individual hazards under IPCC scenarios using Jupiter Intelligence (leveraging GCMs and proprietary models).
  - Exposure: spatially disaggregated GDP consistent with future socio-economic projections, available decennially to 2100.
  - Vulnerability: damage functions translating hazard magnitude into damages—Huizinga et al. (2017) for floods; Eberenz et al. (2021) for TCs.
- Outputs: damage rates (defined as loss of value of assets, expressed in percent of asset value before event) computed at location level (latitude/longitude) and aggregated to country-level expected annual damage rates.

### Key findings (high level)
- General trends:
  - Most countries considered experience increased flood and TC damages by mid-century.
  - For most countries, damages are higher under more severe climate scenarios (positive correlation with global warming).
  - Flood and TC risks are unevenly distributed globally.
- Flood-specific findings:
  - Increasing forward-looking country-level annual damage rates for most countries representing 74-80 percent of global GDP in 2020.
  - Depending on scenario, 58-67 percent of the countries (representing 74-80 percent of global GDP in 2020) display an increase in expected annual country-level damage rate as of 2050 relative to the baseline.
  - Highest damages concentrated in tropical and subtropical areas and in small land area countries (greater asset concentration).
  - Largest relative changes in damage rates located in South America, Africa, and Southeast Asia.
- TC-specific findings:
  - Countries accounting for around 41 percent of 2020 global GDP are expected to experience increased TC damage rates.
  - Among historically exposed countries, 66-67 percent display an increase in TC damage rates from baseline to 2050.
  - Number of countries with increasing TC damage rates grows through the years for SSP2-4.5 and SSP5-8.5, highest for SSP5-8.5.
  - Variability of damage rates increases for more severe scenarios toward the end of the century for countries with increasing risks.
  - Highest TC damage rates concentrated in the Caribbean, South and Southeast Asia, Eastern Africa, and Oceania.

### Methodological choices and limitations noted
- Hazards: Jupiter Intelligence data combine multiple GCMs and proprietary modeling; available for SSP1-2.6, SSP2-4.5, SSP5-8.5. Hazards data under SSP1 RCP 1.9 and SSP3 RCP 7.0 are not readily available.
- Flood damage modeling: uses inundation depth as primary driver and applies Huizinga et al. (2017) continent- and asset-type damage curves.
- TC damage modeling: uses maximum sustained windspeed as primary driver and applies Eberenz et al. (2021) regionally calibrated damage functions.
- Scope limitations: focus on direct damages only (excluding indirect losses), and on two hazards (floods and TCs) owing to data and vulnerability literature limitations. Floods exclude rainfall-induced flooding due to data availability.

### Practical applications and policy relevance
- Diagnostic tool: provides forward-looking, granular country-level indicators of flood and TC physical risks to complement backward-looking EM‑DAT–based indicators.
- Macro-financial analysis: damage estimates can be inputs to macro models to generate macro-financial scenarios incorporating climate physical risks (e.g., for banks’ stress tests).
- Micro- and sectoral applications: damage/loss components can be integrated with borrower-level data for micro approaches to climate risk analysis.
- Adaptation and mitigation appraisal: hazard damages can be used in macroeconomic models to evaluate macroeconomic and financial implications of alternative investment and financing programs.

### Paper organization (as presented)
- Section 2: Literature review.
- Section 3: Data for hazards, vulnerability, and exposures.
- Section 4: Methodology to compute losses from point locations to country aggregation and baseline definition.
- Section 5: Results and comparison with other studies.
- Section 6: Conclusions.
- Annexes: additional information on data, results, comparisons, and limitations (including Jupiter Intelligence Data, GDP downscaling approach, drivers of flood and TC results, scenario-specific results for SSP1-2.6 and SSP5-8.5, NGFS comparison, and limitations).

