Deep winter cyclones over the Eastern Mediterranean are accompanied by extreme western wind events over Israel e.g [1Navarra A, Tubiana L. Regional Assessment of Climate Change in the Mediterranean 2013., 2Lionello P. The clime of the Mediterranean Region 2012.]. A critical effect of these extreme winds is the erosion of the coastline and the need for civil authorities to prepare for this challenge [3Kit K. Marine Policy Plan for Israel 2014.http:// msp-israel.net.technion.ac.il /files/2014 /12/ATOS-report -final.pdf-5Herrmann M, Somot S, Calmanti S, Dubois C, Sevault F. Representation of spatial and temporal variability of daily wind speed and of intense wind events over the Mediterranean Sea using dynamical downscaling: Impact of the regional climate model configuration. Nat Hazards Earth Syst Sci 2011; 11: 1983-2001.
[http://dx.doi.org/10.5194/nhess-11-1983-2011] ]. A possible way is to use weather models; however, this effort demands extensive resources. Hence, statistical downscaling of extreme surface wind events is needed. Global climate models often underestimate the wind at 1000 hPa [5Herrmann M, Somot S, Calmanti S, Dubois C, Sevault F. Representation of spatial and temporal variability of daily wind speed and of intense wind events over the Mediterranean Sea using dynamical downscaling: Impact of the regional climate model configuration. Nat Hazards Earth Syst Sci 2011; 11: 1983-2001.
[http://dx.doi.org/10.5194/nhess-11-1983-2011] -7Lavaysse C, Vrac M, Drobinski P, Lengaigne M, Vischel T. Statistical downscaling of the french mediterranean climate: Assessment for present and projection in an anthropogenic scenario. Nat Hazards Earth Syst Sci 2012; 12: 651-70.
[http://dx.doi.org/10.5194/nhess-12-651-2012] ], and sometimes the 1000 hPa is not available [8Hochman A, Harpaz T, Saaroni H, Alpert P. Synoptic classification in 21st Century CMIP5 predictions over the Eastern Mediterranean with focus on cyclones, accepted for publication Int. J Clim 2017.]. Hence, the ability to infer surface winds from low troposphere data should be considered. This work examines the feasibility to employ data from low troposphere climate models in order to statistically downscale wind according to the semi objective synoptic classification.

The definition of the semi objective synoptic classes [9Alpert P, Osetinsky I, Ziv B, Shafir H. Semi-objective classification for daily synoptic systems: Application to the Eastern Mediterranean climate change. Intern J Climatol 2004; 24: 1001-11.http:// www.tau.ac.il/ ~pinhas/papers/2004 /Alpert_et_al _IJC_2004a.pdf, 10Osetinsky I. Climate changes over the E. Mediterranean-A synoptic systems classification approach. Ph.D. thesis, Tel Aviv University, 2006: 153 pp.] according to General Circulation Models (GCM) data [8Hochman A, Harpaz T, Saaroni H, Alpert P. Synoptic classification in 21st Century CMIP5 predictions over the Eastern Mediterranean with focus on cyclones, accepted for publication Int. J Clim 2017.] and the relation between synoptic classes and historical measured surface winds will enable the classification of future surface wind fields up to the next century. Furthermore, under winter lows over Israel, the ability to directly infer from low troposphere wind data suggests a simple alternative to downscaling extreme western wind events according to the semi objective synoptic classification.

