Introduction
Climate reanalyses are an important source of information for monitoring
climate and for the validation and calibration of numerical weather
prediction (NWP) models but also have vast uses outside of meteorology and
climatology. Because they are carried out using a fixed version of a forecast
model and a data assimilation system which utilises historical observations,
they produce parameters that are physically consistent and often not
routinely observed. Thus, climate reanalyses have the potential to extend the
knowledge gained from current observation networks.
Atmospheric reanalyses originated with the production of datasets by ECMWF
and GFDL for the 1979 Global Weather Experiment .
Since then, the production of global datasets has evolved with many being
updated close to real time (e.g. ERA-Interim (, 1979–present,
∼ 79 km), JRA-55 (, 1958–present,
∼ 65 km), the Modern-Era Retrospective Analysis for Research and
Applications (MERRA (, 1979–present,
∼ 60 km), the NCEP/NCAR global reanalysis (,
1957–1996, ∼ 210 km) and the ERA-40 global reanalysis
(, 1957–2002, ∼ 125 km)).
More recent global reanalyses use atmosphere-land-ocean coupled systems or
earth system models (e.g. ECMWF's ERA-20C and ERA-20CM
(, 1900–2010, ∼ 125 km)).
ECMWF's latest reanalysis product ERA5 (, 1979–present,
∼ 31 km) has entered production. Although not coupled to an ocean
model, it is more advanced in other ways: uncertainty is estimated using a
10-member ensemble for data assimilation, it contains CMIP5
greenhouse gas concentrations, volcanic eruptions,
SSTs and sea-ice concentrations and additional observations and satellite
data.
(a) MÉRA (2.5 km grid spacing) and
(b) ERA-Interim (79 km grid spacing) orographies. Over sea
grid-points the geopotential fields used to plot the model orographies show
“spectral ripples”, which are a reflection of the fact that the ERA-Interim
and HARMONIE models are defined in spectral space.
(a) Time-series of the number of upper-air temperature
observations used by the 3D-Var data assimilation system in HARMONIE-AROME.
(b) Time-series of average first-guess (OB-FG) and analysis (OB-AN)
departure statistics for aircraft temperature observations assimilated by
HARMONIE-AROME where mean OB-FG values are shown in blue and mean OB-AN
values are shown in red. A 7-day running average is used in (a) and
(b) for the period 1981–2014.
Due to computational constraints global reanalyses cannnot be run at the very
high resolutions required to resolve mesoscale processes but are generally
used to provide boundary conditions for regional reanalyses. The advantage of
regional reanalyses is that they can be run at high temporal and spatial
resolution so that focus can be put on near surface parameters, extremes and
frequency distributions. Several regional reanalyses have already been
produced including, for example, the Arctic Reanalysis ASR
(, 2000–2010, 30 km), the Baltic Sea reanalysis
(, 1965–2005, 11 km), COSMO-REA6
(, 1997–2004, 6 km), the HIRLAM reanalysis
(, 1979–2014, ∼ 22 km), the North American
Regional Reanalysis (, 1979–2003, ∼ 32 km)
and the South Asian Regional Reanalysis (, 1998–2002,
∼ 30 km) among many others. The EU European Reanalysis and
Observations for Monitoring project (EURO4M: )
combined results from several model-based European regional reanalyses as
well as observations from satellites and ground-based stations. Its
successor, the ongoing FP7 (Framework Programme 7) funded reanalysis project,
UERRA (Uncertainties in Ensembles of Regional ReAnalyses;
), will include the recovery of historical data and the
estimation of uncertianties in reanalyses.
Time-series of (a) first-guess (O-B) and (b)
analysis (O-A) departure statistics for radiosonde temperature observations
assimilated by HARMONIE-AROME at a selection of standard pressure levels. A
7-day running average is used in (a) and (b) for the period
1981–2014.
Spin-up soil moisture (red) compared with production soil moisture
(blue) for (a) 20 cm below the surface and (b) 300 cm
below the surface. The comparison includes the years 2000, 2005 and 2010.
