The Köppen–Geiger climate classification made simpler (I hope so)

[Charerg]

How do you put it on the map?
Select about 30% of the area?

I suppose that works, though in July the "Cool category" (if we follow the terminology in your prior post) tends to stick quite closely to the coast.

Here are the sample maps that include the new category, following your suggested colour scheme:

Jan:


Jul:





Tbh I haven't thought too much about the placement instructions yet. Though personally I plan to use an alternative method to create the temperature maps anyway. I made a little test continent and created a basic greyscale height map for it:



The scale used is the following:

RGB	Elevation
0	-150 m
1	-125 m
2	-100 m
6	0 m
255	6225 m

And converted into the standard elevation map:


Then as a test I've created a map of surface-level temperatures in July. Again, in greyscale, ranging from -50 to 50 °C (200 to 0 RGB, darker=warmer). Here are the different temperature bands with the continent shown on top, gradient mapped at 10 °C intervals:

Colour	Temp (°C)
Red	30+
Orange	20 to 30
Yellow	10 to 20
Green	0 to 10
Light Blue	-10 to 0
Blue	-20 to -10

Then I've blurred the temperature bands, creating a gradual transition. The elevation map can be directly translated into an "elevation adjustment layer" if the temperature is assumed to drop in a linear fashion with incresed elevation, and then applied as an overlay over the surface-level temp map. The final map (btw, the blurring is not shown here in the background layer, those are the original temp bands):



If you use this method you can easily include as many temperature layers as you like, since the elevation part is essentially automated, though it's a bit more work to create the gradual map of sfc-level temperatures. And of course you need a fairly detailed height map in greyscale for this method to work.

[Azélor]

It's actually 37,5 % following the logic:

Cool Max-Cool Min/Old Cool Max
13-10/18= 37,5%

The proportion is probably different in reality but anywhere between 30 and 40% is a good guess.

---

The temperatures around the Great Lakes look different from my memory. Is it normal for the boundary to follow coastline like that?

[Charerg]

It's actually 37,5 % following the logic:

Cool Max-Cool Min/Old Cool Max
13-10/18= 37,5%

The proportion is probably different in reality but anywhere between 30 and 40% is a good guess.

---

The temperatures around the Great Lakes look different from my memory. Is it normal for the boundary to follow coastline like that?

The temperature certainly follows the coastline in the WorldClim maps. Whether that is normal, hard to say, but if the lakes don't completely freeze then the temperature probably is a bit warmer over the lake than overland in winter.

[Azélor]

I thought about the gradients too. The main problem to overcome is the irregularity between different regions at the same latitude.

[Charerg]

I thought about the gradients too. The main problem to overcome is the irregularity between different regions at the same latitude.

Yes, but since you only need to worry about the surface-level temperatures, I don't think the whole process is necessarily more time consuming. Of course, if my test map was a serious map, I'd have made the temperatures in the interior hotter, cooler around the coast, and so forth (instead of having the temp bands roughly follow latitude lines). But in essence, it's not a different process to the usual one, you just have to make sure to include enough "temperature bands" that the final result can be blurred together.

Basically, first draw the "usual bands" like the 28, 22, 18, 10 °C and so forth, just like when following the usual tutorial, then start adding extra bands halfway in between until there's a sufficient density that everything can be blurred into a gradually transitioning map.

EDIT:
Btw, I think you're right that the Very Cold category is a bit more extensive in my latest maps. Apparently I forgot to reset the boundary back to -10 °C (I used -8 °C as the VCold/Cold boundary in one of those prior tests).

[AzureWings]

So I've been following your work here with great interest, although I'm still working my way through some of the distinctions you've been discussing in the Köppen-Geiger classifications themselves. I've been putting together a Python script to do the classification via the methods you've put forth here on input temperature/precipitation map images, and I'm a little unsure how you've been drawing the distinctions between the temperature types (specifically Ca/Cb/Cc/Da/Db/Dc/Dd) where the definition seems to depend on 'hottest 3/4 months' while the input data has only (approx) hottest and coldest. Are you basing it off of an annual overall average compared against that or some other threshold (for requirements like 'no more than 3 months average >10 °C')?

[Azélor]

We use the data we have.
More specifically, we have the temperatures for the extreme month: January and July. We assume that these months are always the hottest and coldest of the year, even if it's not exactly the case in reality.
We could have more months but it would require even more work. So we stick to 2 month to keep things from getting out of control.
Having only two months means we don't know what's between these extremes.

For example, the difference between Cb and Cc is the number of months the temperature is above 10 C.
Cb has 4+ months, Cc has between 1 and 3 months above 10 C (0 would be a tundra).
By default, in our model, only the coolest possible combination of the C climates has been assigned to Cc (Mild+Cool). It is the only one suitable for Cc but it is also a Cb climate.
In reality, only about a third of these places are really Cc, the rest is Cb. That is because we use only data from 2 months. Also because we use categories and only keep the average temperatures.
So, the temperature could be anything between 18 and 10 in July and anything between 0 and 10 in January.

