Interpret extreme climate impacts from large population simulations — are they invisible or impractical? – Is it good?

Open Access Paper in Environmental Research Letters

T Kelder^8.1N Wanders²K van der Wiel³TI Marjoribanks⁴LJ Slater⁵R l Wilby^first and C Prudhomme^1,6,7

Published March 29, 2022 • © 2022 author(s). Published by IOP Publishing Ltd
Environmental Research Letter, Episode 17, 4

Quote T Kelder et al 2022 Environment. Res. Lett. 17 044052

abstract

Large population climate model simulations can provide a deeper understanding of the characteristics and causes of extreme events than historical observations, due to their larger sample sizes. However, a full assessment of simulated ‘unseen’ events that are more severe than those seen in historical records is complicated by observational uncertainty and variability. natural nature. Therefore, conventional evaluation and correction methods cannot determine whether simulations outside of the observed variability are correct for appropriate physical reasons. Here, we introduce a three-step process to evaluate the realism of simulated extreme events based on model properties (step 1), statistical features (step 2), and reliability. physical reliability of extreme events (step 3). We illustrate these steps for a 2000-year Amazon monthly flood population simulated using the EC-Earth global climate model and the PCR-GLOBWB global hydrological model. EC-Earth and PCR-GLOBWB are suitable for large-scale watersheds like the Amazon, and have simulated ‘unseen’ monthly floods far beyond the observed variability. We found that the realism of these simulations could not be statistically explained. For example, there can be a legitimate discrepancy between simulation and observation due to the occasional time-lapse accumulation of multiple flood peaks, rarely seen in observations. Physical reliability tests are important to assess their realism and show that the Amazon’s invisible monthly floods are produced by an unrealistic bias adjustment for precipitation. We conclude that there is a high sensitivity of simulations outside of the observed variability for the bias correction method, and that physical reliability testing is important to understand what is driving the biases. Extreme events are simulated. Understanding the driving mechanisms of unprecedented events could guide future research by uncovering key flaws in climate models. They can also play an important role in helping decision makers predict unseen impacts by uncovering legitimate motives.

Original content from this work may be used under the terms of Creative Commons Attribution License 4.0. Any further distribution of this work must maintain attribution and title of the work, journal citation and DOI.

Introduce

Extreme weather such as floods, droughts, heat waves, and tornadoes can have large social impacts including death and disease (Gasparrini et al. et al 2015Raymond et al Year 2020), and economic losses (Felbermayr and Gröschl 2014Klomp and Valckx 2014Kousky 2014). Extreme weather can also increase inequality (Dell et al 2012Hallegatte and Rosenberg 2017). In the risk analysis, a full assessment must be made of the impacts that may arise from extreme climate and weather (Sutton et al. 2019). For example, a reliable maximum extreme event is important for risk estimates of potentially disruptive effects (Wilby et al 2011), such as mortality, morbidity, and damage from flooding in major river systems and from dam failures (e.g. Vano et al 2019), or for climate-related shocks to food security (Kent et al 2017). However, the brevity and sparsity of historical records are well-known limitations that confound estimates of the likelihood of extreme events (Alexander 2016Wilby et al 2017). Climate model projections reduce this limitation but may not fully capture the extremes that may arise from climate change using only some composite members (Van der) Wiel et al 2019bMankin et al Year 2020). However, large synthetic simulations ranging from seasonal to multi-decade prediction systems provide a solution for estimating rare events due to their multiple perceptions (Allen et al. 2003van den Brink et al 2005Thompson et al 2017Van der Wiel et al 2019bMankin et al Year 2020Brunner and Slater 2022).

Traditionally, large populations were generated by trained random weather generators based on historical records (e.g. Wilks and Wilby 1999Brunner and Gilleland Year 2020). However, advances in supercomputing and the physical realism of climate models have facilitated the exploitation of large population simulations to simulate events with drivers. physically plausible has not been observed (Coumou and Rahmstorf 2012Stevenson et al 2015Stott et al 2016Kent et al 2019Thompson et al 2019Deser et al Year 2020Kay et al Year 2020Swain et al Year 2020Brunner and Slater 2022). Follow Thompson et al (2017).