(c) and (d) are similar but show soil temperature at the
surface and 20 cm below the surface. As soil temperatures spin-up quickly
only days 1 to 7 are shown. In each subplot the production versus spin-up
differences are shown in green.
Time-series of verification scores for MÉRA and ERA-Interim 3 h
forecasts of (a) MSLP, (b) 2 m temperature,
(c) 10 m wind speed compared with SYNOP observations. The average
error, or bias, is shown in red (MÉRA) and maroon (ERA-Interim) and the
standard deviation of the errors is shown in navy (MÉRA) and light blue
(ERA-Interim). A 1-month running average is applied to the data.
Despite the abundance of reanalyses for Europe, no very high resolution
dataset has been produced for Ireland. To address this, we have run the
HARMONIE-AROME canonical configuration of the ALADIN-HIRLAM system on a
2.5 km horizontal grid for the 35-year period 1981–2015 forced by
ERA-Interim lateral boundary conditions, with the aim of improving our
knowledge of surface parameters and extremes. Inspiration for this project
came from KNMI's production of a 35-year wind atlas for the North Sea
(KNW-atlas; 1997–2013; ) which was also
produced using the ALADIN-HIRLAM system. The Irish reanalysis dataset,
called MÉRA (Met Éireann ReAnalysis) will extend the knowledge gained
from observations as the model grid is much finer than observational coverage
over Ireland. It was carried out using the same domain as has been used
operationally by Met Éireann since 2011. Centred over the Island of
Ireland, the domain covers Ireland, the United Kingdom and an area of
northern France, see Fig. a. The extra
topographical information gained by using the 2.5 km grid is clear when
compared with the ERA-Interim grid (∼ 79 km, Fig. b).
MÉRA data will be available for use in Spring 2017.
This paper provides an overview of the MÉRA project including the shared
ALADIN-HIRLAM system, the data assimilation methods, the observations used
and an initial evaluation. It is laid out as follows. Section 2 provides
details on the model configuration and MÉRA simulations. Section 3
focusses on the data assimilation system. Results of the spin-up simulations
and verification of surface parameters are given in Sect. 4 with discussions
and conclusion in Sect. 5.
MÉRA: model configuration and simulation details
Details regarding the HARMONIE-AROME model configuration and data
assimilation system are given in Sect. ; information on
the simulations is provided in Sect. .
HARMONIE-AROME configuration used for MÉRA.
Model version
HARMONIE-AROME 38h1.2
Domain
540 × 500 grid points (Δx= 2.5 km)
Vertical levels
65 levels up to 10 hPa, first level at 12 m
Forecast cycle
3 h
Data assimilation
Optimal interpolation for surface parameters
3D-Var assimilation for upper air parameters
Observations
Pressure from SYNOP, SHIP and DRIBU
Temperature and winds from AIREP and AMDAR
Winds from PILOT
Temperature,winds and humidity from TEMP
Forecast
3 h forecasts, but a 33 h forecast at 00:00 Z
Summary of output available on pressure, height and sub-surface
levels.
Level type
Parameters
Levels
Pressure
Temperature, wind, cloud,
100, 200, 300, 400, 500, 600,
relative humidity, geopotential
700, 800, 850, 900, 925, 950, 1000 hPa
Height above ground
Temperature, wind, relative humidity
30, 50, 60, 70, 80, 90, 100, 125,
150, 200, 300, 400 m
Sub-surface
Temperature, moisture, ice
0, 20, 300 cm (below the surface)
Surface
Precipitation diagnostics
Surface
Diagnostic
Screen level parameters
2 m for temperatures and 10 m for winds and gusts
Surface
Radiative and non-radiative fluxes
Surface
Top of atmosphere
Radiative and non-radiative fluxes
Nominal top of atmosphere
MÉRA and observed soil temperature 20 cm below the surface for
the period 1981–2014. (a) Soil temperature by day of year,
(b) time series where a 2-month running average is applied,
(c) scatter plot of observed versus MÉRA soil temperatures
(colours are indicative of the number of observations).
