NeurIPS 2019 Workshop
Tackling Climate Change with Machine Learning

About Speakers Schedule Organizers Accepted Works Call for Submissions FAQ


Many in the ML community wish to take action on climate change, yet feel their skills are inapplicable. This workshop aims to show that in fact the opposite is true: while no silver bullet, ML can be an invaluable tool both in reducing greenhouse gas emissions and in helping society adapt to the effects of climate change. Climate change is a complex problem, for which action takes many forms - from designing smart electrical grids to tracking deforestation in satellite imagery. Many of these actions represent high-impact opportunities for real-world change, as well as being interesting problems for ML research.

About the Workshop

Invited Speakers

Jeff Dean (Google AI)
Carla Gomes (Cornell)
Felix Creutzig (MCC Berlin, TU Berlin)
Lester Mackey (Microsoft Research, Stanford)


8:15 - 8:30 - Welcome and opening remarks
8:30 - 9:05 - Jeff Dean (Google AI) Computation + Systems vs Climate Change (Invited talk)
9:05 - 9:15 - Felipe Oviedo: Machine learning identifies the most valuable synthesis conditions for next-generation photovoltaics (Spotlight talk)
9:15 - 9:25 - Valentina Zantedeschi: Cumulo: A Dataset for Learning Cloud Classes (Spotlight talk)
9:25 - 9:35 - Qinghu Tang: Fine-Grained Distribution Grid Mapping Using Street View Imagery (Spotlight talk)
9:35 - 9:45 - Shamindra Shrotriya and Niccolo Dalmasso: Predictive Inference of a Wildfire Risk Pipeline in the United States (Spotlight talk)
9:45 - 10:30 - Coffee break + Poster Session
10:30 - 11:05 - Felix Creutzig (MCC Berlin, TU Berlin): Leveraging digitalization for urban solutions in the Anthropocene (Invited talk)
11:05 - 11:15 - Ashish Kapoor: Helping Reduce Environmental Impact of Aviation with Machine Learning (Spotlight talk)
11:15 - 12:00 - Yoshua Bengio, Andrew Ng, Carla Gomes, Lester Mackey, Jeff Dean: Panel - Climate Change: A Grand Challenge for ML
12:00 - 2:00 - Networking lunch (provided) + Poster Session
2:00 - 2:40 - Carla Gomes (Cornell) Computational Sustainability: Computing for a Better World and a Sustainable Future (Invited talk)
2:40 - 2:50 - Kyle Story: A Global Inventory of Utility-Scale Solar Photovoltaic Power Stations (Spotlight talk)
2:50 - 3:00 - Kiwan Maeng: Enhancing Stratospheric Weather Analyses and Forecasts by Deploying Sensors from a Weather Balloon (Spotlight talk)
3:00 - 3:10 - Daisy Zhe Wang: Measuring Impact of Climate Change on Tree Species: analysis of JSDM on FIA data (Spotlight talk)
3:10 - 3:20 - Adrian Albert: Emulating Numeric Hydroclimate Models with Physics-Informed cGANs (Spotlight talk)
3:20 - 3:30 - Draguna Vrabie: Stripping off the implementation complexity of physics-based model predictive control for buildings via deep learning (Spotlight talk)
3:30 - 4:15 - Coffee break + Poster Session
4:15 - 4:40 - Lester Mackey (Microsoft Research) Improving Subseasonal Forecasting in the Western U.S. (Invited talk)
4:40 - 4:50 - Saumya Sinha: Detecting Avalanche Deposits using Variational Autoencoder on Sentinel-1 Satellite Imagery (Spotlight talk)
4:50 - 5:00 - Jacob Pettit: Increasing performance of electric vehicles in ride-hailing services using deep reinforcement learning (Spotlight talk)
5:00 - 6:00 - John Platt, Jennifer Chayes, James Kelloway, Marta Gonzalez, Matt Horne: Panel - Practical Challenges in Applying ML to Climate Change


David Rolnick (UPenn)
Priya Donti (CMU)
Lynn Kaack (ETH Zürich)
Alexandre Lacoste (Element AI)
Tegan Maharaj (Mila)
Andrew Ng (Stanford)
John Platt (Google AI)
Jennifer Chayes (Microsoft Research)
Yoshua Bengio (Mila)

ElementAI logo

DeepMind logo

Microsoft logo

Accepted Works

(1) Warm-Starting AC Optimal Power Flow with Graph Neural Networks pdf

Frederik Diehl (fortiss)

Abstract: (click to expand) Efficient control of power grids is important both for efficiently managing genera- tors and to prolong longevity of components. However, that problem is NP-hard and linear approximations are necessary. The deployment of machine learning methods is hampered by the need to guarantee solutions. We propose to use Graph Neural Networks (GNNs) to model a power grid and produce an initial solution used to warm-start the optimization. This allows us to achieve the best of both worlds: Fast convergence and guaranteed solutions. On a synthetic power grid modelling Texas, we achieve a mean speedup by a factor of 2.8. This allows us to dispense with linear approximation, leads to more efficient generator dispatch, and can potentially save hundreds of megatons of CO2 -equivalent.

(2) Quantifying Urban Canopy Cover with Deep Convolutional Neural Networks pdf

Bill Cai (Massachusetts Institute of Technology); Xiaojiang Li (Temple University); Carlo Ratti (Massachusetts Institute of Technology )

Abstract: (click to expand) Urban canopy cover is important to mitigate the impact of climate change. Yet, existing quantification of urban greenery is either manual and not scalable, or use traditional computer vision methods that are inaccurate. We train deep convolutional neural networks (DCNNs) on datasets used for self-driving cars to estimate urban greenery instead, and find that our semantic segmentation and direct end-to-end estimation method are more accurate and scalable, reducing mean absolute error of estimating the Green View Index (GVI) metric from 10.1% to 4.67%. With the revised DCNN methods, the Treepedia project was able to scale and analyze canopy cover in 22 cities internationally, sparking interest and action in public policy and research fields.

(3) Using LSTMs for climate change assessment studies on droughts and floods pdf

Frederik Kratzert (LIT AI Lab, Institute for Machine Learning, Johannes Kepler University Linz, Austria); Daniel Klotz (LIT AI Lab, Institute for Machine Learning, Johannes Kepler University Linz, Austria); Johannes Brandstetter (LIT AI Lab, Institute for Machine Learning, Johannes Kepler University Linz, Austria); Pieter-Jan Hoedt (Johannes Kepler University Linz); Grey Nearing (Department of Geological Sciences, University of Alabama, Tuscaloosa, AL United States); Sepp Hochreiter (LIT AI Lab, Institute for Machine Learning, Johannes Kepler University Linz, Austria)

Abstract: (click to expand) Climate change affects occurrences of floods and droughts worldwide. However, predicting climate impacts over individual watersheds is difficult, primarily because accurate hydrological forecasts require models that are calibrated to past data. In this work we present a large-scale LSTM-based modeling approach that - by training on large data sets - learns a diversity of hydrological behaviors. Previous work shows that this model is more accurate than current state-of-the-art models, even when the LSTM-based approach operates out-of-sample and the latter in-sample. In this work, we show how this model can assess the sensitivity of the underlying systems with regard to extreme (high and low) flows in individual watersheds over the continental US.

(4) Deep learning predictions of sand dune migration

Kelly Kochanski (University of Colorado Boulder); Divya Mohan (University of California Berkeley); Jenna Horrall (James Madison University); Ghaleb Abdulla (Lawrence Livermore National Laboratory)

Abstract: (click to expand) Climate change is making many desert regions warmer, drier, and sandier. These conditions kill vegetation, and release once-stable sand into the wind, allowing it to form dunes that threaten roads, farmland, and solar panel installations. With enough warning, people can mitigate dune damages by moving infrastructure or restoring vegetation. Current dune simulations, however, do not scale well enough to provide useful forecasts for the ~5% of Earth's land surface that is covered by mobile sands. We propose to train a deep learning simulation to emulate the output of a community-standard physics-based dune simulation. We will base the new model on a GAN-based video prediction model with an excellent track record for predicting spatio-temporal patterns to model, and use it to simulate dune topographies over time. Our preliminary work indicates that the new model will run up to ten million times faster than existing dune simulations, which would turn dune modelling from an exercise that covers a handful of dunes to a practical forecast for large desert regions.

