Sebastian Lerch
@sebastianlerch.bsky.social
Professor at the Department of Mathematics and Computer Science at the University of Marburg, interested in probabilistic forecasting, statistics, ML, with applications in weather, energy, environmental sciences, and beyond
Forecast are available for 55 meteorological variables mapped to station locations and spatially aggregated forecasts from surrounding grid points, for NWP models initialized at 00 and 12 UTC, in hourly lead times up to 21h. Observations of 6 variables are available at 170 stations.
August 7, 2025 at 4:10 PM
Forecast are available for 55 meteorological variables mapped to station locations and spatially aggregated forecasts from surrounding grid points, for NWP models initialized at 00 and 12 UTC, in hourly lead times up to 21h. Observations of 6 variables are available at 170 stations.
The potential CRPS of the HRES forecast aligns well with the CRPS of the operational IFS ensemble.
June 5, 2025 at 8:48 AM
The potential CRPS of the HRES forecast aligns well with the CRPS of the operational IFS ensemble.
AIWP models show skillful forecasts for lead times of up to 10 days when compared to the ERA5 climatology in terms of the potential CRPS.
June 5, 2025 at 8:48 AM
AIWP models show skillful forecasts for lead times of up to 10 days when compared to the ERA5 climatology in terms of the potential CRPS.
Results on WeatherBench 2 data confirm fast-paced progress, with AIWP models, in particular GraphCast, showing improvements in the potential CRPS over the HRES model
June 5, 2025 at 8:48 AM
Results on WeatherBench 2 data confirm fast-paced progress, with AIWP models, in particular GraphCast, showing improvements in the potential CRPS over the HRES model
In addition to forecast evaluation via proper scoring rules, we also evaluate the forecasts from an economic perspective by considering trading strategies that utilize the multivariate probabilistic information.
June 3, 2025 at 5:37 AM
In addition to forecast evaluation via proper scoring rules, we also evaluate the forecasts from an economic perspective by considering trading strategies that utilize the multivariate probabilistic information.
We propose a generative ML model for multivariate, probabilistic forecasting of time series of electricity prices, and compare to state-of-the-art statistical benchmark models.
June 3, 2025 at 5:37 AM
We propose a generative ML model for multivariate, probabilistic forecasting of time series of electricity prices, and compare to state-of-the-art statistical benchmark models.
Personal update: After almost 10 years at KIT, I will move to the University of Marburg as a professor at the Department of Mathematics and Computer Science in April. I will of course miss the many great colleagues and students at KIT, but am very much looking forward to exciting new opportunities.
March 25, 2025 at 9:02 AM
Personal update: After almost 10 years at KIT, I will move to the University of Marburg as a professor at the Department of Mathematics and Computer Science in April. I will of course miss the many great colleagues and students at KIT, but am very much looking forward to exciting new opportunities.
New preprint: "Learning low-dimensional representations of ensemble forecast fields using autoencoder-based methods" with Jieyu Chen and Kevin Höhlein: arxiv.org/abs/2502.04409. We propose dimensionality reduction methods tailored to ensemble simulations of gridded fields.
February 10, 2025 at 8:02 AM
New preprint: "Learning low-dimensional representations of ensemble forecast fields using autoencoder-based methods" with Jieyu Chen and Kevin Höhlein: arxiv.org/abs/2502.04409. We propose dimensionality reduction methods tailored to ensemble simulations of gridded fields.
We further compare the post-processing approaches to a NN-based direct forecasting model, which predicts PV power based on the weather inputs without the intermediate conversion via the model chain, and achieves almost the same performance.
January 2, 2025 at 8:26 PM
We further compare the post-processing approaches to a NN-based direct forecasting model, which predicts PV power based on the weather inputs without the intermediate conversion via the model chain, and achieves almost the same performance.
Applying post-processing to the PV power predictions obtained as the output of the model chain is the most important contributor to improving the forecasts, whereas the effects of post-processing the weather inputs are negligible.
January 2, 2025 at 8:26 PM
Applying post-processing to the PV power predictions obtained as the output of the model chain is the most important contributor to improving the forecasts, whereas the effects of post-processing the weather inputs are negligible.
In a case study on a benchmark dataset from the Jacumba solar plant in the US, we find that post-processing generally improves the GHI and PV power forecasts. Neural network-based methods achieve slightly better performance than statistical approaches.
January 2, 2025 at 8:26 PM
In a case study on a benchmark dataset from the Jacumba solar plant in the US, we find that post-processing generally improves the GHI and PV power forecasts. Neural network-based methods achieve slightly better performance than statistical approaches.
We investigate the use of post-processing and ML in model chain approaches, where different strategies are possible: Post-processing only the weather inputs, post-processing only the PV power predictions, or applying post-processing in both steps (or none at all).
January 2, 2025 at 8:26 PM
We investigate the use of post-processing and ML in model chain approaches, where different strategies are possible: Post-processing only the weather inputs, post-processing only the PV power predictions, or applying post-processing in both steps (or none at all).
In a case study on a benchmark dataset from the Jacumba solar plant in the US, we find that post-processing generally improves the GHI and PV power forecasts. Neural network-based methods achieve slightly better performance than statistical approaches.
January 2, 2025 at 8:22 PM
In a case study on a benchmark dataset from the Jacumba solar plant in the US, we find that post-processing generally improves the GHI and PV power forecasts. Neural network-based methods achieve slightly better performance than statistical approaches.
We investigate the use of post-processing and ML in model chain approaches, where different strategies are possible: Post-processing only the weather inputs, post-processing only the PV power predictions, or applying post-processing in both steps (or none at all).
January 2, 2025 at 8:22 PM
We investigate the use of post-processing and ML in model chain approaches, where different strategies are possible: Post-processing only the weather inputs, post-processing only the PV power predictions, or applying post-processing in both steps (or none at all).
Open 3-year postdoc position at KIT on ML for probabilistic hydro-meteorological forecasting.
The position is part of an interdisciplinary research project with partners from meteorology, hydrology and federal flood forecasting agencies.
Contact me if you are interested or have any questions!
The position is part of an interdisciplinary research project with partners from meteorology, hydrology and federal flood forecasting agencies.
Contact me if you are interested or have any questions!
December 9, 2024 at 7:36 AM
Open 3-year postdoc position at KIT on ML for probabilistic hydro-meteorological forecasting.
The position is part of an interdisciplinary research project with partners from meteorology, hydrology and federal flood forecasting agencies.
Contact me if you are interested or have any questions!
The position is part of an interdisciplinary research project with partners from meteorology, hydrology and federal flood forecasting agencies.
Contact me if you are interested or have any questions!
Interested in probabilistic forecasting, ensemble post-processing or verification? Submit an abstract to our
EGU 2025 Session NP 5.2!
⏰Deadline: January 15, 2025
EGU 2025 Session NP 5.2!
⏰Deadline: January 15, 2025
November 24, 2024 at 7:08 PM
Interested in probabilistic forecasting, ensemble post-processing or verification? Submit an abstract to our
EGU 2025 Session NP 5.2!
⏰Deadline: January 15, 2025
EGU 2025 Session NP 5.2!
⏰Deadline: January 15, 2025
Of course, results vary over variables and locations (here: CRPSS, using the ECMWF ensemble as a reference model).
March 21, 2024 at 11:36 AM
Of course, results vary over variables and locations (here: CRPSS, using the ECMWF ensemble as a reference model).
Overall, we find that the PH methods and some of the IC methods are able to provide probabilistic forecasts competitive with the operational ECMWF ensemble forecast, and can offer improvements for the first few days of lead time.
March 21, 2024 at 11:36 AM
Overall, we find that the PH methods and some of the IC methods are able to provide probabilistic forecasts competitive with the operational ECMWF ensemble forecast, and can offer improvements for the first few days of lead time.
Post-hoc (PH) UQ approaches, by contrast, utilize statistical or ML methods to supplement deterministic forecasts with uncertainty information and thus turn them into probabilistic forecasts.
March 21, 2024 at 11:36 AM
Post-hoc (PH) UQ approaches, by contrast, utilize statistical or ML methods to supplement deterministic forecasts with uncertainty information and thus turn them into probabilistic forecasts.
Initial condition (IC) approaches generate an ensemble forecast by running a data-driven model multiple times based on a number of different initial conditions. Those can be otained by adding random noise, using NWP ensemble ICs, or data-driven approaches.
March 21, 2024 at 11:35 AM
Initial condition (IC) approaches generate an ensemble forecast by running a data-driven model multiple times based on a number of different initial conditions. Those can be otained by adding random noise, using NWP ensemble ICs, or data-driven approaches.