This work presents a comprehensive meta-analysis aimed at characterizing the intricate landscape of Artificial Intelligence (AI) applications and their impact within the domain of climate change research, both in adaptation and mitigation efforts. Notably, a significant upswing in this interdisciplinary intersection has been observed since 2020. Utilizing advanced topic clustering techniques and qualitative analysis, we have discerned 12 distinct impact and application areas: disaster and risk management, ecosystems and biodiversity, infrastructure and industry, the energy sector, agriculture, healthcare, land use and urban planning, emissions and materials, businesses and information science, wildfires, and water and marine environments. The work furnishes a data-rich panoramic view regarding the functions and roles that AI takes on in these areas. Identified tasks range from predicting and forecasting, to measuring, monitoring, classifying, modelling, and sensing, among others. The intention is to offer valuable guidance to the scholarly community and propel further research endeavors, encouraging meticulous examinations of research trends and gaps in addressing the formidable challenges posed by climate change.

How to cite: Burmester, C. and Scantamburlo, T.: Artificial Intelligence in Climate Change Research: An Overview of Impact and Application Areas, 13. Deutsche Klimatagung, Potsdam, Deutschland, 12–15 Mar 2024, DKT-13-61, https://doi.org/10.5194/dkt-13-61, 2024.

Das Bundesamt für Seeschifffahrt und Hydrografie (BSH) analysiert wöchentlich die Meeresoberflächentemperaturen (sea surface temperature, SST) der Nord- und Ostsee im operationellen Betrieb. Seit 1995 geschieht dies in einem halb-automatisierten Prozess, in welchem täglich erfasste Satellitenmessungen und Temperaturmessungen von Schiffen und Stationen erst manuell geprüft und dann zu Wochenmitteln zusammengefasst werden. Über Lücken in der Fläche wird im aktuellen Prozess mit einem Krigingverfahren interpoliert.

Seit 2023 wird für diesen wöchentlichen Prozess, zunächst für die Nordsee, eine neue Methode entwickelt, welche auf maschinellem Lernen (ML) basiert und ein intelligentes Auffüllen der lückenhaften Beobachtungen ermöglichen soll. Konkret wird hierbei der Ansatz des „Inpaintings“ verwendet - urspünglich in der digitalen Bildbearbeitung verwurzelt. Dabei wird ein „Convolutional Neural Network“ (CNN) mit vollständingen Daten“bildern“, z.B. aus Ozean-Modellen oder Reanalysen trainiert. Das hierbei verwendete neuronale Netzwerk basiert auf einer „partial convolution“ Technik, die auch bei großen, unregelmäßig geformten Datenlücken zuverlässig funktionieren soll.

Hier präsentieren wir erste, vielversprechende Ergebnisse unseres ML-Modells, das mit täglichen SST-Daten aus BSH-eigenen Klimamodellsimulationen und des operationellen Ozeanmodells des BSH trainiert wurde. Zudem wurden aus täglichen Satellitendaten Masken erstellt, die dem ML-Modell realistische Datenlücken vorgeben. Im jetzigen setup liegt der mittlere Temperaturfehler (RMSE) zum Beispiel bei ca. 0.3°C.

Die bestehende wöchentliche SST Analyse hat eine horizontale Auflösung von 20 km, mit der neuen Methode und den hochaufgelösten Trainingsdaten streben wir zukünftig eine höhere Auflösung von ca. 5 km an. Perspektivisch soll uns der ML-Ansatz nicht nur ermöglichen, die wöchentliche SST Analyse weiter zu automatisieren, sondern es soll auch geprüft werden, ob eine neue Rekonstruktion von Nordseetemperaturen für die Zeit vor der Satelliten-Ära möglich ist.

How to cite: Hinrichs, C., Kadow, C., Meuer, J., and Kruschke, T.: Machine-Learning-basierte Analyse und Rekonstruktion von Meeresoberflächentemperaturen der Nordsee, 13. Deutsche Klimatagung, Potsdam, Deutschland, 12–15 Mar 2024, DKT-13-9, https://doi.org/10.5194/dkt-13-9, 2024.

Accurately projecting precipitation changes under anthropogenic global warming is crucial due to the high ecological and socio-economic impacts, especially of extreme events. Earth system model (ESM) simulations that numerically solve the governing equations on a discretized grid are our primary tool for projecting the impacts of changing precipitation characteristics in a warming climate. However, the limited resolution and complexity of current ESMs can introduce systematic errors in the numerical simulations, such as an underestimation of extremes and reduced spatial variability.
Recently, generative machine learning methods have been applied to bias-correct precipitation fields [1,2]. While demonstrating comparable or better results than established statistical approaches, these methods can suffer from training instabilities and require computationally costly retraining for each Earth system model individually. Moreover, they only allow for limited control over the spatial scale at which biases are corrected.
Here, we propose a new approach that promises to address the above issues and can correct different ESMs at a chosen spatial scale. We apply our method to bias-correct and downscale global precipitation simulations from the POEM ESM with three degrees spatial resolution. 
Different approaches to control the spatial consistency between the downscaled fields and the ESM are evaluated, such as noised initial conditions [3] and stabilization constraints [4].
     
References
    
[1] Hess, P., Drüke, M., Petri, S., Strnad, F. M., & Boers, N. (2022). Physically constrained generative adversarial networks for improving precipitation fields from Earth system models. Nature Machine Intelligence, 4(10), 828-839.
    