(5) Learning to Focus and Track Hurricanes

Sookyung Kim (Lawrence Livermore National Laboratory); Sunghyun Park (Korea University); Sunghyo Chung (Kakao Corp.); Joonseok Lee (Google Research); Jaegul Choo (Korea University); Mr Prabhat (Lawrence Berkeley National Laboratory); Yunsung Lee (Korea University)

Abstract: (click to expand) This paper tackles the task of extreme climate event tracking. We propose a simple but robust end-to-end model based on multi-layered ConvLSTMs, suitable for climate event tracking. It first learns to imprint the location and the appearance of the target at the first frame in an auto-encoding fashion. Next, the learned feature is fed to the tracking module to track the target in subsequent time frames. To tackle the data shortage problem, we propose data augmentation based on conditional generative adversarial networks. Extensive experiments show that the proposed framework significantly improves tracking performance of a hurricane tracking task over several state-of-the-art methods.

(6) DeepWind: Weakly Supervised Localization of Wind Turbines in Satellite Imagery pdf

Sharon Zhou (Stanford University); Jeremy Irvin (Stanford); Zhecheng Wang (Stanford University); Ram Rajagopal (Stanford University); Andrew Ng (Stanford U.); Eva Zhang (Stanford University); Will Deaderick (Stanford University); Jabs Aljubran (Stanford University)

Abstract: (click to expand) Wind energy is being adopted at an unprecedented rate. The locations of wind energy sources, however, are largely undocumented and expensive to curate manually, which significantly impedes their integration into power systems. Towards the goal of mapping global wind energy infrastructure, we develop deep learning models to automatically localize wind turbines in satellite imagery. Using only image-level supervision, we experiment with several different weakly supervised convolutional neural networks to detect the presence and locations of wind turbines. Our best model, which we call DeepWind, achieves an average precision of 0.866 on the test set. DeepWind demonstrates the potential of automated approaches for identifying wind turbine locations using satellite imagery, ultimately assisting with the management and adoption of wind energy worldwide.

(7) Streamflow Prediction with Limited Spatially-Distributed Input Data pdf

Martin Gauch (University of Waterloo); Juliane Mai (University of Waterloo); Shervan Gharari (University of Saskatchewan); Jimmy Lin (University of Waterloo)

Abstract: (click to expand) Climate change causes more frequent and extreme weather phenomena across the globe. Accurate streamflow prediction allows for proactive and mitigative action in some of these events. As a first step towards models that predict streamflow in watersheds for which we lack ground truth measurements, we explore models that work on spatially-distributed input data. In such a scenario, input variables are more difficult to acquire, and thus models have access to limited training data. We present a case study focusing on Lake Erie, where we find that tree-based models can yield more accurate predictions than both neural and physically-based models.

(8) Establishing an Evaluation Metric to Quantify Climate Change Image Realism pdf

Sharon Zhou (Stanford University); Sasha Luccioni (Mila); Gautier Cosne (Mila); Michael Bernstein (Stanford University); Yoshua Bengio (Mila)

Abstract: (click to expand) With success on controlled tasks, generative models are being increasingly applied to humanitarian applications. In this paper, we focus on the evaluation of a conditional generative model that illustrates the consequences of climate change-induced flooding to encourage public interest and awareness on the issue. Because metrics for comparing the realism of different modes in a conditional generative model do not exist, we propose several automated and human-based methods for evaluation. To do this, we adapt several existing metrics, and assess the automated metrics against gold standard human evaluation. We find that using Frechet Inception Distance (FID) with embeddings from an intermediary Inception-V3 layer that precedes the auxiliary classifier produces results most correlated with human realism. While insufficient alone to establish a human-correlated automatic evaluation metric, we believe this work begins to bridge the gap between human and automated generative evaluation procedures.

(9) Energy Usage Reports: Environmental awareness as part of algorithmic accountability pdf

Kadan Lottick (Haverford College); Silvia Susai (Haverford College); Sorelle Friedler (Haverford College); Jonathan Wilson (Haverford College)

Abstract: (click to expand) The carbon footprint of algorithms must be measured and transparently reported so computer scientists can take an honest and active role in environmental sustainability. In this paper, we take analyses usually applied at the industrial level and make them accessible for individual computer science researchers with an easy-to-use Python package. Localizing to the energy mixture of the electrical power grid, we make the conversion from energy usage to CO2 emissions, in addition to contextualizing these results with more human-understandable benchmarks such as automobile miles driven. We also include comparisons with energy mixtures employed in electrical grids around the world. We propose including these automatically-generated Energy Usage Reports as part of standard algorithmic accountability practices, and demonstrate the use of these reports as part of model-choice in a machine learning context.

(10) Natural Language Generation for Operations and Maintenance in Wind Turbines pdf

Joyjit Chatterjee (University of Hull); Nina Dethlefs (University of Hull)

Abstract: (click to expand) Wind energy is one of the fastest-growing sustainable energy sources in the world but relies crucially on efficient and effective operations and maintenance to generate sufficient amounts of energy and reduce downtime of wind turbines and associated costs. Machine learning has been applied to predict faults in wind turbines, but these predictions have not been supported by suggestions on how to avert and fix occurring errors. In this paper, we present a data-to-text generation system utilising transformers to produce event descriptions of turbine faults from SCADA data capturing the operational status of turbines, and proposing maintenance strategies. Experiments show that our model learns reasonable feature representations that correspond to expert judgements. We anticipate that in making a contribution to the reliability of wind energy, we can encourage more organisations to switch to sustainable energy sources and help combat climate change.

(11) Predictive Inference of a Wildfire Risk Pipeline in the United States pdf

Shamindra Shrotriya (Carnegie Mellon University); Niccolo Dalmasso (Carnegie Mellon University); Alex Reinhart (Carnegie Mellon University)

Abstract: (click to expand) Wildfires are rare events that present severe threats to life and property. Understanding their propagation is of key importance to mitigate and contain their impact, especially since climate change is increasing their occurrence. We propose an end-to-end sequential model of wildfire risk components, including wildfire location, size, duration, and risk exposure. We do so through a combination of marked spatio-temporal point processes and conditional density estimation techniques. Unlike other approaches that use regression-based methods, this approach allows both predictive accuracy and an associated uncertainty measure for each risk estimate, accounting for the uncertainty in prior model components. This is particularly beneficial for timely decision-making by different wildfire risk management stakeholders. To allow us to build our models without limiting them to a specific state or county, we have collected open wildfire and climate data for the entire continental United States. We are releasing this aggregated dataset to enable further o pen research on wildfire models at a national scale.

(12) FutureArctic - beyond Computational Ecology pdf

Steven Latre (UAntwerpen); Dimitri Papadimitriou (UAntwerpen); Ivan Janssens (UAntwerpen); Eric Struyf (UAntwerpen); Erik Verbruggen (UAntwerpen); Ivika Ostonen (UT); Josep Penuelas (UAB); Boris Rewald (RootEcology); Andreas Richter (University of Vienna); Michael Bahn (University of Innsbruck)

Abstract: (click to expand) This paper presents the Future Arctic initiative, a multi-disciplinary training network where machine learning researchers and ecologists cooperatively study both long- and short-term responses to future climate in Iceland.

