Marta Mrozowska

University of Copenhagen

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Trends

Author Order

Document Type

Co-Authors

Markus Jochum

University of Copenhagen

Swantje Bastin

Max Planck Institute for Meteorology

АВ

А. В. Колдунов

Alfred-Wegener-Institut Helmholtz-Zentrum für Polar- und Meeresforschung

Tim Fischer

GEOMAR Helmholtz Centre for Ocean Research Kiel

Marcus Dengler

GEOMAR Helmholtz Centre for Ocean Research Kiel

Johann Jungclaus

Max Planck Institute for Meteorology

Nils Brüggemann

Max Planck Institute for Meteorology

Roman Nuterman

University of Copenhagen

Oliver Gutjahr

Max Planck Institute for Meteorology

Mia Sophie Specht

Max Planck Institute for Meteorology

Cooperative Institutions

Max Planck Institute for Meteorology

University of Copenhagen

Universität Hamburg

GEOMAR Helmholtz Centre for Ocean Research Kiel

Alfred-Wegener-Institut Helmholtz-Zentrum für Polar- und Meeresforschung

Helmholtz-Zentrum Hereon

Norwegian Institute of Marine Research

Max Planck Institute for Comparative and International Private Law

Laboratoire d’Études en Géophysique et Océanographie Spatiales

University of Cape Verde

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Field

How Well Do High Resolution Ocean Models Simulate Local Turbulent Mixing?

Latest Thinking (2017)

Marta Mrozowska

10.21036/ltpub101164

Cite

Citations (0)

Sensitivity of the tropical Atlantic to vertical mixing in two ocean models (ICON-O v2.6.6 and FESOM v2.5)

Swantje Bastin А. В. Колдунов Florian Schütte Oliver Gutjahr Marta Mrozowska

Abstract. Ocean General Circulation Models still have large upper-ocean biases e.g. in tropical sea surface temperature, possibly connected to the representation of vertical mixing. In earlier studies, the ocean vertical mixing parameterisation has usually been tuned for a specific site or only within a specific model. We present here a systematic comparison of the effects of changes in the vertical mixing scheme in two different global ocean models, ICON-O and FESOM, run at a horizontal resolution of 10 km in the tropical Atlantic. We test two commonly used vertical mixing schemes; the K-Profile Parameterisation (KPP) and the Turbulent Kinetic Energy (TKE) scheme. Additionally, we vary tuning parameters in both schemes, and test the addition of Langmuir turbulence in the TKE scheme. We show that the biases of mean sea surface temperature, subsurface temperature, subsurface currents and mixed layer depth differ more between the two models than between runs with different mixing scheme settings within each model. For ICON-O, there is a larger difference between TKE and KPP than for FESOM. In both models, varying the tuning parameters hardly affects the pattern and magnitude of the mean state biases. For the representation of smaller scale variability like the diurnal cycle or inertial waves, the choice of the mixing scheme can matter: the diurnally enhanced penetration of equatorial turbulence below the mixed layer is only simulated with TKE, not with KPP. However, tuning of the parameters within the mixing schemes does not lead to large improvements for these processes. We conclude that a substantial part of the upper ocean tropical Atlantic biases is not sensitive to details of the vertical mixing scheme.

Mixed layer

Langmuir Turbulence

Ocean dynamics

10.5194/egusphere-2024-2281

Cite

Citations (1)

Bayesian optimization of ocean mixed layer parameterizations

Marta Mrozowska Markus Jochum James Avery Ida Stoustrup Roman Nuterman

Global climate is highly sensitive to tropical sea surface temperature. Accurately representing the tropical SST remains a significant challenge for general circulation and climate models. One of the largest sources of uncertainty is the vertical turbulent mixing. To accurately represent the distribution of ocean mixed layer depths, turbulence closure schemes necessitate careful tuning. This is most commonly done manually by comparing with mixed layer depth climatologies. Advancements in machine learning research introduce a new strategy: automated tuning. Veropt, an add-on to the python ocean model Veros, uses Gaussian processes to emulate an objective function in a multi-dimensional parameter space. We present a surprising combination of changes to the default parameters of the commonly used turbulent kinetic energy (TKE) closure scheme that minimise the model bias in tropical mixed layer depth.

Mixed layer

Bayesian Optimization

10.5194/egusphere-egu24-15235

Cite

Citations (0)

nextGEMS: Output of the WP6 ocean vertical mixing sensitivity runs (timeseries)

Zenodo (CERN European Organization for Nuclear Research) (2023)

Swantje Bastin А. В. Колдунов Helmuth Haak Nikolay Koldunov Sergey Danilov

In work package 6 of the nextGEMS project, several ocean-only model runs were performed with FESOM (Version 2.0) and ICON-O (Version 2.6.6), to test the sensitivity of the upper tropical Atlantic to different settings of the vertical mixing scheme. Two different mixing schemes were tested: TKE and KPP. For TKE, we tested different settings of the c_k parameter (0.1, 0.2 and 0.3), and for KPP different settings of the critical bulk Richardson number (0.3 and 0.27). These runs were done with both ICON-O and FESOM, to enable a comparison of the effects of the vertical mixing settings across different models. From ICON-O only, there are some additional TKE runs available, where we increased the interior ocean background mixing, and switched on the Langmuir turbulence parameterisation. There is also an ICON-O run which uses the FESOM default forcing bulk formulae, to check how much of the differences between the models originates from their different default bulk formulae.