The ability to use the semi objective synoptic class [9Alpert P, Osetinsky I, Ziv B, Shafir H. Semi-objective classification for daily synoptic systems: Application to the Eastern Mediterranean climate change. Intern J Climatol 2004; 24: 1001-11.http:// www.tau.ac.il/ ~pinhas/papers/2004 /Alpert_et_al _IJC_2004a.pdf] as a predictor for hourly surface wind along the central and the southern Israeli coast has been demonstrated [11Berkovic S. Synoptic classes as a predictor of hourly surface wind regimes: The case of the central and southern israeli coastal plains. J App Met Climatol 2016; 55(7): 1533 2016; 55(7): 1533. http://dx.doi.org/10.1175/JAMC-D-16-0093.1]. This approach is based on the idea that a given circulation pattern is often associated with similar local-scale meteorological conditions e.g. [12Zorita E, von Storch H. The analog method as a simple down-scaling technique: Comparison with more complicated methods. J Clim 1999; 12: 2474-89.
[http://dx.doi.org/10.1175/1520-0442(1999)012<2474:TAMAAS>2.0.CO;2] , 13Huth R. Disaggregating climatic trends by classification of circulation patterns. Int J Climatol 2001; 21: 135-53.
[http://dx.doi.org/10.1002/joc.605] ]. The current study extends [11Berkovic S. Synoptic classes as a predictor of hourly surface wind regimes: The case of the central and southern israeli coastal plains. J App Met Climatol 2016; 55(7): 1533 2016; 55(7): 1533. http://dx.doi.org/10.1175/JAMC-D-16-0093.1] in order to check whether it is possible to infer surface winds along the Israeli coast from high level winds (850 or 925 hPa) under the various semi objective synoptic classes and seasons at 0, 12Z. For this purpose, the horizontal wind turning at the low troposphere levels is examined. The study is limited to Beit Dagan, a representative station near the homogeneous central Israeli coastline (Fig. 1). The limitation to a single station is derived from the fact that radiosonde data are scarce, and only available for this station.

Fig. (1) The EM region and the study area showing Beit Dagan (BD) (~ 30 m ASL) and the 4 closest ERA-Interim grid points. Pt3 over the Mediterranean Sea, Pt1 on the seashore like BD but further south (~63 km), Pt4 and Pt2 are inland points over complex topography. Pt2 is on the eastern slope of the Judea Mountain, Pt4 to the North of Samaria Mountain (~ 600-800 m ASL).

The most prevalent geostrophic wind over the east Mediterranean has a westerly component [10Osetinsky I. Climate changes over the E. Mediterranean-A synoptic systems classification approach. Ph.D. thesis, Tel Aviv University, 2006: 153 pp.]. The sea breeze and upslope winds (Fig. 1), blowing from the sea to the land, during daytime have a westerly component, while the land breeze and downslope winds have an easterly component. At 12Z the mesoscale and the synoptic winds have similar directions (W component winds), therefore, the possibility to infer surface wind from higher level wind is expected to be successful. During the night hours the meso wind direction usually opposes the synoptic wind direction [11Berkovic S. Synoptic classes as a predictor of hourly surface wind regimes: The case of the central and southern israeli coastal plains. J App Met Climatol 2016; 55(7): 1533 2016; 55(7): 1533. http://dx.doi.org/10.1175/JAMC-D-16-0093.1, 14Berkovic S. Winter wind regimes over Israel using self-organizing maps. J Appl Meteorol Climatol 2017; 56: 2671-91.
[http://dx.doi.org/10.1175/JAMC-D-16-0381.1] , 15Skibin D, Hod A. Subjective analysis of mesoscale flow patterns in northern Israel. J Appl Meteorol 1979; 18: 329-38.
[http://dx.doi.org/10.1175/1520-0450(1979)018<0329:SAOMFP>2.0.CO;2] ] and a study of the Cross Angle (CA) at 0Z will reveal the relation between the surface wind and 925 or 850 hPa winds at this hour.

The boundary layer height over Israel is 1- 2 km [16Dayan U, Shenhav R, Graber M. The spatial and temporal behavior of the mixed layer in Israel. J Appl Meteorol 1988; 27: 1382-94.
[http://dx.doi.org/10.1175/1520-0450(1988)027<1382:TSATBO>2.0.CO;2] ]. Accordingly, the 925 hPa (~ 800 m) level is around or lower than the mixing layer height, while the 850 hPa (~1.5 km) level is mostly above the mixing height. Zaiter and Al-Jumaily [17Zaiter DS, Al-Jumaily KJ. Characteristics of thermal wind over iraq. Inter J Sci Res Publ 2016; 6: 163-7.] showed that the wind at 900-850 hPa at latitude 32º N, longitude 30º -50º E obeys the thermal wind equation according to NCEP data. We assume similar behavior according to ERA-Interim data. The ability to use reanalysis data was verified by comparing measurements to ERA-Interim data at the three levels according to season, hour and synoptic class. To the best of our knowledge, no such extensive quantitative verification has been made previously for our area. This verification gives an indication of the ability to use model data to infer the surface wind from low troposphere levels along the Israeli coast.

