Probabilistic programming: A review for environmental modellers. (April 2019)

• Record Type: Journal Article
• Authors: Krapu, Christopher; Borsuk, Mark
• Is Part Of: Environmental modelling & software. Volume 114(2019)
• Page Start: 40
• Page End: 48
• Publication Date: 2019-04
• Subjects: Bayesian statistics -- Parameter estimation -- Uncertainty quantification -- Probabilistic programming; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2019.01.014
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: The development process for an environmental model involves multiple iterations of a planning-implementation-assessment cycle. Probabilistic programming languages (PPLs) are designed to expedite this process with general-purpose methods for implementing models, efficiently inferring their parameters, and generating probabilistic predictions. Probabilistic programming exists at the intersection of Bayesian statistics, machine learning, and process-based modelling and therefore can be of value to the environmental modelling community. In this review article, we explain how it can be used to accelerate model development and allow for statistical inference using more complicated models and larger data sets than previously possible. Specific challenges and limitations to employing such frameworks are also raised. We provide guidance to help modellers decide whether incorporating probabilistic programming in their work may improve the efficiency and quality of their analyses. Highlights: Probabilistic programming languages offer modellers generic components for specifying stochastic and deterministic models. Advances in Bayesian estimation methods allow for model fitting and uncertainty quantification in a range of scenarios. Probabilistic programming approaches can help bridge the gap between statistics, environmental modelling and machine learning.