Time-series of verification scores for MÉRA and ERA-Interim
12 h forecasts of (a) geopotential and (b) temperature
compared with radiosonde observations at 500 hPa. The average error, or
bias, is shown as dashed lines for both MÉRA and ERA-Interim. The
standard deviation of the errors is shown as continuous lines. A 2-month
running average is applied to the data.
MÉRA minus gridded rainfall observations for the period
1981–2012 by season (a) MAM, (b) JJA, (c) SON and
(d) DJF.
Bias and standard deviation of the residuals in 24 h precipitation
accumulations for MÉRA and ERA-Interim for the period 1981–2014 (2-month
running averaging is applied).
Hurricane Charley 25/26 August 1986: (a) 09:00 Z,
25 August 1986 to 09:00 Z on 26 August 1986 observed rainfall,
(c) 30 h forecast of the MÉRA rainfall accumulation from
00:00 Z on 25 August 1986, (e) 30 h forecast of the ERA-Interim
rainfall accumulation from 00:00 Z on 25 August 1986. 3 August 1997 bank
holiday precipitation (b) 09:00 Z on 4 August 1997 to 09:00 Z on
4 August 1997 observed rainfall (d) MÉRA rainfall accumulation
from 00:00 Z on 3 August 1997 to 00:00 Z on 5 August 1997.
(f) Era-Interim rainfall accumulation from 00:00 Z on 3 August 1997
to 00:00 Z on 5 August 1997. (e) and (f) are based on
successive 24 h accumulations.
Bias and standard deviations in (a) MSLP, (b) 2 m
temperature, (c) 10 m wind speed for the following model
configurations relative to observations: MÉRA, ERA-Interim (EI),
operational HARMONIE-AROME cycle 38h1 (HAR OP) and operational HIRLAM cycle
7.2 (HIR OP). 1-month running averaging is applied.
Model Configuration
The harmonie-38h1.2 version of ALADIN-HIRLAM system was used to carry out the
MÉRA simulations. This canonical configuration of HARMONIE-AROME was run
on a horizontal grid of 2.5 km spacing, using 65 vertical levels and a model
top of 10 hPa. The data assimilation component of the model is described in
and ; the forecast component is
described in with more recent updates included in
. The SURFEX externalised surface scheme
is used for surface data assimilation and the modelling of surface processes.
A summary of the model details is presented in Table .
Conventional observations (i.e. from synoptic stations, ships, buoys,
radiosonde ascents and aircraft) were assimilated. These observations are the
same as those used by the European Centre for Medium Range Weather Forecasts'
(ECMWF) ERA-Interim reanalysis. Locally available SYNOP observations were
used to fill gaps in the ERA-Interim SYNOP observation archive to ensure
cycle continuity and successful data assimilation each cycle. MÉRA uses a
3 h forecast cycle with surface and upper-air data assimilation
(cf. Sect. ). Three-hour forecasts were produced for each cycle
except the midnight (00:00 Z) cycle when a 33 h forecast was produced. This
provides a precipitation forecast each day that can be evaluated using
locally available observations of daily accummulations of precipitaion. The
33 h forecast was deemed necessary in order to avoid the negative impact of
model spin-up on the quality of short-range (3–4 h) precipitation
forecasts. It is generally acknowledged that model spin-up, in terms of
precipitation, is of the order of a few hours (e.g. ).
A more detailed analysis of this, in the context of MÉRA, is currently
underway.
No significant changes were applied to the default HARMONIE-AROME data
assimilation or forecast configurations used by the MÉRA project.
However, some tuning of the surface drag coefficient, CD, used by
SURFEX, was carried out following correspondence with Xiaohua Yang, personal
communication, (2014). Comparisons of forecast winds and gusts with equivalent
observations were used to evaluate the quality of near surface winds. Based
on the verification results from a series of month-long sensitivity tests,
CD was changed from a value of 0.01 to 0.025.
