`profsea-climate`: Add new Antarctic component by gregrmunday · Pull Request #28 · MetOffice/ProFSea-tool

Greg Munday (gregrmunday) · 2025-12-05T08:58:56Z

I've been working on an Antarctic ice-sheet emulation of ISMIP6 2300 simulations which responds more realistically to overshoot and idealised temperature forcings.

Addresses #27

I've added the component here but it still needs to be worked into the main emulator code!

Greg Munday (gregrmunday) · 2025-12-05T09:21:42Z

A look at how it behaves over a crazy ramp up ramp way down forcing:

Greg Munday (gregrmunday) · 2025-12-05T14:01:26Z

profsea/notebooks/ais_fitting.ipynb

Jennifer Weeks (@mo-jenniferweeks) I've uploaded the AIS emulator fitting notebook here - I'm still changing some things around in it and it might be a bit messy at the moment!

I'm working on fitting the emulator with the 19 common IS models across all scenarios as opposed to 43 across half of the scenarios to see how it compares!

Greg Munday (gregrmunday) · 2025-12-05T14:04:39Z

WIP comparison between AR5 emulator and ISMIP6 emulator:

Plumes are 66% likely range. At the moment it looks a fair bit less sensitive than the AIS emulator to warming, at least over this timescale - although the ramp up in SLE under higher emissions looks a lot more non-linear.

Next step is probably to compare this against AR6 and against the other emulators in SLEIP

Greg Munday (gregrmunday) · 2025-12-05T15:17:14Z

Comparing ISMIP6 emulation including only 14 of the contributing models, which submitted simulations over all 10 forcing scenarios.

Seems to just lower the projection a bit, the spread is almost captured but this set of models happens to be biased towards a smaller response to temperature, and have a high proportion of long-timescale responses.

Hemant Khatri (hmkhatri) · 2025-12-08T15:12:29Z

It is interesting that ISMIP6 response is much smaller than AR5 (and possibly AR6). I think we should keep both options in the code - (1) standard AR5/AR6 (2) improved ISMIP. So, users can choose the algorithm they prefer. Of course, we will need references and some disclaimer about different methods and what causes these differences in projected sea level.

Greg Munday (gregrmunday) · 2025-12-08T15:16:56Z

It is interesting that ISMIP6 response is much smaller than AR5 (and possibly AR6). I think we should keep both options in the code - (1) standard AR5/AR6 (2) improved ISMIP. So, users can choose the algorithm they prefer. Of course, we will need references and some disclaimer about different methods and what causes these differences in projected sea level.

Hmm yes, I'm not totally satisfied with this to be honest. The parameterisation captures the range of the ISMIP6 ensemble well but it does underestimate the median by quite some bit, so I think still a bit of work required to find a better fit to the data!

Greg Munday (gregrmunday) · 2025-12-08T15:18:16Z

It is interesting that ISMIP6 response is much smaller than AR5 (and possibly AR6). I think we should keep both options in the code - (1) standard AR5/AR6 (2) improved ISMIP. So, users can choose the algorithm they prefer. Of course, we will need references and some disclaimer about different methods and what causes these differences in projected sea level.

But I agree that the user should be able to choose which parameterisation to go with. We can use whichever one we feel most confident in as the default, eventually.

Jennifer Weeks (mo-jenniferweeks) · 2026-01-16T10:18:13Z

Greg Munday (@gregrmunday) is it possible to compare with the Edwards ISMIP emulation too, even though it is to 2100 only?

Greg Munday (gregrmunday) · 2026-01-16T11:15:41Z

Greg Munday (@gregrmunday) is it possible to compare with the Edwards ISMIP emulation too, even though it is to 2100 only?

Good idea, I'll assume the data is available with their paper?

Greg Munday (gregrmunday) · 2026-04-03T10:47:11Z

Hemant Khatri (@hmkhatri) this is ready for review!

