Using an ultrahighresolution regional climate ... - Wiley Online Library

6 downloads 0 Views 747KB Size Report
Jan 17, 2013 - *Correspondence to: C. Wang, Met Office Hadley Centre, ... Met Office Unified Model (MetUM) based on the concept of very high-resolution.
Quarterly Journal of the Royal Meteorological Society

Q. J. R. Meteorol. Soc. 139: 1964–1976, October 2013 B

Using an ultrahigh-resolution regional climate model to predict local climatology Changgui Wang,a *† Richard Jones,a† Matthew Perry,b† Catrina Johnsonb† and Peter Clarkc a

Hadley Centre, Met Office, Reading, UK Scientific Consultancy, Met Office, Exeter, UK c Department of Meteorology, University of Reading, UK b

*Correspondence to: C. Wang, Met Office Hadley Centre, Meteorology Building, University of Reading, UK, RG6 6BB. E-mail: [email protected] † The contribution of these authors was written in the course of their employment at the Met Office, UK, and is published with the permission of the Controller of HMSO and the Queen’s Printer for Scotland.

Many applications of risk assessment and Environmental Impact Assessments (EIA) require information about the local mesoscale climate at a very fine spatial scale. Current regional climate models (RCMs) with grid resolutions that normally range between 10 and 50 km cannot resolve the local physiographic features that influence the small-scale features of local climate. This article presents a local climate model (LCM) system with an ultrahigh resolution and demonstrates its application to study the local climatology in a coastal zone where details of orography and coastline are important influences on the climate. The LCM has been developed from the Met Office Unified Model (MetUM) based on the concept of very high-resolution numerical weather prediction. The ultrahigh-resolution regional climate modelling is achieved through operating multiple nesting from a spatial resolution of 12 km down to ∼500 m. In this pilot study of the LCM, RCMs were run at 12, 4, 1.5 and 0.5 km grid resolutions to simulate the mesoscale climate of Weymouth Bay and Portland Harbour along the south coast of the United Kingdom. The period simulated was July-August of the years 1993 to 2002, with model results validated against meteorological station time series. The results demonstrated the added value of the 500 m resolution model compared with the 1.5 km resolution model in modelling very fine local mesoscale climate patterns over the inner domain, including diurnal variations of temperature and winds, such as southwesterly flows, and the effect of wind channelling between the Isle of Portland and the mainland. Lower mean wind speeds are also found along the south-facing coast to the east and in the shelter of Portland in the 500 m resolution model, particularly showing a sheltering zone between Portland and Weymouth Bay. Key Words: mesoscale climate; local climatology; local climate model; regional climate modelling; ultrahighresolution climate modelling; climate change impact Received 10 May 2012; Revised 16 October 2012; Accepted 1 November 2012; Published online in Wiley Online Library 17 January 2013 Citation: Wang C, Jones R, Perry M, Johnson C, Clark P. 2013. Using an ultrahigh-resolution regional climate model to predict local climatology. Q. J. R. Meteorol. Soc. 139: 1964–1976. DOI:10.1002/qj.2081

c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

Ultrahigh-Resolution Regional Climate Model Prediction of Local Climate

1. Introduction Many engineering developments require information about the local mesoscale climate, often at scales much smaller than can be met by conventional observations and at locations where observations are not available. These are often required for Environment Impact Assessments (EIAs) either directly (in which case there is usually a need to run with or without some hypothetical development, such as an airport or industrial complex) or driving downstream environmental risk impact models. So-called ‘downscaling’ of available analysis data using high-resolution regional climate models (RCMs) is a well-recognized approach (Rummukainen, 2010) to solving this problem, whereby the RCM adds skilful local details, particularly mesoscale features, such as land–sea wind circulation or other orographically induced wind systems close to the coast (Feser et al., 2011). Feser et al.(2011) also showed that RCMs add value in describing mesoscale variability compared with the driving global reanalysis but the RCMs need the higherresolved orography or coastlines to achieve more realistic results than the already well described global reanalyses for near-surface wind speed. Salath´e et al.(2008) found that future climate scenarios simulated with a high-resolution climate model (15 km grid cell) showed very different trends in temperature and precipitation over the Pacific Northwest due to resolved terrain features compared with its driving global climate model (GCM). Currently, typical resolutions of RCMs range from 50 km down to 10 km. Mesoscale climatology down to a local scale using RCMs, such as a ∼500 m grid, has not yet been attempted. The realistic representation of climate patterns on the local mesoscale in climate models remains a key challenge. There is also an increasing need for more reliable assessments of local climate using very high-resolution local climate modelling to assess how small-scale highimpact weather extremes may change under global warming. Current RCMs are not sufficient to address these issues, again because of their resolution. The Met Office Unified Model (MetUM) is used for both numerical weather prediction (NWP; Walters et al., 2011) and climate prediction (Hardiman et al., 2010). The first is characterized by a continuous, sophisticated analysis system set up for the whole globe and a limited number of areas followed by short-term forecasts that, inevitably, diverge from real events. The second is characterized by longterm integrations starting from more-or-less arbitrary initial conditions with the purpose of predicting the statistics of weather (i.e. the climate), but not specific events. However, RCMs are also used in a hindcast mode when they are driven by reanalyses (Noguer et al., 1998), where the specific events are reproduced due to the use of observed data for both initial and boundary conditions. In principle, the NWP system, if run long enough, should tend to a ‘climatological’ prediction. These two applications have led to a divergence in tools and systems used to apply configurations of the MetUM. For example, weather forecasts are initialized using sophisticated data assimilation systems that need to be fine-tuned for a given domain, so setting up new domains is uncommon because it is time-consuming. Then limited-area forecasts are nested (sometimes multiple) inside other forecasts, a ‘host model’ that ultimately is a global model, and these are run for relatively short periods at high resolution. In c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