(13) Make Thunderbolts Less Frightening — Predicting Extreme Weather Using Deep Learning pdf

Christian Schön (Saarland Informatics Campus); Jens Dittrich (Saarland University)

Abstract: (click to expand) Forecasting severe weather conditions is still a very challenging and computationally expensive task due to the enormous amount of data and the complexity of the underlying physics. Machine learning approaches and especially deep learning have however shown huge improvements in many research areas dealing with large datasets in recent years. In this work, we tackle one specific sub-problem of weather forecasting, namely the prediction of thunderstorms and lightning. We propose the use of a convolutional neural network architecture inspired by UNet++ and ResNet to predict thunderstorms as a binary classification problem based on satellite images and lightnings recorded in the past. We achieve a probability of detection of more than 94% for lightnings within the next 15 minutes while at the same time minimizing the false alarm ratio compared to previous approaches.

(14) Cumulo: A Dataset for Learning Cloud Classes Best Paper Award

Valentina Zantedeschi (Jean Monnet University); Fabrizio Falasca (Georgia Institute of Technology); Alyson Douglas (University of Wisconsin Madison); Richard Strange (University of Oxford); Matt Kusner (University College London); Duncan Watson-Parris (University of Oxford)

Abstract: (click to expand) One of the greatest sources of uncertainty in future climate projections comes from limitations in modelling clouds and in understanding how different cloud types interact with the climate system. A key first step in reducing this uncertainty is to accurately classify cloud types at high spatial and temporal resolution. In this paper, we introduce Cumulo, a benchmark dataset for training and evaluating global cloud classification models. It consists of one year of 1km resolution MODIS hyperspectral imagery merged with pixel-width `tracks' of CloudSat cloud labels. Bringing these complementary datasets together is a crucial first step, enabling the Machine-Learning community to develop innovative new techniques which could greatly benefit the climate community. To showcase Cumulo, we provide baseline performance analysis using an invertible flow generative model (IResNet), which further allows us to discover new sub-classes for a given cloud class by exploring the latent space.

(15) Targeting Buildings for Energy Retrofit Using Recurrent Neural Networks with Multivariate Time Series pdf

Gaby Baasch (University of Victoria)

Abstract: (click to expand) The existing building stock accounts for over 30% of global carbon emissions and energy demand. Effective building retrofits are therefore vital in reducing global emissions. Current methods for building energy assessment typically rely on walk-throughs, surveys or the collection of in-situ measurements, none of which are scalable or cost effective. Supervised machine learning methods have the potential to overcome these issues, but their application to retrofit analysis has been limited. This paper serves as a novel showcase for how multivariate time series analysis with Gated Recurrent Units can be applied to targeted retrofit analysis via two case studies: (1) classification of building heating system type and (2) prediction of building envelope thermal properties.

(16) Coupling Oceanic Observation Systems to Study Mesoscale Ocean Dynamics

Gautier Cosne (Mila); Pierre Tandeo (IMT-Atlantique); Guillaume Maze (Ifremer)

Abstract: (click to expand) Understanding local currents in the North Atlantic region of the ocean is a key part of modelling heat transfer and global climate patterns. Satellites provide a surface signature of the temperature of the ocean with a high horizontal resolution while in situ autonomous probes supply high vertical resolution, but horizontally sparse, knowledge of the ocean interior thermal structure. The objective of this paper is to develop a methodology to combine these complementary ocean observing systems measurements to obtain a three-dimensional time series of ocean temperatures with high horizontal and vertical resolution. Within an observation-driven framework, we investigate the extent to which mesoscale ocean dynamics in the North Atlantic region may be decomposed into a mixture of dynamical modes, characterized by different local regressions between Sea Surface Temperature (SST), Sea Level Anomalies (SLA) and Vertical Temperature fields. Ultimately we propose a Latent-class regression method to improve prediction of vertical ocean temperature.

Dara Farrell (Graduate of University of Washington)

Abstract: (click to expand) Predicting future sea levels depends on accurately estimating the rate at which ice sheets deliver fresh water and ice to the oceans, and projecting rates of iceberg calving will be improved with more observations of calving events. The background noise environment in a glacial fjord was measured and the data were analyzed. This paper includes an analysis of methods useful for evaluating background noise. It explores the utility of spectral probability density in evaluating background noise characteristics in the frequency domain, models probability density functions of spectral levels and introduces a parameter \(\sigma_T\) that quantifies the character of noise in frequency bands of interest. It also explores the utility of k-medoids clustering as a pre-sorting method to inform the selection of features on which to base the training of more complex algorithms.

(18) Reducing Inefficiency in Carbon Auctions with Imperfect Competition pdf

Kira Goldner (Columbia University); Nicole Immorlica (Microsoft Research); Brendan Lucier (Microsoft Research New England)

Abstract: (click to expand) We study auctions for carbon licenses, a policy tool used to control the social cost of pollution. Each identical license grants the right to produce a unit of pollution. Each buyer (i.e., firm that pollutes during the manufacturing process) enjoys a decreasing marginal value for licenses, but society suffers an increasing marginal cost for each license distributed. The seller (i.e., the government) can choose a number of licenses to auction, and wishes to maximize societal welfare: the total economic value of the buyers minus the social cost. Motivated by emission license markets deployed in practice, we focus on uniform price auctions with a price floor and/or price ceiling. The seller has distributional information about the market, and their goal is to tune the auction parameters to maximize expected welfare. The target benchmark is the maximum expected welfare achievable by any such auction under truth-telling behavior. Unfortunately, the uniform price auction is not truthful, and strategic behavior can significantly reduce (even below zero) the welfare of a given auction configuration. We use tools from theoretical computer science and algorithmic game theory to address the strategic vulnerabilities of these auctions. We describe a subclass of "safe-price" auctions for which the welfare at any Bayes-Nash equilibrium will approximate the welfare under truth-telling behavior. We then show that the better of a safe-price auction, or a truthful auction that allocates licenses to only a single buyer, will approximate the target benchmark. In particular, we show how to choose a number of licenses and a price floor so that the worst-case welfare, at any equilibrium, is a constant approximation to the best achievable welfare under truth-telling after excluding the welfare contribution of a single buyer. This provides a concrete recommendation for how to set the auction parameters in practice in order to achieve guarantees, even in the face of strategic participants.

(19) Reduction of the Optimal Power Flow Problem through Meta-Optimization pdf

Letif Mones (Invenia Labs); Alex Robson (Invenia Labs); Mahdi Jamei (Invenia Labs); Cozmin Ududec (Invenia Labs)

Abstract: (click to expand) We introduce a method for solving Optimal Power Flow (OPF) using meta-optimization, which can substantially reduce solution times. A pre-trained classifier that predicts the binding constraints of the system is used to generate an initial reduced OPF problem, defined by removing the predicted non-binding constraints. Through an iterative procedure, this initial set of constraints is then ex- tended by those constraints that are violated but not represented in the reduced OPF, guaranteeing an optimal solution of the original OPF problem with the full set of constraints. The classifier is trained using a meta-loss objective, defined by the computational cost of the series of reduced OPF problems.

(20) Human-Machine Collaboration for Fast Land Cover Mapping

Caleb Robinson (Georgia Institute of Technology); Anthony Ortiz (University of Texas at El Paso); Nikolay Malkin (Yale University); Blake Elias (Microsoft); Andi Peng (Microsoft); Dan Morris (Microsoft); Bistra Dilkina (University of Southern California); Nebojsa Jojic (Microsoft Research)

Abstract: (click to expand) We propose incorporating human labelers in a model fine-tuning system that provides immediate user feedback. In our framework, human labelers can interactively query model predictions on unlabeled data, choose which data to label, and see the resulting effect on the model's predictions. This bi-directional feedback loop allows humans to learn how the model responds to new data. Our hypothesis is that this rich feedback allows human labelers to create mental models that enable them to better choose which biases to introduce to the model. We implement this framework for fine-tuning high-resolution land cover segmentation models and evaluate it against traditional active learning based approaches. More specifically, we fine-tune a deep neural network -- trained to segment high-resolution aerial imagery into different land cover classes in Maryland, USA -- to a new spatial area in New York, USA. We find that the tight loop turns the algorithm and the human operator into a hybrid system that can produce land cover maps of large areas more efficiently than the traditional workflows.