All model runs are ocean only, forced with hourly ERA5 reanalysis data. The horizontal resolution is 10km (for FESOM, the extratropical regions have a coarser grid). Here we provide high-frequency (3 hourly) time series of the model output, at selected locations in the tropical Atlantic where observational data are available for comparison in the simulated time range (2014 and 2015). We provide the following locations here: 0N, 10W 0N, 23W 11.5N, 23W 15N, 38W 11N, 21.2W 17.6N, 24.3W

10.5281/zenodo.8225705

Cite

Citations (0)

How Does Turbulence Affect the Sea Surface Temperature in the Tropical Ocean?

Latest Thinking (2017)

Marta Mrozowska

Affect

Tropical marine climate

10.21036/ltpub101158

Cite

Citations (0)

Using NIW Observations to Assess Mixed Layer Parameterizations: A Case Study in the Tropical Atlantic

Journal of Geophysical Research Oceans (2024)

Marta Mrozowska Markus Jochum Swantje Bastin Rebecca Hummels А. В. Колдунов

Abstract Tropical sea surface temperature (SST) biases can cause atmospheric biases on global scales, hence SST needs to be represented well in climate models. A major source of uncertainties is the representation of turbulent mixing in the oceanic boundary layer, or mixed layer (ML). In the present study we focus on near‐inertial wave (NIW) induced mixing. The performance of two mixing schemes, Turbulent Kinetic Energy and K‐profile parameterization (KPP), is assessed at two sites (11.5°N, 23°W and 15°N, 38°W) in the tropical Atlantic. At 11.5°N, turbulence observations (eddy diffusivities, shear and stratification) are available for comparison. We find that the schemes differ in their representation of NIWs, but both under‐represent the observed enhanced diffusivities below the observed ML. However, we find that the models do mix below the ML at 15°N when a storm passes nearby. The near‐inertial oscillations remain below the ML for the following 10 days. Near‐inertial kinetic energy (NIKE) biases in the models are not directly correlated with the wind speed, the MLD biases, or the stratification at the ML base. Instead, NIKE biases are sensitive to the vertical mixing scheme parameterization. NIKE biases are lowest when the KPP scheme is used.

Tropical Atlantic

Mixed layer

10.1029/2024jc020985

Cite

Citations (1)

Fast and efficient: Bayesian optimization with GPU acceleration for ocean models

Authorea (Authorea) (2024)

Marta Mrozowska James Avery Aster Stoustrup Roman Nuterman Carl-Johannes Johnsen

Ocean general circulation models (OGCMs) contain numerous parameterizations of sub-grid scale processes. The parameter tuning procedure is rarely reported and often done by hand. We present an automated alternative: Bayesian optimization, a method which has recently emerged as a frontier in expensive black box optimization. VerOpt, a Python package for the ocean model Veros, adapts Bayesian optimization to climate model tuning. We use VerOpt to identify a set of parameter values of the Turbulent Kinetic Energy (TKE) closure scheme that minimize mixed layer depth (MLD) bias in Veros. We present the results of two optimization procedures: TWIN and OBS. The goal is to minimize modeled MLD error relative to a target map. In TWIN, the target is MLD simulated using Veros with a known parameterization. The ratio of two TKE parameters ckcϵ-1, proportional to the critical Richardson number Ric, is the dominant factor in setting the global MLD. After 180 model simulations, the lowest error in the TWIN experiment is 1.18%. In OBS, the target is MLD climatology. The MLD bias is smallest when Ric<1, and the default TKE parameterization falls within this range. We find, however, that altering the TKE parameterization is not sufficient to reduce the significant MLD bias of 42.62%. The OBS experiment results indicate that the TKE scheme parameters are not the dominant source of MLD bias in Veros. We discuss other possible sources of MLD bias, as well as the potential of extending of the optimization procedure to other parameterizations.

Bayesian Optimization

10.22541/essoar.173134338.88249016/v1

Cite

Citations (0)

nextGEMS: Output of the WP6 ocean vertical mixing sensitivity runs (timeseries)

Zenodo (CERN European Organization for Nuclear Research) (2023)

Swantje Bastin А. В. Колдунов Helmuth Haak Nikolay Koldunov Sergey Danilov

10.5281/zenodo.8225706

Cite

Citations (0)