*IMF Working Paper: A multi-country study of forward-looking economic losses from floods and tropical cyclones*

### 3. Data

### 3. Data

### Hazard data
- Damages arise from the interaction of three components: hazard, vulnerability and exposure.
- Hazard data used: floods and TCs severity supplied by Jupiter Intelligence.
- Floods:
  - Data covers river and coastal flood depth projections under different scenarios, using data from GCMs as an input.
  - Jupiter Intelligence’s inland river flooding model uses projected regional changes in extreme streamflow to estimate how flood depth and extent may change in a future climate.
  - Coastal floods use multiple climate projection datasets to estimate effects of sea-level rise, storm surge and tides on coastal inundation, and storm surge and lake levels on lake shoreline inundation.
  - Sea-level rise, long-term lake levels and surge and tides from the same climate scenario are combined to produce a unified estimate of expected water levels; then river and coastal flood results are consolidated.
  - Data provided for mean levels for flood depth and fraction of land flooded in 3 arcseconds grids for 10, 20, 50, 100, 200, and 500-year return periods.
  - Measures provided for a baseline (climate for 10-year period centered on 1995) and projections from 2020-2100 with 5-year increments under three scenarios: SSP1-2.6, SSP2-4.5, and SSP5-8.5 (low, intermediate and very high GHG emission scenarios respectively).
  - Note: Rainfall-induced flooding is currently not available.
  - Note: Arcsecond ≈ 30 meters at the equator.
  - Jupiter’s hazards projections dataset includes global data for 100,000 locations selected to maximize global and country-level GDP coverage using “gridded GDP” projections under SSP2-4.5 by Murakami and others (2021); selected locations provide a coverage of 97 percent of global GDP in year 2040.
- Tropical Cyclones (TCs):
  - Hazards data reflect wind speeds defined as the maximum 1-minute sustained wind speed at 10 meters above ground level.
  - Measure produced using data from GCMs and a synthetic TCs model.
  - The maximum 1-minute sustained windspeed variable is used in damages’ calculation only for areas where TCs have been observed historically.
  - Data include air temperature throughout the lower atmosphere, humidity throughout the lower atmosphere, wind speeds and directions throughout the lower atmosphere, sea- and skin-surface temperatures, surface (sea-level) pressure. Multiple GCMs are used from both CMIP5 and CMIP6.
  - Synthetic TCs tracks can be generated to overcome spatial and temporal limitations of historical data.

### Vulnerability – Damage functions
- Damage functions relate economic damages to climate inputs; they define impacts of specific hazards on real assets and activities.
- Selected damage functions: Huizinga and others (2017) for floods; Eberenz and others (2021) for TCs—chosen for public availability and global consistency.
- Flood damage functions (Huizinga et al., 2017):
  - Piecewise linear functions depicting fractional damage as a function of water depth for categories: buildings and contents (residential, commercial, industrial), transport facilities, infrastructure (roads and railroads) and agriculture.
  - Calibrated using quantitative data from multiple studies.
  - Calibrated for six continents: Europe, Africa, Asia, Oceania, North America, South and Central America.
- TC damage functions (Eberenz et al., 2021):
  - Regionally calibrated; functional form is sigmoidal expressing percentage of damage as a function of wind speed:
    f = v_n^3 / (1 + v_n^3),
    where v_n = MAX[(V − V_thresh), 0] / (V_half − V_thresh).
  - V is the 1 min sustained wind speed at 10 m above ground per storm event.
  - V_thresh is a minimum threshold for impacts; below it no directly wind-induced damage is expected.
  - V_half is the slope parameter (wind speed at which impact reaches 50 percent of exposed asset value).
  - Functional form builds on empirical studies; on physical grounds damage should vary as the cube of wind speed over a threshold value and fraction damaged approaches unity at very high wind speeds.
  - Notes: The red dotted line in the calibration figures represents a windspeed of 25.7 m/s, or 92.5 km/h, the threshold below which damage rate is assumed to be zero.

### Exposures
- Economic exposures proxied using downscaled GDP data.
- Downscaling methodology used: Murakami and others (2021) gridded GDP under different SSPs.
- Global dataset of downscaled GDP at 30 arcminutes is publicly available and spans 1980–2100 by 10-year intervals.
  - 1980–2010: estimated by downscaling actual GDPs by country.
  - 2020–2100: estimated by downscaling projected GDPs under SSP1-2.6; SSP2-4.5; and SSP5-8.5.
- Gridded GDP centroids correspond to latitude/longitude of grid cells.

