Uncertainty estimation with deep learning for rainfall–runoff modeling

Research question

How can deep learning produce and evaluate predictive uncertainty for rainfall–runoff modelling?

Method and findings

Klotz and colleagues developed a benchmarking procedure and evaluated four deep-learning baselines: three using mixture density networks and one using Monte Carlo dropout. They found strong uncertainty-estimation baselines, particularly among the mixture-density approaches, and examined model behaviour beyond aggregate performance.

Limitations

A predictive distribution is evaluated within its data and modelling context. It is not a guarantee that future conditions outside that context are adequately represented.

Engineering interpretation

Read the validation workflow before comparing forecast scores.

Sources & further reading

• View original research: Uncertainty estimation with deep learning for rainfall–runoff modeling ↗Klotz et al. · 31 March 2022
  External source · Checked 24 September 2026

Source findings are distinguished from editorial interpretation. Apply current local criteria and project evidence when making engineering decisions.