Right now, we still concentrate on improvements of the processing chain that we think will have a more global impact – ultimately a question of resources. Happy to discuss more! I could give a presentation on the data set some time if you have a suitable forum for that!!

I guess this could be improved, but we would have to look in detail which factors (snow, ground properties, etc.) are responsible for the misrepresentation and how to change it.

… we use a global landcover product which shows “grassland” there. This class is not important for permafrost in N America, but widely distributed in Central Asia, so the associated model settings are optimized for these globally important permafrost areas.

In the discontinuous zone (Yellowknife, Wrigley), the uncertainty in the permafrost fraction is really high (up to 0.4) as the ground temperatures are generally close to zero. So small changes in e.g. subsurface parameters have a large impact on permafrost fraction. The prairies is a special case…

Especially in the discontinuous and sporadic permafrost regions, we often lack the knowledge on the environmental factors governing permafrost occurrence (which local experts like you have). These not only vary between regions, but we need to represent them using global data sets like landcover.

Hi Brendan, it’s clear that the products have a lot of uncertainties – we do not tweak the output to fit for individual regions with lots of user interest, but rather have a globally consistent algorithm. And the challenge is not only to be right, but to be right for the right reasons.

I can offer similar insights for the other areas you mention, let me know if you are interested! We are very much open to input from local experts in Permafrost_cci - in quite a few cases, this has improved the performance in the next version!

FINALLY IMPORTANT: we provide an explicit uncertainty field in each of the netcdf files, which is around 7-14% for the Yukon North Slope and the northern part of the Peel plateau. Within this uncertainty, continuous permafrost on the Yukon North Slope is actually in agreement with the product.

For parts of the Yukon North Slope and much of the Peel plateau, the 1997 PFR product shows values of 86%, which is quite close to the threshold of 90% for continuous permafrost. Clearly off, but not by much.

In the project, we can only afford to run 7 ensemble members per pixel, so the possible values for the permafrost fraction in the product are separated by 1/7, i.e. 100%, 86%, …, 14% and 0%.

The Permafrost_cci products do not explicitly provide permafrost zones, but permafrost fractions for each 1km pixel. This number is obtained by running the model for different snow, soil and landcover conditions for each pixel and counting up how many of these “ensemble members” show permafrost.

Sebastian Westermann from Norway here, I’m responsible for making these products. I can offer some first-hand insights in the debate, although it is not easy to reduce the rather complex reality to 300-character posts. But I’ll do my best!