1965

contrast, climate forecasts and projections are run for much longer periods (from months to centuries), which limits resolution, and these are essentially independent of initial conditions because a spin up period of the order of a year or more is removed before the simulations are used. Models such as the MetUM can be used for a third application that might loosely be called ‘mesoscale climatology’. In essence, models are used as a substitute for observations. They are used to downscale relatively lowresolution available analysis data and add local details by running the model at a higher resolution than the driving data and using higher resolution surface data (orography, land-use). The assumption is that small-scale structures are deterministically related to (or ‘slaved to’) the large-scale analysis, which is reasonable where small-scale structures are dominated by this surface forcing. This third application of the MetUM clearly sits between forecasting at high resolution and regional climate prediction. Increased resolution together with state-of-the-art physical parametrizations, including improved simulation of turbulent transport and radiation transfer, adapted from and developed within the contemporary numerical weather prediction practice, are expected to bring improvements in the information quality of the fine-scale climate obtained by the model. The local climate model (LCM) is designed as a multiply nested regional climate modelling system that produces the very high-resolution data required in climate assessments or climate change impact studies. The development of the LCM is not intended to investigate resolution-related issues and sensitivities that can be found in other publications (Denis, 2002), but to demonstrate the capability of high resolution in climate modelling. Therefore, the purpose of this article is to present the LCM’s potential by applying it over Weymouth Bay and Portland Harbour, to not only highlight the power of high-resolution modelling for gaining an insight into the local climatology but also to provide guidance on general climate and wind conditions due to channelling and sheltering by the coastline across the Olympic and Paralympic sailing venue during the London 2012 Olympics. The purpose is also to underline the strength of integrating advances in very high-resolution weather forecasting into a regional climate model in order to improve local climate modelling and hence provide a local climate information service for risk assessment and EIAs. 2. The components of the LCM 2.1. Numerical weather forecast model – The MetUM The dynamical core of the MetUM implements the fully compressible, non-hydrostatic, deep atmosphere formulation of the Navier–Stokes equations (Davies et al., 2005). Numerically, the advection of the prognostic variables is treated using a semi-Lagrangian discretization, which follows the air particle trajectories. The exception is density, for which an Eulerian, that is, grid-based, discretization is retained in order to conserve mass. Semi-implicit time integration is used for the highfrequency terms. The model includes a comprehensive set of parametrizations, including the MOSES 2 tiles land surface scheme (Essery et al., 2001), boundary layer (Lock et al., 2000) with the modifications described by Lock (2001) and Brown et al.(2008), mixed-phase cloud microphysics (Wilson and Ballard, 1999) and convection Q. J. R. Meteorol. Soc. 139: 1964–1976 (2013)

1966

C. Wang et al.

Table 1. Regional climate model (RCM) physics comparison between horizontal resolution of 500 m, 1.5 km, 4 km and 12 km configurations of the Met Office Unified Model used in this study. Model property

500m-RCM,1.5km-RCM

4km-RCM

12km-RCM

Vertical levels Grid length (degrees) Time step LBC updating interval Convection scheme Gravity wave drag Prognostic rain Diffusion–turbulence

70 0.0045,0.0135 10 s, 50 s 30 min No No Yes Sub-grid turbulence Smagorinsky type

70 0.036 100 s 1h Gregory and Rowntree, 1999, Robert et al., 2003 Yes No Horizontal diffusion (theta,winds, moisture)

38 0.11 5 min 6h Gregory and Rowntree, 1999 Yes No No

LBC, lateral boundary conditions.