(21) A User Study of Perceived Carbon Footprint pdf

Victor Kristof (EPFL); Valentin Quelquejay-Leclere (EPFL); Robin Zbinden (EPFL); Lucas Maystre (Spotify); Matthias Grossglauser (École Polytechnique Fédérale de Lausanne (EPFL)); Patrick Thiran (EPFL)

Abstract: (click to expand) We propose a statistical model to understand people’s perception of their carbon footprint. Driven by the observation that few people think of CO2 impact in absolute terms, we design a system to probe people’s perception from simple pairwise comparisons of the relative carbon footprint of their actions. The formulation of the model enables us to take an active-learning approach to selecting the pairs of actions that are maximally informative about the model parameters. We define a set of 18 actions and collect a dataset of 2183 comparisons from 176 users on a university campus. The early results reveal promising directions to improve climate communication and enhance climate mitigation.

(22) Design, Benchmarking and Graphical Lasso based Explainability Analysis of an Energy Game-Theoretic Framework pdf

Hari Prasanna Das (UC Berkeley ); Ioannis C. Konstantakopoulos (UC Berkeley); Aummul Baneen Manasawala (UC Berkeley); Tanya Veeravalli (UC Berkeley); Huihan Liu (UC Berkeley ); Costas J. Spanos (University of California at Berkeley)

Abstract: (click to expand) Energy use in buildings account for approximately half of global electricity consumption and a significant amount of CO2 emissions. The occupants of a building typically lack the independent motivation necessary to optimize their energy usage. In this paper, we propose a novel energy game-theoretic framework for smart building which incorporates human-in-the-loop modeling by creating an interface to allow interaction with occupants and potentially incentivize energy efficient behavior. We present open-sourced dataset and benchmarked results for forecasting of energy resource usage patterns by leveraging classical machine learning and deep learning methods including deep bi-directional recurrent neural networks. Finally, we use graphical lasso to demonstrate the explainable nature on human decision making towards energy usage inherent in the dataset.

(23) Predicting ice flow using machine learning pdf

Yimeng Min (Mila); Surya Karthik Mukkavilli (Mila); Yoshua Bengio (Mila)

Abstract: (click to expand) Though machine learning has achieved notable success in modeling sequential and spatial data for speech recognition and in computer vision, applications to remote sensing and climate science problems are seldom considered. In this paper, we demonstrate techniques from unsupervised learning of future video frame prediction, to increase the accuracy of ice flow tracking in multi-spectral satellite images. As the volume of cryosphere data increases in coming years, this is an interesting and important opportunity for machine learning to address a global challenge for climate change, risk management from floods, and conserving freshwater resources. Future frame prediction of ice melt and tracking the optical flow of ice dynamics presents modeling difficulties, due to uncertainties in global temperature increase, changing precipitation patterns, occlusion from cloud cover, rapid melting and glacier retreat due to black carbon aerosol deposition, from wildfires or human fossil emissions. We show the adversarial learning method helps improve the accuracy of tracking the optical flow of ice dynamics compared to existing methods in climate science. We present a dataset, IceNet, to encourage machine learning research and to help facilitate further applications in the areas of cryospheric science and climate change.

(24) DeepClimGAN: A High-Resolution Climate Data Generator pdf

Alexandra Puchko (Western Washington University); Brian Hutchinson (Western Washington University); Robert Link (Joint Global Change Research Institute)

Abstract: (click to expand) Earth system models (ESMs), which simulate the physics and chemistry of the global atmosphere, land, and ocean, are often used to generate future projections of climate change scenarios. These models are far too computationally intensive to run repeatedly, but limited sets of runs are insufficient for some important applications, like adequately sampling distribution tails to characterize extreme events. As a compromise, emulators are substantially less expensive but may not have all of the complexity of an ESM. Here we demonstrate the use of a conditional generative adversarial network (GAN) to act as an ESM emulator. In doing so, we gain the ability to produce daily weather data that is consistent with what ESM might output over any chosen scenario. In particular, the GAN is aimed at representing a joint probability distribution over space, time, and climate variables, enabling the study of correlated extreme events, such as floods, droughts, or heatwaves.

(25) Quantifying the Carbon Emissions of Machine Learning

Sasha Luccioni (Mila); Victor Schmidt (Mila); Alexandre Lacoste (Element AI); Thomas Dandres (Polytechnique Montreal)

Abstract: (click to expand) From an environmental standpoint, there are a few crucial aspects of training a neural network that have a major impact on the quantity of carbon that it emits. These factors include: the location of the server used for training and the energy grid that it uses, the length of the training procedure, and even the make and model of hardware on which the training takes place. In order to approximate these emissions, we present our Machine Learning Emissions Calculator, a tool for our community to better understand the environmental impact of training ML models. We accompany this tool with an explanation of the factors cited above, as well as concrete actions that individual practitioners as well as organizations can take to mitigate their carbon emissions.

(26) Measuring Impact of Climate Change on Tree Species: analysis of JSDM on FIA data pdf Honorable Mention

Hyun Choi (University of Florida); Sergio Marconi (University of Florida); Ali Sadeghian (University of Florida); Ethan White (University of Florida); Daisy Zhe Wang (Univeresity of Florida)

Abstract: (click to expand) One of the first beings affected by changes in the climate are trees, one of our most vital resources. In this study tree species interaction and the response to climate in different ecological environments is observed by applying a joint species distribution model to different ecological domains in the United States. Joint species distribution models are useful to learn inter-species relationships and species response to the environment. The climates’ impact on the tree species is measured through species abundance in an area. We compare the model’s performance across all ecological domains and study the sensitivity of the climate variables. With the prediction of abundances, tree species populations can be predicted in the future and measure the impact of climate change on tree populations.

(27) A Global Census of Solar Facilities Using Deep Learning and Remote Sensing Honorable Mention

Lucas Kruitwagen (University of Oxford); Kyle Story (Descartes Labs); Johannes Friedrich (World Resource Institute); Sam Skillman (Descartes Labs); Cameron Hepburn (University of Oxford)

Abstract: (click to expand) We present a comprehensive global census of solar power facilities using deep learning and remote sensing. We search imagery from the Airbus SPOT 6/7 and European Space Agency Sentinel-2 satellites covering more than 48% of earth’s land-surface using a combination of deep-learning models, image processing, and hand-verification. We locate solar facilities and measure their footprints and installation dates. The resulting dataset of 68,797 facilities has an estimated generating capacity of 209 GW; 78% of this capacity was not previously reported in public databases. These asset-level data are critical for understanding energy infrastructure, evaluate climate risk, and efficiently use intermittent solar energy - ultimately enabling the transition to a predominantly renewable energy system.

(28) Machine Learning for Precipitation Nowcasting from Radar Images pdf

Shreya Agrawal (Google); Luke Barrington (Google); Carla Bromberg (Google); John Burge (Google); Cenk Gazen (Google); Jason Hickey (Google)

Abstract: (click to expand) High-resolution nowcasting is an essential tool needed for effective adaptation to climate change, particularly for extreme weather. As Deep Learning (DL) techniques have shown dramatic promise in many domains, including the geosciences, we present an application of DL to the problem of precipitation nowcasting, i.e., high-resolution (1 km x 1 km) short-term (1 hour) predictions of precipitation. We treat forecasting as an image-to-image translation problem and leverage the power of the ubiquitous UNET convolutional neural network. We find this performs favorably when compared to three commonly used models: optical flow, persistence and NOAA's numerical one-hour HRRR nowcasting prediction.

(29) Enhancing Stratospheric Weather Analyses and Forecasts by Deploying Sensors from a Weather Balloon pdf

Kiwan Maeng (Carnegie Mellon University); Iskender Kushan (Microsoft); Brandon Lucia (Carnegie Mellon University); Ashish Kapoor (Microsoft)

Abstract: (click to expand) The ability to analyze and forecast stratospheric weather conditions is fundamental to addressing climate change. However, our capacity to collect data in the stratosphere is limited by sparsely deployed weather balloons. We propose a framework to collect stratospheric data by releasing a contrail of tiny sensor devices as a weather balloon ascends. The key machine learning challenges are determining when and how to deploy a finite collection of sensors to produce a useful data set. We decide when to release sensors by modeling the deviation of a forecast from actual stratospheric conditions as a Gaussian process. We then implement a novel hardware system that is capable of optimally releasing sensors from a rising weather balloon. We show that this data engineering framework is effective through real weather balloon flights, as well as simulations.

