A Parsimonious Coupled Model of the Equatorial Pacific Surface Temperature Change
Welcome to the seager19 replication documentation!
This project reassembles a parsimonious coupled model of the equatorial Pacific, from Seager et al. 2019 (S19), which they created to explain the bias in the trend in NINO3.4 temperature in CMIP5 models. When forced with ECMWF reanalysis fields, it can reproduce the trend observed in ECMWF/ORAS4 that was forced with the same fields. It shows that the CMIP5 bias in the trend in NINO3.4 from 1958-2017 could be due to a product of the CMIP5 bias in relative humidity and sea surface winds. This is shown through exchanging ECMWF mean fields for CMIP5 multimodel mean fields. The replacements of mean relative humidity, mean wind speed, and both together, lead to increases in the NINO3.4 trend of 0.31±0.03 K, 0.054±0.005 K, and 0.47±0.04 K, respectively. The error bars are from tests with a range of plausible inputs. This is congruent with the difference of 0.478 K between the ECMWF/ORAS4 reanalysis product and the CMIP5 multimodel mean. I investigate how reliable the results from this model might be by varying the free parameters and find that, as far as tested, the model is not overly sensitive to subjective inputs.
It is therefore plausible that the observed bias in the increase in sea surface temperature in the nino3.4 region is caused by excess humidity, and insufficient tropical windspeeds in the mean climatological state. This appears to be largely produced through more humidity and less wind reducing the sensitivity of the surface latent heat flux to increased temperature. Therefore, the relative humidity and windspeed bias from the mean state of the surface atmosphere makes the central and east equatorial Pacific seem to be easier to warm than it is.
This model does not explain why these surface atmosphere biases occur in the first place. For that, we would need to consider the much more complicated processes that are important in the tropical atmosphere, that are connected to the “Double ITCZ Cold Tongue Bias” which has been observed in each subsequent generation of complicated climate models. The model explains that a bias in the mean state of CMIP5/6 propagates into a bias in the sensitivity of CMIP5/6 to forcing, not why this bias in the mean state exists.
The nino3.4 trend bias would probably introduce a nonstationary bias in the various hazards in CMIP5/6 that have a teleconnection to El Nino Southern Oscillation. For example, more tropical cyclones form in the North Atlantic when there is La Nina, due to reduced shear (e.g. Camargo, Emanuel and Sobel 2007). Therefore, the nino3.4 trend bias would lead to an underprediction in the apparent risk from tropical cyclones for the North American East coast in the future projections from CMIP5/6.
See the final report
for more details and tests.
CMIP6 seems to have a smaller bias (at least from the 51 ensemble member that I collected from Pangeo). However, it seems that this bias is not as small as our model would predict. The mean relative humidity and wind speed fields seem to have improved more between the two models than the trend in nino3.4.
The first section seager19 contains the main README.md of the repository. This should provide a reasonable introduction to the repository as a whole.
Here is the current breakdown of the model code by language:
$ cloc --report-file=docs/lang.txt $(git ls-files)
github.com/AlDanial/cloc v 1.74 T=0.16 s (819.0 files/s, 195329.7 lines/s)
--------------------------------------------------------------------------------
Language files blank comment code
--------------------------------------------------------------------------------
Python 59 2288 3362 7839
Fortran 77 15 1360 1378 6170
C 5 493 211 2735
Pascal 1 248 182 1310
Markdown 16 287 0 992
C/C++ Header 8 88 18 365
YAML 14 23 90 333
make 3 46 58 149
Bourne Shell 3 45 33 67
Dockerfile 1 17 31 29
Fortran 95 1 4 11 24
Bourne Again Shell 1 1 0 2
--------------------------------------------------------------------------------
SUM: 127 4900 5374 20015
--------------------------------------------------------------------------------
MRes Proposal:
Seager et al. 2019 [1, hereafter S19] showed that although CMIP5 ensemble members have a positive NINO3.4 trend (towards El Nino) where as the observations show a more negative NINO3.4 trend (towards La Nina).
They showed that the observed trend can be reproduced with a simple coupled physical model. Here, we carry out a parameter sensitivity analysis of the S19 model. Of particular interest might be the S19 model’s sensitivity to the drag coefficient, as S19 note that they chose a much higher value than normal so as to replicate the amplitude of ENSO. This sensitivity analysis could first be achieved using a Gaussian Process (GP) with a radial basis function (RBF) kernel of a given smoothness, as the number of data points will initially be quite small (<10^{4}). S19 is computationally lightweight, allowing for a large number of parallel sensitivity experiments to be run at the same time in order to generate the training dataset for the GP model. The GP model will allow us to rapidly explore the parameter space in between our chosen parameter configurations, in terms of both the mean value and uncertainties. From this initial baseline, we could expand to more sophisticated sensitivity analyses, and/or more complicated model settings.
Scientific questions to be addressed include:
Can we replicate the results displayed in S19? [mostly]
How robust is the model to the parameters chosen? [fairly]
Can the sensitivity of the model to the parameters be understood from the physical processes underlying it? [partially]
How skillful are different emulation functions at fitting input/output of the model? [untested]
Citations:
[1] Seager, R. et al. Strengthening tropical Pacific zonal sea surface temperature gradient consistent with rising greenhouse gases, https://doi.org/10.1038/s41558-019-0505-x (July 2019.)
[2] Tian, B. & Dong, X. The Double-ITCZ Bias in CMIP3, CMIP5, and CMIP6 Models Based on Annual Mean Precipitation. Geophysical Research Letters 47, e2020GL087232. issn: 0094-8276. doi:10.1029/2020GL087232. https://onlinelibrary.wiley.com/doi/abs/10.1029/2020GL087232 (Apr. 2020).
- A Parsimonious Coupled Model of the Equatorial Pacific Surface Temperature Change
- seager19
- Purpose
- Setup
- Get the Github repository
- If you have root access and can install Linux packages
- If you want to make the docker environment yourself
- If you need to install the singularity environment
- Making the environment and testing it works (either in singularity or not)
- Add optional features
- Examples of running the model
- Other handy commands for development of repo
- src package
- Ocean model
- Gallery
- Tutorials
- About