: Recent finds that in during the last two decades induced a higher warming and cloud level rise than that caused by , which threatens and water supply in African montane forests. Scientists call for urgent protection of ecosystems

It was hot this weekend and the cats were sick, so I couldnt sleep. In the feverish state of one seeking sleep but not finding it and finding instead not one, but two vomiting cats, I became obsessed with an idea: how quickly can I make an infilled dataset from I wanted to know, roughly what the tendency was and whether the changes in land and ocean trends made any difference to the general observation that infilling increases the overall warming.

I already had a vague idea of what to do:

1. Use ordinary kriging to fill in the gaps for land and ocean separately. Ordinary Kriging (OK) is nice because it assumes that the field has a non-zero mean, which is estimated at the same time as the interpolation. Plus its sort of like what was done for HadCRUT, so I could borrow approximate parameter values from there.
2. Blend the two using the HadCRUT prescription, but without too much faffing. To simplify this, Id use HadISST2 regridded to 55 lat-lon resolution. Grid cells that were a combination of land and ocean would get an area weighted blend and sea ice covered areas would be declared land. Any land area fraction greater than zero but less then 0.25 would be set to 0.25 (to give small islands greater weight in the blend).

Its a bit of faff setting up the kriging so I used Open AI to do that for me. I would rate it at about 1/10 for this task. It completely failed to solve it. It started off by providing me with a toy example using pykrige on a 55 grid that worked OK, but on being asked to scale that up to a global 55 lat-lon grid, using great circle distances between points, it failed spectacularly. It calculated great circle distances between points, but then never used them in the kriging calculation. It failed to fix this, even after it was pointed out twice. It jiggled things around and invented keywords for functions, but never got close to getting it right and very close to setting fire to my computer: via a misused keyword, it managed to create a situation that generated a 54 Gb array, which it then proceeded to do something computationally intensive to. The temperature of my office increased by 5 degrees and the fan sound woke up the cats.

I then switched tack. I didnt really want the great circle distance because it doesnt work properly with the covariance functions I wanted to use. So, I asked it to use simple Euclidian distances instead. After a false start where it assumed that I wanted lats and lons to be treated as xs and ys, it gave me a function that converted lats and lons to (x,y,z) arrays1, but was still no closer to actually using the distances in the interpolation. It was also switching unbidden between Ordinary Kriging and Universal Kriging, which is fine, but it also threw in Cubic interpolation too. Not fine.

By this point, though it wasnt doing what I wanted, it was doing something, so I asked it to plot out the result as maps because I like to look at pictures. It plotted out the grids, but as simple arrays and not as maps. I had to tell it specifically to use cartopy, then it got the idea. For some reason, when using cartopy, it becomes obsessed with drawing borders between countries. I never ask it to, and sometimes even when I tell it not to, it does it anyway. When I point this out, it says, oh year, my bad, heres a script that doesnt do that. Then it creeps it back in an iteration or two later.

After a few more iterations, it was still using the kriging functions without any of the modifications for calculating distances. It also insisted on using a linear covariance model and it was still having difficulties sticking to my instructions for plotting the results.

Then I ran out of Chat-GPT 4 credits.

I had a choice, wait 24 hours for my credits to refresh, or use the programming skills I have spent 40 years honing2. I could also, I suppose, have gone back to bed at this point, but once I start something like this, I find it very hard to stop.

The pykrige documentation is pretty good and I quickly surmised that I needed to use OrdinaryKriging3D rather than OrdinaryKriging as that would allow me to plug the cartesian coordinates straight in. Pykrige also allows you to specify a covariance model using gstools. This is exactly what I wanted. I like to krige with a Gaussian covariance3 and it was but a moments work to specify a 3D Gaussian covariance model. There were many beguiling options length scales4, anisotropy, angles, variances, nuggets but I stuck to a simple isotropic model with guesstimated variance (3 for land, 1 for ocean), length scale (1200 km) and nugget (0.25).

As a test, I used the pre-blended temperature field in the . The result was encouraging. The gaps were filled and the result looked reasonable. In the early record, the ordinary kriging tends to the mean of the available observations in data voids, which is not desirable when the fields are mostly data voids and the available data are clustered together in the northern hemisphere and highly variable. I used the uncertainty information that comes out of pykrige to mask the output, setting the infilled values to missing where the uncertainty exceeded the variance of the covariance model. That seemed to do a reasonable job.

I wrote a quick algorithm to blend land and ocean using as the sea ice field and land mask5 and then set it up to infill the land and oceans separately before blending them together. Ideally, both would get kriged together, but the variability over land and ocean is quite different. For that matter, variability over land varies by latitude and season, but thats a problem for another day.

