Unexpected results in InVEST Coastal Vulnerability model when comparing SLR scenarios (SSP 1-2.6, 2-4.5)

Software Support
1

Hi everyone,

I am developing a project to quantify the global coastal protection potential provided by mangroves under sea-level rise (SLR) scenarios. I am using the InVEST Coastal Vulnerability model for this analysis. I have created four different SLR scenarios: one for the present (2020s) and three for the year 2070 based on Shared Socioeconomic Pathways (SSPs): 1-2.6 (optimistic), 2-4.5 (intermediate/realistic), and 5-8.5 (pessimistic). All input data remain identical except for the SLR variable.

I would like to share my methodology and some questions regarding the results I obtained.

Methodology:

  1. I downloaded the SLR rasters (1km resolution) from the IPCC AR6 via Google Earth Engine.

  2. In GIS software, I converted the rasters into point shapefiles and clipped them to my Area of Interest (AOI) to reduce unnecessary data points, as the InVEST manual states the model uses the two nearest points for calculations.

  3. To compare scenarios, I calibrated the SLR values. I identified the maximum variation range across all scenarios (SSP 5-8.5) and divided it into 5 percentiles. In QGIS, I created a new rank column in the attribute table using the following logic:

    • CASE

      • WHEN “VALUE” <= 189.2 THEN 1

      • WHEN “VALUE” <= 378.4 THEN 2

      • WHEN “VALUE” <= 567.6 THEN 3

      • WHEN “VALUE” <= 756.8 THEN 4

      • ELSE 5

    • END

  4. After calibrating each SLR shapefile, I ran the InVEST model.

Post-processing:

I categorized the exposure results into risk levels: 1–2.33 (Low Risk), 2.33–3.66 (Moderate Risk), and 3.66–5 (High Risk).

The Issue:

My output analysis showed an unexpected result: the “Realistic” (SSP 2-4.5) scenario showed a higher percentage of the coast in the “Low Risk” category compared to the “Optimistic” (SSP 1-2.6) scenario. This is counterintuitive, as I expected the low-risk areas to decrease as the SSP scenarios became more severe.

Upon review, the input SLR values are correct and follow the expected progression (higher SLR in SSP 2-4.5 than in 1-2.6). I suspect that InVEST is re-calculating its own percentiles even with my pre-calibrated (SLR rank from 1 to 5) values. Since the SSP 2-4.5 scenario might not contain all ranks from 1 to 5 (perhaps only 4 and 5), the model seems to be calculating the index based only on the values present in that specific run.

Proposed Solution: I am considering recalculating the geometric mean in the output attribute table during post-processing by manually joining the raw SLR data with its original 1–5 rank column.

What do you think of this approach? Would you recommend a different workflow for this type of multi-scenario analysis?

Here is the link to my data acess: InVEST forum - Google Drive

Thank you!

2

Hi @marianamendesm

Yes, that’s correct. InVEST has no idea that you already did your own ranking, so it assumes the values given represent actual SLR measurements or rates, and it does its own percentile-based ranking for each model run.

Yes, this is a great approach for comparing across scenarios.