(30) Automatic data cleaning via tensor factorization for large urban environmental sensor networks pdf

Yue Hu (Vanderbilt University); Yanbing Wang (Vanderbilt University); Canwen Jiao (Vanderbilt University); Rajesh Sankaran (Argonne National Lab); Charles Catlett (Argonne National Lab); Daniel Work (Vanderbilt University)

Abstract: (click to expand) The US Environmental Protection Agency identifies that urban heat islands can negatively impact a community’s environment and quality of life. Using low cost urban sensing networks, it is possible to measure the impacts of mitigation strategies at a fine-grained scale, informing context-aware policies and infrastructure design. However, fine-grained city-scale data analysis is complicated by tedious data cleaning including removing outliers and imputing missing data. To address the challenge of data cleaning, this article introduces a robust low-rank tensor factorization method to automatically correct anomalies and impute missing entries for high-dimensional urban environmental datasets. We validate the method on a synthetically degraded National Oceanic and Atmospheric Administration temperature dataset, with a recovery error of 4%, and apply it to the Array of Things city-scale sensor network in Chicago, IL.

(31) Machine Learning-based Estimation of Forest Carbon Stocks to increase Transparency of Forest Preservation Efforts pdf

Björn Lütjens (MIT); Lucas Liebenwein (Massachusetts Institute of Technology); Katharina Kramer (Massachusetts Institute of Technology)

Abstract: (click to expand) An increasing amount of companies and cities plan to become CO2-neutral, which requires them to invest in renewable energies and carbon emission offsetting solutions. One of the cheapest carbon offsetting solutions is preventing deforestation in developing nations, a major contributor in global greenhouse gas emissions. However, forest preservation projects historically display an issue of trust and transparency, which drives companies to invest in transparent, but expensive air carbon capture facilities. Preservation projects could conduct accurate forest inventories (tree diameter, species, height etc.) to transparently estimate the biomass and amount of stored carbon. However, current rainforest inventories are too inaccurate, because they are often based on a few expensive ground-based samples and/or low-resolution satellite imagery. LiDAR-based solutions, used in US forests, are accurate, but cost-prohibitive, and hardly-accessible in the Amazon rainforest. We propose accurate and cheap forest inventory analyses through Deep Learning-based processing of drone imagery. The more transparent estimation of stored carbon will create higher transparency towards clients and thereby increase trust and investment into forest preservation projects.

(32) Identify Solar Panels in Low Resolution Satellite Imagery with Siamese Architecture and Cross-Correlation pdf

Zhengcheng Wang (Tsinghua University); Zhecheng Wang (Stanford University); Arun Majumdar (Stanford University); Ram Rajagopal (Stanford University)

Abstract: (click to expand) Understanding solar adoption trends and their underlying dynamics requires a comprehensive and granular time-series solar installation database which is unavailable today and expensive to create manually. To this end, we leverage a deep siamese network that automatically identifies solar panels in historical low-resolution (LR) satellite images by comparing the target image with its high-resolution exemplar at the same location. To resolve the potential displacement between solar panels in the exemplar image and that in the target image, we use a cross-correlation module to collate the spatial features learned from each input and measure their similarity. Experimental result shows that our model significantly outperforms baseline methods on a dataset of historical LR images collected in California.

(33) DeepRI: End-to-end Prediction of Tropical Cyclone Rapid Intensification from Climate Data pdf

Renzhi Jing (Princeton University); Ning Lin (Princeton University); Yinda Zhang (Google LLC)

Abstract: (click to expand) Predicting rapid intensification (RI) is extremely critical in tropical cyclone forecasting. Existing deep learning models achieve promising results, however still rely on hand-craft feature. We propose to design an end-to-end deep learning architecture that directly predict RI from raw climate data without intermediate heuristic feature, which allows joint optimization of the whole system for higher performance.

(34) Autonomous Sensing and Scientific Machine Learning for Monitoring Greenhouse Gas Emissions pdf

Genevieve Flaspohler (MIT); Victoria Preston (MIT); Nicholas Roy (MIT); John Fisher (MIT); Adam Soule (Woods Hole Oceanographic Institution); Anna Michel (Woods Hole Oceanographic Institution)

Abstract: (click to expand) Greenhouse gas emissions are a key driver of climate change. In order to develop and tune climate models, measurements of natural and anthropogenic phenomenon are necessary. Traditional methods (i.e., physical sample collection and ex situ analysis) tend to be sample sparse and low resolution, whereas global remote sensing methods tend to miss small- and mid-scale dynamic phenomenon. In situ instrumentation carried by a robotic platform is suited to study greenhouse gas emissions at unprecedented spatial and temporal resolution. However, collecting scientifically rich datasets of dynamic or transient emission events requires accurate and flexible models of gas emission dynamics. Motivated by applications in seasonal Arctic thawing and volcanic outgassing, we propose the use of scientific machine learning, in which traditional scientific models (in the form of ODEs/PDEs) are combined with machine learning techniques (generally neural networks) to better incorporate data into a structured, interpretable model. Our technical contributions will primarily involve developing these hybrid models and leveraging model uncertainty estimates during sensor planning to collect data that efficiently improves gas emission models in small-data domains.

(35) VideoGasNet: Deep Learning for Natural Gas Methane Leak Classification Using An Infrared Camera

Jingfan Wang (Stanford University)

Abstract: (click to expand) Mitigating methane leakage from the natural gas system have become an increasing concern for climate change. Efficacious methane leak detection and classification can make the mitigation process more efficient and cost effective. Optical gas imaging is widely used for the purpose of leak detection, but it cannot directly provide detection results and leak sizes. Few studies have examined the possibility of leak classification using videos taken by the infrared camera (IR), an optical gas imaging device. In this study, we consider the leak classification problem as a video classification problem and investigated the application of deep learning techniques in methane leak detection. Firstly we collected the first methane leak video dataset - GasVid, which has ~1 M frames of labeled videos of methane leaks from different leaking equipment, covering a wide range of leak sizes (5.3-2051.6 g\ce{CH4}/h) and imaging distances (4.6-15.6 m). Secondly, we studied three deep learning algorithms, including 2D Convolutional Neural Networks (CNN) model, 3D CNN and the Convolutional Long Short Term Memory (ConvLSTM). We find that 3D CNN is the most outstanding and robust architecture, which was named VideoGasNet. The leak-non-leak detection accuracy can reach 100%, and the highest small-medium-large classification accuracy is 78.2% with our 3D CNN network. In summary, VideoGasNet greatly extends the capabilities of IR camera-based leak monitoring system from leak detection only to automated leak classification with high accuracy and fast processing speed, significant mitigation efficiency.

(36) Detecting Avalanche Deposits using Variational Autoencoder on Sentinel-1 Satellite Imagery pdf

Saumya Sinha (University of Colorado, Boulder); Sophie Giffard-Roisin (University of Colorado Boulder); Fatima Karbou (Meteo France); Michael Deschatres (Irstea); Nicolas Eckert (Irstea); Anna Karas (Meteo France); Cécile Coléou (Meteo France); Claire Monteleoni (University of Colorado Boulder)

Abstract: (click to expand) Avalanche monitoring is a crucial safety challenge, especially in a changing climate. Remote sensing of avalanche deposits can be very useful to identify avalanche risk zones and time periods, which can in turn provide insights about the effects of climate change. In this work, we use Sentinel-1 SAR (synthetic aperture radar) data on the French Alps for the exceptional winter of 2017-18, with the goal of automatically detecting avalanche deposits. We address our problem with an unsupervised learning technique. We treat an avalanche as a rare event, or an anomaly, and we learn a variational autoencoder, in order to isolate the anomaly. We then evaluate our method on labeled test data, using an independent in-situ avalanche inventory as ground truth. Our empirical results show that our unsupervised method obtains comparable performance to a recent supervised learning approach that trained a convolutional neural network on an artificially balanced version of the same SAR data set along with the corresponding ground-truth labels. Our unsupervised approach outperforms the standard CNN in terms of balanced accuracy (63% as compared to 55%). This is a significant improvement, as it allows our method to be used in-situ by climate scientists, where the data is always very unbalanced (< 2% positives). This is the first application of unsupervised deep learning to detect avalanche deposits.

