Abstract ID: 046
Deep learning for post-processing global probabilistic forecasts on sub-seasonal time-scales
Lead Author: Nina Horat
Karlsruhe Institute of Technology, Germany
Keywords: Post-processing, Machine Learning, Probabilistic Forecasting
Abstract: Reliable sub-seasonal forecasts for precipitation and temperature are crucial to many sectors including agriculture, public health and renewable energy production. Since the forecast skill of numerical weather forecasts for lead times beyond two weeks is limited, we have developed probabilistic post-processing approaches that combine machine learning methods with meteorological process understanding to improve sub-seasonal forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF).
We investigate four post-processing approaches based on convolutional neural network (CNN) architectures which are able to process global numerical weather forecasts and aim to exploit the predictive information in the spatial structure of the forecasts. The CNN models utilize global forecast fields of multiple meteorological variables as input, and predict tercile probabilities for biweekly averaged temperature and accumulated precipitation for weeks 3-4 and 5-6. The model architectures and the training strategy are optimized to deal with the low signal-to-noise ratio in sub-seasonal forecasts and the limited amount of training data. All proposed post-processing models provide skilful probabilistic predictions and improve over climatology and the respective ECMWF baseline forecast in terms of the ranked probability score for both lead times.
We further investigated different strategies for incorporating time series of slowly varying components of the climate system into the CNN models. While in principle, additional teleconnection information should enable more skilful forecasts, we will discuss the potential and challenges of combining exogenous times series information with spatial forecasts in deep learning models.
Co-authors:
Sebastian Lerch (KIT)
