http://nsidc.org/arcticseaicenews/
Artic sea lows are usually in september. Will resolve yes if new record reached. 2012 record
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ2,513 | |
| 2 | Ṁ443 | |
| 3 | Ṁ435 | |
| 4 | Ṁ217 | |
| 5 | Ṁ171 |
Average Arctic sea ice extent for September 2023 was 4.37 million square kilometers (1.69 million square miles), placing it fifth lowest in the 45-year satellite record (Figure 1a). Following the annual minimum of 4.23 million square miles (1.63 million square kilometers), the growth in Arctic sea ice extent has been slower than average (Figure 1b). The image plot for October 2 from the Advanced Microwave Scanning Radiometer 2 (AMSR2) instrument provided by the University of Bremen shows that with the cessation of melt, the ice edge has become more sharply defined (Figure 1c).
Certainly looks like we have a minimum, and that it's meaningfully higher than the record low. @ThatGuy How long should we wait for finalized data / something weird?
Created an extent market here:
https://manifold.markets/ScottSupak/what-will-be-the-arctic-sea-ice-min
Data/model talk: I had time to do cross-validation to tune different hyper-parameters along with different cutoffs, and validation to test against a different cutoff with with the new hyper-parameters. These gave me a better idea of when the model fails. It usually fails during extremes such as 2012, and despite how much adjustment of the hyper-parameters is done, it creates a large discontinuity for the immediate next forecast day after the last observation, rather than keeping a smooth trend, this leads it to over-estimate the sea-ice extent and end up with a minimum that is out of the 99.5% range. The 99.5% interval is not a good judge of the accuracy since trends for sea-ice extent have plenty of variability but it is helpful for finding a useful spread most of the time. I used whether the actual minimum for different forecast date cutoffs falls in the 99.5% interval for cross-validation as a measure for the model's accuracy in predicting the minimum. I also repeat the runs for the current forecast to see if there is any chance of a new record minimum falling into the bounded range. Calculations for two forecast ranges for cross-validation accuracy percent: https://github.com/JRPdata/sea-ice-extent/blob/main/cross-valdiation-sea-ice-tuning.xlsx
The model's accuracy (having two different forecast ranges and parameters) at predicting (during Late July-early September) the sea-ice extent minimum falls in the range of between 75% to 90%; after ~8-25, it is 85-90%.
New run (tuned) with latest data:
Minimum Forecast Value: 4.841
Lower Bound: 3.565
Upper Bound: 5.805
Since it is now configured for (only) yearly trends, here is the (seasonal) yearly trend for the Prophet model (for reference). Since it is a historical model it doesn't know anything else (including El Nino) so it will likely underestimate the amount of downward trend and variability:
Notebook (it is large since it has all cross the validation output) (doesn't show the prophet plots on github, unfortunately):
https://github.com/JRPdata/sea-ice-extent/blob/main/sea-ice-extent-forecast-validate.ipynb
@parhizj
Update with last two days of data. Forecast for market hasn't changed.
Minimum Forecast Value: 4.825
Lower Bound: 3.643
Upper Bound: 5.772
@parhizj
None of the 20 runs had it go below the record minimum (not updating bet):
Minimum Forecast Value: 4.83
Lower Bound: 3.524
Upper Bound: 5.902
@parhizj
This is the first run I've had that gets a lower bound near the record, but it's still not within range (none of the 20 runs did either). The trend based on my previous posts has the predicted lower bound getting pushed slightly more downwards. There is still no abnormal discontinuity yet in the first day of the projection so that is good. Based on the tuning/validation I did previously, I'll wait until August 4th before increasing my NO bet again.
Minimum Forecast Value: 4.834
Lower Bound: 3.458
Upper Bound: 5.837
@parhizj
Minimum Forecast Value: 4.822
Lower Bound: 3.584
Upper Bound: 5.798
Still predicting NO, but since the market has more confidence at this date for this type of prediction (market is at 18%, where I only have 80% confidence, so I can not go below 20%), I am selling my NO shares.
@parhizj
Minimum Forecast Value: 4.833
Lower Bound: 3.627
Upper Bound: 5.855
Still predicting NO but the market is too confident for me to buy more NO or go against it.
Edit: The minimum was set in 2012 and was ~3.41 10^6 km^2 (this is a bit higher than the daily data).
I experimented with Prophet (the Facebook package on github) to try to forecast this with only historical data (99.5% CI) (for reference it falls out of the interval for 99.95%), using daily historical data from nsdic.org above (requires annoyingly alot of pre-processing of data).
Given that this is purely prediction from historical data and not a physical model, it's predictive value is unknown, especially with this year's conditions, so I won't assign a point probability to the market resolving YES/NO. I'll only assign a ~60% confidence in this analysis being right, and adjust my bet size to this. Since the low predicted is higher than the record, I am betting NO.
Project with notebook is here: https://github.com/JRPdata/sea-ice-extent
Minimum Forecast Value: 4.943
99.5% Interval:
Lower Bound: 3.715
Upper Bound: 5.961@parhizj
I don't know if it will change drastically with July-August so I'll try to check to see if the prediction changes much (which it hasn't much after only 2 more days of data):
Minimum Forecast Value: 4.943
99.5% Interval:
Lower Bound: 3.828
Upper Bound: 5.943
@parhizj The record minimum should say ~3.41 not 1.31, none of the edits I've put through end up going through.... The 1.31 million was in miles^2 not km^2.
Its looking Eh. Could still be something
Isn’t this a confirmed YES already as of February 14, 2023?
https://nsidc.org/arcticseaicenews/2023/02/antarctic-sea-ice-extent-sets-a-new-record-low/
Antarctic sea ice extent appears to have broken the record low set last year. With a couple more weeks likely left in the melt season, the extent is expected to drop further before reaching its annual minimum.
@MartinModrak Yes, wrong pole. This explains some of the YES bets, but there are enough other small YES bets to make me suspect something funny is going on. Maybe the market is linked in a place where people either have non-obvious info suggesting this will happen, or think they have such info?
@StevenK agree that it shouldn't be that far from your market on the warmest year ever.... I have no idea what's going on....
@MartinModrak The connection to global mean temperature seems pretty loose. According to GISTEMP, every year since 2012 has been warmer than 2012 (and a few years before 2012 were warmer as well), but 2012 still had the record low arctic sea ice.
@StevenK There hasn't been a record since 2012, which seems to have been a weird outlier. Yes, it's getting warmer over time, but warm years like 2016 and 2020 didn't set a record, so why assign a high probability to a record in 2023?
@StevenK EL nino was declared and current trend is hard to diagnose. I guess it might be a long shot.