(37) Fine-Grained Distribution Grid Mapping Using Street View Imagery pdf

Qinghu Tang (Tsinghua University); Zhecheng Wang (Stanford University); Arun Majumdar (Stanford University); Ram Rajagopal (Stanford University)

Abstract: (click to expand) Fine-grained distribution grid mapping is essential for power system operation and planning in the aspects of renewable energy integration, vegetation management, and risk assessment. However, currently such information can be inaccurate, outdated, or incomplete. Existing grid topology reconstruction methods heavily rely on various assumptions and measurement data that is not widely available. To bridge this gap, we propose a machine-learning-based method that automatically detects, localizes, and estimates the interconnection of distribution power lines and utility poles using readily-available street views in the upward perspective. We demonstrate the superior image-level and region-level accuracy of our method on a real-world distribution grid test case.

(38) Bayesian optimization with theory-based constraints accelerates search for stable photovoltaic perovskite materials

Armi Tiihonen (Massachusetts Institute of Technology)

Abstract: (click to expand) Bringing a new photovoltaic technology from materials research stage to the market has historically taken decades, and the process has to be accelerated for increasing the share of renewables in energy production. We demonstrate Bayesian optimization for accelerating stability research. Convergence is reached even faster when using a constraint for integrating physical knowledge into the model. In our test case, we optimize the stability of perovskite compositions for perovskite solar cells, an efficient new solar cell technology suffering from limited lifetime of devices.

(39) Increasing performance of electric vehicles in ride-hailing services using deep reinforcement learning pdf

Jon Donadee (LLNL); Jacob Pettit (LLNL); Ruben Glatt (LLNL); Brenden Petersen (Lawrence Livermore National Laboratory)

Abstract: (click to expand) New forms of on-demand transportation such as ride-hailing and connected autonomous vehicles are proliferating, yet are a challenging use case for electric vehicles (EV). This paper explores the feasibility of using deep reinforcement learning (DRL) to optimize a driving and charging policy for a ride-hailing EV agent, with the goal of reducing costs and emissions while increasing transportation service provided. We introduce a data-driven simulation of a ride-hailing EV agent that provides transportation service and charges energy at congested charging infrastructure. We then formulate a test case for the sequential driving and charging decision making problem of the agent and apply DRL to optimize the agent's decision making policy. We evaluate the performance against heuristic policies and show that our agent learns to act competitively without any prior knowledge.

(40) Optimizing trees for carbon sequestration

Jeremy Freeman

Abstract: (click to expand) In the IPCC models of climate change mitigation, most scenarios ensuring less than 2ºC of warming assume deployment of some form of “negative emissions technology,” alongside dramatic reductions in emissions and other major societal changes. Proposed negative emissions technologies include bioenergy with carbon capture and storage, enhanced weathering of minerals, direct air capture, and afforestation / reforestation. Among these technologies, the use of trees for carbon sequestration through photosynthesis is well established, requires little energy, has comparable sequestration potential, and can be deployed at scale for relatively low cost. The primary constraint on using trees for sequestration is land, which is limited and increasingly subject to competitive demand. Thus, maximizing the capacity and long-term stability of every hectare used for planting would bolster the critical role of trees in a broad negative emissions strategy. Here, we propose to build a new data resource and optimization tool that leverages modern measurements and machine learning to help address this need.

(41) Toward Resilient Cities: Using Deep Learning to Downscale Climate Model Projections

Muge Komurcu (MIT); Zikri Bayraktar (IEEE)

Abstract: (click to expand) Climate projections from Earth System Models (ESM) are widely used to assess climate change impacts. These projections, however, are too coarse in spatial and temporal resolution (e.g. 25-50 kms, monthly) to be used in local scale resilience studies. High-resolution (<4 km) climate projections at dense temporal resolution (hourly) from multiple Earth System models under various scenarios are necessary to assess potential future changes in climate variables and perform meaningful and robust climate resilience studies. Running ESMs in high-resolution is computationally too expensive, therefore downscaling methods are applied to ESM projections to produce high-resolution projections. Using a regional climate model to downscale climate projections is preferred but dynamically downscaling several ESM projections to < 4km resolution under different scenarios is currently not feasible. In this study, we propose to use a 60 year dynamically downscaled climate dataset with hourly output for the Northeastern United States to train Deep Learning models and achieve a computationally efficient method of downscaling climate projections. This method will allow for more ESM projections to be downscaled to local scales under more scenarios in an efficient manner and significantly improve robustness of regional resilience studies.

(42) Stripping off the implementation complexity of physics-based model predictive control for buildings via deep learning pdf

Jan Drgona (Pacific Northwest National Laboratory); Lieve Helsen (KU Leuven); Draguna Vrabie (PNNL)

Abstract: (click to expand) Over the past decade, model predictive control (MPC) has been considered as the most promising solution for intelligent building operation. Despite extensive effort, transfer of this technology into practice is hampered by the need to obtain an accurate controller model with minimum effort, the need of expert knowledge to set it up, and the need of increased computational power and dedicated software to run it. A promising direction that tackles the last two problems was proposed by approximate explicit MPC where the optimal control policies are learned from MPC data via a suitable function approximator, e.g., a deep learning (DL) model. The main advantage of the proposed approach stems from simple evaluation at execution time leading to low computational footprints and easy deployment on embedded HW platforms. We present the energy savings potential of physics-based (also called 'white-box') MPC applied to an office building in Belgium. Moreover, we demonstrate how deep learning approximators can be used to cut the implementation and maintenance costs of MPC deployment without compromising performance. We also critically assess the presented approach by pointing out the major challenges and remaining open-research questions.

(43) Towards self-adaptive building energy control in smart grids pdf

Juan Gómez-Romero (Universidad de Granada); Miguel Molina-Solana (Imperial College London)

Abstract: (click to expand) Energy consumption in buildings greatly contributes to worldwide CO2 emissions and thus any improvement in HVAC operation will greatly help tackling global climate change. We are putting forward a proposal for self-adaptive energy control in smart grids based on Deep Learning, Deep Reinforcement Learning and Multi-Agent technologies. Particularly, we introduce the concept of Deep Neural Simulation Model (DNSM) as a way of generating digital twins of buildings in which the agent can test and learn optimal operations by itself and by collaborating with other agents. Not only do we expect a reduction on energy consumption and an increment on the use of renewable sources, but also a reduction on the cost of controlling energy in buildings.

(44) Machine learning identifies the most valuable synthesis conditions for next-generation photovoltaics Best Paper Award

Felipe Oviedo (MIT) and Zekun Ren (MIT)

Abstract: (click to expand) Terawatts of next-generation photovoltaics (PV) are necessary to mitigate climate change. The traditional R&D paradigm leads to high efficiency / high variability solar cells, limiting industrial scaling of novel PV materials. In this work, we propose a machine learning approach for early-stage optimization of solar cells, by combining a physics-informed deep autoencoder and a manufacturing-relevant Bayesian optimization objective. This framework allows to: 1) Co-optimize solar cell performance and variability under techno-economic revenue constrains, and 2) Infer the effect of process conditions over key latent physical properties. We test our approach by synthesizing 135 perovskite solar cells, and finding the optimal points under various techno-economic assumptions.