### Methodology (calculation of damages)
- Focus: direct damages only (exclude indirect damages such as business interruption, spillovers, second round effects).
- Damage rates: loss of value of assets expressed in percent of value before hazard.
- Damage rates from floods and TCs calculated separately; caution when combining due to correlation (e.g., TCs may lead to flooding).
- Notation (as provided):
  - depth_{c,i,RP,t,s}, frac_{c,i,RP,t,s}, wind_{c,i,RP,t,s} denote flood depth, fraction of land flooded, maximum sustained windspeed respectively in country c, location i, return period RP, projection year t and scenario s.
  - Damage function for floods: df_floods(depth).
  - Damage function for storms: df_storms(wind).
- Flood damage specifics:
  - For each gridded cell, exposure measure divided into built-up and non-built-up areas using Copernicus Global Land Operations “Vegetation and Energy” (CGLOPS-1) for 2019 (BuiltUp-CoverFraction-layer at 100 m resolution), downsampled to 5 arcminutes via weighted average of contributing pixels.
  - For built-up areas: combine residential, commercial, and industrial damage functions by equally weighting each function.
  - For non-built-up areas: consider agriculture and infrastructure damage functions.
- Location-level damage rate formulas:
  - Floods:
    d_{c,i,RP_j,t,s} = frac_{c,i,RP_j,t,s} * df_floods(depth_{c,i,RP_j,t,s}).
  - TCs:
    d_{c,i,RP_j,t,s} = df_storms(wind_{c,i,RP_j,t,s}).
- Hazard data available for return periods RP = {10, 20, 50, 100, 200, 500} denoted RP_10, RP_20, RP_50, RP_100, RP_200, RP_500.
- Expected annual damage rate: trapezoidal rule to approximate area under the damage-return curve.
  - Discretized damage variable d_{c,i,t,s} defined piecewise across severity intervals and return periods (as in source).
  - Location-level expected annual damage rate:
    E[d_{c,i,t,s}] = Σ_{j ∈ RP̅} (1/RP_j − 1/RP_{j+1}) * (d_{c,i,RP_j,t,s} + d_{c,i,RP_{j+1},t,s}) / 2  + d_{c,i,RP_500,t,s} / RP_500,
    where RP̅ = {10,20,50,100,200,500}.
- Country-level aggregation:
  - For locations i = {1,...,n} in country c, gridded GDP at a location is GDP_{c,i,t,s}, total GDP is GDP_{c,t,s}.
  - Country-level expected damage rate for year and scenario:
    E[D_{c,t,s}] = Σ_{i=1}^{n} E[d_{c,i,t,s}] * (GDP_{c,i,t,s} / GDP_{c,t,s}).
- Outcome metrics:
  - Damage rate in a given year and scenario: E[D_{c,t,s}].
  - Change in levels relative to baseline: E[D_{c,t,s}] − E[D_c] (baseline).
  - Relative change: (E[D_{c,t,s}] / E[D_c] (baseline)) − 1.
  - Baseline defined as hazard for the 10-year period centered around 1995 and GDP in 2000.

### Results — Floods damages (high-level findings)
- Projection horizons showcased: 2050 and 2100.
- Flood risks projected to increase for most countries by mid-century.
- Depending on scenario, 58-67 percent of the countries (representing 74-80 percent of global GDP in 2020) display an increase in the country-level expected damage rate in 2050 relative to the baseline.
- Median country-level damage rates (for countries with increasing flood risks):
  - 2050: 0.28-0.35 percent (depending on scenario).
  - 2100: 0.36-0.41 percent (depending on scenario).
- For countries with decreasing flood risks:
  - 25-33 percent of countries (representing 18-25 percent of global GDP in 2020) display a decrease in flood risks; the rest display no change.
  - Median damage rate for these countries:
    - 2050: 0.22-0.32 percent (depending on scenario).
    - 2100: 0.25-0.27 percent (depending on scenario).
- Table 3 (key statistics for flood damages) — counts and medians:
  - 2050 SSP1-2.6: 107 countries increasing (0.35), 60 countries decreasing (0.22).
  - 2050 SSP2-4.5: 115 countries increasing (0.28), 52 countries decreasing (0.32).
  - 2050 SSP5-8.5: 123 countries increasing (0.32), 45 countries decreasing (0.32).
  - 2100 SSP1-2.6: 121 countries increasing (0.36), 47 countries decreasing (0.26).
  - 2100 SSP2-4.5: 117 countries increasing (0.36), 51 countries decreasing (0.27).
  - 2100 SSP5-8.5: 124 countries increasing (0.41), 44 countries decreasing (0.25).
- Variability:
  - Variability of damage rates increases for more severe scenarios at the end of the century for countries with increasing risks; number of extreme observations increases (many are small land area countries such as small island countries).
  - Distribution for countries with decreasing damage rates is more stable across time and scenarios.
- Geographic patterns (SSP2-4.5 focus; similar findings for SSP1-2.6 and SSP5-8.5):
  - Top-10 countries with largest damage rates are mostly in tropical and sub-tropical regions and most have small land area.
  - Several countries experience significant relative increases in flood damages by 2050 and by 2100.
    - Median relative change: 8.46 percent by 2050; 17.37 percent by 2100.
    - Median levels change: 0.02 percentage points in 2050; 0.03 percentage points in 2100.
  - Largest relative changes in damage rates: South America, Africa, Southeast Asia.
- Drivers:
  - Country-level damage rate changes over time and scenarios mainly driven by corresponding changes of hazard severity (flood depth).
  - For most countries damage rates increase over time and with more severe scenarios due to increasing average flood depth.
  - Some high-latitude regions (e.g., Canada and some parts of Europe) are expected to experience drier climates which can lead to a decrease in riverine flood risks.
- Notes:
  - Results for other scenarios and additional details are available in annexes (Annex III, Annex V referenced in source).