Section 2 will describe the method and data used in this study, section 3 will briefly describe the synoptic classes, and sections 4-6 will display the results of the CA histograms and verify ERA-Interim wind prediction. Summary and conclusions follow in sections 7 and 8.

Time series of radiosonde data over Israel are scarce and limited to Beit Dagan (BD), a station on the Israeli coastal plain with relatively flat topography. Therefore, the study is limited to this location at the hours 0, 12Z. BD radiosonde data at 0, 12Z at the 1000, 925 and 850 hPa pressure levels from the years 1974-2012 was downloaded from weather.uwyo.edu [18 weather.uwyo.edu weather.uwyo.edu http ://weather.uwyo.edu/ upperair/]. The available variables of the radiosonde data are: pressure, height, temperature, humidity and wind. Wind speeds were archived in full knots, therefore after converting into m/sec units the wind speed uncertainty is ±0.25 m/s.

A dataset of the synoptic classification for the investigated location and time period has been performed by our research group members according to the published synoptic classification of Alpert et al. [9Alpert P, Osetinsky I, Ziv B, Shafir H. Semi-objective classification for daily synoptic systems: Application to the Eastern Mediterranean climate change. Intern J Climatol 2004; 24: 1001-11.http:// www.tau.ac.il/ ~pinhas/papers/2004 /Alpert_et_al _IJC_2004a.pdf] and Osetinsky [10Osetinsky I. Climate changes over the E. Mediterranean-A synoptic systems classification approach. Ph.D. thesis, Tel Aviv University, 2006: 153 pp.].

The relation between synoptic classes and surface winds was previously defined [11Berkovic S. Synoptic classes as a predictor of hourly surface wind regimes: The case of the central and southern israeli coastal plains. J App Met Climatol 2016; 55(7): 1533 2016; 55(7): 1533. http://dx.doi.org/10.1175/JAMC-D-16-0093.1]. This has led us to a further study to investigate whether the 1000 hPa wind can be inferred from the 925 or 850 hPa wind according to synoptic class and season at each hour (12Z or 0Z). The surface pressure at BD is mostly 1000 -1010 hPa, and therefore 1000hPa wind is very similar to the surface wind.

The synoptic classification is the semi–objective classification derived by Alpert et al. [9Alpert P, Osetinsky I, Ziv B, Shafir H. Semi-objective classification for daily synoptic systems: Application to the Eastern Mediterranean climate change. Intern J Climatol 2004; 24: 1001-11.http:// www.tau.ac.il/ ~pinhas/papers/2004 /Alpert_et_al _IJC_2004a.pdf] determined according to temperature, geopotential, and horizontal wind components at 1000 hPa over the East Mediterranean (EM) from NCEP/NCAR data. The frequent synoptic classes, presented in Table 1, were selected in order to ensure a large number of instances (> 30) under each season and synoptic class at each hour. A description of the synoptic classes is given in the following section.

Table 1
The 14 most frequent (out of total 19) synoptic classes according to Alpert et al. 2004, Osetinsky 2006.

In order to study the low troposphere winds along the Israeli coast, the following methodology was performed:

Firstly, the wind turning was studied according to the Cross Angle (CA) between the wind vectors of the two levels 1000, 925 hPa or 1000, 850 hPa (abbreviated as CA925, CA850). The CA is the angle created due to the subtraction of the lower level wind vector from the upper level wind vector. The wind may be veering or backing (clockwise or counterclockwise), therefore, the sign of the turning is kept and designated by + or – respectively. This definition is after the well-known thermal wind relationship between geostrophic wind direction vertical shear and horizontal temperature advection [19Holton JR. An Introduction to Dynamic Meteorology 4^th ed. 2004.-23Wang S, O’Neill LW, Jiang Q, et al. A regional real time forecast of marine boundary layers during VOCALS-Rex. Atmos Chem Phys 2011; 11: 421-37.
[http://dx.doi.org/10.5194/acp-11-421-2011] ]. Since wind direction can be determined meaningfully only when the wind speed is higher than 0.5 m/s, CA925 or CA850 was calculated if the wind speed at the two relevant levels were greater than 0.5 m/s. In order to facilitate the verification of wind direction uniformity at two levels a small CA criterion has been set. A previous study of surface wind steadiness along the Israeli coast according to synoptic class [11Berkovic S. Synoptic classes as a predictor of hourly surface wind regimes: The case of the central and southern israeli coastal plains. J App Met Climatol 2016; 55(7): 1533 2016; 55(7): 1533. http://dx.doi.org/10.1175/JAMC-D-16-0093.1] has shown that a high degree of wind steadiness is obtained under winter lows, and then the standard deviation of the wind direction is ~ 60°. Therefore, our criterion for a small CA (SC) value is the % of events bounded by 60°. The maximal CA in this study is 180º. Under all the studied events, the pressure at the station-level was equal to, or greater than 1000 hPa (1000-1027 hPa).