(45) Helping Reduce Environmental Impact of Aviation with Machine Learning pdf Best Paper Award

Ashish Kapoor (Microsoft)

Abstract: (click to expand) Commercial aviation is one of the biggest contributors towards climate change. We propose to reduce environmental impact of aviation by considering solutions that would reduce the flight time. Specifically, we first consider improving winds aloft forecast so that flight planners could use better information to find routes that are efficient. Secondly, we propose an aircraft routing method that seeks to find the fastest route to the destination by considering uncertainty in the wind forecasts and then optimally trading-off between exploration and exploitation. Both these ideas were previously published in [5] and [8] and contain further technical details.

(46) Machine Learning for Generalizable Prediction of Flood Susceptibility pdf

Dylan Fitzpatrick (Carnegie Mellon University); Chelsea Sidrane (Stanford University); Andrew Annex (Johns Hopkins University); Diane O’Donoghue (kx); Piotr Bilinski (University of Warsaw)

Abstract: (click to expand) Flooding is a destructive and dangerous hazard and climate change appears to be increasing the frequency of catastrophic flooding events around the world. Physics-based flood models are costly to calibrate and are rarely generalizable across different river basins, as model outputs are sensitive to site-specific parameters and human-regulated infrastructure. Statistical models trained primarily from remotely-sensed Earth observation data could reduce the need for extensive in-situ measurements. In this work, we develop generalizable, multi-basin models of river flooding susceptibility using geographically-distributed data from the USGS stream gauge network. Machine learning models are trained in a supervised framework to predict two measures of flood susceptibility from a mix of river basin attributes, impervious surface cover information derived from satellite imagery, and historical records of rainfall and stream height. We report prediction performance of multiple models using precision-recall curves, and compare with performance of naive baselines. This work on multi-basin flood prediction represents a step in the direction of making flood prediction accessible to all at-risk communities.

(47) A Deep Learning-based Framework for the Detection of Schools of Herring in Echograms pdf

Alireza Rezvanifar (University of Victoria); Tunai Porto Marques (University of Victoria ); Melissa Cote (University of Victoria); Alexandra Branzan Albu (University of Victoria); Alex Slonimer (ASL Environmental Sciences); Thomas Tolhurst (ASL Environmental Sciences ); Kaan Ersahin (ASL Environmental Sciences ); Todd Mudge (ASL Environmental Sciences ); Stephane Gauthier (Fisheries and Oceans Canada)

Abstract: (click to expand) Tracking the abundance of underwater species is crucial for understanding the effects of climate change on marine ecosystems. Biologists typically monitor underwater sites with echosounders and visualize data as 2D images (echograms); they interpret these data manually or semi-automatically, which is time-consuming and prone to inconsistencies. This paper proposes a deep learning framework for the automatic detection of schools of herring from echograms. Experiments demonstrated that our approach outperforms a traditional machine learning algorithm that uses hand-crafted features. Our framework could easily be expanded to detect more species of interest to sustainable fisheries.

(48) Emulating Numeric Hydroclimate Models with Physics-Informed cGANs pdf Honorable Mention

Ashray Manepalli (terrafuse); Adrian Albert (terrafuse, inc.); Alan Rhoades (Lawrence Berkeley National Lab); Daniel Feldman (Lawrence Berkeley National Lab)

Abstract: (click to expand) Process-based numerical simulations, including those for climate modeling applications, are compute and resource intensive, requiring extensive customization and hand-engineering for encoding governing equations and other domain knowledge. On the other hand, modern deep learning employs a significantly simpler and more efficient computational workflow, and has been shown impressive results across a myriad of applications in the computational sciences. In this work, we investigate the potential of deep generative learning models, specifically conditional Generative Adversarial Networks (cGANs), to simulate the output of a physics-based model of the spatial distribution of the water content of mountain snowpack - the snow water equivalent (SWE). We show preliminary results indicating that the cGAN model is able to learn diverse mappings between meteorological forcings and SWE output. Thus physics based cGANs provide a means for fast and accurate SWE modeling that can have significant impact in a variety of applications (e.g., hydropower forecasting, agriculture, and water supply management). In climate science, the Snowpack and SWE are seen as some of the best indicative variables for investigating climate change and its impact. The massive speedups, diverse sampling, and sensitivity/saliency modelling that cGANs can bring to SWE estimation will be extremely important to investigating variables linked to climate change as well as predicting and forecasting the potential effects of climate change to come.

(49) Predicting Arctic Methane Seeps via Satellite Imagery pdf

Olya (Olga) Irzak (Frost Methane Labs); Amber Leigh Thomas (Stanford); Stephanie Schneider (Stanford); Catalin Voss (Stanford University)

Abstract: (click to expand) The arctic has seen significant warming and releases of methane, a potent greenhouse gas have been reported. We aim to apply computer vision to satellite imagery in order to quantify geological methane emissions from the permafrost as well as track and predict their change due to increasing temperatures.

(50) GeoLabels: Towards Efficient Ecosystem Monitoring using Data Programming on Geospatial Information pdf

David Dao (ETH); Johannes Rausch (ETH Zurich); Ce Zhang (ETH)

Abstract: (click to expand) Monitoring, Reporting and Verification (MRV) systems for land use play a key role in the decision-making of climate investors, policymakers and conservationists. Remote sensing is commonly used for MRV but practical solutions are constrained by a lack of labels to train machine learning-based downstream tasks. GeoLabels is an automated MRV system that can rapidly adapt to novel applications by leveraging existing geospatial information and domain expertise to quickly create training sets through data programming. Moreover, GeoLabels uses dimensionality reduction interfaces, allowing non-technical users to create visual labeling functions.

(51) A deep learning approach for classifying black carbon aerosol morphology

Kara Lamb (Cooperative Institute for Research in the Environmental Sciences)

Abstract: (click to expand) Black carbon (BC) is a sub-micron aerosol sourced from incomplete combustion which strongly absorbs solar radiation, leading to both direct and indirect climate impacts. The state-of-the-art technique for characterizing BC is the single particle soot photometer (SP2) instrument, which detects these aerosols in real time via laser-induced incandescence (L-II). This measurement technique allows for quantification of BC mass on a single particle basis, but time-resolved signals may also provide constraints on BC morphology, which impacts both its optical properties and atmospheric lifetime. No methods currently exist to use this information. I propose applying a deep learning based approach to classify the fractal dimension of single BC particles from time-resolved L-II signals. This method would provide the first on-line measurement technique for quantifying BC morphology. These observations could be used to improve representations of BC optical properties and atmospheric processing in climate models.

(52) Forecasting El Niño with Convolutional and Recurrent Neural Networks pdf

Ankur Mahesh (ClimateAi); Maximilian Evans (ClimateAi); Garima Jain (ClimateAi); Mattias Castillo (ClimateAi); Aranildo Lima (ClimateAi); Brent Lunghino (ClimateAi); Himanshu Gupta (ClimateAi); Carlos Gaitan (ClimateAi); Jarrett Hunt (ClimateAi); Omeed Tavasoli (ClimateAi); Patrick Brown (ClimateAi, San Jose State University); V. Balaji (Geophysical Fluid Dynamics Laboratory)

Abstract: (click to expand) The El Niño Southern Oscillation (ENSO) is the dominant mode of variability in the climate system on seasonal to decadal timescales. With foreknowledge of the state of ENSO, stakeholders can anticipate and mitigate impacts in climate-sensitive sectors such as agriculture and energy. Traditionally, ENSO forecasts have been produced using either computationally intensive physics-based dynamical models or statistical models that make limiting assumptions, such as linearity between predictors and predictands. Here we present a deep-learning-based methodology for forecasting monthly ENSO temperatures at various lead times. While traditional statistical methods both train and validate on observational data, our method trains exclusively on physical simulations. With the entire observational record as an out-of-sample validation set, the method’s skill is comparable to that of operational dynamical models. The method is also used to identify disagreements among climate models about the predictability of ENSO in a world with climate change.