(Gregory and Rowntree, 1990), with additional downdraft and momentum transport. Edwards and Slingo (1996) implemented the two-stream approximation to radiative transfer and incorporated scattering in both the long-wave and short-wave regions of the spectrum in the MetUM for calculating radiative fluxes. The MetUM developments in recent years, together with its configurations, have been detailed by Walters et al.(2011). The limited-area model runs on a rotated latitude–longitude horizontal grid with Arakawa C staggering and a terrain-following hybrid-height vertical coordinate with Charney–Philips staggering. As the model is non-hydrostatic, it can be run at a high horizontal resolution (e.g. 4 km and 1.5 km, etc.), which will preserve the important features, for example orography and coastline (Wang et al., 2009, 2012). The nested RCMs in the LCM are designed to have grid resolutions ranging from 12 km to 500 m, and four resolutions are used in this study: 12km-RCM, 4km-RCM, 1.5km-RCM and 500m-RCM. The 12km-RCM used here is a limited-area version of the latest Met Office Hadley Centre Global Environmental Model (HadGEM3-RA) and the model physics is described by Walters et al.(2011). The 4km-RCM and 1.5km-RCM are introduced from the mesoscale NWP models assessed by Lean et al.(2008) at the Met Office and these mesoscale NWP models were also used for downscaling global NWPs dynamically by Wang et al.(2009, 2012) in order to explore the local coastline effect on atmospheric refractivity. The model physics in the 4kmRCM is very similar to the 12km-RCM model (Table 1). There are a few notable differences (Table 1) in the 1.5kmRCM and the 500m-RCM; in particular, the 4km-RCM and 12km-RCM use a convection scheme based on Gregory and Rowntree (1990) and Robert (2003). The 12km-RCM is run without any horizontal diffusion, whereas subgrid turbulence Smagorinsky type diffusion (Ansumali et al., 2004) is applied in the 1.5km-RCM and the 500m-RCM. The 12km-RCM assumes that rain falls directly to the ground, without being advected by the winds. This approximation is increasingly poor at higher resolution, and so prognostic rain is used in the 1.5km-RCM and 500m-RCM. All nested RCMs use the MOSES-2 land surface scheme, which was introduced into mesoscale by Best et al.(2000).

Figure 1. Dynamic downscaling steps in the local climate model.

higher resolution local-scale climate information. The 1.5km-RCM nested between the 4km-RCM and 500m-RCM is not only for providing physically consistent boundaries for the 500m-RCM but also for comparison purposes for this study. In essence, the 12km-RCM to 1.5km-RCM in the LCM mimic the dynamic nesting models used by Wang et al.(2009), but are run in climate mode by initializing each RCM once at the beginning for a longer climate simulation period. Therefore, each RCM in the LCM is a dynamic model used for downscaling that consists of three components: ‘host’ model for providing boundary conditions, locally specific land surface information and its own physics-based equations to resolve the model based on the first two components. The results in each RCM are comparatively local predictions that are informed by both local specifics and its ‘host’ model. The steps of dynamic downscaling from a GCM or reanalysis towards a grid resolution of ∼500 m in the LCM for each climate simulation period are depicted in Figure 1. 2.3.

Climatology database – input large-scale climatologies

To run a full mesoscale model for a continuous period, a database containing sufficient continuous threedimensional data is required within the LCM for downscaling, for example the ERA40 (Uppala et al., 2005) or ERA-INTERIM reanalyses (Dee et al., 2011). For general 2.2. Dynamic downscaling climatology, the period covered should be as long as possible (ideally 30 years or more) but for many applications The LCM is built up using the one-way nested RCMs 5–10 years would be regarded as adequate to define aspects described above to downscale GCMs or reanalysis to of mean climatology. The data need to be quickly accessible grid resolution of 12 km and then cascade down to grid in order to provide the boundary conditions for the model resolutions of 4 km, 1.5 km and ∼500 m (Table 1) to obtain generating the mesoscale climatology. This means, ideally, c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