Simulation Details
ERA-Interim model-level analysis and forecast data were used for the
HARMONIE-AROME lateral boundary conditions (LBC). The ERA-Interim IFS
configuration uses a T255 spherical-harmonic representation for dynamical
fields, a reduced Gaussian grid with a horizontal grid spacing of
approximately 79 km for surface fields, and 60 vertical levels with a model
top at 0.1 hPa. Information from the ERA-Interim LBCs is read by the
HARMONIE-AROME forecast model every 3 h using one-way nesting. The
downscaling ratio, the ratio of the driving model grid spacing to the limited
area model (LAM) grid spacing, is approximately 32 : 1. Ideally, the
resolution of the LBC grid spacing and the LAM grid spacing should be as
close as possible to avoid the problem of reflections at
outflow boundaries when there is a mismatch in grid spacings as discussed in
. showed that errors are small and confined
to the boundaries for downscaling ratios of up to 4 : 1. Global climate
simulations have been successfully downscaled using a downscaling factor of
17 . In addition, recent projects to produce an extreme wind
climatology for The Netherlands and a wind/wave climatology
for Ireland using ERA-Interim and HARMONIE-AROME have
shown that the approach taken in this study, i.e. nesting HARMONIE-AROME
directly with ERA-Interim LBCs, is effective.
Seven separate simulations were set up to run for five years at a time, with
a one year spin-up period for each simulation. We therefore have 6 spin-up
years (1985, 1990, 1995, 2000, 2005, 2010) that overlap with corresponding
production years, which were used to evaluate the spin-up process of sub-soil
parameters. A spin-up period of 1 year was deemed necessary to allow deep
soil parameters to reach an equilibrium (cf. Sect. ). Each
simulation was run on ECMWF's Cray XC30 system, cca. The output data are,
temporarily, stored in ECMWF's data handling system, ECFS. The project has
produced approximately 750 TB of forecast and observation feedback data,
200 TB of which will be archived, with the forecast data stored as GRIB files
and the observation feedback as ODB files. MÉRA will continue to be
updated in real-time.
Three-hourly analysis output is available. Forecast model output is available
for each forecast hour up to 33 h for the 00:00 Z forecast and to 3 h
otherwise. A small subset of the surface (SURFEX) output is available at
analysis times and for each 3 h forecast while upper-air data are available
on pressure levels and a selection of near-surface levels. The analysis and
forecast output data are summarised in Table .
Data Assimilation
Data assimilation is used to estimate the initial conditions of the surface
and the atmosphere for the HARMONIE-AROME forecast model using meteorological
observations and a background field provided by short-range forecasts.
The surface data assimilation produces an analysis using optimum
interpolation (OI), a weighted least squares fit to observations and a
background field. CANARI (Code d'Analyse Nécessaire á ARPEGE pour ses
Rejets et son Initialisation; ), is used to calculate
analysis increments at each grid-point. The SURFEX OI data assimilation
scheme is then used to adjust the soil moisture and temperature in the
uppermost two levels of the surface. A snow analysis is carried for each
06:00 Z cycle by the CANARI OI scheme. Sea surface temperatures and sea ice
concentrations are taken directly from the ERA-Interim LBCs.
Three dimensional variational data assimilation (3D-Var) is used to produce
the most likely state of the atmosphere, i.e. the analysis, by minimising a
cost function – the sum of deviations of the analysis from observations and
deviations of the analysis from the background field, weighted by respective
errors as in . The data assimilation cycle in MÉRA
is carried out every 3 h using the 3 h forecast from the previous cycle and
observations combined with error covariance information. The climatological
background error covariances, generally referred to as structure functions,
used in the cost function calculation were derived using an ensemble of
HARMONIE-AROME forecasts as in . Forecasts were produced once
per day for a 1-month period by downscaling four ECMWF Ensemble Data
Assimilation members. Differences between the 6 h HARMONIE-AROME forecasts
were used to derive the structure functions. This was the recommended
approach for the ALADIN-HIRLAM system at the start of the MÉRA project.
