-
Quantifying Uncertainty in Discrete-Continuous and Skewed Data with Bayesian Deep Learning
Authors:
Thomas Vandal,
Evan Kodra,
Jennifer Dy,
Sangram Ganguly,
Ramakrishna Nemani,
Auroop R. Ganguly
Abstract:
Deep Learning (DL) methods have been transforming computer vision with innovative adaptations to other domains including climate change. For DL to pervade Science and Engineering (S&E) applications where risk management is a core component, well-characterized uncertainty estimates must accompany predictions. However, S&E observations and model-simulations often follow heavily skewed distributions…
▽ More
Deep Learning (DL) methods have been transforming computer vision with innovative adaptations to other domains including climate change. For DL to pervade Science and Engineering (S&E) applications where risk management is a core component, well-characterized uncertainty estimates must accompany predictions. However, S&E observations and model-simulations often follow heavily skewed distributions and are not well modeled with DL approaches, since they usually optimize a Gaussian, or Euclidean, likelihood loss. Recent developments in Bayesian Deep Learning (BDL), which attempts to capture uncertainties from noisy observations, aleatoric, and from unknown model parameters, epistemic, provide us a foundation. Here we present a discrete-continuous BDL model with Gaussian and lognormal likelihoods for uncertainty quantification (UQ). We demonstrate the approach by develo** UQ estimates on `DeepSD', a super-resolution based DL model for Statistical Downscaling (SD) in climate applied to precipitation, which follows an extremely skewed distribution. We find that the discrete-continuous models outperform a basic Gaussian distribution in terms of predictive accuracy and uncertainty calibration. Furthermore, we find that the lognormal distribution, which can handle skewed distributions, produces quality uncertainty estimates at the extremes. Such results may be important across S&E, as well as other domains such as finance and economics, where extremes are often of significant interest. Furthermore, to our knowledge, this is the first UQ model in SD where both aleatoric and epistemic uncertainties are characterized.
△ Less
Submitted 24 May, 2018; v1 submitted 13 February, 2018;
originally announced February 2018.
-
Physics-guided probabilistic modeling of extreme precipitation under climate change
Authors:
Evan Kodra,
Singdhansu Chatterjee,
Stone Chen,
Auroop R. Ganguly
Abstract:
Earth System Models (ESMs) are the state of the art for projecting the effects of climate change. However, longstanding uncertainties in their ability to simulate regional and local precipitation extremes and related processes inhibit decision making. Stakeholders would be best supported by probabilistic projections of changes in extreme precipitation at relevant space-time scales. Here we propose…
▽ More
Earth System Models (ESMs) are the state of the art for projecting the effects of climate change. However, longstanding uncertainties in their ability to simulate regional and local precipitation extremes and related processes inhibit decision making. Stakeholders would be best supported by probabilistic projections of changes in extreme precipitation at relevant space-time scales. Here we propose an empirical Bayesian model that extends an existing skill and consensus based weighting framework and test the hypothesis that nontrivial, physics-guided measures of ESM skill can help produce reliable probabilistic characterization of climate extremes. Specifically, the model leverages knowledge of physical relationships between temperature, atmospheric moisture capacity, and extreme precipitation intensity to iteratively weight and combine ESMs and estimate probability distributions of return levels. Out-of-sample validation shows evidence that the Bayesian model is a sound method for deriving reliable probabilistic projections. Beyond precipitation extremes, the framework may be a basis for a generic, physics-guided approach to modeling probability distributions of climate variables in general, extremes or otherwise.
△ Less
Submitted 18 July, 2017;
originally announced July 2017.