Q. J. R. Meteorol. Soc. 139: 1964–1976 (2013)

Ultrahigh-Resolution Regional Climate Model Prediction of Local Climate

1967

that the data are stored on a disc system attached to the computer that will use it, although a sufficiently fast, reliable and automated off-line archive system may suffice. 2.4. Domain definition and ancillary file generation When initiating a new study, first a set of domains must be designed, nested inside each other, covering the desired area with a given resolution using the MetUM’s rotated latitude–longitude coordinate system. This often needs a certain amount of iteration. Once a domain is chosen, some ancillary files must be produced before the model can run. These include two types of variable. (1) Fixed properties, such as land fraction, land/sea Figure 2. The structure of the local climate model components. mask, land surface height and parameters relating to unresolved orography, land-use (specifically, at variables to be able to generate initial and boundary present, fractions of the nine Met Office surface data. This database must be easy, fast and cheap to exchange scheme (MOSES) surface tiles, vegetation access (see section 2.3). parameters and soil properties (Essery et al., 2001)). (2) A utility to create initial data for a specified domain (2) Initial or continuously updated values of parameters and time. that vary with time. These include soil temperatures, (3) A subsystem that automatically updates the LBC, soil moisture (prescribed as initial conditions), seaancillary files and model nesting. surface temperature and sea ice, plus ozone and (4) A subsystem that automates model output archiving, perhaps other atmospheric constituents (which are resubmission and post-processing. either updated or provided as varying through the (5) A failure recovery system. annual cycle), although these usually are taken from climatologies that are acceptably accurate. Figure 2 shows the structure of the LCM components and Some variables, such as land-use, may be regarded as Figure 3 shows the top LCM interface, which includes spanning the two categories, as they may vary with time. utilities of defining domains, generating and editing When setting up, it has to be decided which data source ancillaries, computing system settings and submitting to use for these variables. In particular, the sea-surface jobs, etc. temperatures (SST), soil temperatures and soil moisture in operational models are superior to available, low-resolution 3. Application of the LCM over Weymouth and Portland climatologies, although the quality of soil moisture is Harbour probably more questionable outside the UK. Data such as orography and land-use may be available for individual 3.1. Geographical features and synoptic conditions sites at higher resolution or quality than ‘standard’ global data sets. If so, the LCM is designed to import such data. Weymouth is on the southern coast of England and on Finally, the ability of the LCM to edit data manually is the northern shore of the English Channel (Figure 4). essential – for example, to change land-use in a particular The coastline in this area consists of bays that form a area. Thus the LCM provides functionality for manual rather smooth wave-like geometry (Figure 5). To the east adjustment of land fraction and land/sea mask, orography of Weymouth, cliffs rise up sharply from the coast up to and land-use, with some additional facilities to adjust soil an elevation of 120 m. To the west is a lower lying area moisture and temperature. behind Chesil Beach, a tombolo that connects the mainland to the Isle of Portland, a headland that reaches an elevation 2.5. Initial and lateral boundary condition of 150 m. Portland Harbour, in the bay between Portland and Weymouth, is one of the largest man-made harbours Once a set of domains is defined, the model needs in the world. The complex orography and coastline provide initial atmospheric and land surface conditions as well as a challenging and interesting scenario on which to test the atmospheric variables, such as lateral boundary condition high-resolution modelling capabilities. (LBC), corresponding to the full length of the period to The mean synoptic condition over northwestern Europe be simulated. The analysis data described in section 2.3 are is controlled mainly by the position of the Icelandic low and used to generate these initial and LBC data. With these data the Azores high and the path taken by Atlantic depressions in place the model is then able to simulate the required approaching Europe. In summertime, the Icelandic low is period. In general this is done as a sequence of shorter, for generally weakened and the Azores high usually sits in a example monthly, periods at the end of which a full set of relatively northerly position. This gives, on average, westerly restart conditions is produced. This allows the model to be synoptic winds in northwestern Europe. At Weymouth Bay restarted easily after, for example, when the computer needs and Portland Harbour, the prevailing wind direction is to be shut down or suffers a power failure. To perform these westerly or southwesterly. Weymouth Bay is sheltered from long integrations, the LCM has been developed to contain these predominant prevailing winds by the Isle of Portland. the following features. The mild seas that almost surround the tied island produce (1) A database of continuous data defining the three- a temperate maritime climate with a small variation in daily dimensional atmospheric structure with sufficient and annual temperatures. c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

Q. J. R. Meteorol. Soc. 139: 1964–1976 (2013)

1968

C. Wang et al.

Figure 3. The initial page of the local climate model interface.

wind by acting as a barrier, forcing the wind to change direction both horizontally and vertically. This can produce a number of phenomena, for instance downslope winds around mountains and wind increase due to channelling or wave-like structures. In coastal areas, these types of effects very often occur around islands and in channels where the wind increases in some places and decreases in others. To simulate the impact of those factors on climate dynamically within this small coastal area of Weymouth and Portland Harbour, an ultrahigh horizontal resolution RCM, for example 500 m, is needed to resolve the local mesoscale features within the region of interest. To this end, four nested domains were designed using the LCM and the resolution in the inner domain focusing on Weymouth and Portland Harbour was 500 m. Figure 5 shows that the 500mFigure 4. The nested domains over Weymouth. RCM (Figure 5(a)) has a detailed structure of topography as compared with the 1.5km-RCM (Figure 5(b)). The strategy adopted in this pilot study was to perform 3.2. Modelling strategy high-resolution MetUM numerical simulations of local weather events across a local region using a series of The local coastal climatology is decided not only by the nested grids with 100 × 100 points, with domains that synoptic weather conditions, but also, to a large extent, it progressively focus on the Weymouth Bay area (Figure 4) can be affected by local forcing. Several factors affect the and with designed RCMs in the LCM. The 12 km model was weather in coastal areas and therefore are of importance initialized using ERA-40 reanalysis data from the European to the local climate, such as local terrain, surface friction, Centre for Medium-Range Weather Forecasts (ECMWF). horizontal temperature gradients and the vertical structure The numerical simulations were run for periods of 1 month of the atmosphere. Local terrain has a direct impact on the from 20 July to 19 August; 10 of these 1 month simulations c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