Lots of changes throughout this one:

Optimisation and refactor of the Global class
Added the ISMIP6Emulator to the Global class
Updated probabilistic_wrapper.py and fixed some bugs
Fixed bugs in, improved the performance of, and improved the usability of the Spatial class

I would like to think about making all the auxiliary data required to run ProFSea easily available in some shape or form. One approach would be like how we do it in SpeedyWeather, with a separate repo for data (call it something like ProFSeaAssets). I'm sure we could compress all the CMIP6 and GRD/GIA patterns small enough to do that.

In that case we could include some code to download it locally when a ProFSea user calls Spatial() the first time, then after that it's already local. This could be a good idea as not every ProFSea user will want to produce spatial projections, they may just want to quickly make some global ones in which case having to download the extra spatial-related data is a pain.

If we can compress the spatial data very small, then we could just include it in the repo, but this could be less ideal.

Hemant Khatri (hmkhatri) · 2026-04-07T08:39:31Z

Greg Munday (@gregrmunday) Thanks for the pull request. I will review it this week.

Hemant Khatri (hmkhatri) · 2026-04-10T08:54:09Z

profsea/emulator/antarctica.py

Greg Munday (@gregrmunday) Could you add some comments about the approach in antarctica.py ? This is the main part of the calculation and some notes would be useful

For sure! Will upload in the next commit

Added this in the latest commit. I don't think we need to go into excessive detail here as it'll be detailed in the paper.

Hemant Khatri (hmkhatri)

Greg Munday (@gregrmunday) thanks for the pull request. I have made some suggestions. Could you please look into those?

Also, it would be useful to add an option (wherever possible) to choose from different available methods for calculating a component. It provides more clarity to a user on what exactly is running behind ProFSea.

See issue #38

Hemant Khatri (hmkhatri) · 2026-04-10T08:59:14Z

profsea/emulator/gmslr.py

        self.random_sample = random_sample
        self.T_percentile_95 = T_percentile_95
        self.OHC_percentile_95 = OHC_percentile_95
-        self.cum_emissions_total = cum_emissions_total


I suggest retaining this. If we want to use AR5 method for icesheets, then we would need cumulative emissions.
Similar to glaciermip: bool | int = 2, could provide option to choose between ISMIP or AR5 method.

Alright I've kept this in. I agree we should come up with a nice way for users to choose how they want to build the model up with different components...

Hemant Khatri (hmkhatri) · 2026-04-10T09:04:06Z

profsea/emulator/gmslr.py

+        if "antarctica" in self.active_components:
+            wais_path = Path(__file__).parent / "aux_data" / "wais_params.nc"
+            eais_path = Path(__file__).parent / "aux_data" / "eais_params.nc"
+            aispen_path = Path(__file__).parent / "aux_data" / "aispen_params.nc"
+            self.wais_model = AntarcticaISMIP6(wais_path)


Is ActarcticaISMIP6 the only method for antractica component or is there an option to run with the older method? Better to retain both, in case one wants to run compare the two.

It's not the only way, the user can specify antsmb and antdyn, although maybe this should be simplified with a general antarctica_ar5 component which models both.

Hemant Khatri (hmkhatri) · 2026-04-10T09:12:40Z

profsea/emulator/gmslr.py

+        self.scenario = scenario
+        self.T_change = np.asarray(T_change)
+        self.OHC_change = np.asarray(OHC_change)
+        self.cum_emissions_total = cum_emissions_total


There is no option to supply cum_emissions_total. Sugegst keeping this input for now.

ProFSea-tool/profsea/emulator/gmslr.py

Lines 135 to 155 in dd7ef43

def __init__(

self,

end_yr: int,

seed: int = 1234,

nt: int = 100,

nm: int = 1000,

tcv: float = 1.0,

glaciermip: bool | int = 2,

parallel: bool = True,

input_ensemble: bool = True,

output_percentiles: list | np.ndarray = None,

random_sample: bool = False,

T_percentile_95: np.ndarray = None,

OHC_percentile_95: np.ndarray = None,

palmer_method: bool = True,

active_components: list[str] = None,

) -> None:

self.end_yr = end_yr

self.nt = nt

self.nm = nm

I've re-added this

Hemant Khatri (hmkhatri) · 2026-04-10T09:25:55Z

profsea/emulator/gmslr.py

+                    ),
+                ]
+            else:
+                raise KeyError("glaciermip must be 1 or 2")


It seems glaciermip = 3 is a valid option. I guess valid options for glaciermip input are False, 1, ,2 or 3. Is this correct?
Even if glaciermip is set to False, the else condition kicks in and computes its contribution.

