A5060759568- Meteorological Phenomena and Simulations
- Climate variability and models
- Energy Load and Power Forecasting
- Hydrology and Drought Analysis
- Sociology and Education Studies
- Hydrological Forecasting Using AI
- Forecasting Techniques and Applications
- Wind and Air Flow Studies
- Education Methods and Technologies
- Data Analysis with R
- Statistical Methods and Inference
- Monetary Policy and Economic Impact
- Solar Radiation and Photovoltaics
- Statistical Methods and Bayesian Inference
- Atmospheric and Environmental Gas Dynamics
- Financial Risk and Volatility Modeling
- Market Dynamics and Volatility
- Neural Networks and Applications
- Corporate Management and Leadership
- Vocational Education and Training
- Psychology, Coaching, and Therapy
- Precipitation Measurement and Analysis
- Air Quality Monitoring and Forecasting
- Statistics Education and Methodologies
- Advanced Statistical Methods and Models
Karlsruhe Institute of Technology
2016-2025
Heidelberg Institute for Theoretical Studies
2016-2025
Chalmers University of Technology
2024
DIPF | Leibniz Institute for Research and Information in Education
2023
NOAA Global Systems Laboratory
2022-2023
ETH Zurich
2018
Norwegian Computing Center
2017
Free University of Bozen-Bolzano
2017
Heidelberg University
2013-2015
University of Applied Sciences Mainz
2015
Ensemble weather predictions require statistical post-processing of systematic errors to obtain reliable and accurate probabilistic forecasts. Traditionally, this is accomplished with distributional regression models in which the parameters a predictive distribution are estimated from training period. We propose flexible alternative based on neural networks that can incorporate nonlinear relationships between arbitrary predictor variables forecast automatically learned data-driven way rather...
Statistical postprocessing techniques are nowadays key components of the forecasting suites in many National Meteorological Services (NMS), with for most them, objective correcting impact different types errors on forecasts. The final aim is to provide optimal, automated, seamless forecasts end users. Many now flourishing statistical, meteorological, climatological, hydrological, and engineering communities. methods range complexity from simple bias corrections very sophisticated...
Probabilistic forecasts in the form of probability distributions over future events have become popular several fields including meteorology, hydrology, economics, and demography. In typical applications, many alternative statistical models data sources can be used to produce probabilistic forecasts. Hence, evaluating selecting among competing methods is an important task. The scoringRules package for R provides functionality comparative evaluation based on proper scoring rules, covering a...
In public discussions of the quality forecasts, attention typically focuses on predictive performance in cases extreme events. However, restriction conventional forecast evaluation methods to subsets observations has unexpected and undesired effects, is bound discredit skillful forecasts when signal-to-noise ratio data generating process low. Conditioning outcomes incompatible with theoretical assumptions established methods, thereby confronting forecasters what we refer as forecaster's...
In weather forecasting, nonhomogeneous regression is used to statistically postprocess forecast ensembles in order obtain calibrated predictive distributions. For wind speed forecasts, the model given by a truncated normal distribution where location and spread are derived from ensemble. This paper proposes two alternative approaches which utilize generalized extreme value (GEV) distribution. A direct apply GEV family, while regime switching approach based on median of ensemble incorporates...
Ensembles of forecasts are obtained from multiple runs numerical weather forecasting models with different initial conditions and typically employed to account for forecast uncertainties. However, biases dispersion errors often occur in ensembles: they usually underdispersive uncalibrated require statistical post‐processing. We present an Ensemble Model Output Statistics (EMOS) method calibration wind‐speed based on the log‐normal (LN) distribution we also show a regime‐switching extension...
Postprocessing ensemble weather predictions to correct systematic errors has become a standard practice in research and operations. However, only few recent studies have focused on postprocessing of wind gust forecasts, despite its importance for severe warnings. Here, we provide comprehensive review comparison eight statistical machine learning methods probabilistic forecasting via postprocessing, that can be divided three groups: State the art techniques from statistics (ensemble model...
The most mature aspect of applying artificial intelligence (AI)/machine learning (ML) to problems in the atmospheric sciences is likely post-processing model output. This article provides some history and current state science with AI for weather climate models. Deriving from discussion at 2019 Oxford workshop on Machine Learning Weather Climate, this paper also presents thoughts medium-term goals advance such use AI, which include assuring that algorithms are trustworthy interpretable,...
Probabilistic solar forecasts may take the form of predictive probability distributions, ensembles, quantiles, or interval forecasts. State-of-the-art approaches build on input from numerical weather prediction (NWP) models and post-processing with statistical machine learning methods. We propose a probabilistic benchmark based deterministic forecast clear-sky irradiance, introduce new methods for that merge techniques modern neural networks, discuss spatio-temporal scenario forecasts,...
Abstract. Non-homogeneous regression is a frequently used post-processing method for increasing the predictive skill of probabilistic ensemble weather forecasts. To adjust seasonally varying error characteristics between forecasts and corresponding observations, different time-adaptive training schemes, including classical sliding window, have been developed non-homogeneous regression. This study compares three such approaches with sliding-window approach application near-surface air...
