Towards a monitoring system of temperature extremes in Europe

Extreme-temperature anomalies such as heat and cold waves may have strong impacts on human activities and health. The heat waves in western Europe in 2003 and in Russia in 2010, or the cold wave in southeastern Europe in 2012, generated a considerable amount of economic loss and resulted in the death of several thousands of people. Providing an operational system to monitor extreme-temperature anomalies in Europe is thus of prime importance to help decision makers and emergency services to be responsive to an unfolding extreme event. In this study, the development and the validation of a monitoring system of extreme-temperature anomalies are presented. The first part of the study describes the methodology based on the persistence of events exceeding a percentile threshold. The method is applied to three different observational datasets, in order to assess the robustness and highlight uncertainties in the observations. The climatology of extreme events from the last 21 years is then analysed to highlight the spatial and temporal variability of the hazard, and discrepancies amongst the observational datasets are discussed. In the last part of the study, the products derived from this study are presented and discussed with respect to previous studies. The results highlight the accuracy of the developed index and the statistical robustness of the distribution used to calculate the return periods.

on the most extreme events (the thresholds are defined according to the yearly maximums), and it does not 73 take into account the Tmin that has a strong human impact according to WMO (2015). 74 In this study we propose an operational system to monitor heat and cold waves based on an adapted index 75 inspired by the previous studies. In section 2, data and methods are presented and the uncertainties related to 76 the observations are assessed. Then, the climatology in term of occurrence, intensity and duration of the waves 77 are presented in section 3. This represents the baseline of the monitoring system that will become operational 78 and embedded in the EDO system. Finally, concluding remarks are provided in section 4. In this study we use daily Tmax and Tmin from three different datasets. The first one is based on the 2m 83 temperature datasets provided by the European National Weather Services, which, in turn, is used as an input 84 for the LisFlood hydrological model (De Roo et al., 2000). The observations are gridded onto a regular lat/lon 85 grid of one square degree. The use of gridded observation data is motivated i) to focus on large scales heat/cold 86 waves and ii) to compare the station data with reanalysis. This LisFlood product will be eventually used in the the year is cut in two periods: The extended summer period, when heat waves could have more impacts (6 111 hottest month over Europe, from April to September), and the extended winter period to focus on the cold 112 waves (from October to March). Note that also during the summer (winter) period, cold (heat) waves may 113 occur but they are not considered here. The independent calculation of the daily quantiles of observed Tmin 114 and Tmax is done by applying a leave-one-out method to avoid inhomogeneities (Zhang et al. 2005). The year 115 studied is removed from the climatology. The data without this year is exploited to perform the observed 116 cumulative distribution function (CDF). To remove artefacts due to the relative small sampling (21 years), a 117 window of 11 days centred on the day studied is exploited. The daily temperatures are transformed into 118 quantile by this procedure to create two daily temperature quantiles from 1995 to 2015, derived from the CDF 119 of Tmin and Tmax independently.

120
The main difference with the previous studies is the use of both Tmax and Tmin, rather than Tmax only or the 121 daily mean temperature. Then a hot day is defined when simultaneously the daily quantiles of Tmax and Tmin 122 are above quantile 0.9 during the extended summer (from April to September). The same definition is applied 123 for cold days when the two quantiles are lower than quantile 0.1 from October to March. The occurrences are 124 strongly influenced by these thresholds. As this study aims at quantifying the intensity of waves regarding the 125 climatology and at assessing with robust scores the forecast of these events, it is not possible to focus only on 126 the most extreme cases. So these thresholds (quantiles 0.9 and 0.1) are chosen as compromise between the 127 need to have a minimum number of events and the definition of extremes. They are also used in a large number 128 of other studies (WMO, 2015, Hirschi et al., 2011. Note that in order to discuss the sensitivity of using the 129 intersection of Tmin and Tmax rather than one temperature value per day, the same methodology has also 130 been applied using only Tmin and only Tmax to determined hot and cold days. the ratio is about 60% (i.e. 226 out of 376 heat waves are indicated in the last column. The use of the two temperatures tends to reduce drastically the 147 number of events (from 44 or 51 to 16.9 on average during the period) but also their durations (5.11 or 5.3 148 days to 4.8). The continental regions appear less affected by this reduction than coastal regions (not shown).

149
In analogy, Table 2 shows the same data for the case of cold waves.

150
Once a wave is detected, two main characteristics are recorded: the duration (in days) and the intensity. To 151 take into account different characteristics and to assess the sensitivity of the methods, the latter is calculated 152 by three different methods. The first one is based on the sum of the quantiles above (or under) the threshold 153 during the detected wave.  Tmin and the differences in the Tmin definition amongst National Weather Services. Indeed, over certain 195 countries, Tmin is measured during night time between 1800LT and 0600LT the following day, elsewhere 196 from 0000LT to 2400LT or from 0600LT day d to 0600LT day d+1, which could result in a delay of one day. The total occurrences of heat and cold waves during the 21 years are calculated using the definition presented 210 in section 2. This is performed independently for the three datasets to provide information on the robustness 211 of the results. As shown in Table 1 and 2 cold waves are more frequent than heat waves for the three datasets

229
The main difference between the datasets is the higher occurrence of both heat and cold waves for ERAI than 230 the other datasets. This could be an effect of the coarser resolution in time and space of the reanalysis compared 231 to the ground observations that tends to smooth the temporal evolutions of the temperature anomalies and so 232 of the quantiles. Due to that lower temporal variability, the chance to get long term anomalies is increased 233 when using ERAI as compared to the other datasets.

