GC33J-2731 Spatial Temperature-Precipitation Compound Events in Europe: Weather Generator vs. Regional Climate Models, Future vs. Present Climate

Martin 1,2, Petr Stepanek 2, Jiri Miksovsky 2,3, Ondrej Lhotka 1,2, Jan Meitner 2, Petr Skalak 2, (1) Institute of Atmospheric Physics CAS, , Czech Republic, (2) Global Change Research Institute CAS, , Czech Republic, (3) Charles University, Faculty of Mathematics and Physics, Prague, Czech Republic

www.ufa.cas.cz/dub/prasce/2018-agu-spagetta~mad.pdf

1. Abstract To justify the use of Weather 2. Data and Target Regions 3. SPAGETTA Weather Generator Generators (WGs) in climate change impact PREC occurrence ~ Markov chain studies, WGs are validated for their ability to Observations ~ E-OBS [v.13.1; TAVG & PREC; 0.5º data PREC amount ~ Gamma distribution represent statistical structure of the real weather regridded into 1º; calibr.period = 1971-2000] data. The validation indices may include Non-PREC variables ~ AR(1) model 19 RCM simulations: 8 regions: characteristics of (a) probability distribution To spatialise WG, two parallel multivariate AR models control - CORDEX database functions of weather variables (mean, variability, - EUR-44 data regridded the spatial coherence of precipitation and non-precipitation quantiles, extremes), (b) temporal structure into “88” res. variables. SPAGETTA is calibrated using weather data from a set (persistence, occurrence of spells of specific - RCP85 emissions of sites distributed either regularly (grids) or irregularly (stations) weather), (c) spatial structure (for spatial WGs), - baseline = 1971-2000 in space. Then it may produce synthetic time series having spatial and (d) relationships between variables. - future = 2071-2100 and temporal structure similar to the calibration data. This contribution shows results obtained by the To generate series representing the future climate, WG parametric spatial weather generator parameters are modified using the climate change scenarios SPAGETTA. We compare its performance for 8 derived from climate simulations made with RCMs or GCMs. European regions with the results based on 19 RCM simulations taken from the CORDEX 4. Present experiment (made for 8 regions) database. The WG and RCMs are validated for their ability to reproduce compound spatial • calibration (using E-OBS) of the bivariate multi-grid WG temperature-precipitation events: days and spells • generation of 300-year multi-grid synthetic weather series of spatially extensive hot-dry, hot-wet, cold-dry and • comparing lag-0 and lag-1 spatial auto-correlations of PREC cold-wet weather. The events are analysed from and TEMP derived from EOBS, WG and RCMs [ BOX 5 ] reference weather data (EOBS), synthetic series (produced by WG), and RCM baseline & future • comparing spatial TEMP-PREC indices from EOBS, climate simulations. WG, RCM-baseline and RCM-future weather series. [ BOX 6 ]

5. Dependence of lag-0 and lag-1 spatial auto correlations on the inter-grid distance (region = cEUR /Central Europe/) Input daily TEMP & PREC time series: (a) EOBS = E-OBS v.13.1 1971-2000 [grey dots] Inter-grid distance (for Lag-1 graphs): considering the lag-1 (b) SYN = SPAGETTA-generated synthetic 300y series\[green dots] correlation relates today’s values in G1 grid with yesterday’s (c) RCM = 1971-2000 series: - individual RCMs: symbols represent means for separate distance intervals values in G2, than the negative (positive) value of the distance - averages of all 19 RCMs: large orange circles indicates than G1 lies eastward (westward) from G2

Lag-0 COR Lag-1 COR TEMP

Lag-0 COR(TEMP) Lag-1 COR(TEMP)

Lag-0 COR(PREC) Lag-1 COR(PREC) PREC 6. “Spatial” dry / wet / hot / cold / “compound” days & spells D R Y W E T (baseline & future) Wet/Dry Day = PREC ≥ 0.05 mm / PREC < 0.05 mm in N>50% of grids

Hot/Cold Day = T ≥ Thi / T ≤ Tlo in N>33 % of grids (Thi = A + 1.282 x S, Tlo = A  1.282 x S; A & S are baseline-climate mean & std of T for a th given day of the year; Tlo and Thi are 10 and 90th percentiles of N(0,1) distribution xxx Spell = continuous sequence of xxx days

. Graphs show (colours of text correspond to colours of bars in the graphs): a) EOBS and SPAGETTA: - means (+/- std) of annual counts of xxx days - means (+/- std) of annual longest xxx spell b) For RCM ensemble: - A(19RCMs) ± A(s) / 19RCMs-FUT ± A(s) = averages of the means ± averages of STDs of 19 RCMs’ annual characteristics (baseline /future) - A(19 RCMs) ± S(RCMs) / 19RCMs-FUT ± S(RCMs) = average ± std of the means of RCMs’ annual characteristics (baseline / future) “Days” H O T ”Spells” e v n t s “Days” C o m p o u n d C o m p u C O L D C O L “Spells”

7. Comments & Conclusion WG outperforms RCMs in reproducing hot-dry, hot-wet and cold-wet Days; RCMs are better in reproducing wet days, cold-dry days and hot-wet spells. a) BOX 5: In some results, the inter-RCM variability is very small (e.g. “hot days” and “dry • WG perfectly reproduces spatial temperature autocorrelations. However, as days”), and generally smaller than internal RCM uncertainty. these correlations are included in a set of WG parameters, the perfect fit “only” confirms correctness of the generator’s source code and appropriateness of applied numerical approaches. b) BOX 6: RCMs-based Climate change impacts on spatial Temp-Prec events • WG underestimates precipitation auto-correlations, which is a result of using the • frequency of spatial hot events will increase everywhere, most of all in Wilks’ spatialisation approach, in which the precipitation occurrence patterns are Mediterranean (especially in IBER & TURK) modelled by AR(1) process. • correspondingly, frequency of spatial cold events will decrease • Similarly to WG, RCMs also fit temperature autocorrelations and underestimate • change in frequency in spatial dry / wet events is less significantly manifested. PREC autocorrelations. Generally, dry events will occur more often (except for Scandinavia), wet events will become less frequent a) BOX 6: Validation of WG and RCMs (baseline climate) • change in frequency of spatial compound events is controlled by the temperature As a result of some similarity between WG’s and RCMs’ abilities to reproduce change: hot-dry & hot-wet events will come more often, while cold-dry and cold- temperature and precipitation autocorrelations (shown in BOX 5), they also went will become less frequent similarly well reproduce the “Days” and “Spells”. Note that results in BOX 6 relate • In contrast with the baseline climate, inter-RCM variability is mostly larger than to 8 European regions (while only cEUR is treated in BOX 5]. internal RCM variability

Acknowledgements: The experiment was made within the frame of two projects funded by Czech Science Foundation (projects nos. 16-04676S and 18-15958S) and also supported within the SustES project (CZ.02.1.01/0.0/0.0/16_019/0000797).