Future changes in hourly extreme precipitation, return levels, and non-stationary impacts in Türkiye

Abstract

Extreme weather events have increased in severity and frequency throughout the majority of geographical regions. Once rare events like unusually hot, wet, or dry conditions have become common in some regions. For instance, heavy downpour events are occurring more frequently and with greater severity. The Mediterranean basin is one of the locations identified in the literature as a climate change hotspot by research that looked at the SREX and AR6 regions. The study of this inland sea, which borders numerous communities in Europe and Africa, shows that the Mediterranean basin's drought conditions have gotten worse and more protracted. Furthermore, these conditions are predicted to get worse in future climate simulations, as measured by a decrease in annual total precipitation and a rise in extreme event intensity. Türkiye, the basin's easternmost member, also follows the trend observed around the Mediterranean Sea. The amount of precipitation in Türkiye's central and southern regions has decreased and is expected to continue to do so, although the northern boundary shows a different pattern. Similarly, the coastal and interior regions respond to climate change differently. Contrary to what has been observed and expected for the annual total precipitation, the magnitude of the severe precipitation has grown, affecting the majority of the nation's land areas. Considering that, we should pay more attention to extremes than averages since the variability in severe occurrences mirrors the primary impact of climate on our society. When a research subject involves block maxima or minima (BM) of time series, the Generalized Extreme Value (GEV) model is the most frequently used among EVT methods. So far, stationary GEV practices (for instance, stationary return levels) have laid the foundation for creating mitigation and adaptation measures in the context of organizational infrastructures and facilities. In that, the statistical features of extreme occurrences, such as mean and variability, are thought to remain consistent across time under stationarity. However, the quantitative characteristics of extremes are not static (non-stationary) in the face of human-induced climate change, land degradation, and urban densification. Studies in Türkiye employing GEV primarily uses daily and multi-day values of the climatic parameters, with relatively few studies using sub-daily (e.g., hourly) evaluation of extremes, one of the major factors in urban planning. Yet, those who take into account sub-daily samples of hydrometeorological parameters prefer not to use and disregard climate projections. In addition, they only utilize the intensity index and do not employ the frequency or persistence indexes for extreme value modeling. This study recognizes the usefulness of using sub-daily (hourly) samples of several precipitation indices (intensity, frequency, and persistence), using climate projections, and identifying the non-stationary impact on the future values of hourly intense precipitation episodes. For these purposes, the Consortium for Small-scale Modeling (COSMO-CLM), established by the CLM-Community of the German Meteorological Service (DWD), which is capable of carrying out long-run simulations that serve climate modelers in the industry and academia, is utilized. Using the MPI-ESM-LR as the initial-boundary condition, this study undertakes two subsequent downscaling procedures, first from the native resolution down to 0.44° and then to 0.11° for the periods 1985-2005 (reference), 2061-2080 (first projection), 2081-2100 (second projection). The eastern Mediterranean is represented by the first downscaling as its main (outer) domain, while the area around Türkiye is subjected to secondary downscaling activity (inner domain), as that nation likewise suffers from worsening drought, decreased precipitation, and more intense heat. Quantile Delta Mapping (QDM) is chosen as the principal bias-correction method to modify observed partialities in the total precipitation variable of COSMO-CLM simulations. This is done by using ERA5-Land as the benchmark data and computing the bias coefficients in relation to the reference period. Following that, extreme precipitation indexes are calculated and investigated for hourly total precipitation data centered on three main titles: Intensity Index (InI), Frequency Indexes (FrI), and Persistence Indexes (PeI). FrI and PeI are further categorized into two sub-categories, percentile and absolute. The suggested extreme precipitation indexes are then submitted to twi-formed GEV analysis. First, stationary Generalized Extreme Value (S-GEV) models are constructed for all three 20-year periods. Here, the return levels of each indicator during the reference and projection timeframes are examined using different return periods (20 and 50 years) to determine the potential risks associated with future climate change. Second, the GEV analysis is forced to incorporate non-stationarity by embedding a linear trend in its location parameter (NS-GEV) based on an analysis of the combined projection periods (40 years, 2061-2100), and the non-stationary impact is investigated. According to the original index results, the northern, western, and eastern parts of Türkiye witness an increase in the InI of up to 40%–50% in the projection periods. Large portions of the Mediterranean coastline and a small portion of the Aegean coast (Izmir), however, are only subject to a decrease in InI of no more than 20% during both projection periods, with sporadic exceptions. Both frequency-based metrics show a general south-to-north contrast over time in terms of percent changes in projection periods. However, the indicated latitudinal rate of change in the