234
As the total number of occurrences is the sum of all individual waves, the distribution of the wave durations 235 is needed to complete the picture. Fig. 6 displays the spatial variability of the last quartile of the wave durations 236 recorded for each grid point. It appears that the difference between the durations of heat and cold waves 237 between the three different datasets is much lower than the difference of occurrence discussed previously ( in agreement for the intensities and the spatial variabilities. It is interesting to highlight that these intensities 283 are not well correlated to the occurrence, i.e., a region with more cases does not necessarily record the most 284 extreme events (Fig. 4 and 5). The relative short period of study (21 years) can generate some artefacts over 285 regions that recorded extraordinary events (especially true for Russia).

286
To assess the uncertainties relative to the methodology used, Fig. 10

Return periods 298
As the purpose of this study is to provide a methodology that is useable for a monitoring system that must be 299 robust and understandable for users and decision makers, the information will be provided in terms of return 300 periods. This product will quantify, at monthly time scale, the intensity of the cold or heat waves that have 301 occurred. To build this indicator, all the days defined as cold or heat waves are summed for different 302 accumulation periods (from monthly to seasonally, see Table 3). Monthly values characterize either one 303 specific event as defined previously or several consecutives cases. As indicated by WMO (2015)  and heat waves. In that section, the results are produced using LisFlood dataset, which has been validated in 317 the previous section, but similar results are obtained with the two other datasets.

319
The boxplots (in Fig. 11) indicate the relationships between intensities and return periods over each grid point 320 in Europe. According to size of the inter quartiles, it appears the large spatial variabilities over the domain.

321
For instance, heat waves with intensities of 20 (10)  The spatial variabilities are then analysed in more detail with a region classification. This classification is a 328 simplification of the one reported in the EEA report (2016) that takes into account the climatology of the 329 regions (Continental, Mediterranean, Oceanic, Scandinavian, small panels in Fig. 11). Over these regions, the 330 calculation of return periods are assessed and compared (coloured dots in Fig. 11 Based on these calculations, the monthly intensities are transformed into return periods, resulting in more 343 comprehensible information for users. In Fig. 12, the intensities of the heat and cold waves with a return period 344 of 10 years are plotted using I2 and I3. These values are sensitive to the distribution (number and intensities) 345 of the waves recorded during the 21 years analysed. That is why we observe the increase of intensity over 346 western Russia in Fig. 12 (left panels) where the waves are more frequent (Fig. 4 and 5) and more intense 347 (Fig. 9). The same results with I3 show a different behaviour Fig. 12 (right panels), mainly due to the change 348 of the most intense waves recorded and plotted Fig. 10. The potential impacts of these heat and cold waves 349 will be calculated regarding the absolute intensities and the return periods. But we could expect that identical France using I3 does not give extreme value regarding the intensities recorded in continental regions.

353
Nevertheless, the equivalent return value over France is larger than 50 years (not shown), in agreement with 354 Barriopedro et al. (2011) andTrigo et al. (2005), which suggest the potential strong risk associated.

355
Given the short period used in this study, the return periods cannot be as accurate as those ones reported in 356 previous studies, nevertheless, the information allows identifying the most extreme situations. The same 357 information is also available for the 2-month and seasonal time scales (not shown). The developed 358 methodology will allow to provide a robust and understandable indicator that is standardized by the local 359 climatology.

371
This is probably due to the good agreement in between the two datasets for Tmin and Tmax. Using ERAI 372 some differences appeared mainly due to the coarser resolution of the original grid and the use of only 4 values 373 per day to define Tmin and Tmax. In this case, the persistency and the spatial correlation were increased, 374 generating less spatial distinction and more intense waves than using the first two datasets. However, the main 375 results are in overall agreement for all three datasets and show an increased hazard for heat and cold waves in 376 the continental part of Europe. The data are also in agreement when transforming the intensities into return 377 periods. These relationships will be used operationally in the EDO website to provide robust and 378 comprehensible information for decision makers and users.

379
In perspective, these datasets and results should be compared to the results derived from forecast products in                 Nat. Hazards Earth Syst. Sci. Discuss., doi:10.5194/nhess-2017-181, 2017 Manuscript under review for journal Nat. Hazards Earth Syst. Sci. Discussion started: 29 May 2017 c Author(s) 2017. CC-BY 3.0 License. Figure 11 Return periods of monthly intensities of heat (top) and cold (bottom panels) waves for two intensities 609 (I2, left panels and I3, right panels). Boxes assess the spatial variability for the grid points. Coloured dots 610 indicate the return period calculated over the regions defined in the small panels.