[2] Harris, L., McRae, A. T., Chantry, M., Dueben, P. D., & Palmer, T. N. (2022). A generative deep learning approach to stochastic downscaling of precipitation forecasts. Journal of Advances in Modeling Earth Systems, 14(10), e2022MS003120.

[3] Bischoff, T., & Deck, K. (2023). Unpaired Downscaling of Fluid Flows with Diffusion Bridges. arXiv preprint arXiv:2305.01822.

[4] White, A., Kilbertus, N., Gelbrecht, M., & Boers, N. (2023). Stabilized Neural Differential Equations for Learning Constrained Dynamics. arXiv preprint arXiv:2306.09739.

How to cite: Hess, P., Gelbrecht, M., Aich, M., Pan, B., White, A., Bathiany, S., and Boers, N.: Bias correction and downscaling of precipitation simulations from Earth system models with generative machine learning, 13. Deutsche Klimatagung, Potsdam, Deutschland, 12–15 Mar 2024, DKT-13-44, https://doi.org/10.5194/dkt-13-44, 2024.

Reservoir computer (RC) is an attractive neural computing framework that can well predict the dynamics of nonlinear systems, which is promissing to address the nonlinear nature of climatic time series. While in dynamical systems theory, estimating recurrent analog (RA) trajectory in state space is as a reliable method to fit and predict the nonlinear time series. Here we would present an investigation on comparing performances of the RC and RA in predicting climatic time series. We find that the RC outperforms the RA in case of the complete observations for a dynamical system, and a combination between the RC and RA can significantly improve their ability to predict the system in case of partial observation. Additionally, we extend their comparision to the framework of inferring causality from the time series, i.e., a RC-based causality detection framework proposed by us, and the convergent cross mapping causality method based on the RA. The results demonstrate that, in terms of accuracy, computation efficiency and robustness to the noise, the causality method based on the RC outperforms that on the RA. These results could provide indications for future developments and applications of the RC in addressing climatic time series. 

How to cite: Huang, Y., Fu, Z., and Franzke, C.: A comparison between Reservoir Computing with Recurrent Analogs in predicting dynamical systems, 13. Deutsche Klimatagung, Potsdam, Deutschland, 12–15 Mar 2024, DKT-13-25, https://doi.org/10.5194/dkt-13-25, 2024.

Event-to-event differences of the El Niño Southern Oscillation (ENSO) result in different patterns of extreme climate conditions globally, which requires ENSO forecasts to not only accurately predict the likelihood of an event but also its type. The high autocorrelation of tropical sea surface temperature anomalies (SSTA) allows sub-seasonal to seasonal (S2S) forecasts of ENSO. Recent studies have suggested that skillful multi-year predictions may even be possible after strong El Niño events.

The Linear Inverse Model (LIM) has been shown to produce state-of-the-art ENSO forecasts. LIM describes the dynamics of the slower-varying ocean as stochastically forced by the rapidly varying atmosphere with its deterministic dynamics assumed to be multivariate linear. Due to the linearity assumption, however, LIM is unable to capture observed asymmetries of ENSO that raise the question of whether predictable nonlinearities must be accounted for or may be treated stochastically. 

In this study, we combine deep neural networks (DNN) with the LIM to assess the role of predictable nonlinearities and non-Markovianity in the evolution of monthly tropical SSTA. The different models are tested on SSTA and sea surface height anomalies (SSHA) data from the CESM2 preindustrial control run, where we observe that modeling nonlinearities significantly enhances the forecast accuracy, particularly in the western tropical Pacific within the 9 to 18-month range. Our results further indicate that the asymmetry of warm and cold events is the main source of nonlinearity that improves the forecast skill.

How to cite: Schlör, J., Jannik Thümmel, J., Capotondi, A., Newman, M., and Goswami, B.: A Hybrid Deep Learning Model for El Niño Southern Oscillation Dynamics , 13. Deutsche Klimatagung, Potsdam, Deutschland, 12–15 Mar 2024, DKT-13-41, https://doi.org/10.5194/dkt-13-41, 2024.

Machine learning (ML) is becoming increasingly important in climate research and Earth system modeling. Our goal is to better understand how different ML methods compare, and to enable researchers to make informed decisions when choosing machine learning tools for a given task. In this work, we explore the problem of learning the dynamics of a system from observed data without prior knowledge of the laws governing the system. Our extensive simulation study focuses on ordinary differential equation (ODE) problems that are specifically designed to reflect key aspects of various ML tasks for dynamical systems - namely, chaos, complexity, measurement uncertainty, and variability in measurement intervals. The study evaluates a variety of methods, including neural ODEs, transformer networks, Gaussian processes, echo state networks, and spline-based estimators. Our results show that the relative performance of the methods tested varies widely depending on the specific task, highlighting that no single method is universally superior. Although our research is predominantly in low-dimensional settings, in contrast to the high-dimensional nature of many climate science challenges, it provides insightful comparisons and understanding of how different approaches perform in learning the dynamics of complex systems.

How to cite: Schötz, C., White, A., and Boers, N.: Comparing Machine Learning Methods for Dynamical Systems, 13. Deutsche Klimatagung, Potsdam, Deutschland, 12–15 Mar 2024, DKT-13-47, https://doi.org/10.5194/dkt-13-47, 2024.