The correlation between the wind speed at the two levels 1000, 925 hPa or 1000, 850 hPa was verified according to scatter plots (not shown) and the Pearson correlation was calculated (python software pearsonr function from scipy.org [24 Scipy.org https:// docs.scipy.org /doc/scipy -0.14.0/reference /generated/ scipy.stats.pearsonr.html]). The Pearson correlation coefficient measures the linear relationship between two data sets and requires that each dataset be normally distributed.

Secondly, a comparison between ECMWF (European Centre for Medium-Range Weather Forecasts) ERA-Interim (European Reanalysis) analyzed wind [25Apps.ecmwf.int. ERA Interim, Daily. [online] 2017. Available at: http:// apps.ecmwf.int/ datasets/data /interim-full- daily/, 26Berrisford P, Dee DP, Poli P, et al. The ERA-interim archive version 2.0 2011. Available from: http:// www.ecmwf.int/ en/elibrary/ 8174-era-interim -archive-version-20] and the radiosonde data was performed by bilinear interpolation of data at the grid points closest to BD at each pressure level. The native model resolution is ~ 80 km. The downloaded interpolated data horizontal resolution is 0.75° (~ 83 km). (Fig. 1) displays the location of BD and its closest ERA-Interim grid points. Point3 (Pt3) is over the Mediterranean Sea, point1 (Pt1) is on the seashore as is BD, point4 (Pt4) and point2 (Pt2) are inland points over complex topography. The terrain height at BD (lat 32.0073, lon 34.8138) is 31m ASL. The Samaria and Judea mountains run north to south, east of BD. The peak height is 400 meters above sea level.

The comparison of ERA-Interim vs. BD radiosonde horizontal wind at 1000, 925 and 850 hPa was performed firstly according to seasons and hours, and secondly according to seasons, synoptic classes and hours. Scatter plots of wind speed and direction presenting measurements versus reanalysis values were drawn; however, they are too numerous to be shown. The statistical parameters that were derived from the scatter plots - bias, RMSE, correlation and the linear interpolation slope (a) and intercept (b) - will be summarized and presented. For the second comparison which included the synoptic classes only RMSE will be presented.

This description of synoptic groups and classes was given in [9Alpert P, Osetinsky I, Ziv B, Shafir H. Semi-objective classification for daily synoptic systems: Application to the Eastern Mediterranean climate change. Intern J Climatol 2004; 24: 1001-11.http:// www.tau.ac.il/ ~pinhas/papers/2004 /Alpert_et_al _IJC_2004a.pdf-11Berkovic S. Synoptic classes as a predictor of hourly surface wind regimes: The case of the central and southern israeli coastal plains. J App Met Climatol 2016; 55(7): 1533 2016; 55(7): 1533. http://dx.doi.org/10.1175/JAMC-D-16-0093.1] and is briefly repeated here for the benefit of the reader. Table 1 summarizes the main 4 groups and their classes. Full presentation of the synoptic classes is given in Fig. (2) [11Berkovic S. Synoptic classes as a predictor of hourly surface wind regimes: The case of the central and southern israeli coastal plains. J App Met Climatol 2016; 55(7): 1533 2016; 55(7): 1533. http://dx.doi.org/10.1175/JAMC-D-16-0093.1].