I set it going and it looked to be running fairly fast. The 1850s took about 20 seconds to process, but it got progressively slower as time went on and coverage increased6. While it was all processing, I wrote another programme to calculate time series and let me look at the output as it was produced.

Everything was working fine, but there are outliers in the data. I knew this would be an issue as the dataset isnt infilled. Along the edge of the SST field and in other areas with few observations, you get outliers. Where there are few observations, you dont benefit from the noise-quelling powers of the mean. Also, there are few or no other independent observations to compare them to, making quality control difficult. In the most extreme cases, the only obs available as a comparison were those made by the same ship and also likely erroneous. While these aberrant values are confined to one or two grid boxes, theyre not too bothersome, but if an extremely erroneous value gets extrapolated across half a hemisphere, then you have a problem. This could show up as noise in the global means which would be annoying but not defeat the aim of the exercise. Unfortunately, there seemed to be a preponderance of large positive anomalies which had a net warming effect once they got spread around.

As a quick fix, I removed SST grid cells that had an anomaly with an absolute magnitude greater than 5 degrees and reran the kriging. Visual inspection of the fields suggested this did a reasonable job without removing anything vital, like ENSO. There were still issues with rogue ship tracks crossing several gridcells, but they are far less common than individual outlier gridcells7 and theres no quick fix.

Then I waited. The sun was coming up, the cats were sleeping, and I had proper work to do.

With a few false starts, an unplanned Microsoft update, and an IO issue with python and netcdf, it took a few hours to process all the fields and plot them out.

I used the IPCC method for calculating global means calculate an area-weighted average of available data for each hemisphere and then take a simple arithmetic mean of the hemispheres to get a global average which avoids giving to much weight to the northern hemisphere early on. I also had to reuse some of the sea ice fields to get the series completely up to date. The netcdf file I have for HadISST2 ice ends in August 2020. I used 2019 ice concentrations for for missing fields in the 2020-2023 period.

After all that, it looks like some modest infilling (blue line) does increase the overall warming by a few hundredths of a degree relative to the unfilled dataset (red line) mostly due to a slightly cooler 19th century. Theres also a small hole at the pole effect post 2005 including the 2015/2016 cool/warm pattern8, but both effects are very small thousandths of a degree rather than hundreds. Interestingly, the infilling cools the 19th and early 20th century, but actually warms World War 2 slightly, so the early 20th century warming increases again.

This is very noddy analysis and should be taken with a large pinch of salt. Mostly, I was interested in seeing the broader tendencies and familiarising myself with the dataset a little bit. Its always nice to play with a new dataset to get a feel for its peculiarities. Also, I had a chance to do something quasi useful with ChatGPT4. In this case it sucked.

Anyway, enjoy the movie bottom row is unfilled data, top row is infilled:

DCENT is available from under a license.

-fin-

1. It used a radius of one for the earth, but I could live with that.
2. Connoisseurs of the literally interpreted metaphor will tell you correctly that honing something for 40 years wont leave much of the thing honed.
3. Not for any intrinsically scientific reasons, but because it looks pretty.
4. PLURAL!
5. Not a great choice as a land mask it must be said, but regridded to 55, its not completely awful. It does miss all the Pacific islands mind
6. Its one of those paradoxes of interpolation: the more data you have, the less the interpolation algorithm has to do, but the longer it takes to do it. In kriging this is because it calculates using matrices of all the pairs of data points (expensive) and then projects that onto the unobserved locations (cheap).
7. In HadCRUT infilling, the interpolation has access to the error covariances, so ship tracks are treated as a single entity, as far as an entity can be encoded in a covariance matrix at that resolution. This is generally fairly effective at discerning between real and artefactual variability. Unfortunately, off the shelf kriging solutions dont generally allow this kind of information as an input and anyway, DCENT doesnt currently have the necessary measurement and sampling uncertainty information though the paper says this is on the way.
8. Infilling cools 2015 slightly counter to the general trend and warms 2016 slightly.

#climate #climateChange #globalWarming #uncertainty

: USA's arm gave more than $500K to rightwing orgs promoting denial, anti-abortion and anti-LGBTQ+ agendas, including several involved in Project 2025, with aims to weaken environmental laws Story via Environmental Health News

Quite a while ago, when I learned the heatwave last year in Greece had fucked the olive groves, I said - Oh, that's going to hit the price of .

Went into Lidl last Friday and bought a bottle of very ordinary, basic olive oil. 7.29.

This is about three-and-a-half times what it cost a year ago.

With Spain, Italy, and the whole Mediterranean circumference getting cooked it's likely this will get worse until what little can be produced there will be a luxury only item.

"On the cutting board"

From memory:

Homage to the green fish pulled from the blue ocean - facing a human knife.

