Overestimated Impact in CLIMADA Despite Calibrated Exposure and Flattened Impact Functions (SIDR Case Study)

[ketanpednekar]

Summary

I am modeling the 2007 SIDR cyclone using CLIMADA's TC module, LitPop exposure via API, and district-level damage data from Bangladesh. I've scaled exposure values to match actual damage totals and manually flattened impact functions to reflect realistic vulnerability tiers.

Despite this, modeled impact is still 3,000–15,000% higher than actual damage across all districts.

---

What I have Done

• Used IBTrACS track for SIDR
• Fetched LitPop exposure via CLIMADA API
• Spatially joined exposure to 5 affected districts
• Scaled exposure totals to match actual damage
• Assigned district-wise impact functions:
  • Satkhira: steep (now flattened to MDD 0.7–0.85)
  • Barisal/Khulna: moderate (MDD 0.5–0.7)
  • Patuakhali/Barguna: gentle (MDD 0.35–0.6)
• Ran impact calculation and compared modeled vs actual damage

---

Results

District	Actual Damage (USD)	Modeled Impact (USD)	Error (%)
Satkhira	90M	13.85B	+15,296%
Khulna	110M	14.25B	+12,859%
Barisal	100M	8.28B	+8,182%
Patuakhali	130M	8.04B	+6,087%
Barguna	150M	4.69B	+3,032%

---

Screenshot attached..

---

Question

Is this a known issue with TC modeling in CLIMADA? Are there recommended practices for further calibrating impact or validating hazard intensity?

Would appreciate any insights or suggestions. Happy to share code snippets or rerun diagnostics if needed.

Thanks!

[peanutfun]

Hi @ketanpednekar and thank you for raising an issue. Without your source, it is hard to follow what exactly went wrong. However, I do not understand your approach to calibration. Why did you "scale exposure totals to match actual damage"? The exposure should be considered an invariant of the model and cannot be deduced from the acutal damage.

Please note that Climada contains a module for impact function calibration, with a tutorial. I suggest you use this and try to calibrate impact functions on the original, unscaled exposure. I would be interested to see the results.