the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 27 Jun 2025
Feedback-based sea level rise impact modelling for integrated assessment models with FRISIAv1.0
Abstract. Global warming is expected to lead to a substantial rise in coastal sea levels by the end of the century, which imposes future impacts and adaptation challenges on the coastal zone. Capturing the socio-economic costs of sea level rise (SLR) is therefore an important component of climate impacts in integrated assessment models (IAMs). However, there is a lack of process-based models of SLR impacts with a focus on global, time-varying dynamics. Current SLR impact models often follow a cost-benefit analysis approach, fail to represent diverse pathways of SLR impacts, or do not include coastal adaptation. Here, we present the Feedback-based knowledge Repository for Integrated assessments of Sea level rise Impacts and Adaptation version 1.0 (FRISIAv1.0), a model designed for process-based, non-equilibrium IAMs that follows a system dynamics approach. FRISIA's SLR component is based on existing models of SLR, while its impact component is a substantially modified adaptation of the Coastal Impact and Adaptation Model (CIAM) for use in globally or regionally aggregated models. While a reduced-feedback version of FRISIA approximately reproduces CIAM results, the integration of additional feedbacks in FRISIA leads to emerging new behaviours, such as a potential peak and decline in SLR-driven storm surge damages in the early 22nd century, due to economic feedbacks in the coastal zone. When coupling FRISIA to an IAM, global GDP is reduced by 1.6–6.1 % (17th–83rd percentile range) under a global SLR of 0.8 m by 2100 and no coastal adaptation, which is in the range of previous studies. The coupling of a diverse set of SLR impact streams from FRISIA into a system dynamics IAM has the advantage of leading to a wide range of socio-economic consequences that go beyond just a reduction in global GDP, such as an effect on inflation. Our simulations highlight the benefits of accounting for dynamic coastal feedback and coupling diverse SLR impact and cost strains to IAMs, and showcase that FRISIAv1.0 is a useful tool for doing so.
- Preprint
(4961 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2025-1875', Anonymous Referee #1, 29 Jul 2025
-
RC2: 'Comment on egusphere-2025-1875', Anonymous Referee #2, 31 Jul 2025
The authors present version 1 of the Feedback-based knowledge Repository for Integrated assessments of Sea level rise Impacts and Adaptation (FRISIAv1.0). FRISIA is designed for use in non-equilibrium Integrated Assessment Models (IAMs) and draws from published approaches for sea level as well as sea level impacts and adaptation modeling. FRISIA integrates dynamic feedbacks to more comprehensively capture economic damages from sea level rise impacts and has been coupled to and tested in a newly developed IAM framework.
The authors have to be commended for taking on a very complex task and successfully developing a tool that is able to estimate socio-economic damages from sea level rise in a transparent and plausible way. The main challenge for these tools is the development of reasonable simplifications and assumptions that allow for an efficient modeling chain while not neglecting necessary real-world boundary conditions. In this regard, I see a couple of issues that would need to be added and clarified. FRISIA will represent a much needed additional tool for estimating coastal damages from sea level rise in a research landscape that has been lacking diversity in modeling approaches.
Main comments:Overview figure
While the manuscript follows a logical structure it is hard for the reader to stay on top of the model structure given the number of components and assumptions introduced. In my view, it is necessary to include a schematic at the beginning of the model description that provides an overview of all FRISIA components and the coupling to FRIDA. This would establish a clear visual framework and it would be easier to link the following subsections to the overall modeling flow.Adequate introduction of adopted assumptions and concepts
Because FRISIA builds on existing modeling approaches, not all underlying assumptions and concepts are introduced in sufficient detail. This is particularly true for assumptions adopted from the CIAM and DIVA models. For example, FRISIA uses the DIVA coastal segments but it is never clarified what these segments consist of. Table A1 includes information in this regard but it has to be clear upfront in the text which components constitute a coastal segment. Importantly, it is not clear that coastal segments exclude any classification according to coastal morphology which, of course, is key for modeling coastal flooding (see below). Another example is coastal assets. While “asset” is a common term in economic modeling, it is very important to be clear at the beginning of subsection 3.2.2 which coastal asset types are tracked by FRISIA.Coastal flooding
Arguably, the most important component of the FRISIA modeling chain is the projection of a reasonable aggregated coastal flooding indicator. This, however, is not captured with the adequate amount of detail. Even if a previously publish method (CIAM) is applied, it has to clear how coastal flooding is derived from the sea level rise projection which represent mean water level heights. To adequately model coastal flooding, three components are needed: sea level rise, extreme sea level or storm surge height, and wave height. The latter is crucial for coastal damage estimates but is neglected far too often. Including information from a wave model is not feasible for a light-weight impact model like FRISIA, of course. But this topic has to be resolved better. Would it be possible to add to the effective surge height parameter a term that would account for region-dependent typical wave heights during storm surges? I would suggest to include a new dedicated subsection 3.2.1 that would introduce the key concepts and discuss simplifications and caveats in detail.Shoreline types