ProFSea-tool/profsea/emulator/gmslr.py

Lines 600 to 615 in dd7ef43

else:

raise KeyError("glaciermip must be 1 or 2")

else:

glparm = [

dict(name="Marzeion", factor=4.96, exponent=0.685),

dict(name="Radic", factor=5.45, exponent=0.676),

dict(name="Slangen", factor=3.44, exponent=0.742),

dict(name="Giesen", factor=3.02, exponent=0.733),

]

cvgl = 0.20 # random methodological error

ngl = len(glparm) # number of glacier methods

r_per_model = self.nm // ngl

r_remainder = self.nm % ngl

r = self.rng.standard_normal(self.nm)

I am confused about which valid options are available for a user.

I've removed the glaciermip3 option - I thought about including results from the latest glaciermip but I think this should be tackled in its own PR later down the line.

Hemant Khatri (hmkhatri) · 2026-04-10T09:28:18Z

profsea/emulator/spatial.py

                "Provide namelist of global projection components. "
                "Currently allowed components are expansion, antdyn, "
-                "antsmb, glacier, greenland and landwater")
+                "antsmb, wais, eais, glacier, greenland and landwater")


Need some info on how antsmb, antdyn are different from wais, eais.

I've update this error message to include more info!

Hemant Khatri (hmkhatri) · 2026-04-10T09:34:41Z

profsea/emulator/spatial.py

        self.components = {}
-        component_path = Path(components_dir) / f"{scenario}_global.nc"
-        ds_component = xr.load_dataset(component_path)
+        component_path = Path(components_dir)


Is this correct? Where is the filename of global.nc used?

You're right - this is poorly named, I think the user should provide the exact path to the file here.

Hemant Khatri (hmkhatri) · 2026-04-10T09:47:01Z

profsea/probabilistic_wrapper.py



-def run_fair(baseline_start: int, baseline_end: int, scenarios: list) -> tuple[np.ndarray]:
+def load_fair_forcing(


Now sure about having load_fair_forcing function inside ProFSea. I suggest keeping ProFSea more general rather than tied to any specific climate emulator.

There is code for running FaIR as well here. I would suggest not to keep this in ProFSea. A user should run FaIR (or any other simple climate model) separately and supply inputs as needed.

I think this is a fair point (no pun intended) but I do think it's super useful to have a script with FaIR baked in lying around somewhere in the repo, even if users don't need to rely on it all the time. We need to think a bit more carefully about what scripts are useful to provide along with the generally usable Global and Spatial objects, and where we should keep them.

Added new antarctic component

e6481bd

Greg Munday (gregrmunday) added enhancement ✨ New feature or request antarctica 🇦🇶 emulator 👾 labels Dec 5, 2025

Improve performance + add ais_params.nc

a29f565

Add AIS and other notebooks

6deca3b

Greg Munday (gregrmunday) self-assigned this Dec 5, 2025

Greg Munday (gregrmunday) commented Dec 5, 2025

View reviewed changes

Greg Munday (gregrmunday) added 2 commits December 7, 2025 15:07

Updated code

2c09229

Dev updates

198652a

Greg Munday (gregrmunday) added 2 commits December 9, 2025 12:44

model update

f4e68c9

Merge branch 'profsea-climate' into gm/add_AIS_emulator

18701d9

Greg Munday (gregrmunday) marked this pull request as draft December 9, 2025 13:20

Resolve merge conflicts

7b37ee1

Greg Munday (gregrmunday) changed the title ~~Add new Antarctic component~~ profsea-climate: Add new Antarctic component Jan 16, 2026

Greg Munday (gregrmunday) added 7 commits January 16, 2026 16:49

Merge branch 'profsea-climate' into gm/add_AIS_emulator

19f399b

More AIS fitting...

