Identifying Relations Between Deep Convection and the Large‐Scale Atmosphere Using Explainable Artificial Intelligence

Abstract

Precipitation from convective systems accounts for the majority of precipitated water in the tropics and therefore is a vital part of the hydrological cycle. Clouds formed by convective processes strongly affect the radiative heat balance of the atmosphere, which through energy constraints, feeds back directly on the hydrological cycle. This makes convection a key process in connecting the energy and water cycles on Earth. While the importance of convection is well recognized, computational constraints require convection to be parametrized in most global weather and climate models (e.g., Arakawa, 2004). Hence, an important question regarding convection is, how the relatively small convective cloud scales, and the heating produced in them, relate to larger atmospheric scales. An example of such a relationship is the well-known relation of convective precipitation to atmospheric moisture content (e.g., Bretherton et al., 2004;Holloway & Neelin, 2009;Schiro et al., 2018). At short timescales, observational studies have also revealed relationships between convective precipitation and various measures of (in) stability (e.g.,

Drivers of mesoscale convective aggregation and spatial humidity variability in the tropical western Pacific

“…In the boreal winter/spring period, where clustered convection is the norm, the major change is that the surface LH flux feedback reverses and acts against clustering, which is again in agreement with idealized modelling studies [16,25], although a separate analysis (Fig. S5) shows this is mostly driven by surface wind feedbacks here, rather than the changing boundary layer structure as found in idealized studies [25], and coinciding with previous observational studies [28]. These traits are mirrored when looking at the reversal versus organized conditions (c and d), with the reversal state looking much like the homogeneous condition that predominate in the summer/autumn period.…”

A climatology of stratospheric gravity waves induced by tropical cyclones

“…The importance of each explanatory variable can be obtained as Fig. 4c by Layer-wise Relevance Propagation (LRP) 45 . The background wind in the stratosphere (70-30 hPa), especially at 30 hPa, is the most important variable for the TC-SGWI pattern.…”

“…The "neural network" algorithm 57 , hereafter NN for reference, has been widely used in the meteorological eld 45,58 and in studies of SGWs 59,60 . The NN is appropriate for large data volumes and can identify the nonlinear relationships among the data.…”

Comparison of Climate Model Large Ensembles With Observations in the Arctic Using Simple Neural Networks

“…To evaluate how the ANN is classifying each temperature map with the correct GCM, we use a method of explainable machine learning called layer-wise relevance propagation (LRP; Bach et al, 2015;Montavon et al, 2017Montavon et al, , 2018. First introduced by Toms et al ( 2020) for applications in the geosciences, LRP has now been used in a wide range of studies across atmospheric and climate sciences for attempting to understand the decision-making process of neural networks (e.g., Gordon et al, 2021;Hilburn et al, 2020;Mayer & Barnes, 2021;Retsch et al, 2022;Sonnewald & Lguensat, 2021). Importantly for its use in this work, LRP has also been shown to be an effective technique for extracting regional patterns of forced climate change that are collectively found between climate models and observations (e.g., Barnes et al, 2020;Labe & Barnes, 2021;Madakumbura et al, 2021;Rader et al, 2022).…”