Testing the Reliability of Interpretable Neural Networks in Geoscience Using the Madden-Julian Oscillation

We test the reliability of two neural network interpretation techniques, backward optimization and layerwise relevance propagation, within geoscientific applications by applying them to a commonly studied geophysical phenomenon, the Madden-Julian Oscillation. The Madden-Julian Oscillation is a multi-scale pattern within the tropical atmosphere that has been extensively studied over the past decades, which makes it an ideal test case to ensure the interpretability methods can recover the current state of knowledge regarding its spatial structure. The neural networks can, indeed, reproduce the current state of knowledge and can also provide new insights into the seasonality of the Madden-Julian Oscillation and its relationships with atmospheric state variables. The neural network identifies the phase of the Madden-Julian Oscillation twice as accurately as linear regression, which means that nonlinearities used by the neural network are important to the structure of the Madden-Julian Oscillation. Interpretations of the neural network show that it accurately captures the spatial structures of the Madden-Julian Oscillation, suggest that the nonlinearities of the Madden-Julian Oscillation are manifested through the uniqueness of each event, and offer physically meaningful insights into its relationship with atmospheric state variables. We also use the interpretations to identify the seasonality of the Madden-Julian Oscillation, and find that the conventionally defined extended seasons should be shifted later by one month. More generally, this study suggests that neural networks can be reliably interpreted for geoscientific applications and may there

[1]  Amy McGovern,et al.  Making the Black Box More Transparent: Understanding the Physical Implications of Machine Learning , 2019, Bulletin of the American Meteorological Society.

[2]  John M. Wallace,et al.  Three-Dimensional Structure and Evolution of the Moisture Field in the MJO , 2015 .

[3]  J. Hurrell,et al.  Viewing Forced Climate Patterns Through an AI Lens , 2019, Geophysical Research Letters.

[4]  Bradford S. Barrett,et al.  QBO Influence on MJO Amplitude over the Maritime Continent: Physical Mechanisms and Seasonality , 2019, Monthly Weather Review.

[5]  Jonathan A. Weyn,et al.  Can Machines Learn to Predict Weather? Using Deep Learning to Predict Gridded 500‐hPa Geopotential Height From Historical Weather Data , 2019, Journal of Advances in Modeling Earth Systems.

[6]  Hod Lipson,et al.  Understanding Neural Networks Through Deep Visualization , 2015, ArXiv.

[7]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[8]  Eric D. Maloney,et al.  Moisture Modes and the Eastward Propagation of the MJO , 2013 .

[9]  Chidong Zhang,et al.  QBO‐MJO Connection , 2018 .

[10]  Paul E. Roundy,et al.  Applications of a Multiple Linear Regression Model to the Analysis of Relationships between Eastward- and Westward-Moving Intraseasonal Modes , 2004 .

[11]  Wojciech Samek,et al.  Methods for interpreting and understanding deep neural networks , 2017, Digit. Signal Process..

[12]  Paul E. Roundy,et al.  Modulation of the global atmospheric circulation by combined activity in the Madden-Julian oscillation and the El Niño-Southern Oscillation during boreal winter. , 2010 .

[13]  A. Ingersoll,et al.  Triggered Convection, Gravity Waves, and the MJO: A Shallow-Water Model , 2012, 1210.5533.

[14]  Harry H. Hendon,et al.  Stratospheric control of the Madden-Julian oscillation. , 2017 .

[15]  Alexander Binder,et al.  Explaining nonlinear classification decisions with deep Taylor decomposition , 2015, Pattern Recognit..

[16]  Geoffrey E. Hinton,et al.  Deep Learning , 2015, Nature.

[17]  Da Yang,et al.  Testing the Hypothesis that the MJO is a Mixed Rossby–Gravity Wave Packet , 2011 .

[18]  Frederic Vitart,et al.  The Impact of the Stratosphere on the MJO in a Forecast Model , 2020, Journal of Geophysical Research: Atmospheres.

[19]  Klaus M. Weickmann,et al.  A Comparison of OLR and Circulation-Based Indices for Tracking the MJO , 2014 .

