Answers are grounded in this paper and cite their sources.
We congratulate the authors for their effort and for their tour de force. Our comments are listed below. Over-smoothing extreme events: significant advancements in high-resolution regional climate models (RCMs) have improved their ability to simulate extreme events—specifically extreme rainfall. While it is valid to use observational data to model extreme events, the outright criticism of climate models may overlook the significant progress made in model refinement, bias correction techniques, and real-time data assimilation. The authors are encouraged to acknowledge such advancements in numerical weather predication (NWP)-based modelling of extreme events (He et al., 2023). The authors note a larger temporal change in the marginal behaviour of extreme temperatures compared with mean temperature levels. Comparing changes in extremes to mean levels requires careful consideration of statistical context and scaling. While extremes can change more dramatically, interpreting this in isolation might be misleading without a comparative analysis of variance and distribution shifts (Easterling et al., 2000). We urge the authors to dedicate a section on physical explanations and/or hypotheses, such as the impact of El Niño and other climate oscillations on the temperature trends (Chandran et al., 2016). The Matérn covariance function fails to fully capture the dominant spatial dependencies of extreme temperature events, especially in a geographically diverse region like Ireland (which the authors acknowledge). Alternative methods, as simple as geographically temporally weighted regression (GTWR), can account for both spatial and temporal dependencies with other key variables such as topography, landcover/land use, and precipitation (Wehbe et al., 2018). The precipitation-temperature feedback is a well-documented and dominant physical process (Nie et al., 2018) that merits expanding the analyses to include precipitation extremes. The ERA-Interim product ingests real-time station observations of temperature along with other observational data types through data assimilation (Dee et al., 2011). Therefore, a correlation between the station and RCM output may propagate across this data flow—from initialization (GCM) to forcing (RCM). The authors are encouraged to address this possibility, in addition to a discussion on scale-based uncertainties between the local/station-based measurements and the gridded RCM data at ∼10 km resolution. Using observational data to capture nonstationarity is unfeasible to Middle East and sub-Saharan Africa. Such regions necessitate a hybrid approach combining high-resolution climate model data with observational records (Almazroui, 2020). The authors assert that they can vary the smoothness and scale parameters of the Matérn kernel so that they depend over time. We claim that this is not feasible, unless the hypothesis of temporal stationarity is violated. Conflict of interests: None declared.
No takes yet. Share an insight, caveat, or question.
Porcu et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: