Deep echo state networks with uncertainty quantification for spatio‐temporal forecasting
Глубокие сети с эхо-состояниями и количественной оценкой неопределенности для пространственно-временного прогнозирования
2018-12-18
SCID: 54.1/6bwg8p48
Discuss with AI
deep echo state networkshierarchical Bayesian frameworkreservoir computingspatio-temporal forecastinguncertainty quantification
Figures from the paper
Abstract (AI)
Abstract Long‐lead forecasting for spatio‐temporal systems can entail complex nonlinear dynamics that are difficult to specify a priori. Current statistical methodologies for modeling these processes are often highly parameterized and, thus, challenging to implement from a computational perspective. One potential parsimonious solution to this problem is a method from the dynamical systems and engineering literature referred to as an echo state network (ESN). ESN models use reservoir computing to efficiently compute recurrent neural network forecasts. Moreover, multilevel (deep) hierarchical models have recently been shown to be successful at predicting high‐dimensional complex nonlinear processes, particularly those with multiple spatial and temporal scales of variability (such as those we often find in spatio‐temporal environmental data). Here, we introduce a deep ensemble ESN (D‐EESN) model. Despite the incorporation of a deep structure, the presented model is computationally efficient. We present two versions of this model for spatio‐temporal processes that produce forecasts and associated measures of uncertainty. The first approach utilizes a bootstrap ensemble framework, and the second is developed within a hierarchical Bayesian framework (BD‐EESN). This more general hierarchical Bayesian framework naturally accommodates non‐Gaussian data types and multiple levels of uncertainties. The methodology is first applied to a data set simulated from a novel non‐Gaussian multiscale Lorenz‐96 dynamical system simulation model and, then, to a long‐lead United States (U.S.) soil moisture forecasting application. Across both applications, the proposed methodology improves upon existing methods in terms of both forecast accuracy and quantifying uncertainty.
Key Findings
1
Across a simulated non-Gaussian multiscale Lorenz-96 system and long-lead U.S. soil moisture forecasting, the methods improve forecast accuracy and uncertainty quantification over existing approaches.
2
Introduces deep ensemble echo state networks (D-EESN) that efficiently forecast high-dimensional nonlinear spatio-temporal processes.
3
Provides bootstrap ensemble and hierarchical Bayesian versions, both generating forecasts with associated uncertainty measures.
4
The Bayesian D-EESN framework accommodates non-Gaussian observations and multiple levels of uncertainty.
Research Object
spatio-temporal environmental processes, specifically a non-Gaussian multiscale Lorenz-96 dynamical system and long-lead U.S. soil moisture
Research Subject
long-lead spatio-temporal forecasting accuracy and uncertainty quantification for complex nonlinear dynamics across multiple spatial and temporal scales
Publication Details
Publication Date
2018-12-18
Journal
Publisher
ISSN
Cited by
91
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest