Include option to run brownfield optimisation

[OskarVagero]

Currently, there is limited possibility to run the model with existing electricity generation and transmission infrastructure. Typically, only hydropower and some Nuclear infrastructure is considered in the hybrid-greenfield approach.

We have been discussing adding the feature of running a brownfield optimisation where:

1. Existing VREs (wind and solar) are included.
2. Existing transmission infrastructure exists and can also be expanded (which is different from running with fx_trans == 'ON')

There are some methodological choices one needs decide upon to make this happen.

[OskarVagero]

For the first point, one initial step is to find data on existing VREs. A few tools/databases exist:

1. powerplantmatching
2. Global Energy Monitor
3. Open Power System Data
4. DTU's wind power database (not publicly available at the moment)

From some initial analysis, powerplantmatching does not seem capable of cross-referencing large datasets of solar and wind power plants (see this issue and section 2.4 of Gotzens et al. (2019))

Since especially wind and solar units are comparatively small in terms of unit-wise capacity (ranging from very-low kW to low MW scale), but huge in terms of deployed numbers (e.g. more than 1.7 million single units only in Germany [24]), they would massively impede the matching process. Moreover, since most of the single solar panels and wind turbines do not have a specific name (except larger wind and solar parks), there is no data in the name column and, consequently, no input for the string comparator available. Therefore, all wind and solar units are being filtered as part of the data mending, enabling us to keep the entire process computationally manageable.

However, for a further usage of the derived power plant data in modelling approaches, PPM is able to concatenate given wind and solar units from the OPSD renewable data package [24] to the final dataset at the end of the matching process.

From an initial look, GEM, which has full coverage of all highRES countries does not look too bad, with some exceptions. OPSD data is not available for all countries, but could act as a complementary data source for those countries where it is available. On the other hand, the DTU database seems to have similar levels of installed capacity as OPSD, but covering all countries, although it falls short for some countries, such as Sweden, Norway and Finland.

Note that the DTU database is only in a preliminary version, and that this can be analysed fully once it's ready (early July 2024)

[OskarVagero]

An alternative approach worth looking into is to distribute the country-level data (which should be available with good confidence) based on some heuristic.

PyPSA-Eur says the following:

Existing wind and solar capacities are retrieved from IRENA annual statistics and distributed among the nodes in a country proportional to capacity factor. (This will be updated to include capacity distributions closer to reality.)

Some testing of how the heuristic compares to real distributions should be explored.

[guillerval]

An alternative approach worth looking into is to distribute the country-level data (which should be available with good confidence) based on some heuristic.

PyPSA-Eur says the following:

Existing wind and solar capacities are retrieved from IRENA annual statistics and distributed among the nodes in a country proportional to capacity factor. (This will be updated to include capacity distributions closer to reality.)

Some testing of how the heuristic compares to real distributions should be explored.

I had a look at the heuristic, and it looks "fine". Some considerations:

• As Oskar said, the heuristic distributes the country-level capacities into nodes in a country based on capacity factors. I followed the heuristic defined in Pypsa, and I defined it for our case:
  • I considered the same number of nodes as regions in a country (NUTS2) (I also did for NUTS3, but it is similar)
  • I defined the annual mean capacity factor per region (I selected 2020), and then I calculated the total capacity factor per country (I summed the capacity factor per region in each country)
  • I divided the annual mean capacity factor per region of a country by the corresponding total capacity factor per country.
  • Finally, I have a factor (fraction) per region in a country
• The heuristic was only defined for the nuts2 regions in our cutout (e.g. Iceland wasn't considered because the cutout doesn't contain that part)
  • I assumed Iceland with a factor of 1 (only one region in NUTS2)
• The plot only focused on the onshore area.
  • I couldn't apply the heuristic to the offshore area because I didn't know how many nodes/regions I should consider.

The plot of the existing capacities applying the heuristic. Left side: heuristic, and right side: DTU database

And here is the difference between both. Red means DTU is higher, and the blue heuristic is higher (white means no difference)

• The heuristic usually overestimates the installed capacities in some region (more blue)
• It tries to prioritize the region with better capacity factors, but it assigns more capacity to regions where it is not possible to install (because of topography)
  • It could be improved by excluding some grid cells from the cutout (maybe?).