There are two main seasons in the EM-the dry season in summer (mid-June–September) and the wet season in winter (December–February)-as well as the “transition months” (April–May, October–November). Five synoptic groups characterize this climate. A single group is dominant during the summer, and four groups prevail during the winter and the transition months. The fifth group “sharav lows” is rare and was not included in our study.

1. The Persian trough – abbreviated PT-is a persistent synoptic class during the summer months (mid-June–September). This trough extends from the Asian monsoon through the Persian Gulf along southern Turkey and the Aegean Sea. The depth of the PT changes during the summer. Deep troughs reach Greece while shallow troughs do not reach the Mediterranean Sea. There is practically no rain during the summer. Since the synoptic situation is static, a remarkable daily periodicity is obtained in the EM during the summer. The flow over the area is mainly determined by mesoscale and microscale processes. Northwesterly synoptic winds flow over the EM as a result of the PT and the NW Etesian winds. During early morning hours, when the temperature gradient between air and sea is minimal, the synoptic wind is mostly W-SW.

2. The winter Lows (Ls) and Cyprus Lows (CLs) form the dominant synoptic group during winter (Fig. 2, first row). They are responsible for the largest amount of rain in the EM. CLs are the main contribution to the winter lows, each contributing rainfall during two to three consecutive days. The name CL is given to Mediterranean cyclones that reach the EM; they need not be located over Cyprus itself to be designated CLs. Deep lows are characterized by strong (max ~10 m/s) westerly winds.

3. Highs:

a) The “Siberian highs” (winter highs) occur mainly during the cool season. Siberian highs are highs of a northern origin: Europe (e.g., Romania, Turkey and Caucasia) or Siberia (the winter Asiatic monsoon) itself. The extension of the Siberian high is rare, occurring once every few years. It is accompanied by cold, dry northeasterly winds that last for several days. Winter highs bring fair weather, inducing cold, dry northerly to easterly winds over Israel.

b) The “subtropical high” (the High to the west of Israel [H_W]) may occur throughout the year. It is an extension of the Azores high, which prevails over North Africa. When this high is extended further to the east and reaches Israel, it brings NW winds with humid air from the sea.

4. The Red Sea Trough (RST), a trough over the Red Sea and its surroundings, is an extension of a low over the Sudan, which may occur during the transition months or during winter. It is characterized by a dry easterly flow east of the trough axis. When an upper trough stretches over the EM, the RST develops northward and deepens. As a result, rain clouds may develop. Convective storms may evolve and cause floods over southern Israel and Sinai, especially with ground heating during the day.

Fig. (2) Typical examples for the EM frequent (more than 5 days/year) synoptic classes as defined by Alpert et al. 2004. Each map is a SLP (hPa) map from a representative day (Osetinsky 2006). The frequency of each synoptic class (number of days per year) is denoted above each SLP map. The winter lows are presented in the first row: Low east deep (LE-D), Deep Cyprus low from the north (CLN-D), cold low from the west (LW). Highs are presented in the second row: High from the north (HN), High from the west (HW) and high over Israel (HC). Troughs are presented in the third row: Medium Persian trough (PT-M), Red sea trough to the east (RSTE) and Red sea trough center (RSTc). The subfigures were downloaded from http:// www.esrl.noaa.gov/ psd/ data/ gridded/ data.ncep.reanalysis.html

The fifth group “sharav lows” is rare and was not included in our study.

The subdivision of the four major synoptic groups provides the 14 most frequent classes referred to in this work. The divisions are made according to the location/depth and the field gradients of the classes over the EM region.

The CL group includes the following classes: The Deep CL to the north (CLN-D) and the Shallow CL to the north (CLN-S). The L group includes these classes: the Cold low to the west (LW), the Deep low to the east (LE-D) and the Shallow low to the east (LE-S).

The RST group includes these classes: the RST with an eastern axis (RSTE) and the RST with a central axis (RSTC). The location of the axis is defined relative to the EM coastline.

PTs are divided into three classes: deep, medium and shallow (respectively, PT-D, PT-M and PT-W). Siberian highs (winter highs) are divided into three classes: north, east and over Israel (respectively, HN, HE and HC). The summer subtropical high is known as H_W. The names are based on the location of the pressure gradient.