What we have agreed is that Cc climate usually have an average temperature 13 and below in July. Making sure Ireland stays Cb but barely.

We try to stick to the official criteria of Koppen. The same is true for precipitations. They can be found here : http://koeppen-geiger.vu-wien.ac.at/pdf/Paper_2006.pdf
But we use the 0 C isotherm instead of 3. There is no consensus about this. Also, the distinction between ice caps and tundra is the 0 isotherm, so we will be keeping it.

[Charerg]

So I've been following your work here with great interest, although I'm still working my way through some of the distinctions you've been discussing in the Köppen-Geiger classifications themselves. I've been putting together a Python script to do the classification via the methods you've put forth here on input temperature/precipitation map images, and I'm a little unsure how you've been drawing the distinctions between the temperature types (specifically Ca/Cb/Cc/Da/Db/Dc/Dd) where the definition seems to depend on 'hottest 3/4 months' while the input data has only (approx) hottest and coldest. Are you basing it off of an annual overall average compared against that or some other threshold (for requirements like 'no more than 3 months average >10 °C')?

The a class has the requirement that the hottest month must be above 22 °C, so the distribution is relatively easy to determine. Although in reality it also demands 4 months min. with avg temp above 10 °C as you noted, but it's very rare that any climate would be that hot in summer and still have less than 4 months below 10 °C.

In the system used in the tutorial, any climate that has the hottest month's temperature above 22 °C is classified as a for this reason (as long as it doesn't qualify as A).

The d class is similar in the sense that it also has the requirement that the coldest month is -38 °C or below. And if it's that cold, it can be safely assumed that it has max. 3 months above 10 °C (as long as it has a cooler summer than 22 °C in summer).

All in all, I'd say the distribution of a and d climates matches reality almost exactly, even when using these more limited criteria to determine the distribution. Unfortunately, this doesn't work so well for determining the boundary between b and c climates, since that is based solely on the number of months above 10 °C. If you've read all the walls of text posted lately, you might have noted that I suggested the addition of a new temperature category to try and improve the situation, especially regarding the distribution of Cc climates.

At present, a climate that has a Mild (10-18 °C) hot season and Cool (0-10 °C) cold season will be classified as Cc. With the suggested extra temp category it would be Cool (10-13 °C) hot season and Chill (0-10 °C) cold season, which is a better approximation, but it's still pretty rough. So you could say that it's based on the annual mean temp, but it's a specific combination of the limited temperature categories we're working with, not an exact numerical threshold.

I also experimented a bit if the Dc distribution could be similarly improved by either adding extra categories or changing the borders of existing ones, but unfortunately the experiments had limited success. Although the Dc distribution does roughly follow an annual mean temperature isotherm (it's close to 2 °C or below mean annual temp, in this map), I wasn't able to find a combination that works.

[AzureWings]

Thank you! I've been trying to replicate your output from the script-fu version back on page 21, since you so kindly provided input data easily in reach along with the corresponding output, although I realize now the result there isn't totally current to the better representations you've been aiming for over the course of later discussion. The b/c distinction was exactly one of the major trouble spots I've been having - with a lot of the temperature bound conditions I was trying to use Cc/Dc would eat all the Cb/Db or vice-versa (Cfb still seems to be eating all of Cfc, actually, but western Europe at least seems to be more accurate with the suggested 13 °C threshold than without).

My current output is looking like this:


I've been using the R = 2 * ((10 * (T_ann - 10)) + (3 * P)), P = % hot months precipitation / total annual precipitation, to determine the threshold for aridity, although I'm finding that many places seem to be coming out a bit drier than they should (that's my other biggest problem, along with other more localized issues like the Dwd and steppe climates showing up in the Nunavut area). That might be because I should be using the defined aridity categories to get thresholds instead; I'll give that a try in a bit.

Striking the balance between getting as accurate as possible and still having the input data be feasible to generate for speculative worlds does seem like a bit of a conundrum. Writing the script to check multiple months' data would be (relatively) straightforward, but brewing up twelve temperature maps of a given fictional world seems like it'd get a bit maddening. I almost think I prefer the idea of the extra 'length of growing season' map you considered earlier if it can sort Cb/Cc and Db/Dc out cleanly; while it's still extra data to make it doesn't seem quite so onerous.

[Charerg]

If you used the maps from page 21 as the input data, those don't have the Mild (10-18 °C) category split into Mild (13-18 °C) and Cool (10-13 °C), so it's no surprise that Cc doesn't appear. Btw, are you using the 6 or 8 precipitation category maps as the source map?

Also, you're better off looking at actual köppen maps as references, since the "generated maps" will inevitably contain inaccuracies due to the various simplifications and limited categories used.

I wouldn't worry too much about the occasional Dw and BS in Alaska and Canada. It's basically a result from using data from just two months. And if you check out a high-res köppen map of Alaska, there actually is some BS there, so I'd guess the region is quite dry overall.

Edit:
I should probably mention that the newest version of the script (using the 6 precipitation categories) is available on page 34.