Program Committee

Andrew Ross (Harvard)
Aneesh Rangnekar (RIT)
Ashesh Chattopadhyay (Rice)
Ashley Pilipiszyn (Stanford)
Bolong Cheng (SigOpt)
Christian Schroeder (Oxford)
Clement Duhart (MIT)
Dali Wang (Oak Ridge National Lab)
David Dao (ETH)
Di Wu (McGill)
Dimitrios Giannakis (Courant Institute, NYU)
Duncan Watson-Parris (Oxford)
Evan Sherwin (Stanford)
Femke van Geffen (FU Berlin)
Gege Wen (Stanford)
George Chen (CMU)
Greg Schivley (Carbon Impact Consulting)
Han Zou (UC Berkeley)
Hari Prasanna Das (UC Berkeley)
Hillary Scannell (University of Washington)
Joanna Slawinska (University of Wisconsin-Milwaukee)
Johan Mathe (Frog Labs)
Jonathan Binas (Mila, Montreal)
Jussi Gillberg (Aalto University)
Kalai Ramea (PARC)
Karthik Kashinath (Lawrence Berkeley National Lab)
Kate Duffy (Northeastern)
Kelly Kochanski (CU Boulder)
Kevin McCloskey (Google)
Kris Sankaran (Mila)
Lea Boche (EPRI)
Loubna Benabbou (Mohammadia School of Engineering, Mohammed V University)
Mahdi Jamei (Invenia Labs)
Max Callaghan (MCC Berlin)
Mayur Mudigonda (UC Berkeley)
Melrose Roderick (CMU)
Mohammad Mahdi Kamani (Penn State)
Natasha Jaques (MIT)
Neel Guha (CMU)
Niccolo Dalmasso (CMU)
Nikola Milojevic-Dupont (MCC Berlin)
Pedram Hassanzadeh (Rice)
Robin Dunn (CMU)
Sajad Haghanifar (University of Pittsburgh)
Sanam Mirzazad (EPRI)
Sandeep Manjanna (McGill)
Sasha Luccioni (Mila)
Sharon Zhou (Stanford)
Shubhankar Deshpande (CMU)
Sookyung Kim (Lawrence Livermore National Lab)
Soukayna Mouatadid (University of Toronto)
Surya Karthik Mukkavilli (Mila)
Telmo Felgueira (IST)
Thomas Hornigold (Oxford)
Tianle Yuan (NASA)
Tom Beucler (Columbia & UCI)
Vikram Voleti (Mila, Montreal)
Volodymyr Kuleshov (Stanford)
Yang Song (Oak Ridge National Lab)
Ydo Wexler (Amperon)
Zhecheng Wang (Stanford)
Zhuangfang Yi (Development Seed)

About NeurIPS

NeurIPS (formerly written “NIPS”) is one of the premier conferences on machine learning, and includes a wide audience of researchers and practitioners in academia, industry, and related fields. It is possible to attend the workshop without either presenting at or attending the main NeurIPS conference.

Call for Submissions

We invite submissions of short papers using machine learning to address problems in climate mitigation, adaptation, or modeling, including but not limited to the following topics:

All machine learning techniques are welcome, from kernel methods to deep learning. Each submission should make clear why the application has (or could have) positive impacts regarding climate change. We highly encourage submissions which make their data publicly available. Accepted submissions will be invited to give poster presentations, of which some will be selected for spotlight talks.

The workshop does not record proceedings, and submissions are non-archival. Submission to this workshop does not preclude future publication. Previously published work may be submitted under certain circumstances (see the FAQ).

All submissions must be through the submission website. Submissions will be reviewed double-blind; do your best to anonymize your submission, and do not include identifying information for authors in the PDF. We encourage, but do not require, use of the NeurIPS style template (please do not use the “Accepted” format as it will deanonymize your submission).

We will be awarding $30K in cloud computing credits, sponsored by Microsoft AI for Earth, as prizes for top submissions. Winners will be announced at the workshop.

Please see the Tips for Submissions and FAQ, and contact with questions.

Submission tracks

There are two tracks for submissions. Submissions are limited to 3 pages for the Papers track, and 2 pages for the Proposals track, in PDF format (see examples here). References do not count towards this total. Supplementary appendices are allowed but will be read at the discretion of the reviewers. All submissions must explain why the proposed work has (or could have) positive impacts regarding climate change.

PAPERS track

Work that is in progress, published, and/or deployed

Submissions for the Papers track should describe projects relevant to climate change that involve machine learning. These may include (but are not limited to) academic research; deployed results from startups, industry, public institutions, etc.; and climate-relevant datasets.

Submissions should provide experimental or theoretical validation of the method presented, as well as specifying what gap the method fills. Algorithms need not be novel from a machine learning perspective if they are applied in a novel setting. Details of methodology need not be revealed if they are proprietary, though transparency is highly encouraged.

Submissions creating novel datasets are welcomed. Datasets should be designed to permit machine learning research (e.g. formatted with clear benchmarks for evaluation). In this case, baseline experimental results on the dataset are preferred, but not required.


Detailed descriptions of ideas for future work

Submissions for the Proposals track should describe detailed ideas for how machine learning can be used to solve climate-relevant problems. While less constrained than the Papers track, Proposals will be subject to a very high standard of review. No results need to be demonstrated, but ideas should be justified as extensively as possible, including motivation for why the problem being solved is important in tackling climate change, discussion of why current methods are inadequate, and explanation of the proposed method.

Tips for submissions

Travel Grants

We are excited to announce limited travel grants, sponsored by Microsoft Research. Travel grant applications can be submitted at, and are due October 3.

We also encourage workshop participants to apply for NeurIPS 2019 travel grants and other grants (e.g. Google Conference and Travel Scholarships) for which they may be eligible. If you are aware of additional scholarships that may be relevant to workshop attendees, please contact the workshop organizers so we can make this information available.

Frequently Asked Questions

Q: How can I keep up to date on this kind of stuff?
A: Sign up for our mailing list!

Q: I’m not in machine learning. Can I still submit?
A: Yes, absolutely! We welcome submissions from many fields. Do bear in mind, however, that the majority of attendees of the workshop will have a machine learning background; therefore, other fields should be introduced sufficiently to provide context for the work.

Q: What if my submission is accepted but I can’t attend the workshop?
A: You may ask someone else to present your work in your stead.

Q: Do I need to use LaTeX or the NeurIPS style files?
A: No, although we encourage it.

Q: It’s hard for me to fit my submission on 2 or 3 pages. What should I do?
A: Feel free to include appendices with additional material (these should be part of the same PDF file as the main submission). Do not, however, put essential material in an appendix, as it will be read at the discretion of the reviewers.

Q: What do I do if I need an earlier decision for visa reasons?
A: Contact us at and explain your situation and the date by which you require a decision and we will do our best to be accommodating.

Q: Can I send submissions directly by email?
A: No, please use the CMT website to make submissions.

Q: The submission website is asking for my name. Is this a problem for anonymization?
A: You should fill out your name and other info when asked on the submission website; CMT will keep your submission anonymous to reviewers.

Q: Do submissions for the Proposals track need to have experimental validation?
A: No, although some initial experiments or citation of published results would strengthen your submission.

Q: The submission website never sent me a confirmation email. Is this a problem?
A: No, the CMT system does not send automatic confirmation emails after a submission, though the submission should show up on the CMT page once submitted. If in any doubt regarding the submission process, please contact the organizers. Also please avoid making multiple submissions of the same article to CMT.

Q: Can I submit previously published work to this workshop?
A: If it was previously published in a non-ML venue, YES! If it was previously published in an ML venue, NO! If you are unsure, contact This policy is as per the NeurIPS workshop guidelines: “Workshops are not a venue for work that has been previously published in other conferences on machine learning or related fields. Work that is presented at the main NeurIPS conference should not appear in a workshop, including as part of an invited talk… (Presenting work that has been published in other fields is, however, encouraged!)”

Q: Can I submit work to this workshop if I am also submitting to another NeurIPS 2019 workshop?
A: Yes. We cannot, however, guarantee that you will not be expected to present the material at a time that conflicts with the other workshop.