This work has considered two aspects pertaining to the feasibility of statistically downscaling surface wind according to synoptic class, season and hour from 925, 850 hPa wind with the stress on extreme wind events: 1. wind veering and wind speed correlation in the low stratosphere from radiosonde data 2. the ability of ERA-Interim analysis to predict low stratosphere wind. The CA925, CA850 and wind speed correlations were calculated from radiosonde data at Beit Dagan (BD) a station on the Israeli coastal plain. Afterwards, a comparison of ERA-Interim to radiosonde horizontal wind at 1000, 925 and 850 hPa levels was performed according to season, hour (0, 12Z) and synoptic class. The comparison was performed at each pressure level individually in order to facilitate comparison to open literature works and to avoid error cancellation. To the best of our knowledge no previous quantitative verification of wind profile qualities has been performed for this area.

Two extreme categories were found. Under the first category, “low troposphere inversion”, strong thermal inversion prevents efficient interaction between these two levels and the CA values are as high as ±180º. 75-100% of the CA925 under PT’s and Hw magnitudes were smaller than 60°; however, only 49-70% CA850 magnitudes under Hw were smaller than 60°. Very low correlation between wind speeds at 1000 and 925 hPa levels was found under PT-W, PT-M and Hw during summer, the wind speed at each of these levels is relatively weak 4± 2 m/s at most of the events (according to scatter plots of wind speed, not shown). In the second “cyclone activity” category, as expected, strong vertical mixing creates higher resemblance between the winds along the vertical profile. In this category, 65-93% of the CA925 and CA850 magnitudes are smaller than 60º. Under extreme wind events of winter lows at 925 and 850 hPa at 12Z 72-100% of the CA925 or CA850 magnitudes are smaller than 60º. High correlation is found between the wind speeds at the two sets of levels (0.6-0.8).

At 0Z, due to increased thermal stability next to the surface, larger values of CA850 and CA925 were obtained under all synoptic classes, and only 49-82% of CA925 and 35-68% of CA850 magnitudes were smaller than 60°. Extreme wind events under winter lows at 0Z still show relatively high CA values, 60-88% of CA925 and only 42-74% of CA850 magnitudes are lower than 60°. Therefore direct inference of wind direction from 925, 850 hPa data at 0Z under winter lows is questionable. On the other hand, wind speed inference may still be possible, due to high correlation between wind speeds at the two sets of levels during the night.

To summarize, an inspection of the wind speed correlation and CA at 12Z reveals that under winter lows the wind speed correlation is high and the turning of the wind is in general small, which may allow derivation of wind close to the ground from the higher levels. At 0Z, high correlation of wind speed at the two sets of levels was found under winter lows, i.e. wind speed inference may be possible. At 12Z relatively small CA925 and very low wind speed correlation were found under high to the west and PTs, i.e. wind direction inference may be possible.

A thorough study of the ERA-Interim prediction of wind speed and direction at 1000, 925, 850 hPa according to season and hour has shown good prediction of the wind direction at 925 and 850 hPa (RMSE 20°-60°) at 12Z and (RMSE 20°-40°) at 0Z. At 1000 hPa at 12Z good prediction is obtained for all synoptic classes (20°- 60°), especially under PT’s and Hw during JJA (20°-30°). However, at 0Z, the quality of the prediction is lower (RMSE 60°-90°) for all seasons and synoptic classes. Referring to the wind speed prediction, the RMSE is 1-2 m/s for all seasons and synoptic classes except winter lows with 2-3 m/s RMSE at 0, 12Z at all levels. According to scatter plots, the reanalysis under-predicts the wind speed for all seasons. The slope is 0.5-0.8 at all seasons and levels except JJA. During JJA at 1000 hPa the mesoscale effects were not detected by the analysis and constant wind speed ~ 4 m/s is obtained. i.e., for 1000 hPa during JJA at 12Z better prediction of wind direction and poorer prediction of wind speed is obtained.

According to the CA calculations, wind speed correlation and the analysis verification, inference of surface wind at 12Z from 925 or 825 hPa winds may be possible under winter lows. Wind direction inference may be possible at 12Z from 925 hPa winds under high to the west and Persian troughs. Wind speed inference may be possible under lows at 0Z. As a result of the under estimation of the wind speed at the higher levels (925, 850 hPa), wind speed should be inferred by interpolation, according to historical data from measurements or high resolution model. Under other synoptic classes, surface winds may be estimated according to statistics of measurements (Berkovic 2016). Around 50% of the 3 hourly events under the frequent synoptic classes can be characterized by their average wind surface data with relatively small standard deviation (std) of wind direction (std < 60º) and speed (std ~ 1-2 m/s).

