Periodic Reporting for period 1 - LISTEN (Lost In translation: Strengthening communication skills between real world and climaTe modEls for seasonal to decadal predictioN)
Berichtszeitraum: 2019-02-01 bis 2021-01-31
Near-term climate information is increasingly incorporated in the decision-making process of many socio-economic sectors. Attracting interest of society on the changing climate, as well as advising the stakeholder on how to benefit from climate information, along with learning their needs are clear priorities to mitigate climate change impacts. Parallel to these priorities, there is still work that needs to be done by the scientific community to provide the best near-term climate information, for it to be a valuable instrument for the decision process.
The main tools to provide near-term climate information are Global Climate Models (GCMs). GCMs depict complex real-world processes providing their simplified representation through a set of equations and algorithms. Climate models are therefore imperfect and suffer from systematic errors. Model imperfections lead to differences between the model and observed mean state and thus complicate the initialisation task of near-term climate prediction systems.
While in the long term the goal is to reduce the biases through model improvements and increased computer power, medium-term solutions are also needed. LISTEN aims at contributing to the climate prediction community effort to compensate for the model inadequacy by enhancing the transfer of observed information to the model during the forecast initialisation. This objective is achieved through the implementation of innovative initialisation techniques explicitly designed to tackle specific limitations detected in the methods currently in use and are applied to initialise decadal predictions.
The study is developed in the context of a common framework in order to ensure efficient use and dissemination of the data and the findings within the scientific community.
In addition, the forecast skill assessment is carried out with a special focus on the analysis of large-scale recurrent patterns of variability (weather regimes).
The main tools to provide near-term climate information are Global Climate Models (GCMs). GCMs depict complex real-world processes providing their simplified representation through a set of equations and algorithms. Climate models are therefore imperfect and suffer from systematic errors. Model imperfections lead to differences between the model and observed mean state and thus complicate the initialisation task of near-term climate prediction systems.
While in the long term the goal is to reduce the biases through model improvements and increased computer power, medium-term solutions are also needed. LISTEN aims at contributing to the climate prediction community effort to compensate for the model inadequacy by enhancing the transfer of observed information to the model during the forecast initialisation. This objective is achieved through the implementation of innovative initialisation techniques explicitly designed to tackle specific limitations detected in the methods currently in use and are applied to initialise decadal predictions.
The study is developed in the context of a common framework in order to ensure efficient use and dissemination of the data and the findings within the scientific community.
In addition, the forecast skill assessment is carried out with a special focus on the analysis of large-scale recurrent patterns of variability (weather regimes).
LISTEN designs and implements two innovative initialisation techniques: the first one is the quantile matching method, that aims at tackling the drift and the potential inconsistencies between the observed/model distribution of variability. The drift is avoided by choosing an initial state that belongs to the model attractor (i.e. the ensemble of all the model trajectorie) and the variability amplitude incompatibilities are corrected by matching the observed and model statistical distributions.
The most remarkable impacts of the quantile matching technique are found in the North Atlantic. First of all, the quantile matching predictions overcome the issue of the deep convection collapse in the Labrador Sea, which the standard full field predictions experience. As an effect of the correct representation of the convection, the skill of the barotropic stream function in the Western subpolar North Atlantic sector is significant throughout the whole forecast period and is the highest compared to that found in historical simulations and full field decadal predictions. Also the Atlantic Meridional Overturning Circulation is skillfully predicted by the quantile matching predictions throughout the whole forecast time, and the main improvements of sea surface temperature and ocean heat content skill are found in the North Atlantic subpolar gyre sector.
The second innovative experiment implemented in LISTEN is the analogue method, which attempts to address the issue of the geographical mismatch between the model and the observed variability modes.
The analogue method consists in choosing an initial state that belongs to the model attractor and whose amplitude of the main variability modes is as close as possible to the amplitude of the corresponding reference modes, at the initialisation time. The initialisation method has been implemented and the production of the decadal climate predictions initialised with the analogue method is currently ongoing.
The analysis of the weather regimes has shown a remarkable ability of the quantile matching predictions in reproducing the regimes patterns. Moreover, the predictions also show a good performance in capturing the regime frequency and persistence, with a tendency to slightly overestimate short events and underestimate longer events.
The most remarkable impacts of the quantile matching technique are found in the North Atlantic. First of all, the quantile matching predictions overcome the issue of the deep convection collapse in the Labrador Sea, which the standard full field predictions experience. As an effect of the correct representation of the convection, the skill of the barotropic stream function in the Western subpolar North Atlantic sector is significant throughout the whole forecast period and is the highest compared to that found in historical simulations and full field decadal predictions. Also the Atlantic Meridional Overturning Circulation is skillfully predicted by the quantile matching predictions throughout the whole forecast time, and the main improvements of sea surface temperature and ocean heat content skill are found in the North Atlantic subpolar gyre sector.
The second innovative experiment implemented in LISTEN is the analogue method, which attempts to address the issue of the geographical mismatch between the model and the observed variability modes.
The analogue method consists in choosing an initial state that belongs to the model attractor and whose amplitude of the main variability modes is as close as possible to the amplitude of the corresponding reference modes, at the initialisation time. The initialisation method has been implemented and the production of the decadal climate predictions initialised with the analogue method is currently ongoing.
The analysis of the weather regimes has shown a remarkable ability of the quantile matching predictions in reproducing the regimes patterns. Moreover, the predictions also show a good performance in capturing the regime frequency and persistence, with a tendency to slightly overestimate short events and underestimate longer events.
LISTEN has introduced novel medium-term options to reduce the effect of the model error and its consequent bias. The two innovative initialisation techniques aim at tackling some specific limitations of the well-known existing methods, for multi-annual to decadal climate predictions.
LISTEN main findings highlight the potential to improve the prediction skill by optimising the transfer of observed information to the model, at the initialisation time. The promising results from the techniques implemented have paved the way to explore further refinements of these methodologies.
Furthermore, LISTEN has offered a favourable opportunity for substantial progress in near-term climate predictions and the outcome of the action may also have an impact on future climate services.
LISTEN main findings highlight the potential to improve the prediction skill by optimising the transfer of observed information to the model, at the initialisation time. The promising results from the techniques implemented have paved the way to explore further refinements of these methodologies.
Furthermore, LISTEN has offered a favourable opportunity for substantial progress in near-term climate predictions and the outcome of the action may also have an impact on future climate services.