It has been since noted that the derivation of structure functions would
benefit from forecast data sampling both daily and seasonal variations; i.e.
by using forecast data spanning the diurnal cycle and from different seasons.
Also, longer forecast integrations may be required to produce more realistic
convective-scale spectra as noted in . Observation error
covariances are prescribed in the observation preparation process.
Conventional observations are assimilated using the HARMONIE-AROME 3D-Var data
assimilation system. In this a minimisation is followed by a blending process
whereby the large scales produced by the analysis are combined with the
meso-scale features available in the background field. This process is
repeated every 3 h for each forecast cycle. Figure a shows data
counts of the number of upper-air temperature observations used by the 3D-Var
for the years 1981 up to the end of 2015. During this period there is a
notable increase in the number of aircraft observations. However, there were
also sharp drops in the numbers of upper-air observations available for
assimilation during 2003 and December 2014.
The difference between observation values (OB) and the equivalent first-guess
(short-range forecast) values (FG) used by 3D-Var (i.e. the first-guess
departures) are used to monitor and assess the quality of the MÉRA data
assimilation. The 3D-Var minimization process should produce an analysis (AN)
which is closer to the observations than the model background. The departures
for aircraft observations are shown in Fig. b for the same
period as the data count plots in Fig. a. It is clear that OB-AN
is closer to zero than OB-FG. An improvement in the average upper-air
temperature departures is noticeable from around 2000 onwards due to the
significant increase in the number of (AMDAR) aircraft observations available
for assimilation. The improvement in upper-air temperature analyses can also
be seen when radiosonde observations are compared with model first-guess
values and analysis values. Figure shows a time-series of
average temperature OB-FG (Fig. a) and OB-AN
(Fig. b) profiles for the years 1981–2015. There is a
noticeable improvement in the quality of analyses below 200 hPa for the
entire reanalysis period.
Forecast Model
This section summarises an initial validation of MÉRA output including
an analysis of the spin-up period, surface parameters, upper-air paramaters
and 24 h rainfall accumulations including comparisons to ERA-Interim where
possible. A short discussion on model performance attribution is also
included.
Model Spin-up
As described in Sect. 2, a spin-up period of 1 year was deemed necessary, but
not proven a priori, to allow deep soil parameters reach an equilibrium. To
investigate the length of spin-up required, soil moisture was compared for
the year-long periods of 2000, 2005 and 2010, for both the spin-up and
corresponding production year. Production output is treated as the “truth”.
It is estimated that it takes approximately three months for soil moisture at
20 cm below the surface (Fig. a) and nearly an entire year
for soil moisture at 300 cm below the surface (Fig. b) to
reach equilibrium. Note that the value of 300 cm is in fact arbitrary and
taken to mean the bottom of the soil layer. Soil temperatures reach
equilibrium much quicker (e.g. of the order of a few days for soil
temperatures 20 cm below the surface; see Fig. c and d).
It is important to note that the soil moisture/temperature values in
Fig. were computed by taking areal averages over the land
grid points in the domain (sea grid points do not have values).
Verification of Surface Parameters
The forecast model performance was validated by comparing observed surface
parameters and MÉRA output. Here we compare MÉRA 3 h forecasts with
synoptic observations available over the MÉRA domain i.e. observations
from Ireland, the UK and northern France. In addition, the MÉRA outputs
are also compared to the corresponding ERA-Interim reanalysis fields. The
inclusion of ERA-Interim data in the comparisons means that only 3 h
forecasts from the 00:00 and 12:00 UTC runs are compared as 3 h forecasts
are unavailable for other ERA-Interim forecasts cycles.
