Natural hazards – capturing weather patterns across climates and scales


CNDS fellows have published a recent study in the NATURE journal Scientific Reports on the challenges of spatio-temporal precipitation generation under various climate and data related conditions, using a model called TripleM, which is now also available to the public.

Precipitation is the most important component of the water cycle. Multiple studies rely on the availability of long precipitation time series for different types of impact analyses such as in the earth sciences, ecology and climate research. Such impact analyses include but are not limited to flood hazard assessment, agricultural production or public health. Long precipitation time series can be simulated with stochastic algorithms, usually termed weather generators. Weather generators are mathematical algorithms that can extrapolate observed ground observation weather time series with similar statistical characteristics to the observations.

The simulation of precipitation fields is challenging due to the complex intermittent character of precipitation in space and time. For this reason, various algorithms have been proposed and applied. As strategic evaluations are still missing, the knowledge in spatio-temporal precipitation generation remains fragmented. “All the really interesting different weather generators that have been published over the last few years have specific advantages and disadvantages, and we are currently facing a situation where a strategic evaluation of the different model philosophies in an application context is required, to get a better picture of what the different algorithms can or cannot provide to the various scientific disciplines”, says Dr. Korbinian Breinl.

CNDS fellows collaborated in this study with national and international peers of Uppsala University, SLU, the University of Zurich (Switzerland) and the United Nations University (Germany). They suggested a new evaluation framework for testing weather generation across spatial scales, climates and assuming data scarcity as faced in various regions around the globe using their own precipitation model TripleM. Three large-scale study areas in the United States with a solid database were chosen to cover a range of different climate types (Figure 1). The team took a very first step towards narrowing the knowledge gap. “We also want to be as transparent as possible with our work and made the source code available to support the development of future algorithms”, says Prof. Giuliano Di Baldassarre, co-author of the study and director of CNDS. The team considers their study to be a call upon scientists for testing various weather generators using common evaluation standards and for making their source codes accessible to contribute to this task. 

Figure 1. The three study areas in the United States including the rain gauges providing the observation data (GHCN-Daily - for the weather generation experiments.
Figure 1. The three study areas in the United States including the rain gauges providing the observation data (GHCN-Daily - for the weather generation experiments. 

The TripleM precipitation model is a crucial component of the ongoing project STEEP STREAMS (funded by the Swedish Research Council FORMAS within WaterJPI, ERA-Net Cofund WaterWorks 2014), where hydrologists of Uppsala University in collaboration with Italian and Portuguese researchers from Trento and Lisbon are investigating how future changes of weather patterns will lead to changes in the hydrology of small steep Alpine catchments that are highly vulnerable to flash flooding and hyperconcentrated flows. The results of the STEEP STREAMS project will support the development of new design criteria for defence structures to mitigate the impact of such extreme events in the future.

Download the paper here:

Breinl, K., Di Baldassarre, G., Lopez, M. G., Hagenlocher, M., Vico, G., & Rutgersson, A. (2017). Can weather generation capture precipitation patterns across different climates, spatial scales and under data scarcity?. Scientific Reports7.

Access the model here:

Contact: Korbinian Breinl