Q. J. R. Meteorol. Soc. 139: 1964–1976 (2013)

Ultrahigh-Resolution Regional Climate Model Prediction of Local Climate

(a)

1969

(b)

Wyke Regis

B A

Portland Isle of Portland

m 0

20

40

60

m 80

100

140

0

20

40

60

80

100

140

Figure 5. Orography of the Weymouth Bay area, the locations of weather stations (triangles) and points A and B where model outputs have been analysed. (a) The 500m-RCM and (b) the 1.5km-RCM. (a)

(b)

B

B A

A

m/s 6

7

8

9

m/s 10

11

12

6

7

8

9

10

11

12

Figure 6. Daytime (0800–2000 UTC) 10 m mean wind speed (knots) over the 500m-RCM domain for the period of 1993–2002. (a) The 500m-RCM and (b) the 1.5km-RCM.

were run, for the years 1993 to 2002. Denis et al.(2002) presented a ‘big brother’ experiment in which an RCM with grid resolution of 45 km was downscaled from a global model in which they showed that a 24 h spin up time is enough for fine-scale atmospheric features responding to fast dynamical and thermodynamical forcings. Wang et al.(2012) also carried out a sensitivity analysis of the initial-time problem where 12 h spin up time was used for obtaining a mesoscale response. Therefore, each simulation period included 1 day spin up time. This short spin up time would not allow deep soil moisture fields to come into equilibrium but near-surface soil moisture can approach equilibrium in a few days. In this case where the climate is being simulated of a small coastal region with atmospheric humidity that will be controlled largely by the adjacent seas, the deep soil moisture would have little effect.

Therefore, this article focuses on the 500m-RCM while making comparison with the 1.5km-RCM to show whether the ultrahigh resolution of the 500m-RCM added values that gained more details for local climate information. 3.3.1. Temporal and spatial variability of wind speeds

Figure 6 shows the spatial variations in daytime mean wind speed at 10 m height over the inner domain. In the 500mRCM (Figure 6(a)) the effect of wind channelling through Portland Harbour, between Portland and the mainland, can be seen as a small area that has a higher mean wind speed compared with the rest of Weymouth Bay. Lower mean speeds and sheltering are found along the south-facing coast to the east, particularly showing the sheltered zone between Portland and Weymouth Bay. The influencing effect of welldescribed altitude and terrain shape (Figure 5(a)) also can 3.3. Modelling results be seen clearly by the high wind speeds along the ridge of The authors have applied well-tuned mesoscale NWPs used high ground and cliffs to the north and east of Weymouth. at the Met Office (Wang et al., 2009, 2012) for running These fine details and mesoscale features are not obvious in the climate mode in the LCM, except for the 500m-RCM. the 1.5km-RCM (Figure 6(b)). c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

Q. J. R. Meteorol. Soc. 139: 1964–1976 (2013)

1970

C. Wang et al.

(b)

15

10 Frequency (%)

Wind speed (knots)

(a) 12

8 6 4

10

5

2 0

0 8 9 10 11 12 13 14 15 16 17 18 19 20 mean

1 3 5 7 9 11 13 15 17 19 21 23 25

Hour of day (UTC) (d) 15

12 10

Frequency (%)

Wind speed (knots)

(c)

Wind speed (knots)

8 6 4

10

5

2 0

0 8 9 10 11 12 13 14 15 16 17 18 19 20 mean

1 3 5 7 9 11 13 15 17 19 21 23 25

Hour of day (UTC)

Wind speed (knots)

Figure 7. Day time (0800–2000 UTC) mean wind speed (10 m height) and histogram of wind frequencies (%) in intervals of 2 knots over the period of 1993–2002 at the locations of A (black bars) and B (grey bars) for ((a) and (b)) the 500m-RCM and ((c) and (d)) the 1.5km-RCM.