One limitation of using model simulations is the potential for biases, which can sometimes create unrealistic extreme events. Many techniques have been developed to detect potential systematic climate model deviations (Eyring et al. et al 2016, 2019), comparing the simulated extremes with the observations (Weigel et al Year 2021), and to evaluate the consistency between simulated and observed distributions of extreme events (Thompson et al 2017, 2019Kelder et al Year 2020Suarez-Gutierrez et al Year 2021). However, none of these procedures can determine whether the models are correct for appropriate physical reasons.

Bias correction (or data adjustment) methods are widely used to reduce model bias, especially when combining climate model simulations with impact models (Warszawski et al. et al 2014), but it is not necessary to correct the simulations for suitable physical reasons (Maraun et al 2017). For example, a mismatch between simulation and observation may be caused by uncertainty in the observations and natural variability, but not by model bias (Addor and Fischer et al. 2015Casanueva et al Year 2020). Therefore, current evaluation and correction methods are not designed for simulated invisible events. As a result, large population simulations with extreme events outside the range of observed variability raise an important question: to what extent can such outliers be trusted? Are events Can not see or unrealistic?

In this paper, we present a framework for testing whether the conclusions about unseen events obtained from large meta-analyses are correct. Our three steps to assess the realism of simulated events fall outside the range of observed variability (Fig. first) was inspired by the event allocation protocol for climate change (Philip et al Year 2020). Step 1 is to examine the properties of the model and evaluate whether the system representation is capable of representing the processes involved leading to extreme events. Step 2 is to evaluate the statistical characteristics of a large set of simulations (whether from global climate models or regional climate models) by evaluating the consistency of model distributions. simulated with observations. Bias correction is an integral part of the evaluation of statistical subjects as it is a common practice (e.g. Warszawa et al 2014) but can affect the simulated distribution of extreme events and impacts. Therefore, we evaluate the statistical features for both the raw value and the corrected bias. Step 3 is to evaluate the physical reliability of the model simulations. Although some studies examine the physical processes that lead to extreme events – such as telemetry and earth-atmosphere interactions (Van der Wiel et al 2017Thompson et al 2019Vautard et al 2019Kay et al Year 2020) – it is not easy to establish material credibility (Philip et al Year 2020), especially for unseen events.

We demonstrate our framework using a flood case study in the Amazon. In 2009 and 2012, flooding in the Amazon led to the spread of disease, food, and water insecurity (Davidson). et al 2012Hofmeijer et al two thousand and thirteenMarengo and Espinoza 2016Bauer et al 2018). At the time, the 2009 flood was the most severe on record for 107 years, but three years later it became the second-highest flood in 110 years, dramatically altering estimates of the likelihood of a flood. out. Although the Amazon phase record is one of the longest in the world, the ∼100-year record streak is still too short to estimate reliable, worst-case events.

To sample more flood events than those available from the historical record, we use EC-Earth large synthetic global climate model simulations combined with a global hydrological model (equilibrium water) PCR-GLOBWB from an earlier study (Van der Wiel et al 2019b). EC-Earth and PCR-GLOBWB are modern global models that have been applied in many multi-model comparison studies, such as in the Paired Inter-Model Comparison Project (e.g. Taylor’s) et al 2012Samaniego et al 2019Vagrants et al 2019), and has been validated globally (Hazeleger et al 2012Sutanudjaja et al 2018), including Amazon stream (van Schaik et al 2018). Here, we extend previous studies by assessing whether extremes simulated beyond the historical record are likely to be unseen or simply unrealistic. We do this by: examining the ability of EC-Earth and PCR-GLOBWB to simulate extreme Amazon flooding (Step 1); then evaluate the statistical consistency of these large population simulations with observations using raw data or bias-corrected simulations (Step 2); explore the physical dynamics behind the largest simulated flood (Step 3).

Read the full article here.