Improving Environmental Model Calibration and Prediction
Abstract
First, we have continued to develop tools for efficient global optimization of environmental models. Our algorithms are hybrid algorithms that combine evolutionary strategies for global search with derivative-free methods for local search in novel and efficient ways. Results thus far are promising and are pointing the way toward practical hybrid optimization tools for environmental models. Second, we are applying function approximation techniques to improve the efficiency of Monte Carlo Markov Chain simulations used in Bayesian optimization of environmental models. Very simple preliminary experiments have yielded 20-30% reduction in computational effort for well converged estimates of model parameter probability distributions.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 18, 2011
- Accession Number
- ADA544854
Entities
People
- Jeffrey S. Baggett