Implement real estate vulnerability model

[joemoorhouse]

This issue is for the implementation of a real estate model with riverine and coastal flooding hazards.

Hi @floriangallo, please can you put the link to the paper here? Plus any thoughts you have about which bits to implement. Let's track all that in this issue and I'll code it up.

[floriangallo]

Hi @joemoorhouse
Here's the link to the paper containing damage curves for different building types on different continents. I suggest we start with textile facilities in Asia, I managed to extract a few thousands lines from the Open Apparel Registry, lets' discuss on Thursday.
https://publications.jrc.ec.europa.eu/repository/handle/JRC105688

[joemoorhouse]

From a catch-up today:

Implement all damage functions in:
copy_of_global_flood_depth-damage_functions__30102017.xlsx
Handle standard deviation using

1. Beta distribution
2. Truncated Gaussian distribution

Asset:
PhysCountry (assumed populated as on-boarding task)
Latitude
Longitude

class RealEstate
Type (property): Building/Residential, Building/Commercial, Building/Industrial, Transport, Infrastructure/Roads, Agriculture?

The general rule is: add derived Asset class if we find ourselves adding a property that is meaningless for other assets, e.g. 'number of floors' would be irrelevant for roads, so need different class

[joemoorhouse]

Had a quick call with @DavideFerri yesterday. We want to separate our model from damage function / uncertainty config, config represented as separate JSON. I will work on that part. Davide is looking at supporting truncated Gaussian as well as Beta uncertainties.

[joemoorhouse]

@floriangallo, on subject of country/continent look-up, from lon/lat, geopandas "naturalearth_lowres" looks pretty appropriate. I am adding the look-up to physrisk, although only to be used for on-boarding; as discussed, I think we want to make country an input properly of the asset. Also that avoids adding geopandas as a dependency - which does draw in a lot of other libraries.

[joemoorhouse]

Thanks for the tip @ryan-vaughn-jupiter
A more precise map here:
https://gadm.org/download_world.html

[joemoorhouse]

@ryan-vaughn-jupiter, I'll also tag you on the associated PRs as discussed. Idea is also to demo a financial impact calculation based on this, although not going as far as credit modelling (yet).

[floriangallo]

Hi @joemoorhouse , as I missed the last weekly meeting, I was wondering how things were progressing, did you have a chance to try to run the scripts with the apparel facilities? Any issue with that?
Thanks

[joemoorhouse]

Hi @floriangallo, sorry for late reply; Easter intervened! I have run through end to end with the full set of assets (~50,000) or so, more as a check on performance: need to do a bit more testing of the figures. Main/only issue is not having the value of each asset in that set. We could just use placeholder values (e.g. the typical ones) on the assumption that we would have the values in future. Or else we can capture an uncertainty in the value here as well: another conditional probability. Any preference/thoughts?

[floriangallo]

Hi @joemoorhouse, my turn to provide a late answer... Yep, true that the lack of value is a problem for the financial assessment. I guess I'd go with placeholder values (assuming that any user who would want to full financial risk assessment would have some production numbers or valuation). If they don't, having similar, typical values for each asset would provide a equally-weighted risk, which is wrong, of course, but logically and understandably wrong, I'd say!

[joemoorhouse]

Complete.