frequency of extreme precipitation (FrI-percentile) becomes even more profound and tauter at the end of the century compared to the frequency of wet hours (FrI-absolute), as the positive (negative) percent anomalies in the north (south) approach - and in some places exceed - %80. The percent change in persistence-based metrics is likewise built up so that the south-to-north contrast (negative to positive) is much more closely packed in percentile-based indexes, much like the future climate projections of frequency-based metrics. According to S-GEV analysis, the intensity index shows that through the end of the century, 20-year return level (RL) values will climb practically nationwide. The highest of these is simulated in eastern Southeast Anatolia in addition to the coastal areas of the Mediterranean, Aegean, Marmara, and Black Seas. Istanbul, one of the most populous cities in the world, is located in the Marmara region, which experiences a noticeable increase in the 20-year RL during the latter projection period. Both frequency indices show a latitudinal variation in the projections for the future, particularly for the years 2081-2100, in contrast to results for the intensity index, which show a nationwide increase in 20-year RL. The 20-year RL along the southern and western parts significantly declines in comparison to 1985–2005, whereas the RL values in the northern regions increase. Percent changes in the persistence indices' 20-year RL over the projection periods show that the latitudinal gradient is not as systematic as it is in frequency-based metrics. Particularly over the zone that divides the south and north, the noise is noticeably higher, and the distribution is noticeably more uneven. Compared to 2061–2080, the acceleration of the RL shift is more noticeable in 2081-2100, as if the negative percent change is moving north over time. A quick inquiry reveals the pattern similarity between the 20-year and 50-year RLs (such as the south-to-north gradient and greater values along the elevated locations). The only distinction between the return periods for the complete indexes can be seen in their magnitudes, where the 50-year RL values are, as one might expect, greater than the 20-year RL quantities. In that, the percent change of InI, FrI, and PeI with negative and positive leanings increases in both directions and changes color to a darker shade. As the return period lengthens during the latter period (2081-2100), this shift is amplified even more, with some grids and indexes crossing %80-%100 bands, especially along the southern and northern coasts, in the Marmara region, and in the interior regions of Anatolia. In the subsequent GEV analysis (non-stationary impact analysis), the FrI-percentile index shows stronger negative non-stationary impacts from a 20-year RL perspective, containing substantial portions of the nation with percent reductions of more than 10% (brown shades), and at certain areas (southern Türkiye), nearing 20%. The regions that show the most detrimental non-stationary impact include those in western Türkiye, the west-middle Taurus Mountains, eastern southeast Anatolia, and the Marmara region. On the other hand, a minimal positive non-stationary impact (%5, at most) is seen along the Black Sea shore. Similarly, for the FrI-absolute index, a negative non-stationary impact is also perceptible across much of Türkiye, with sporadic positive values dispersed. Positive values are interspersed among those experiencing negative non-stationary impact (approximately %15–%20), whereas negative impacts are widespread across a large portion of the Mediterranean coast, eastern Marmara, and Southeast Anatolia regions (about %10–%20). Regarding the persistence-based indexes, parallel impact patterns can be seen in both PeI-percentile and PeI-absolute. In that, the Mediterranean-Aegean coasts and Southeast Anatolia approach negative non-stationary impact values above 10%. On the other hand, the country's interior and northern regions, especially the eastern Aegean region, show positive non-stationary impacts in which the non-stationary RL predominates over its stationary counterpart. Last but not least, the InI index stands out since it has the most uniform non-stationary impact. It has no systematic behavior, the smallest magnitudes of the indicated influence (no more than 5% in both directions), and the absence of any non-stationary impact across the majority of the country. Not surprisingly, all of the indexes show the same pattern as a 20-year-based return period when considering non-stationary influence from a 50-year-based return period viewpoint. The magnitude of the non-stationary impacts is the only simulated difference between them. Specifically, the non-stationary impact grows stronger and more pronounced as the return period lengthens, exceeding 20% at some grids. These discoveries have effects on the frequency of flood episodes. A state of more severe flash flood episodes may be present throughout the majority of the country. Extreme precipitation events may occur more frequently and last longer, especially in northern regions, which are home to several metropolitan agglomerations. This might put pressure on already-vulnerable infrastructure. Also, there is a chance that these events will become less predictable at the extremes of the statistical spectrum (i.e., how intense they will be or how long the extreme rainfall episode will last). This would put more pressure on the system's supporting infrastructure and have an impact on the decision-making processes. Overall, the study's conclusions contain several limitations either, including the choice of climate model settings, the bias correction strategies, and the global warming scenario. Future research should address these limitations and consider potential opportunities.