Nicole Ludwig

Jannik Thuemmel, Matthias Karlbauer, Sebastian Otte et al.

Deep learning has gained immense popularity in the Earth sciences as it enables us to formulate purely data-driven models of complex Earth system processes. Deep learning-based weather prediction (DLWP) models have made significant progress in the last few years, achieving forecast skills comparable to established numerical weather prediction models with comparatively lesser computational costs. In order to train accurate, reliable, and tractable DLWP models with several millions of parameters, the model design needs to incorporate suitable inductive biases that encode structural assumptions about the data and the modelled processes. When chosen appropriately, these biases enable faster learning and better generalisation to unseen data. Although inductive biases play a crucial role in successful DLWP models, they are often not stated explicitly and their contribution to model performance remains unclear. Here, we review and analyse the inductive biases of state-of-the-art DLWP models with respect to five key design elements: data selection, learning objective, loss function, architecture, and optimisation method. We identify the most important inductive biases and highlight potential avenues towards more efficient and probabilistic DLWP models.

Nina Effenberger, Nicole Ludwig

Climate change will impact wind and therefore wind power generation with largely unknown effect and magnitude. Climate models can provide insights and should be used for long-term power planning. In this work we use Gaussian processes to predict power output given wind speeds from a global climate model and compare the aggregated predictions to actual power generation. Analyzing past climate model data supports the use of CMIP6 climate model data for multi-decadal wind power predictions and highlights the importance of being location-aware. Our predictions up to 2050 reveal only minor changes in yearly wind power generation. We find that wind power projections of the two in-between climate scenarios SSP2-4.5 and SSP3-7.0 closely align with actual wind power generation between 2015 and 2023. Our analysis also reveals larger uncertainty associated with Germany's coastal areas in the North as compared to Germany's South, motivating wind power expansion in regions where future wind is likely more reliable. Overall, our results indicate that wind energy will likely remain a reliable energy source in the future.

Nina Effenberger, Nicole Ludwig

Wind power and other forms of renewable energy sources play an ever more important role in the energy supply of today's power grids. Forecasting renewable energy sources has therefore become essential in balancing the power grid. While a lot of focus is placed on new forecasting methods, little attention is given on how to compare, reproduce and transfer the methods to other use cases and data. One reason for this lack of attention is the limited availability of open-source datasets, as many currently used datasets are non-disclosed and make reproducibility of research impossible. This unavailability of open-source datasets is especially prevalent in commercially interesting fields such as wind power forecasting. However, with this paper we want to enable researchers to compare their methods on publicly available datasets by providing the, to our knowledge, largest up-to-date overview of existing open-source wind power datasets, and a categorization into different groups of datasets that can be used for wind power forecasting. We show that there are publicly available datasets sufficient for wind power forecasting tasks and discuss the different data groups properties to enable researchers to choose appropriate open-source datasets and compare their methods on them.

Kornelius Raeth, Nicole Ludwig

Sample-based generative models are increasingly used for probabilistic forecasting in high-stakes decision settings, yet their training objectives are blind to the decision maker's cost structure. These models are commonly trained with strictly proper scoring rules, such as the energy score, which allocate their training signal in proportion to data density, with no awareness of where forecast errors are most costly for downstream decisions. We therefore propose decision-aware training for sample-based generative models, augmenting the energy score objective with a differentiable decision loss that directly penalises the cost incurred by acting on the model's forecast. This combined loss is theoretically grounded, as the decision loss is itself a proper scoring rule. We validate our method on one synthetic and two real-world tasks, showing targeted improvements in cost-sensitive regions while retaining full probabilistic forecasts.