Similary and as already indicated before, the manuscript lacks information on how different shoreline morphologies are distinguished or not. Coastal flooding manifests very differently if a storm surge hits mangroves, sandy beaches, a steep rocky coast, or an urban coastal environment equipped with/without flood defenses. Again, even if the model cannot distinguish between all the relevant shoreline types, it is important to describe and discuss the underlying FRISIA assumptions. This should happen before the coastal protection subsection and could potentially be merged with the subsection clarifying how the effective surge height parameter is derived.Economic damages to coastal industries like fisheries and tourism
In the compilation of cost formulations, I miss the description of costs that would incur from damages to coastal economic sectors like fisheries or tourism. I believe that this is not covered under coastal assets per se. These costs or damages may be better captured in the feedback component as described in section 5, but it seems that they are also not explicitly captured there. As described, the feedbacks of reduced future investment, slower growth, reparation or failing to protect are insufficient to capture these damages, or they would at least have to be relabeled if such damages were captured via an existing GDP reduction term. Please elaborate on this matter and clarify.
Minor Comments:L153, It is unclear how the sea level component-wise “scale factors” are varied in the ensemble projections. Please clarify and also include a table in the Appendix that shows the parameters serving as “scale factors” for each sea level component and the established ranges.
L170, This is, at the latest, where it has to be clarified what the DIVA coastal segments consist of, and that they don’t really account for any morphological information, which one would intuitively assume.
Figure 1, I suggest to use shading for the FRISIA projected ranges, slightly higher opacity for no-MICI and less opacity for MICI setups. Make all three panels equally wide and start in 1990, this should still allow for a readable legend.
L223, Please explain why the 50-year time horizon is chosen, and not 30 years, for example.
L224, It is not clear to me why it is necessary to derive the expected SLR in 50 years, SLR50, using GSAT and CO2 emissions. The sea level model delivers SLR50 until 2150 if we assume that the projections are produced until 2200, as shown in the sea level figures.
L231, The current effective flood height S will be much better understood after adequately introducing and explaining how it is derived (see above major comment).
L247, Please describe here in more detail which assets types are tracked (see above major comment).
L351, Would more recently published information change the weights?
L367-371, This feels more like a paragraph for the discussion/conclusion.
Figure 3, Why do the surge fatality projections in panels a-c show a kink in 2100? Please explain and discuss in the text.
Figure 4, There is not a sufficient description and explanation provided in the text why relocation costs under SSP5-8.5 show a peak in 2100 (panel f).
L478, “reasonable” may be a bit strong given the large number of assumptions, but the model does produce estimates that are comparable with other existing study results.
L560, This is a very important finding which could be elevated more in the conclusion.
L620-624, I do not understand the seemingly arbitrary choice of 0.8 m (0.78 to 0.82 m) of global mean SLR in 2100, also because this would be closer to the more pessimistic AR6 projections under SSP5-8.5 than what could be considered “realistic” if a current policy interpretation was used (SSP2-4.5 would be closest then). If it is the intention to use SLR under very high emissions, then I would suggest to actually reference the medium confidence AR6 projected range of 0.77m (0.63 to 1.01 m) as target for the ensemble selection.
L651, In the conclusion, it is at least important to discuss in more detail the overall challenges of adequately estimating future damages from sea level rise, also because it is so hard to integrate non-economic damages in such assessments.
Code availability and documentationI have cloned the FRISIA code repository and successfully ran the example scripts. It would be useful if the authors expanded the example script to also save key results in a dedicated output directory to increase usability.
Citation: https://doi.org/10.5194/egusphere-2025-1875-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,057 | 45 | 12 | 1,114 | 14 | 30 |
- HTML: 1,057
- PDF: 45
- XML: 12
- Total: 1,114
- BibTeX: 14
- EndNote: 30
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
In this paper, the authors present FRISIA 1.0, a spatially highly aggregated model of sea-level rise impacts on coastal assets, population, and adaptation costs. The paper builds upon the prior CIAM model (Diaz, 2016; Wong et al., 2022) by incorporating impacts on asset prices as well as on population and adaptation, thus providing a feedback mechanism by which the coastal impacts of sea-level rise can affect coastal assets beyond the effects of relocation costs. The model is calibrated against CIAM, but whereas CIAM considers benefits and costs along 12,000+ coastal segments, this model – designed for use in highly aggregated integrated assessment models – operates with 1-7 aggregate regions.