*Source: wpiea2024141-print-pdf - 3. Data*

### 5.2 Tropical cyclones damages

### 5.2 Tropical cyclones damages

### Overview of projected damages
- Projections calculated for 89 countries in regions with historical tropical cyclone (TC) records.
- Depending on the scenario, 66-67 percent of the exposed countries (with the exposed areas representing 40-42 percent of global GDP in 2020) display an increase in country-level expected damage rate in 2050 relative to the baseline.
- In the sample, 16-17 percent (with exposed areas representing 4-5 percent of global GDP in 2020) display a decrease in TC risks.
- The number of countries with increasing damage rates is highest for SSP5-8.5.
- When only certain parts of a country are exposed to TCs, only the GDP within these grids is accounted for.

### Key statistics (Table 4)
- 2050 SSP1-2.6: 59 countries with increasing TC risks (median country-level damage rate 0.11 percent); 15 countries with decreasing TC risks (median 0.0019 percent).
- 2050 SSP2-4.5: 60 countries with increasing TC risks (median 0.11 percent); 14 countries with decreasing TC risks (median 0.0006 percent).
- 2050 SSP5-8.5: 59 countries with increasing TC risks (median 0.12 percent); 15 countries with decreasing TC risks (median 0.0025 percent).
- 2100 SSP1-2.6: 57 countries with increasing TC risks (median 0.12 percent); 17 countries with decreasing TC risks (median 0.0015 percent).
- 2100 SSP2-4.5: 60 countries with increasing TC risks (median 0.13 percent); 14 countries with decreasing TC risks (median 0.0012 percent).
- 2100 SSP5-8.5: 63 countries with increasing TC risks (median 0.16 percent); 11 countries with decreasing TC risks (median 0.0007 percent).

### Distributional patterns and extremes
- Variability of damage rates increases for more severe scenarios at the end of the century for countries that will see an intensification of risk.
- The number of extreme observations is overall stable for countries with increasing and decreasing damage rates.
- All extreme values shown in Figure 10 (left-hand panel) are island countries; small island countries are among the top five countries by damage rate for 2050 and 2100 in the different scenarios.
- Note: While flood risks are expected to decrease for some countries, other hazards such as droughts and wildfires might be increasing; results should not be interpreted as physical risks decreasing overall for these countries.

### Geographical concentration and regional highlights
- Under SSP2-4.5, TC damages concentrate in certain regions; similar general findings hold for SSP1-2.6 and SSP5-8.5.
- Countries with the highest damage rates are in the Caribbean, South and Southeast Asia, Eastern Africa, and Oceania.
- Most countries historically exposed to TCs experience increases in TC-related damage rates.
- Countries show a relative change median of 6.95 percent by 2050, which reaches 14.20 by the end of the century.
- North and Central and South America, and countries along the coast from Eastern Africa to Oceania, show the largest changes.
- High changes in levels of damage rates are observed in countries in the Caribbean, Southeast Africa, and Southeast Asia.
- Drivers of TCs hazard damages are provided in Annex IV.