27da332

Update comparison notebook

b2ad506

Update for latest version

a9eff7a

Mo' updates

41f3284

Smol fix

cd616ce

Optional spatial sampling

eff2802

Greg Munday (gregrmunday) added 4 commits March 9, 2026 11:28

Remove import

e80366e

Add params

792ad55

Add dim check and reformat

4670d38

Fix bugs in probabilistic_wrapper.py

4f29b6d

Greg Munday (gregrmunday) marked this pull request as ready for review April 3, 2026 10:21

Greg Munday (gregrmunday) requested a review from Hemant Khatri (hmkhatri) April 3, 2026 10:21

Remove unnecessary notebooks

763db36

Fix source attribute bug

dd7ef43

Hemant Khatri (hmkhatri) reviewed Apr 10, 2026

View reviewed changes

Hemant Khatri (hmkhatri) requested changes Apr 10, 2026

View reviewed changes

Greg Munday (gregrmunday) added 3 commits April 10, 2026 16:34

Add comments and reformat

64cdbb3

Smol fix

11e388e

Address review comments

dfef6e9

Greg Munday (gregrmunday) requested a review from Hemant Khatri (hmkhatri) April 10, 2026 16:02

	def __init__(
	self,
	end_yr: int,
	seed: int = 1234,
	nt: int = 100,
	nm: int = 1000,
	tcv: float = 1.0,
	glaciermip: bool \| int = 2,
	parallel: bool = True,
	input_ensemble: bool = True,
	output_percentiles: list \| np.ndarray = None,
	random_sample: bool = False,
	T_percentile_95: np.ndarray = None,
	OHC_percentile_95: np.ndarray = None,
	palmer_method: bool = True,
	active_components: list[str] = None,
	) -> None:

	self.end_yr = end_yr
	self.nt = nt
	self.nm = nm

	else:
	raise KeyError("glaciermip must be 1 or 2")
	else:
	glparm = [
	dict(name="Marzeion", factor=4.96, exponent=0.685),
	dict(name="Radic", factor=5.45, exponent=0.676),
	dict(name="Slangen", factor=3.44, exponent=0.742),
	dict(name="Giesen", factor=3.02, exponent=0.733),
	]
	cvgl = 0.20 # random methodological error

	ngl = len(glparm) # number of glacier methods

	r_per_model = self.nm // ngl
	r_remainder = self.nm % ngl
	r = self.rng.standard_normal(self.nm)



		def run_fair(baseline_start: int, baseline_end: int, scenarios: list) -> tuple[np.ndarray]:
		def load_fair_forcing(

Conversation

Greg Munday (gregrmunday) commented Dec 5, 2025

Uh oh!

Greg Munday (gregrmunday) commented Dec 5, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Greg Munday (gregrmunday) commented Dec 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Greg Munday (gregrmunday) commented Dec 5, 2025

Uh oh!

Hemant Khatri (hmkhatri) commented Dec 8, 2025

Uh oh!

Greg Munday (gregrmunday) commented Dec 8, 2025

Uh oh!

Greg Munday (gregrmunday) commented Dec 8, 2025

Uh oh!

Jennifer Weeks (mo-jenniferweeks) commented Jan 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Greg Munday (gregrmunday) commented Jan 16, 2026

Uh oh!

Greg Munday (gregrmunday) commented Apr 3, 2026

Uh oh!

Hemant Khatri (hmkhatri) commented Apr 7, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Hemant Khatri (hmkhatri) left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Greg Munday (gregrmunday) commented Dec 5, 2025 •

edited

Loading

Jennifer Weeks (mo-jenniferweeks) commented Jan 16, 2026 •

edited

Loading