Non‐stationary control of the <scp>NAO</scp> on European rainfall and its implications for water resource management

Water resource forecasting generally centres on understanding hydrological variability over coming months or years, so that water managers can prepare for extremes such as droughts floods (Chang & Guo, 2020; Hao et al., 2018). Some systems seek to project further into the future allow long-term planning of infrastructure and resilience climate change (Svensson 2015). These rely directly indirectly outputs from Global Climate Models (GCMs; gridded reanalysis datasets) forecast conditions (Bhatt Mall, 2015; Ionita Nagavciuc, 2020). In North Atlantic region, in particular Western Europe, Oscillation (NAO) is used an indicator hydrometeorological given its leading control winter rainfall totals (Hurrell Deser, 2010; Scaife 2008, 2014). A dipole pressure anomalies Atlantic, NAO's positive phase (greater than average gradient; NAO+) results wetter northwest Europe with dryer southwest (Rust 2018; Trigo 2004). Its negative (weaker NAO−) inverse effect (Folland shown by correlation coefficients Figure 1). Given this relationship, and, considering role groundwater drought development (e.g., reduced recharge) generation late winter/early spring floods, NAO offers a potential explanatory variable when behaviour some extremes. Many studies have investigated relationship between Index (NAOI) recorded variables including river flow (Burt Howden, 2013), (Lavers 2015), seasonal (West 2019), calculated standardized precipitation index (Moreira 2016). Furthermore, growing number detected multi-annual periodicities 2020), level (Neves 2019) align weak NAOI Loon, 1997), suggesting cycles may affect occurrence Europe. GCMs historically been ineffective at predicting dynamics within scenarios (Smith 2016); however, recent developments high-performance computing enabled effects (Świerczyńska-Chlaściak Niedzielski, 2020) decadal (Athanasiadis Dunstone 2016) timescales. high-resolution datasets becoming more readily available (Sun 2018), regression analyses purposes achievable. As such, there renewed interest resources community developing medium- using NAOI, aim improving preparedness (Hall Hanna, Svensson This particularly useful water-scarce regions southern England Iberian Peninsula where long-range has largest benefit strategic (Stein The current use relationships (herein referred NAOI-P) relies heavily assumption strength direction sufficiently stationary be applicable scenarios. However, research itself NAOI-P non-stationary multi-decadal timescales (Vicente-Serrano López-Moreno, 2008). yet assessment their utility perspective, relevant projections. We discuss non-stationarity undertaking rolling 10-year window estimates period past 125 years (1889–2016) across we identify five zones similar behaviours provide insight application management. Two paper; data NCAR Analysis Section (Hurrell, 2003), Precipitation Center (Schneider 2018) used. are Hurrell station-based (DJFM) 2003). dataset covers 1864–2019. taken GPCC's Full Data Monthly Product Version 2018 which estimated, 0.5° resolution specifically generated investigate climatological 1891–2016. processed according same analysis. Grid cells −13° 20° Longitude 35°–70° Latitude were represent simple Pearson's r produced full-period (Figure analysis was each grid cell 1891–2016, coefficient n-9 n-year representing coefficient. k-medoids cluster (analogous k-means median case) then undertaken rolling-window correlation-series. Each series not normalized prior clustering metric, (in-part) aiming how directional characterized Clustering draw together generalized areas NAOI-winter purposes. Five clusters chosen gave greatest spatial coherence clusters, thereby indicating it appropriate inform decision making. Medoids centroids) cluster. dissimilarity matrix also assessed understand range (most dissimilar) selected nation-based regions. characteristic either 1) clustered correlations 2) domain. Significance thresholds all (full-period rolling) 95% confidence interval. Means variances medoids dissimilar pairs which, addition qualitative temporal will assess non-stationarity. indicate relationship. Clusters show varies considerably time space, transitioning higher latitudes 1 2), lower latitude 3 4). consistent Relative stability correlation, containing two clusters: Transitional greater variance, paper seeks question whether (NAOI-P) Results few considered scale. degree utilized studies, find direct implications efficacy Previous consistently highlighted having strongest Tsanis Tapoglou, 2019). Our agree these areas: example, Cluster (NW Scotland, Ireland Norway) remains strong, majority assessed, 4 (Portugal Spain), remain similarly strong but correlation. NW UK Ireland, highest (Totz 2017) abundance, flood risk high-flows management, rather drought. Portugal Spain, most (Estrela 2012), relative stationarity supports management generalization (particularly Spain) entirely timescale 2f, see considerable fluctuations caution should still applying large While existing (as mentioned previously), our exhibits demand new approaches manage scarcity, (Bryan 2019; Folland Recent (Prudhomme 2017; 2015) focused prediction (rainfall, flow, level) one future, many (at least part) 2, 5, cover transitional exhibit stronger those (Clusters work shows moderate, often significant, correlations; western northern Germany (Riaz 2017), France (Massei 2007). Indeed, found skilful (and other variables) through historical despite reported here (Ionita Moreira 2016; Rasouli Rousi 2014; apparent contradiction (Pauling 2006; Vicente-Serrano 2008) explained common periods results, general throughout series. For increased seen medoid, dissimilarly 2b), south 2c), Italy 2e), 1970 present. previously, high-resolution, gridded, calibration only decades Such representative true here. naturally draws longer-term validity utilize without accounting nature. Counter this, discounted due poor 1): instance, 2007) southeast suggest up 30 stay state significance (95% CI) 5 England) suggests if mechanisms inversion understood, could powerful (short medium-term) ruled out policy practice need flexible, employing frequent reviews forecasts, utilization changing control. Variability subject ongoing focus atmospheric sciences, investigating multidecadal (30 year) timescales, López-Moreno (2008) inversions result eastward shift centre (Azores) towards eastern Mediterranean. Atmospheric literature seems interannual shifts Rosby wave (Luo Gong, Peterson 2003; Zhang 2011), exact mechanics being explored (Huang If critical account happen, required full extent different importantly, fully inverted represented GCMs. Further, shed light uncertainties understanding, summarized Blösch al. (2019). With improved non-stationarities European weather driving systems, better positioned answer questions around cycle paper), drought- flood-rich arise 9), models adapted extrapolate 19). Answers continued sustainable climate. supported Natural Environment Research Council (grant numbers NE/M009009/1 NE/L010070/1) British Geological Survey (Natural Council). J.P.B. publishes permission Executive Director, (NERC). M.O.C. gratefully acknowledges funding Independent Fellowship (NE/P017819/1). support findings study CORD https://doi.org/10.17862/cranfield.rd.13949381. re-analysis publicly https://climatedataguide.ucar.edu/climate-data.