-
DeepSD: Generating High Resolution Climate Change Projections through Single Image Super-Resolution
Authors:
Thomas Vandal,
Evan Kodra,
Sangram Ganguly,
Andrew Michaelis,
Ramakrishna Nemani,
Auroop R Ganguly
Abstract:
The impacts of climate change are felt by most critical systems, such as infrastructure, ecological systems, and power-plants. However, contemporary Earth System Models (ESM) are run at spatial resolutions too coarse for assessing effects this localized. Local scale projections can be obtained using statistical downscaling, a technique which uses historical climate observations to learn a low-reso…
▽ More
The impacts of climate change are felt by most critical systems, such as infrastructure, ecological systems, and power-plants. However, contemporary Earth System Models (ESM) are run at spatial resolutions too coarse for assessing effects this localized. Local scale projections can be obtained using statistical downscaling, a technique which uses historical climate observations to learn a low-resolution to high-resolution map**. Depending on statistical modeling choices, downscaled projections have been shown to vary significantly terms of accuracy and reliability. The spatio-temporal nature of the climate system motivates the adaptation of super-resolution image processing techniques to statistical downscaling. In our work, we present DeepSD, a generalized stacked super resolution convolutional neural network (SRCNN) framework for statistical downscaling of climate variables. DeepSD augments SRCNN with multi-scale input channels to maximize predictability in statistical downscaling. We provide a comparison with Bias Correction Spatial Disaggregation as well as three Automated-Statistical Downscaling approaches in downscaling daily precipitation from 1 degree (~100km) to 1/8 degrees (~12.5km) over the Continental United States. Furthermore, a framework using the NASA Earth Exchange (NEX) platform is discussed for downscaling more than 20 ESM models with multiple emission scenarios.
△ Less
Submitted 8 March, 2017;
originally announced March 2017.
-
Intercomparison of Machine Learning Methods for Statistical Downscaling: The Case of Daily and Extreme Precipitation
Authors:
Thomas Vandal,
Evan Kodra,
Auroop R Ganguly
Abstract:
Statistical downscaling of global climate models (GCMs) allows researchers to study local climate change effects decades into the future. A wide range of statistical models have been applied to downscaling GCMs but recent advances in machine learning have not been explored. In this paper, we compare four fundamental statistical methods, Bias Correction Spatial Disaggregation (BCSD), Ordinary Least…
▽ More
Statistical downscaling of global climate models (GCMs) allows researchers to study local climate change effects decades into the future. A wide range of statistical models have been applied to downscaling GCMs but recent advances in machine learning have not been explored. In this paper, we compare four fundamental statistical methods, Bias Correction Spatial Disaggregation (BCSD), Ordinary Least Squares, Elastic-Net, and Support Vector Machine, with three more advanced machine learning methods, Multi-task Sparse Structure Learning (MSSL), BCSD coupled with MSSL, and Convolutional Neural Networks to downscale daily precipitation in the Northeast United States. Metrics to evaluate of each method's ability to capture daily anomalies, large scale climate shifts, and extremes are analyzed. We find that linear methods, led by BCSD, consistently outperform non-linear approaches. The direct application of state-of-the-art machine learning methods to statistical downscaling does not provide improvements over simpler, longstanding approaches.
△ Less
Submitted 13 February, 2017;
originally announced February 2017.
-
Network science based quantification of resilience demonstrated on the Indian Railways Network
Authors:
Udit Bhatia,
Devashish Kumar,
Evan Kodra,
Auroop R. Ganguly
Abstract:
The structure, interdependence, and fragility of systems ranging from power grids and transportation to ecology, climate, biology and even human communities and the Internet, have been examined through network science. While the response to perturbations has been quantified, recovery strategies for perturbed networks have usually been either discussed conceptually or through anecdotal case studies…
▽ More
The structure, interdependence, and fragility of systems ranging from power grids and transportation to ecology, climate, biology and even human communities and the Internet, have been examined through network science. While the response to perturbations has been quantified, recovery strategies for perturbed networks have usually been either discussed conceptually or through anecdotal case studies. Here we develop a network science-based quantitative methods framework for measuring, comparing and interpreting hazard responses and as well as recovery strategies. The framework, motivated by the recently proposed temporal resilience paradigm, is demonstrated with the Indian Railways Network. The methods are demonstrated through the resilience of the network to natural or human-induced hazards and electric grid failure. Simulations inspired by the 2004 Indian Ocean Tsunami and the 2012 North Indian blackout as well as a cyber-physical attack scenario. Multiple metrics are used to generate various recovery strategies, which are simply sequences in which system components should be recovered after a disruption. Quantitative evaluation of recovery strategies suggests that faster and more resource-effective recovery is possible through network centrality measures. Case studies based on two historical events, specifically the 2004 Indian Ocean tsunami and the 2012 North Indian blackout, and a simulated cyber-physical attack scenario, provides means for interpreting the relative performance of various recovery strategies. Quantitative evaluation of recovery strategies suggests that faster and more resource-effective restoration is possible through network centrality measures, even though the specific strategy may be different for sub-networks or for the partial recovery.
△ Less
Submitted 15 October, 2015; v1 submitted 5 August, 2015;
originally announced August 2015.