Summary In Bayesian inference, predictive distributions are typically in the form of samples generated via Markov chain Monte Carlo or related algorithms. this paper, we conduct a systematic analysis how to make and evaluate probabilistic forecasts from such simulation output. Based on proper scoring rules, develop notion consistency that allows assess adequacy methods for estimating stationary distribution underlying We then provide asymptotic results account salient features posterior...
In order to enable the transition towards renewable energy sources, probabilistic forecasting is of critical importance for incorporating volatile power sources such as solar into electrical grid. Solar methods often aim provide predictions irradiance. particular, many hybrid approaches combine physical information from numerical weather prediction models with statistical methods. Even though can useful at intra-day and day-ahead forecast horizons, ensemble forecasts multiple model runs are...
An influential step in weather forecasting was the introduction of ensemble forecasts operational use due to their capability account for uncertainties future state atmosphere. However, are often underdispersive and might also contain bias, which calls some form post-processing. A popular approach calibration is model output statistics (EMOS) resulting a full predictive distribution given variable. this univariate post-processing may ignore prevailing spatial and/or temporal correlation...
Ensemble weather forecasts based on multiple runs of numerical prediction models typically show systematic errors and require postprocessing to obtain reliable forecasts. Accurately modeling multivariate dependencies is crucial in many practical applications, various approaches have been proposed where ensemble predictions are first postprocessed separately each margin then restored via copulas. These two-step methods share common key limitations, particular, the difficulty include...
Abstract Subseasonal weather forecasts are becoming increasingly important for a range of socioeconomic activities. However, the predictive ability physical models is very limited on these time scales. We propose four postprocessing methods based convolutional neural networks to improve subseasonal by correcting systematic errors numerical prediction models. Our operate directly spatial input fields and therefore able retain relationships generate spatially homogeneous predictions. They...
Abstract Artificial intelligence (AI)-based data-driven weather forecasting models have experienced rapid progress over the last years. Recent studies, with trained on reanalysis data, achieve impressive results and demonstrate substantial improvements state-of-the-art physics-based numerical prediction across a range of variables evaluation metrics. Beyond improved predictions, main advantages are their substantially lower computational costs faster generation forecasts, once model has been...
Abstract. Statistical postprocessing of medium-range weather forecasts is an important component modern forecasting systems. Since the beginning data science, numerous new methods have been proposed, complementing already very diverse field. However, one questions that frequently arises when considering different in framework implementing operational relative performance for a given specific task. It particularly challenging to find or construct common comprehensive dataset can be used...
Ensemble model output statistics (EMOS) is a statistical tool for post-processing forecast ensembles of weather variables obtained from multiple runs numerical prediction models in order to produce calibrated predictive probability density functions (PDFs). The EMOS PDF given by parametric distribution with parameters depending on the ensemble forecasts. We propose an calibrating wind speed forecasts based weighted mixtures truncated normal (TN) and log-normal (LN) distributions where...
Probabilistic forecasts in the form of probability distributions over future events have become popular several fields including meteorology, hydrology, economics, and demography. In typical applications, many alternative statistical models data sources can be used to produce probabilistic forecasts. Hence, evaluating selecting among competing methods is an important task. The scoringRules package for R provides functionality comparative evaluation based on proper scoring rules, covering a...
Abstract Precipitation is affected by soil moisture spatial variability. However, this variability not well represented in atmospheric models that do consider transport as a three-dimensional process. This study investigates the sensitivity of precipitation to uncertainty representation terrestrial water flow. The tools used for investigation are Weather Research and Forecasting (WRF) Model its hydrologically enhanced version, WRF-Hydro, applied over central Europe during April–October 2008....
Weather forecasts are typically given in the form of forecast ensembles obtained from multiple runs numerical weather prediction models with varying initial conditions and physics parameterizations. Such ensemble predictions tend to be biased underdispersive thus require statistical postprocessing. In model output statistics (EMOS) approach, a probabilistic is by single parametric distribution parameters depending on members. This article proposes two semi-local methods for estimating EMOS...
Abstract Deep-learning (DL) postprocessing methods are examined to obtain reliable and accurate probabilistic forecasts from single-member numerical weather predictions of integrated vapor transport (IVT). Using a 34-yr reforecast, based on the Center for Western Weather Water Extremes West-WRF mesoscale model North American West Coast IVT, dynamically/statistically derived 0–120-h IVT under atmospheric river (AR) conditions tested. These compared with Global Ensemble Forecast System (GEFS)...
Model diagnostics and forecast evaluation are closely related tasks, with the former concerning in-sample goodness (or lack) of fit latter addressing predictive performance out-of-sample. We review ubiquitous setting in which forecasts cast form quantiles or quantile-bounded prediction intervals. distinguish unconditional calibration, corresponds to classical coverage criteria, from stronger notion conditional as can be visualized quantile reliability diagrams. Consistent scoring...