Topological data analysis and machine learning for recognizing atmospheric river patterns in large climate datasets

Abstract

Abstract. Identifying weather patterns that frequently lead to extreme weather events is a crucial first step in understanding how they may vary under different climate change scenarios. Here, we propose an automated method for recognizing atmospheric rivers (ARs) in climate data using topological data analysis and machine learning. The method provides useful information about topological features (shape characteristics) and statistics of ARs. We illustrate this method by applying it to outputs of version 5.1 of the Community Atmosphere Model version 5.1 (CAM5.1) and the reanalysis product of the second Modern-Era Retrospective Analysis for Research and Applications (MERRA-2). An advantage of the proposed method is that it is threshold-free – there is no need to determine any threshold criteria for the detection method – when the spatial resolution of the climate model changes. Hence, this method may be useful in evaluating model biases in calculating AR statistics. Further, the method can be applied to different climate scenarios without tuning since it does not rely on threshold conditions. We show that the method is suitable for rapidly analyzing large amounts of climate model and reanalysis output data.

The Atmospheric River Tracking Method Intercomparison Project (ARTMIP): Quantifying Uncertainties in Atmospheric River Climatology

“…Latitude-independent relative methods use one threshold based on the climatology of a given region and can be expected to produce larger gradients in AR statistics, which will likely be more similar to results produced by absolute methods. Furthermore, this paper includes one method based on machine learning (TDA_ML; Muszynski et al, 2019), which defies many of the threshold-based groupings outlined above. It is currently employed over the western Unites States but could be readily applied to other regions.…”

“…For example, both "Lavers" and "Ramos" use time-dependent percentiles based on climatological IVT, and "Viale" imposes the restriction that an AR must be associated with a frontal system. The machine learning technique, "TDA_ML," produces AR frequencies below the median, possibly because the algorithm employed by Muszynski et al (2019) exhibits a fairly strong resolution-dependent decrease in the "sensitivity score" (the proportion of identified ARs that are correctly identified) as resolution increases, with over 25% of features being misclassified (relative to their training data set) as non-ARs at high resolution. Muszynski et al (2019) hypothesize that this is due to an interaction between the decrease in smoothness of the IWV field as resolution increases and the underlying topology-based method that they use to identify potential ARs.…”

“…Interestingly, the AR frequency from the machine learning method, "TDA_ML," is characterized by a maximum of just 2% near 39°N and declines to 0% north of 45°N, where many methods produce their larger frequency values. Muszynski et al (2019) note that this method frequently produces false negatives (i.e., it fails to detect ARs) when an AR merges with another AR or "some other event with high concentration of water vapor and similar topological structure, such as an extratropical cyclone." This would happen with higher frequency in the more active storm track at latitudes north of 45°N, which may explain the rapid drop-off in AR detection associated with this method.…”

“…Of course, in addition to using different identification and tracking methods, many researchers examine different regions, using different data sets, and different periods of records, to do so. More recently, machine learning techniques have been developed to identify and track ARs (e.g., Mudigonda et al, 2017;Muszynski et al, 2019;Radić et al, 2015).…”

Detection Uncertainty Matters for Understanding Atmospheric Rivers

“…Discussions following the 2 nd ARTMIP Workshop suggested that differences among AR tracking algorithms might reflect differences in expert opinion about what constitutes the boundary of ARs; resolving this question would require experts to hand-label ARs. Unrelated, but concurrent, advances in Computational Climate Science have demonstrated the utility of modern machine learning methods for tracking weather phenomena (Mudigonda et al 2017;Muszynski et al 2019;Kurth et al 2018). These developments also highlight the need for high-quality data to train machine learning methods: expert-labeled datasets.…”