### Figures referenced
- Figure 10: Tropical cyclones’ damage rates for countries with increasing (left-hand panel) and decreasing (right-hand panel) damages relative to the baseline (In percent).
- Figure 11: Tropical cyclones’ damage rate for 2050 under SSP2-4.5 (In percent).
- Figure 12: Tropical cyclones’ damage rate relative and levels change between the baseline and 2050 under SSP2-4.5 — Relative change (In percent) and Levels change (Percentage points).
- Source noted for figures: IMF staff calculations.

*Source: wpiea2024141-print-pdf - 5.2 Tropical cyclones damages*

### Annex Figure 2: Floods’ depth levels change in meters between baseline and 2050 under SSP2-4.5

### Annex Figure 2: Floods’ depth levels change in meters between baseline and 2050 under SSP2-4.5

### Major findings on floods and depth changes
- For most countries, damage rates increase over time and as scenarios become more severe, indicating a positive correlation between global flood risks and global warming.
- An important driver of increased damage rates is the change in (average) flood depth over time, which generally displays increasing trends across time and scenarios.
- The GDP distribution in these cases does not tend to change these increasing trends in most countries.

### Notable exceptions and heterogeneous outcomes
- Some countries experience decreases in flood damages over time and with more severe scenarios; examples include parts of Canada and some parts of Europe expected to experience drier climates leading to decreased riverine flood risks.
- Guatemala: average flood depth increases across time and scenarios (in 2100), but the GDP distribution shifts damage rates, resulting in lower damages under SSP5-8.5 than under SSP1-2.6 — indicating more extreme SSP scenarios can be associated with higher flood depth in areas with lower GDP.
- Belize: damages do not display a clear trend across time and scenarios. Drivers include:
  - Interplay between river flood and coastal flood trends (aggregated in the data) that can have different directions.
  - Tropical inland river flood depths can decrease due to decreased precipitation trends (example given for Belize).
  - Coastal water levels due to sea level rise tend to increase the most and in a non-linear fashion in the second half of the century, producing non-linear aggregate trends.

### Data and methodological drivers of heterogeneity
- Aggregation of coastal and river flood metrics can produce non-linear trends because coastal flood fraction tends to increase for all countries by the end of the century while river flood fraction changes are more heterogeneous.
- Flood metrics under different SSPs for a given year and return period are generally expected to be monotonic, but this can fail locally because:
  - SSP scenarios map to atmospheric CO2 levels, but translation to temperature, precipitation, runoff, and extreme flooding is non-linear.
  - Coarse global climate model resolution and required downscaling to local streams, plus averaging across multiple climate models, can change the orderings of SSPs for individual locations.

### Illustrative country examples (as discussed)
- Bangladesh: increasing trends in average flood depth and damage rates over time and scenarios (referenced in Annex Figure 4).
- Canada: drier-climate regions with potential decreases in riverine flood risk (referenced in Annex Figure 5).
- Guatemala: increasing average flood depth but lower damages under SSP5-8.5 than SSP1-2.6 due to GDP distribution shifts (referenced in Annex Figure 6).
- Belize: no clear trend due to mixed river/coastal flood dynamics and regional precipitation trends (referenced in Annex Figure 7).

### Comparison with NGFS (selected quantitative consistency metrics)
- Floods: both this study and NGFS agree on the sign of relative change for an average of 65 percent of countries (maximum of 70 percent and minimum of 60 percent depending on year and scenario).
  - Both approaches show most countries would be negatively impacted by climate change: an average of 64 percent of countries in this study (max. 68 percent and min. 52 percent) and 60 percent of countries in NGFS estimates (max of 63 percent and min. 53 percent) across years and scenarios.
  - For the median relative change, NGFS estimates tend to be higher with a gap of -5.79 percentage points on average between this study and NGFS, and the smallest gap for SSP5-8.5 is -3.17 percent.
- Tropical cyclones (TCs): consistency in sign of relative change between this study and NGFS averages 55 percent across years and scenarios (maximum of 57 percent and minimum of 51 percent).
  - NGFS median relative changes tend to be higher across years and scenarios with the gap between estimates of 3.34 percentage points.
  - Geographic divergence: NGFS expects increasing TC damages in Europe and Northwest Africa and decreasing trends in Southeast Africa and Oceania; differences may stem from synthetic TC generation methods.