Deadly in Made Worse by
Extreme rainfall made 10% heavier by human-caused climate change triggered landslides that killed hundreds.
, caused by the burning of , allows the atmosphere to retain more moisture, which contributes to the severity and intensity of rain. Heavier rain over a longer period of time can waterlog the soil, making a hillside heavier and more susceptible to slippage.

I didn't know how toxic cars are to society. I mean I knew they are pretty toxic, but I didn't know exactly how. US minimum parking laws are a 'great' example of a particular way car policies screw over society. Watch it if you want to feel incredulous.

"Even desert plants known for their resilience are burning and dying in the heat"

Global warming could cause mass bridge collapse in US, media reports say

Anomalous temperatures caused by climate change could lead to the collapse of every fourth steel bridge in the United States by 2050, Report informs, citing The New York Times

"Expect the costs of to stay uncertain. Yet they are no longer quite the terra incognita William Nordhaus described. Despite their flaws, the methods agree on one thing: carries far heavier costs than Mr Nordhaus first imagined"

How a fracking-adjacent technology can store renewable energy underground without lithium batteries

...

Switzerland, GenevaUnited for fresh water and climate change.

The impact of climate change is become increasingly visible in Switzerland as we witness the shrinking of glaciers and experience unprecedented temperatures.

The Swiss Learning Venture focuses on learning, exchanging, and lobbying together about fresh water and climate change.

Toegankelijke introductie van necropolitiek van John the Duncan, met voorbeelden hoe overheden dat gebruiken in hun beleid voor klimaatverandering, Covid-19, mensen met een uitkering, alleenstaande moeders, vluchtelingen, gehandicapten, ....

How Will We Solve The Worlds Water Wars An Ancient Spanish Court Offers One Answer
--
<-- shared media article
--

Grenada is the first country in the world to activate the so-called hurricane clause in August, which suspends debt payments for a few months. The next installments in November 2024 and May 2025 will thus be postponed, amounting to a total of 12.5 million dollars (11.3 million euros). This small Caribbean country, 150 kilometres north of the Venezuelan coast, was devastated by a powerful cyclone in early July

Update on the launch of my new paleoclimate focused podcast, the Raised Beaches Podcast. It took a bit longer to edit than expected, since I decided to re-record a lot of it, but I am putting the finishing touches on. It should launch on Thursday or Friday!

As Jasper preps later return to class, experts urge more schools to plan for climate emergencies

Better plans, policies needed since we know floods, fires and more will happen elsewhere too, says expert

In tomorrows podcast episode, I will be chatting with Dr. Jay Hakes about the potential climate outcomes of the 2024 US presidential election.

The decisions made now will impact our planet until 2029!

Check out/follow the podcast to be sure you don't miss this exceptional episode.

Link:

Yes, yes - there were warm summers in the past too ... and 4 K (temperature differences are always given in Kelvin K 4 C ) aren't that much .... always these climate alarmists ... - by the way, they are paid by YOUR tax money and are called SCIENTISTS

Via TW because I miss it here

is warming a bit faster than other places due to , but Africans are being hit significantly worse by climate change, losing 2 to 5 % of their GDP and facing proportionally higher adaptation costs.

Extreme weather events, whose frequency is increasing with climate change, is speeding the deterioration of bridges and other essential infrastructure. This article focuses on the U.S.
By 2050, extreme heat could cause up to 25% of steel bridges to collapse. Huge temperature swings are also a problem.

Daikin Industries, the worlds largest manufacturer of air conditioners, is looking to expand in India as it responds to consumer demand for whats become a necessity as temperatures soar.

Five added to UK red list of most concern

The new entries include the Arctic tern, known for its incredible migration, the giant skua, or pirate of the sea, and two types of gull.

The UK is known for its colonies of nesting in huge numbers on cliffs - but populations are plummeting amid a host of pressures, from change to a lack of food.

Seabirds are in trouble because of , unsustainable , offshore development, and .

03.09.2024 - 08:00 Uhr
Chart des deutschen Strommix ber die letzten 6 Stunden.

Seeing is believing: your neighbours choice to go solar might have influenced you more than you think

Japan swelters through hottest summer while parts of China log warmest August on record Extreme heat The Guardian

Cory Doctorow on marshmallow longtermism:

"In reinvesting his fathers companys earnings Charles Koch passed one marshmallow test, and it made him a very wealthy man. But in choosing not to believe climate science and instead risking the end of a habitable Earth and the human race he belongs to, he failed a much more important marshmallow test."

Down to Earth: Rare to hit parts of the FRANCE 24 English

See also

Personally I'm convinced we won't survive in the next 5 years.

2050 Net Zero Goal: What Kind Of Progress Have Airlines Made So Far

Yesterday it was 35 here and Ohtani is nearing 50...