Major comments
Deeper comparison to the literature: Overall, I like this paper, but find it could benefit from a deeper comparison to the literature. Most especially, the effects on asset prices seem ex nihilo; there are no references to economic theory papers justifying the formalism, nor any comparison to the growing econometric literature (e.g., see review by Contat et al., 2024 on the empirical relationship between climate risk and real estate prices).
The authors would also benefit by a comparison to Depsky et al. (2023), who construct a coastal dataset similar in structure to DIVA but with updated inputs and then build a CIAM-like model in Python with improved representation of several processes and a data-informed estimate of the relocation cost parameter. Similar to the use of CIAM in the GIVE model described in the manuscript, the “DSCIM-Coastal” model of Depsky et al. (2023) was used for the same recent Social Cost of Greenhouse Gas (SC-GHG) estimation exercise as part of the Data-driven Spatial Climate Impact Model (DSCIM, Climate Impact Lab 2023). DSCIM-Coastal contains an updated and open-source alternative to the DIVA input data used by the authors, as well as an updated Python implementation of the model that the authors modify and reimplement in Python.
The authors would also benefit by a comparison to the more theoretically grounded dynamic spatial integrated assessment model of Desmet et al. (2021), who do not explicitly account for protection but do model flows of capital and population in response to sea-level rise, and like this paper show losses in exposed areas peaking and declining over time.
Appropriateness of CIAM parameterization: FRISIA applies formalism similar to those of CIAM, but rather than operating at a highly resolved coastal segment scale, it applies them at a global or World Bank-region scale. The appropriateness of doing this could be better justified. (Maybe this justification is simply – we are modeling for insights rather than for numbers, and we are fine with our flooding costs being within a factor of X of CIAM results.) As an integration test, the comparison to CIAM in Figures 3-4 does not always give the greatest confidence that FRISIA is a good emulator of CIAM (e.g., compare relocation costs in Figure 3).
To the extent the goal is to emulate the results of CIAM (after adjustment for the difference in the treatment of initial flood protection height), some quantitative performance metrics might be useful.
Treatment of relative SLR: In line 224, the authors refer to ‘relative SLR’, but then immediately say they implement only “global mean SLR”. At the same time, Table A1 suggests that some factor – whose derivation is not described anywhere – is used to localize the SLR projections. This should be clarified throughout.
Declining GDP per capita: It is a little hard to trace the relative effects of declining asset growth (decreasing GDP growth along coasts) and relocation (decreasing population) on GDP per capita (e.g., discussion around line 515). It would be helpful to look at this in greater detail.
Minor comments
Throughout: The authors refer to timeseries of ‘global SLR’; given that this is a univariate time series, I assume they mean ‘global-mean SLR’.
Throughout: The authors refer to “coastal GDP” and cite the SSP database, but do not explain where this coastal-specific GDP data comes from. To my knowledge, only country-level GDP estimates are contained in the SSP database. Is it the same population density-based downscaling approach used in Diaz, 2016?
Line 222: I am struggling to understand the reasoning behind building a second model for SLR, trained on the estimates from the first model. I assume this has something to do with the dynamic nature of the model, in which new SLR scenarios can be endogenously formed at each time step. However, this reasoning is not immediately clear in the manuscript. It would be helpful to include a sentence or two explaining why the outputs of the first SLR model cannot be used directly.
Line 994: “MICI” is probably not an appropriate shorthand for factors driving high-end mass loss from Greenland; see Fox-Kemper et al. (2021):
Code and Documentation
I've verified the code can be cloned from the repo and run successfully. I've not otherwise examined the code. I have the following comments:
The README file needs to provide adequate description to install code and run example. I suggest adding a requirements file. (Requires numby, pandas, matplotlib, netCDF4)
The following commands worked for me to set up the environment and run the example script:
conda create \--name frisia
conda activate frisia
conda install numpy pandas matplotlib netcdf4
git clone https://github.com/lnnrtrmm/FRISIA.git
cd FRISIA
python EXAMPLE\_runFRISIA.py
In general, the README file is quite minimal, and I would argue insufficient to constitute the “user manual” expected to accompany GMD model description papers. The only documentation of the code is the two class-level doc strings in the source code.
EXAMPLE_runFRISIA.py should produce output.
It would also be helpful to have notebook(s) and/or scripts that fully replicate the results shown in the text, especially Figs. 3–6.