To demonstrate that the 500m-RCM can simulate smallscale climate details, we present results from two locations that are 8 km apart, one in Portland Harbour and one in Weymouth Bay (locations A and B in Figure 5). The mean wind speed by hour of the day for the two locations, covering the period from 0800 to 2000 UTC, is shown in Figure 7. The overall mean in Figure 7(a) shows that the winds simulated at location A by the 500m-RCM are much stronger than those at location B. Consistent with this, a climatology of the frequency of different winds speeds (Figure 7(b)) shows that location B has a markedly greater frequency of lower wind classes of less than 9 knots, whereas location A has higher frequencies in high wind classes from above 13 knots. These differences can be explained by their locations with respect to local orographic features. Location A is in Portland Harbour and even on calmer days wind funnels through the gap between the Isle of Portland and the mainland (see section 3.3.3), increasing the wind speed. In contrast to the 500mRCM, the 1.5km-RCM (Figure 7(c) and (d)) presented very similar patterns at these two locations; both showed weaker wind strength during the night and early morning hours though location A has stronger winds but with smaller magnitude of different. 3.3.2.

Wind direction frequencies

The prevailing wind across the area comes from the west and southwest. The analysis of wind direction at 10 m height from the 500m-RCM shows that location A has about 60% from this direction whereas location B has 45%. An important secondary direction is winds from the east, especially for location B, which has 13% of winds coming along the cliffs from this direction. These features can be seen in the wind roses for these locations (Figure 8), which divide the wind direction into 12 sectors. Although the highest wind speeds generally come from the westsouthwest, some high wind speeds can also come from the east and eastnortheast, especially for location B. Figure 8 also shows that there c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

are 10% more winds from the southsouthwest at location A than at location B, due to channel funnelling effects. The 1.5km-RCM also simulated the west and southwest prevailing winds (Figure 8(b)) at both locations, but failed to pick up more frequent high winds at location A, instead it produced 10% more high winds (>24 knots) at location B. The 1.5km-RCM also failed to simulate the wind from the southsouthwest at location A. 3.3.3.

Local features

Figure 9 shows examples of a sea breeze and wind funnelling during the summer. Figure 9(a) and (b) provides an example of a sea breeze, which often develops in Weymouth Bay during afternoons in the summer. Pressure gradients caused by the difference in temperature between land and sea cause a flow of air from the sea onto the land to set in. In this example, the wind direction over the land is northwesterly, whereas over the sea it is southwesterly, with quite strong winds in the bay area. Later in the afternoon, the sea breeze can veer to become more westerly, before dying out as the temperature gradient reduces. Both the 500m-RCM (Figure 9(a)) and 1.5km-RCM (Figure 9(b)) resolved this coastal feature well. Figure 9(c) and (d) shows an example of wind variations around Portland as well as funnelling through Portland Harbour, which on an otherwise calm day generates stronger wind speeds in Weymouth Bay and also has an effect on wind direction, as the wind diverts around the headland of the Isle of Portland. The 500m-RCM (Figure 9(c)) demonstrated its ability to simulate these features, with stronger winds in Weymouth Bay through Portland Harbour and more variations of winds around Portland than can be seen in the 1.5km-RCMs (Figure 9(d)). Such features were also exhibited in the spatial variations of mean wind speed (Figure 6), where the 500m-RCM (Figure 6(a)) simulated enhanced winds passing through Portland Harbour and then Q. J. R. Meteorol. Soc. 139: 1964–1976 (2013)

Ultrahigh-Resolution Regional Climate Model Prediction of Local Climate

N

(a)

1971

N 30% 25% 20% 15% 10% 5%

W

30% 25% 20% 15% 10% 5%

E

W

E

Wind Speed (kts) 0-4 4-8 8 - 12 12 - 16 16 - 20 20 - 24 24 - 28 28 - 30

S

S

N

(b)

N 30% 25% 20% 15% 10% 5%

W

30% 25% 20% 15% 10% 5%

E

W

E

S

S

Location A

Location B

Figure 8. Wind roses for the locations of A (left) and B (right), showing the frequency of wind speed and direction between 0900 and 1500 UTC for July and August from 1993–2002. (a) The 500m-RCM and (b) the 1.5km-RCM. Table 2. Summary statistics of the differences between observations (Obs) and high resolution modelling output. Observations used were from the Portland (1993 to 1998) and Isle of Portland (1994 to 2002) weather stations for the period 20 July to 19 August. Data are hourly except that the rainfall was daily and the cloud fraction and visibility at Isle of Portland only had 3 h data available during 1994 to 1999 (see section 4.3). Variable

Locations

Wind speed (knots) Temperature (◦ C) Relative humidity (%) Cloud (fraction) Visibility (km) Rainfall (mm day−1 )

Port IsleP Port IsleP Port IsleP Port IsleP Port IsleP Port IsleP Wyke

Number of observations 4455 6017 4455 5985 4454 5985 4369 3659 4450 3618 184 231 310