[20]  Harry H. Hendon,et al.  Influence of the QBO on the MJO During Coupled Model Multiweek Forecasts , 2019, Geophysical Research Letters.

[21]  Eric Maloney,et al.  The Consistency of MJO Teleconnection Patterns: An Explanation Using Linear Rossby Wave Theory , 2018, Journal of Climate.

[22]  Imme Ebert-Uphoff,et al.  Evaluation, Tuning and Interpretation of Neural Networks for Meteorological Applications , 2020, Bulletin of the American Meteorological Society.

[23]  Noah D. Brenowitz,et al.  Prognostic Validation of a Neural Network Unified Physics Parameterization , 2018, Geophysical Research Letters.

[24]  Harry H. Hendon,et al.  The Life Cycle of the Madden–Julian Oscillation , 1994 .

[25]  Tianjun Zhou,et al.  Precursor Signals and Processes Associated with MJO Initiation over the Tropical Indian Ocean , 2013 .

[26]  Duane E. Waliser,et al.  A Unified Moisture Mode Framework for Seasonality of the Madden–Julian Oscillation , 2017, Journal of Climate.

[27]  P. R. Julian,et al.  Detection of a 40–50 Day Oscillation in the Zonal Wind in the Tropical Pacific , 1971 .

[28]  Prabhat,et al.  Application of Deep Convolutional Neural Networks for Detecting Extreme Weather in Climate Datasets , 2016, ArXiv.

[29]  M. Wheeler,et al.  An All-Season Real-Time Multivariate MJO Index: Development of an Index for Monitoring and Prediction , 2004 .

[30]  Douglas W. Nychka,et al.  Interpretable Deep Learning for Spatial Analysis of Severe Hailstorms , 2019, Monthly Weather Review.

[31]  Changhyun Yoo,et al.  Modulation of the boreal wintertime Madden‐Julian oscillation by the stratospheric quasi‐biennial oscillation , 2016 .

[32]  Patrick T. Haertel,et al.  Zonal and Vertical Structure of the Madden–Julian Oscillation , 2005 .

[33]  John M. Wallace,et al.  Three-Dimensional Structure and Evolution of the MJO and Its Relation to the Mean Flow , 2014 .

[34]  Imme Ebert-Uphoff,et al.  Physically Interpretable Neural Networks for the Geosciences: Applications to Earth System Variability , 2019, Journal of Advances in Modeling Earth Systems.

[35]  Hong Chen,et al.  Universal approximation to nonlinear operators by neural networks with arbitrary activation functions and its application to dynamical systems , 1995, IEEE Trans. Neural Networks.

[36]  Daehyun Kim,et al.  The MJO as a Dispersive, Convectively Coupled Moisture Wave: Theory and Observations , 2016 .

[37]  Pierre Gentine,et al.  Deep learning to represent subgrid processes in climate models , 2018, Proceedings of the National Academy of Sciences.

[38]  Min Dong,et al.  Seasonality in the Madden–Julian Oscillation , 2004 .

[39]  B. Liebmann,et al.  Description of a complete (interpolated) outgoing longwave radiation dataset , 1996 .

[40]  Joy Merwin Monteiro,et al.  Interpreting the upper level structure of the Madden‐Julian oscillation , 2014 .

[41]  Bin Zhao,et al.  The Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2). , 2017, Journal of climate.

[42]  Achim Zeileis,et al.  BMC Bioinformatics BioMed Central Methodology article Conditional variable importance for random forests , 2008 .

[43]  Harry H. Hendon,et al.  Organization of convection within the Madden-Julian oscillation , 1994 .

[44]  Scott Powell,et al.  Successive MJO propagation in MERRA‐2 reanalysis , 2017 .

[45]  Alexander Binder,et al.  On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation , 2015, PloS one.

[46]  T. Matsuno,et al.  Quasi-geostrophic motions in the equatorial area , 1966 .

[47]  Benjamin A. Toms,et al.  The Global Teleconnection Signature of the Madden‐Julian Oscillation and Its Modulation by the Quasi‐Biennial Oscillation , 2020, Journal of Geophysical Research: Atmospheres.