Detection of Atmospheric Rivers with Inline Uncertainty Quantification: TECA-BARD v1.0

Abstract

Abstract. It has become increasingly common for researchers to utilize methods that identify weather features in climate models. There is an increasing recognition that the uncertainty associated with choice of detection method may affect our scientific understanding. For example, results from the Atmospheric River Tracking Method Intercomparison Project (ARTMIP) indicate that there are a broad range of plausible atmospheric river (AR) detectors, and that scientific results can depend on the algorithm used. There are similar examples from the literature on extratropical cyclones and tropical cyclones. It is therefore imperative to develop detection techniques that explicitly quantify the uncertainty associated with the detection of events. We seek to answer the question: given a ‘plausible’ AR detector, how does uncertainty in the detector quantitatively impact scientific results? We develop a large dataset of global AR counts, manually identified by a set of 8 researchers with expertise in atmospheric science, which we use to constrain parameters in a novel AR detection method. We use a Bayesian framework to sample from the set of AR detector parameters that yield AR counts similar to the expert database of AR counts; this yields a set of 'plausible' AR detectors from which we can assess quantitative uncertainty. This probabilistic AR detector has been implemented in the Toolkit for Extreme Climate Analysis (TECA), which allows for efficient processing of petabyte-scale datasets. We apply the TECA Bayesian AR Detector, TECA-BARD v1.0, to the MERRA2 reanalysis and show that the sign of the correlation between global AR count and El Nino Southern Oscillation depends on the set of parameters used.

Uncertainties in Atmospheric River Lifecycles by Detection Algorithms: Climatology and Variability

“…About 13 ARDTs detect 5–15 objects globally at any given time step, which is consistent with the manual analysis by Newell et al. (1992) and the range of global AR objects manually identified by experts (O'Brien et al., 2020). The lowest counts are from TECA_bard_v1 (#8 in Table 1) and Reid500 (#19), where the former includes a set of “plausible” AR detectors (O'Brien et al., 2020) and the latter uses a scheme that combines the thresholds commonly used in the literature (Reid et al., 2020).…”

“…(1992) and the range of global AR objects manually identified by experts (O'Brien et al., 2020). The lowest counts are from TECA_bard_v1 (#8 in Table 1) and Reid500 (#19), where the former includes a set of “plausible” AR detectors (O'Brien et al., 2020) and the latter uses a scheme that combines the thresholds commonly used in the literature (Reid et al., 2020). The highest count is from CONNECT500 (#2), which captures some tropical disturbances that may not be associated with ARs or are entrained by ARs over their lifecycle (e.g., tropical cyclones) (Figure ; also in Lora et al.…”

Uncertainties in Atmospheric River Life Cycles by Detection Algorithms: Climatology and Variability

“…The average number of objects per time step varies from 6 to 42 (Figure 1a), where the lowest number is from TECA_bard_v1 which includes a set of "plausible" AR detectors and the highest number is from CONNECT500 which captures some tropical disturbances that may not be associated with ARs or are entrained by ARs over their life cycle (e.g., tropical cyclones) (not shown). About 10 algorithms detect 10-20 global AR objects at any given time step, which is consistent with the number from a manual analysis by Newell et al (1992) and the number range of expert-identified global AR objects (O'Brien et al, 2020).…”

The Shifting Scales of Western U.S. Landfalling Atmospheric Rivers Under Climate Change

“…Further, as mentioned in the introduction, global climate models have had long-standing precipitation biases, particularly nonconvergence of extreme precipitation at more refined horizontal resolutions, which may account for some of the precipitation mismatch between VR-CESM and ERA5. Last, as shown by recent studies by Rutz et al (2019) andO'Brien et al (2020), AR algorithm parametric and structural uncertainty is important, particularly in the context of climate change. Future work will aim to address some of these limitations and isolate other impacts associated with changes in landfalling ARs.…”