Forecast Correction for Environmental Risk
Bayesian quantile methods for aligning observations, retrospective products, and forecast products from different systems, with evaluation tied to the information available at each forecast origin.
Research
I develop Bayesian methods for forecasting, uncertainty quantification, and interval estimation in dynamic data.
Current projects use climate, environmental, and energy data to study forecast correction, posterior synthesis, temporal evaluation, and reproducible statistical software.
Dynamic Quantile State-Space Models
Extended dynamic quantile linear models with MCMC, Laplace-delta variational Bayes, diagnostics, forecasting, and posterior predictive synthesis.
Bayesian Quantile Deep Echo State Networks for Nonlinear Time Series
Bayesian quantile forecasting with fixed nonlinear recurrent features, shrinkage priors, simulation studies, multi-quantile reporting, and held-out forecast comparisons.
Mean-Tilted Intervals
MTI work separates fixed-content and tolerance-interval targets from regression and dynamic-model extensions, with generalized-Bayes computation as the common thread.
Best Poster Prize
ISBA 2026 World Meeting
My poster Bayesian quantile-based correction and synthesis of climate products received a Best Poster Prize at the ISBA 2026 World Meeting in Nagoya, Japan. The work presents a Bayesian quantile workflow for correcting forecast products and synthesizing corrected quantile forecasts into a posterior predictive distribution. The case study uses local environmental observations and NOAA/NWS forecast guidance near Big Trees.
- Recognition
- Best Poster Prize, ISBA 2026 World Meeting.
- Authors
- Antonio De Leon, Raquel Prado, and Bruno Sansó
- Focus
- Forecast correction, quantile dynamic modeling, posterior synthesis, and state-space modeling.
Selected Papers & Software
-
JSS / CRAN v1.1.2
De Leon, A., Barata, R., Prado, R., Sansó, B. exdqlm: An R Package for Estimation and Analysis of Flexible Dynamic Quantile Linear Models. Manuscript submitted to the Journal of Statistical Software; package on CRAN; arXiv:2607.22760.
-
arXiv / Environmetrics
De Leon, A., Prado, R., Sansó, B. Bayesian Quantile-Based Correction and Synthesis of Hydrologic Products. arXiv preprint; manuscript submitted to Environmetrics. Related work received the EnviBayes Student Paper Competition award and the ISBA 2026 Best Poster Prize. arXiv:2608.11222. Poster PDF.
-
arXiv preprint
De Leon, A., Prado, R., Sansó, B. Bayesian Quantile Deep Echo State Networks for Nonlinear Time Series. Bayesian quantile regression with fixed deep-reservoir features, MCMC and variational inference, simulation studies, and retrospective streamflow and electricity-price comparisons. arXiv:2609.17579 · PDF · HTML · DOI.
-
arXiv preprint
De Leon, A., Prado, R., Sansó, B. Mean-Tilted Intervals: A Generalized-Bayes Approach to Fixed-Content and Tolerance Intervals. Preprint on fixed-content interval targets, tolerance actions, and generalized-Bayes computation. arXiv:2607.26098.
-
arXiv preprint
De Leon, A., Prado, R., Sansó, B. Mean-Tilted Intervals: Regression and Dynamic Models. Generalized-Bayes inference for interval endpoints in regression and dynamic state-space models. arXiv:2610.13960.
-
Published
De Leon, A., Lobato, I. N. (2024). Evidence of non-fundamentalness in OECD capital stocks. Empirical Economics. DOI.
Code and Reproducibility
Package source, manuscript repositories, and selected implementation examples are listed on the Software page. Each item is labeled by release or manuscript status.