The macroeconomic effects of adapting to high-end sea-level rise via protection and migration

Sea-level rise (SLR) is identified as a major global climate risk, affecting both developing and developed regions. Rising mean and extreme sea levels will continue for centuries and threaten densely populated coasts, deltas and small island states. Adaptation options include advance, protection, retreat, accommodation and ecosystem-based solutions. While coastal protection can be effective and cost-efficient in dense urban areas, high SLR can trigger coastal retreat and out-migration. The macroeconomic literature often assumes no adaptation and focuses on protection, underrepresenting retreat/migration. This paper addresses these gaps by assessing global macroeconomic effects of SLR adaptation—including coastal migration—up to 2050, emphasizing high-end SLR. The research questions are: (i) How do different adaptation responses (protection, autonomous retreat, and their combination) affect macroeconomic outcomes? (ii) How do accumulated indirect effects over time compare with direct damages? (iii) How does including autonomous adaptation in the reference scenario change estimated benefits of planned protection?

Two strands dominate macroeconomic SLR assessments: (1) fully integrated IAMs (e.g., DICE, FUND, MAGICC-linked) with simplified adaptation representations; and (2) soft-linked frameworks that couple detailed sectoral impact models to macroeconomic models (e.g., CGE). Prior studies often assume no adaptation and emphasize coastal protection; retreat/migration is rarely modeled and typically stylized (e.g., FUND under perfect foresight) or static (e.g., Pycroft et al., with migration as forced consumption). Desmet et al. include migration but omit protection and sea-flood costs, yielding high migrant numbers and focusing on spatial agglomeration losses. The literature suggests protection is cost-effective in dense areas, accommodation is limited under high SLR, and that SLR-induced displacement can be substantial even under optimal protection. There is growing evidence that considering a broader set of impacts (including migration) increases estimated welfare losses from SLR and that economic losses may be amplified via indirect and dynamic effects in the economy.

Scenario design and horizon: The analysis is embedded in the RCP–SSP framework and runs to 2050. Three SLR scenarios (relative to 2015) are used: (1) RCP8.5–SSP5 with high-end ice-melting sensitivity: 0.39 m by 2050 (1.62 m by 2100); (2) RCP8.5–SSP5 with medium ice-melting: 0.19 m by 2050 (0.63 m by 2100); (3) RCP4.5–SSP2: 0.16 m by 2050 (0.45 m by 2100). Each is combined with four adaptation cases: (a) no additional adaptation; (b) planned adaptation only (publicly financed sea dikes); (c) autonomous adaptation only (reactive coastal migration/retreat); (d) combined planned and autonomous adaptation. Comparisons are made relative to SSP-specific baselines without climate change. DIVA coastal impact and adaptation model: DIVA assesses global coastal flood damages, protection costs, and migration on 12,148 coastal segments. Exposure is derived from SRTM elevation and GRUMP population; assets are computed from population and sub-national GDP per capita with an assets-to-GDP ratio of 2.8. Extreme water levels (GTSR) shift with SLR; local relative SLR includes glacial isostatic adjustment and delta subsidence. Current protection levels follow Lincke & Hinkel with thresholds for zero protection in low-density areas. Flood damages use depth–damage functions and expected annual damages. Protection is modeled as hard infrastructure (dikes) with national unit costs, 1% annual maintenance, and no failure below design height. Migration is assumed only from unprotected areas once land falls below the 1-in-1-year flood level (reactive retreat), with migrants relocating within-country to non-exposed inland. Migration costs are valued at three times GDP per capita per migrant, capturing moving and rebuilding capital. DIVA does not model non-coastal zones explicitly and assumes agglomeration dispersion effects are neutral. COIN-INT CGE model: COIN-INT is a global, multi-regional, multi-sectoral recursive-dynamic CGE model (GAMS/MPSGE), calibrated to GTAP9 (base year 2011) and projecting in 5-year steps to 2050. It covers 21 world regions (with higher resolution in Europe) and 21 sectors. EU regions feature separate private and public households; non-EU regions use a representative regional household. Investment follows fixed savings rates; production uses nested CES functions. Model coupling and impact channels: DIVA outputs (land loss, expected annual flood damages, capital stock, people flooded, protection investment and maintenance costs, migration costs) are converted from PPP to MER and passed to COIN-INT. Six channels implement SLR impacts in COIN-INT: (1) capital stock losses from sea-flood damages reducing capital accumulation; (2) temporary labour supply losses (2/48 of annual labour for each person flooded); (3) cropland loss from gradual land loss; (4) sea-dike investment as forced public investment crowding out other government consumption and not adding to productive capital; (5) dike maintenance as forced government consumption; (6) migration costs as capital losses reducing capital accumulation. Reconstruction investment is GDP-neutral (crowds out other investment). One GCM (HadGEM2-ES) provides climate forcing consistent with average SLR ranges. Amplification ratio (AR): Annual GDP loss relative to baseline divided by annual direct costs (sum of capital, labour, dike investment and maintenance, and land-loss costs) measures how indirect and dynamic effects amplify or absorb direct damages over time.