*Source: IMF staff*

### Annex VIII. Limitations

### Annex VIII. Limitations

### Hazards data: model uncertainty and coverage
- There is a very wide range of climate models giving rise to sizable model uncertainty.
- Jupiter Intelligence adopts data from different GCMs from CMIP6 and CMIP5 forcings and selects specific models by considering:
  - availability of predictors, scenarios, and data;
  - ensemble members;
  - model quality.
- Jupiter Intelligence provides uncertainty bounds that consider different types of uncertainty depending on the hazard:
  - hydrological model parameter uncertainty for flood depth;
  - Generalized Extreme Value (GEV) fitting uncertainty for the wind.
- Future extensions of the study could take these uncertainty bounds into account.
- Spatial resolution and geographical coverage affect results:
  - Flood data used: spatial resolution of 3 arcseconds (90 meters by 90 meters at the equatorial sea level).
  - Availability of 100,000 locations globally implies the geographical coverage is not always ideal; having additional locations would improve the analysis.
    - Note: the locations were selected to maximize coverage of economic activity, where a gridded GDP coverage above 95 percent is achieved with 100K locations.
- Tropical cyclone (TC) hazard data:
  - Spatial resolution is 15 arcminutes before bilinear interpolation.
  - Effective coverage for TCs is much higher than for floods because the risk cannot change within a single GDP cell.

### Assumptions in damage estimation and their effects
- Lowest return period available from Jupiter Intelligence is 1-in-10 years.
  - Expected damage rate for return periods below 1-in-10 years is set to zero, implying an underestimation of damages when low-severity events cause recurrent losses.
- Correlation of hazard data across locations is assumed to be zero, which implies an overestimation of damages.
- The overall net effect of these assumptions on damages is unclear.
- Improvements suggested:
  - Provide hazards data for additional return periods.
  - Provide information on spatial correlation.

### Aggregation of multiple hazards and compound events
- Aggregating impacts of various hazards is challenging; most literature focuses on single hazards.
- Recognition is growing of the need to consider multi-hazard and cascading effects (Wald and others (2020)).
- Compound floods (interaction of coastal, river, and pluvial floods) can influence overall hazard and risk (Zscheischler and others, 2018).
- Jupiter Intelligence combines fluvial and coastal flood outputs by taking the maximum depth of inundation between river and coastal floods (accounting for sea level rise, storms and tide).
  - This approach evaluates the two hazards independently and is reasonable if the two are not covariate.
  - Where fluvial and coastal flooding can be coincident, the joint hazard distribution becomes important and the present methods may underestimate true flood risk.
- Floods and TC damages are calculated in isolation in this study.
  - Caution advised when combining damages from both floods and TCs together.
- For some locations, it can be important to consider joint flood and wind damages; this study does not do so.

### Exposure data limitations
- The downscaled GDP exposures are a proxy for real economic exposures.
  - Benefits: forward-looking, scenario dependent, globally available.
  - Ideal alternative: granular geospatial data on buildings, infrastructures and other relevant economic exposures when available for specific countries.

### Damage functions and data-source limitations
- Damage functions are simplified expressions of economic damages as a function of climate inputs.
- Use of regionally calibrated damage functions entails limitations:
  - TC damage functions adopted from Eberenz and others (2021) are calibrated using reported damages from EM-DAT data; thus EM-DAT limitations (biases, under-reporting for some countries) extend to associated damage functions.
  - Eberenz and others (2021) note considerable differences in uncertainty across model regions, with the largest uncertainties for the North-West Pacific regions.
- Damage functions used focus on estimating direct damages and exclude indirect damages (for example, damages from business interruption).
  - Accounting for indirect channels would increase damages and is left for future research.

### Adaptation and mitigation
- The study does not explicitly account for adaptation and mitigation measures.
  - Excluding future adaptation and mitigation measures and associated uncertainties provides an upper-bound estimate of damages, useful to illustrate the benefits of these measures.
  - For specific countries where data and plans on these measures are available, the methodology could be extended to incorporate them.
  - Exception: only for floods in the USA, England, the Netherlands and Germany the hazard data accounts for the assumption that locations currently protected by levees will be protected up to and including a 100-year flood in the future.

*Annex VIII. Limitations — IMF Working Paper*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024141-print-pdf.pdf_