16. 16
› Sinks:
• Barents Sea
• Nordic Seas
• Northern and Central North Sea
• Atlantic shelf
› Sources:
• Southern North Sea and English Channel
• Baltic Sea
• Mediterranean
Air-Sea flux 2010-2020
FCO2 / gC yr-1 m-2
17. 17
› Increasing sink strength:
• Northern Atlantic shelf
• Barents Sea
Time Series of air-sea CO2 flux
18. 18
› We can use Random Forest regressions to produce
precise maps of coastal pCO2 and air-sea CO2 flux
› Large variability between regions
› Most regions on the European shelf are a net sink for
CO2
› Some of these are increasing in strength
Conclusions
20. 20
Thanks to everyone who
contributed to the SOCAT
database!
Do you have any
questions?
21. ICOS
Integrated Carbon Observation System
› Europeisk forskningsinfrastruktur
› Målinger av CO2 I atmosfære, hav og land økosystemer
› 34 atmosfærs stasjoner
› 21 hav stasjoner
› 71 Økosystem stasjoner
21
22. ICOS Norge - Marine network
22
Kronprins Haakon (KPH)
Nuka Arctica (until April 2020) /
Tukuma Arctica (from October 2020)
Trans Carrier
GO Sars
SOOP-Bergen Kirkenes
Editor's Notes
Meike Becker
av co2 i atmosfæren allerede været rundt 480ppm og ikke 410 som der er nå
Problem:
We have several different estimates for the open ocean carbon sink, but only few for coastal regions
Coastal regions are very divers
Observations are relatively scarce
simple regression tree divides the input data into subgroups based on different characteristics in the driver data
At each branch fork, called node, the routine divides the node into two child nodes, trying to minimize the dissimilarity within the child nodes
The end nodes that are not divided any further are called leaves.
Danger of overfitting
Solution: use more than one tree
Things to keep in mind:
Relatively conservative estimates
Overfitting when using too small leave sizes