Figure shows the MSLP, 2 m temperature and 10 m wind speed
verification results for the years 1981–2014. Each subplot shows the bias
and standard deviation of the MÉRA and ERA-Interim outputs relative to
observations. A 1-month running average is applied in each case for
visualisation purposes. The number of observations is also shown in black on
each plot. Overall, the results indicate a consistent model performance over
the time period. In terms of the standard deviation of the errors, MÉRA
out-performs ERA-Interim for MSLP (Fig. a), 2 m temperature
(Fig. b) and 10 m wind speed (Fig. c). The MÉRA
dataset captures the 10 m wind speeds much better than ERA-Interim
(Fig. c smaller bias) as expected from a model with improved
orographic representation. The biases in 2 m temperatures relative to
observations are lower than the corresponding ERA-Interim biases for the same
reason (Fig. b). HARMONIE-AROME uses ERA-Interim lateral boundary
conditions. This places constraints on the MÉRA large scale pressure
patterns. Thus, the differences in biases and standard deviations in MSLP for
MÉRA and ERA-Interim are small (Fig. a). The obvious shift in
biases around 1993 is still under investigation and has not yet been pinned
down. However, it is thought to be due to some of the observations for
Ireland available on the GTS.
Figure a shows a comparison of MÉRA and observed soil
temperatures 20 cm below the surface. This parameter is not included in Met
Éireann's operational verification; hence the results presented here
represent the first soil temperature verification of HARMONIE-AROME done for
Ireland. Soil temperatures are very important for agriculture. It is clear
that MÉRA captures the variability in 20 cm soil temperatures but biases
exist of up to +1 ∘C in winter/spring and -1 ∘C in
summer/autumn. This is evident in both Fig. a and b. Although
there are biases of ±1 ∘C the fact that the bulk of the points on the scatter plot in
Fig. c is reassuring.
Verification of Upper-air Parameters
The quality of MÉRA upper-air parameters are validated by comparing 12 h
forecast data with radiosonde observations at 12:00 Z each day. A comparison
to the equivalent ERA-Interim forecast data was also done. Again, bias and
standard deviation are used as a measure of forecast quality. MÉRA
temperature, wind, humidity and geopotential height forecasts
compare favourably with ERA-Interim at all heights. In general, for comparisons
above 300 hPa and below 850 hPa there are only small differences in bias
and standard deviation. However, at levels above 850 hPa and below 300 hPa
MÉRA forecasts clearly outperform ERA-Interim; see Fig. a
and b as typical examples.
Verification of 24 h Rainfall Accumulations
Precipitation forecasts produced by MÉRA are compared with observations
of 24 h accumulations of precipitation recorded by Met Éireann's
voluntary rainfall network (09:00 to 09:00 Z). Seasonal mean biases in daily
precipitation are mostly positive and within 1 mm –
see Fig. . Larger negative biases over mountains are due to
the fact that the 2.5 km grid spacing in HARMONIE-AROME cannot account for
mountain peaks contained within a grid box. Biases in spring (Fig. 8a) are
slightly larger than the other seasons, mostly within 2 mm, and are thought
to be as a result of magnitude and positional errors in convective rainfall.
Fig. shows a time-series of biases and standard deviations of
the errors in 24 h rainfall accumulations (09:00 to 09:00 Z) relative to
observations. ERA-Interim forecasts of precipitation are also included on the
figure. 36 h ERA-Interim forecasts only cover the period 00:00 to 06:00 Z.
For this reason, the 09:00 to 09:00 Z precipitation forecasts shown for
ERA-Interim consist of the sum of the 09:00 to 24:00 Z forecast and the
00:00 to 09:00 Z forecast. As expected, the higher resolution MÉRA shows
an improvement over the coarser resolution ERA-Interim, which underestimates
precipitation at the Irish climatological stations by up to ∼ 2 per
day. Again, this can be attributed to both model resolution and mesoscale
physics parametrizations which resolve convection in MÉRA. Note, however,
that the standard deviation of the model errors is similar for MÉRA and
ERA-Interim.
The improvement in precipitation forecasts is one of the main advantages of
very high resolution reanalyses. Because of this, a qualitative overview of
two extreme rainfall case study examples are also included in this section.