Biasa

Observations mean 9.71 11.25 17.61 17.01 79.13 84.00 0.54 0.48 17.33 22.43 1.03 1.01 1.30

Standard deviation

RMSE

M1

M2

Obs

M1

M2

M1

M2

−1.41 −0.96 −0.37 −0.65 −1.30 −1.60 −0.07 −0.03 5.80 6.20 0.10 0.48 0.03

−0.83 −3.69 −1.23 −0.31 1.89 −3.95

4.95 5.76 2.27 1.80 11.50 10.70 0.32 0.39 10.60 12.80 3.90 3.18 4.17

4.00 5.05 1.86 1.66 10.20 11.20 0.38 0.36 9.12 11.40 3.80 3.82 5.27

4.56 3.93 1.15 1.88 9.26 11.90 0.13 0.12 9.06 12.21 3.20 4.56 6.26

4.03 4.44 1.55 1.47 10.40 9.90 0.34 0.41 14.80 18.00 5.28 4.28 6.46

3.89 5.77 2.28 1.25 9.32 10.51

0.04 0.50 0.22

4.80 5.02 7.39

IsleP, Isle of Portland; Port, Portland; M1, 500m-RCM; M2, 1.5km-RCM; RMSE, root-mean-square error. a Bias = model − observations.

gradually easing off in Weymouth Bay but the 1.5km-RCM (Figure 6(b)) showed no evidence of funnelling. 4. Verification Figure 5(a) shows the locations of meteorological stations in the area. Of these, the stations with hourly observations are Portland (1993 to 1998) and Isle of Portland (1994 to 2002), while daily observations are available from Wyke Regis. Observations from these stations were used to verify the c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

accuracy of climatological values obtained from the LCM. Variables used in this article are wind speed, temperature, relative humidity, cloud cover and rainfall. A summary of the root mean square error (RMSE) is listed in Table 2. 4.1. Wind Speed and wind direction Table 2 shows that the simulated wind speed values from the 500m-RCM are slightly lower than the observed values on average. The biases at Portland and Isle of Portland are Q. J. R. Meteorol. Soc. 139: 1964–1976 (2013)

1972

C. Wang et al.

(a)

(c)

50

50

50

100

100

50

150

50

150

5 m/s 16

50

50

50

5 m/s 18

20

22

24

12

(b)

14

16

18

20

22

24

(d)

50

50

50

50

100

100

50

150

50

150

5 m/s 16

50

50

5 m/s 18

20

22

24

12

14

16

18

20

22

24

Figure 9. Examples of ((a) and b)) sea breeze at 1600 UTC on 8 August 1993 and ((c) and (d)) wind funnelling through Portland Harbour at 1100 UTC on 6 August 2000. Wind speed and direction are shown by arrows and air temperature by grey shading. ((a) and (c)) The 500m-RCM and ((b) and (d)) the 1.5km-RCM.

−1.41 and −0.91 respectively, and their corresponding root mean square error (RMSE) is about 4 knots at each site. The diurnal cycle, which has a greater range at Portland compared with Isle of Portland, is reasonably well modelled (Figure 10). Figure 10(a) compares the diurnal cycle of the observed wind speeds with those produced from the 500mRCM over the same period at the two sites. The bias is greatest in the early hours of the morning when it reaches 2 knots at Portland. Both wind strength and direction from the 1.5km-RCM at Portland are closer to observations (Table 2, Figure 10(a)). However, simulated values of wind speed at Isle of Portland are very low (Table 2, Figure 10(a)) and not realistic when compared with the observations, due to the orography not being as well resolved in the model. Such simulated low winds also can be seen in Figure 6(b). The bias in 500m-RCM at Portland could be an artefact of well-resolved shelter winds (Figure 6(a)) when extracting time-series at a point, because the 500m-RCM captured wind changes in magnitude and direction well at the open land of Isle of Portland when compared with the 1.5km-RCM. Table 3 shows the frequencies of wind speeds at 4 knots intervals. The spreads of simulated wind speeds are slightly lower than those in the observations. There is a tendency, as might be expected, for the model wind speed to be more concentrated in the normal range of 4–12 knots. Both models and observations show less frequency in the c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

extreme range of >16 knots at Portland. However, at Isle of Portland, the 1.5km-RCM simulated more events in the 20 Mean Maximum