- Without further adaptation under high-end SLR (RCP8.5–SSP5, high ice melt), macroeconomic losses are large: by 2050, GDP is ~4.5% below baseline in Italy and Northern Europe, and up to ~11% in Emerging Economies of Asia; China, Oil Exporters, Australia–New Zealand, and South-East Asia face ~7–9% losses. - Planned adaptation (sea dikes) reduces 2050 GDP losses to <1% in European regions and <2.5% in Rest-of-World (ROW) regions. - Autonomous adaptation only (reactive coastal migration) limits 2050 GDP losses to ≤3% in all regions, with many regions <1%. - Combined planned + autonomous adaptation is at least as effective as planned-only in all ROW regions and often more effective; 2050 GDP losses do not exceed ~1% in ROW, while in Europe results are similar to planned-only because densely populated areas are protected, reducing the need to migrate. - Protection-only can be macroeconomically less effective than migration-only in some developing regions (e.g., India, South-East Asia) due to high, unproductive capital tied up in dikes that depresses future capital accumulation. - Autonomous adaptation is reactive: GDP-loss trajectories under migration resemble no-adaptation early on and flatten or reverse after ~2035 as SLR triggers relocation and lowers subsequent flood losses. - Amplification ratios (ARs) rise over time, often >1, reaching up to ~17 (Netherlands, 2050), indicating indirect and intertemporal effects (via weakened capital accumulation) dominate over direct costs. Some regions show negative ARs where GDP gains occur despite positive direct costs (e.g., due to trade/comparative advantages). - Sensitivity to ice-melt: Under RCP8.5–SSP5 with medium ice melt, no-adaptation GDP losses are much lower (up to ~2% in Europe, ~4% in ROW by 2050). With adaptation, losses can be reduced to levels similar to the high-end case’s adapted outcomes. - Including autonomous adaptation in the reference dramatically lowers the estimated benefits of planned protection. Example (Fig. 7): In ECA under high-end SLR, planned protection reduces GDP losses by 10.8 percentage points versus no adaptation, but only by 2.5 percentage points when autonomous migration is included as the reference. - Policy-relevant implication: Future economy-wide damages (GDP losses) substantially exceed contemporaneous direct damages due to propagation through capital dynamics; immediate adaptation yields significant long-term macroeconomic benefits.

The study demonstrates that adaptation to SLR yields sizable macroeconomic benefits, but residual GDP losses remain under high-end SLR even with adaptation. Combining planned protection with autonomous retreat often outperforms protection-only in developing regions by avoiding large, unproductive protective capital stocks and enabling reallocation of assets inland. Rising amplification ratios highlight that indirect and dynamic macroeconomic effects (reduced capital accumulation from both damages and unproductive adaptation investments) dominate long-run costs, underscoring the need for early action. Comparing with prior soft-linked CGE analyses, results are quantitatively consistent where scenarios align, though differences in covered impact channels limit direct comparability. Importantly, using a realistic autonomous adaptation reference (rather than an implausible no-adaptation counterfactual) markedly reduces the estimated benefits of planned protection, cautioning against overstatement of protective adaptation in cost–benefit analyses. Distributional and public–private burden sharing issues are not captured by aggregate GDP metrics; planned measures burden public budgets, while autonomous measures burden private actors with heterogeneous capacities. Overall, findings stress immediate, context-specific adaptation portfolios and the inclusion of autonomous responses in macroeconomic assessment baselines.

This paper provides a global macroeconomic assessment of SLR adaptation that jointly considers planned protection (dikes) and autonomous coastal migration within a soft-linked DIVA–COIN-INT framework to 2050. Key contributions are: (i) showing that combining protection and autonomous retreat can minimize GDP losses relative to protection-only, especially in developing regions; (ii) demonstrating that indirect and intertemporal effects amplify direct damages over time (ARs > 1), implying that early adaptation substantially reduces long-run macroeconomic costs; and (iii) evidencing that including autonomous adaptation in the reference scenario greatly lowers the apparent benefits of planned protection, avoiding overestimation in economic appraisals. Future research should broaden impact channels (e.g., erosion, salinization, ecosystem services, cultural losses), assess distributional and fiscal dimensions of adaptation, incorporate co-benefits of protection, test sensitivity of protection placement rules, explore alternative autonomous adaptations beyond migration, account for spatial agglomeration dynamics explicitly, and extend analyses beyond 2050 with multiple climate models.

- Outcome metric: Aggregate GDP obscures distributional impacts across households, sectors, and between public and private actors. - Autonomous adaptation scope: Modeled solely as reactive internal migration triggered at the 1-in-1-year flood level; limited empirical parameterization and potential bias from the chosen threshold and migration unit cost. - Protection placement: Assumption based on asset density; no systematic sensitivity testing of which areas are protected. - Impact coverage: Excludes direct costs from erosion, salinization, ecosystem changes, and intangible losses (culture, biodiversity); omits co-benefits of protection (e.g., public space). - Modeling assumptions: Dikes treated as non-productive capital; reconstruction assumed GDP-neutral (crowding out). Agglomeration dispersion effects assumed net-neutral outside the coastal zone. - Climate and horizon: Single GCM (HadGEM2-ES) used; analysis limited to 2050 though SLR and flood risks grow substantially beyond. - Migration: Assumes only internal (within-country) relocation to non-exposed areas and neutral macro effects of changing spatial structures. - DIVA coverage: Non-coastal zones not explicitly modeled; land loss implemented as reduced cropland in CGE. - Baseline uncertainties: Socioeconomic pathways (SSPs) entail substantial uncertainties affecting results.