The first, caused by an offshoot depression from Hurricane Charley on
25/26 August 1986, brought rainfall amounts of over 200 mm to parts of the east
coast of Ireland (Fig. a). It was badly forecast by models at
the time. The improvement in the 30 h rainfall forecast by MÉRA
(Fig. c) compared to ERA-Interim (Fig. e) is of
the order of 100 mm, more over higher orography. Note that the observations
cover a 24 h period from 09:00 Z on 25 August whereas the MÉRA and
ERA-Interim accumulations are for the 30 h period starting from 00:00 Z on
25 August. While MÉRA data are available for the same period as the
observations, only up to 30 h forecasts are available for ERA-Interim at
00:00 and 12:00 Z. Thus, it requires 2 ERA-Interim forecast cycles to cover
the same period as the observations and for simplicity, we chose to show the
30 h forecast (i.e. 00:00 Z on 25 August to 06:00 Z on 26 August) for the
model performance comparison.
The second case study, heavy rainfall 3/4 August 1997 (bank holiday in
Ireland), was also badly forecast by models at the time which failed to
forecast any precipitation for Ireland. Instead, accumulations of over
200 mm were observed over the southern half of the country
(Fig. b). As for the “Hurricane Charley” case, MÉRA
performed remarkably well, capturing the extremes, both in terms of intensity
and location. The MÉRA and ERA-Interim rainfall totals shown in
Fig. d and f represent accumulations of the successive 24 h
totals covering the period 00:00 Z on 3 August 1997 to 00:00 Z on 5 August
1997.
Improved Performance Attribution
While it is clear that HARMONIE-AROME performs better than ERA-Interim for
3 h forecasts of 2 m temperature and 10 m wind speed, how much of the
improvement is due to model resolution (and thus better orographic
representation) and/or model processes and parametrizations cannot be deduced
from Fig. . To address this, we compared 3 h forecasts of MSLP,
2 m temperature and 10 m wind speed, for the period February 2013 to
February 2014, from MÉRA, ERA-Interim, HARMONIE-AROME cycle 37h1 (current
operational version at Met Éireann) and HIRLAM (version 7.2, also
operational) (Fig. ). The Irish operational set-up of
HARMONIE-AROME cycle 37h1 is not configured to use 3D-Var; instead surface
data assimilation with upper-air blending is used. 4D-Var data assimilation
is used in the operational HIRLAM version 7.2 model ,
which is run on a ∼ 10 km grid. For MSLP, 2 m temperature and 10 m
wind speed MÉRA and HARMONIE-AROME cycle 37 perform better than HIRLAM
which performs better than ERA-Interim. The improvements seen in MÉRA and
HARMONIE-AROME cycle 37 relative to HIRLAM are due to a combination of
resolution and improved model physics. MÉRA has lower errors in 10 m
wind speeds (Fig. c) than the operational HARMONIE-AROME both
due to the use of 3D-Var and the fact that the surface drag coefficient was
tuned. Biases and standard deviations in MSLP (Fig. a) and
2 m temperatures (Fig. b) for MÉRA and HARMONIE-AROME
cycle 37 are comparable but again are noticably better than HIRLAM and
ERA-Interim.
Conclusions
In this article we have described how the HARMONIE-AROME model has been used
to produce a climate reanalysis dataset MÉRA for Ireland. We have carried
out a preliminary evaluation of the dataset and shown that the HARMONIE-AROME
data assimilation system and forecast model perform well and consistently
when compared with point observations. MÉRA's advantage over ERA-Interim
was also illustrated. A more thorough validation of the entire dataset is
underway, with the aim of quantifying all biases in the dataset. This will
enable improvements to be made to Met Éireann's operational NWP suite
and will also help in the design of a proposed high resolution ensemble
forecasting system.
MÉRA is the highest resolution, freely available reanalysis dataset
covering Ireland and will have uses in research, food and agriculture,
renewable energy, ecology, planning, economics and hydrology. A future
regional reanalysis for Ireland would gain from using the latest version of
HARMONIE-AROME, a larger domain, the assimilation of more observation types
and the generation of an ensemble of reanalyses with ensemble data
assimilation. The use of ERA-5 for lateral boundary conditions as well as a
coupled ocean-atmosphere system should also produce an improved reanalysis
dataset.