Other objective statistical downscaling methods and other predictors such as mixing height, wind, temperature and humidity should be considered. A few examples of objective statistical algorithms [45Trzaska S, Schnarr E. A review of downscaling methods for climate change projections. CIESIN report 2014. http:// www.ciesin.org /documents/ Downscaling_ CLEARED_000.pdf, 46Vandal T, Bhatia U, Ganguly AR. Statistical downscaling in climate with state-of-the-art scalable machine learning Ch.4. in Large-Scale Machine Learning in the Earth Sciences Edited by Ashok N. Srivastava, Ramakrishna Nemani and Karsten Steinhaeuser 2017.] are SOM [14Berkovic S. Winter wind regimes over Israel using self-organizing maps. J Appl Meteorol Climatol 2017; 56: 2671-91.
[http://dx.doi.org/10.1175/JAMC-D-16-0381.1] , 47Horton DE, Johnson NC, Singh D, Swain DL, Rajaratnam B, Diffenbaugh NS. Contribution of changes in atmospheric circulation patterns to extreme temperature trends. Nature 2015; 522(7557): 465-9.
[http://dx.doi.org/10.1038/nature14550] [PMID: 26108856] , 48Yin C. Application of SOM to statistical downscaling of major regional climate variables 2011. Available from: http:// researchcommons. waikato.ac.nz/ handle/ 10289/ 5733], Bayesian inference methods [43Salameh T, Drobinski P, Vrac M, Naveau P. Statistical downscaling of near-surface wind over complex terrain in southern France. Meteorol Atmos Phys 2009; 103: 253-65.
[http://dx.doi.org/10.1007/s00703-008-0330-7] , 49Manor A, Berkovic S. Bayesian Inference aided analog downscaling for near-surface winds in complex terrain. Atmos Res 2014; 164-165: 27-36.
[http://dx.doi.org/10.1016/j.atmosres.2015.04.014] ], non-parametric regression based on generalized additive models [40Zecchetto S, Accadia C. Diagnostics of T1279 ECMWF analysis winds in the Mediterranean basin by comparison with ASCAT 12.5 km winds. Q J R Meteorol Soc 2014; 140: 2506-14.
[http://dx.doi.org/10.1002/qj.2315] ] and Cumulative Distribution Function transform [7Lavaysse C, Vrac M, Drobinski P, Lengaigne M, Vischel T. Statistical downscaling of the french mediterranean climate: Assessment for present and projection in an anthropogenic scenario. Nat Hazards Earth Syst Sci 2012; 12: 651-70.
[http://dx.doi.org/10.5194/nhess-12-651-2012] ]. Dynamical downscaling may provide better prediction of mesoscale effects and will enable direct prediction of surface winds at the expense of much heavier computational effort at higher horizontal resolution (1-3 km). High resolution models in climate projections have started to be employed in recent years [50Mizielinski MS, Roberts MJ, Vidale PL, et al. High-resolution global climate modelling: The UPSCALE project, a large-simulation campaign. Geosci Model Dev 2014; 7: 1629-40.
[http://dx.doi.org/10.5194/gmd-7-1629-2014] -52Flaounas E, Drobinski P, Bastin S. Dynamical downscaling of IPSL-CM5 CMIP5 historical simulations over the Mediterranean: Benefits on the representation of regional surface winds and cyclogenesis. Clim Dyn 2012.
[http://dx.doi.org/10.1007/s00382-012-1606-7] ]. According to Li et al. [53Li H, Kanamitsu M, Hong SY, Yoshimura K, Cayan CR, Misra V. A high-resolution ocean-atmosphere coupled downscaling of the present climate over California. Clim Dyn 2014; 42: 701-14.
[http://dx.doi.org/10.1007/s00382-013-1670-7] ] high resolution and ocean-atmosphere coupled models are necessary to predict future climate in coastal areas.