Portland

Isle of Portland

Wyke Regis

M1

Obs

M1

Obs

M1

Obs

70 17 3 4 3 1 2 1.1 29.0

78 6 3 8 3 0 2 1.0 36.2

69 12 2 9 3 5 0 1.5 21.6

76 8 4 7 2 2 1 1.0 24.4

71 13 5 3 5 1 2 1.2 75.6

74 8 5 5 4 3 1 1.3 50.6

observations, with higher values on average returned by the manual observer. Figure 10(e) shows for 1999–2002 that the model does not produce enough cloud at both sites. Model bias becomes larger during the daytime at Portland by a fraction of 0.1 and during the night-time by a fraction of 0.15 at Isle of Portland compared with the manual observations from Portland. Overall, despite the difficulties caused by the observations, the model has well demonstrated its ability to simulate the trend of cloud variations, with a good representation of the diurnal cycle. There is a reasonably large bias in average visibility between the model and observations (Table 2). However, of most interest in the visibility data are occurrences of low visibility, and especially visibility of less than 1000 m, which is defined as fog. Figure 10(f) shows that the diurnal cycle of observed fog occurrence is reasonably well represented by the model. The main difference at Portland is that there is much 4.3. Cloud cover and visibility more fog observed in the afternoon and early evening. At the For cloud and visibility verification, the authors compared Isle of Portland, the model has more fog than observations only the cloud and visibility in the 500m-RCM with during the night. Also, unlike in the observations, the model observations to see how the model performed. In the generally has more fog at the Isle of Portland than at the mesoscale NWP of MetUM, the visibility is estimated by Portland site. using a representative value of relative humidity in the grid-box together with prognostic aerosol content (Clark 4.4. Rainfall et al., 2008). Observations of cloud cover and visibility from Portland and Isle of Portland were used to verify the hourly General statistical properties are shown in Table 2, cloud fraction and visibility values from the model. Manually while Table 5 lists the modelled maximum daily rainfall observed hourly data from Portland are available from 1993 amounts together with percentage frequencies of ranges to 1998. At Isle of Portland, 3 h manual observations are of daily rainfall totals compared with those observed at available during daytime hours from 1994 to 1998, and from stations over the simulation period. Note that each of the 1999 to 2002 the continuing 3 h manual observations are verification stations covers a slightly different period. In supplemented by hourly observations for the intervening addition, convection parametrizations in both models are hours made automatically using a ceilometer (for cloud) off (Table 1), hence there is no distinguishing between the large-scale and convective rainfall. Table 2 shows that and a weather sensor (for visibility). Local cloud cover and visibility, which normally are not the simulated average daily rainfall amounts in the 500mwell simulated in coarse-resolution models, were reasonably RCM correspond well with those from the observations at well represented by the 500m-RCM (Table 2, Figure 10e). Portland and Wyke Regis, although there is a bias of 0.5 mm There is only a small bias between average cloud cover per day at the Isle of Portland. The 1.5km-RCM seems to values from the model and from observations of less than simulate wetter conditions overall. Figure 11 shows the spatial variations in average monthly 0.1, or 10% cover (Table 2). The average error is 0.03. There are difficulties with accurately observing cloud cover rainfall from the high-resolution domain compared with amount, which means that the differences shown here are observed rainfall for land points. Figure 11(a) shows not necessarily due to model error. In particular, there the simulated average monthly rainfall from the 500 m are systematic differences between manual and automatic resolution model. The range of values of around 30–70 mm c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

Q. J. R. Meteorol. Soc. 139: 1964–1976 (2013)

Ultrahigh-Resolution Regional Climate Model Prediction of Local Climate

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

1975

Dorchester

Weymouth

Ringstead

Portland

30

40

50

60

70

-30

-20

-10

0

10

20

-30

-20

-10

0

10

20

-30

-20

-10

0

10

20

Figure 11. Average monthly rainfall (mm) over the 500 m domain for the period 20 July to 19 August, 1993–2002: (a) average rainfall of the 500m-RCM, (b) regridded average rainfall of the 500m-RCM to 5 km resolution, (c) regridded average rainfall of the 1.5km-RCM to 5 km resolution, (d) the 1.5km-RCM over the 500m-RCM domain, (e) average rainfall from the regridded observations in 5 km resolution, (f) the 500m-RCM bias, (g) the 1.5km-RCM bias, (h) bias of the 1.5km-RCM to the 500m-RCM.

compares very well with the averages over the same month-long period for this area derived from daily regridded observed precipitation (Figure 11(e)) at 5 km resolution (Perry et al., 2009). In particular, the modelled rainfall amounts along the coast, such as at Weymouth and Ringstead areas, compare well. Rainfall amounts from the model over Portland are slightly higher than those from regridded observations, but very close to observations. Figure 11(a) also shows that higher amounts of rainfall over the Dorchester area are simulated and the same feature can be seen from the observations (Figure 11(e)). However, the coverage and locations of monthly averaged simulated higher rainfall amounts over the Dorchester area are slightly different from the regridded observations. It is also to be seen that the wet bias in the model is close to the domain boundary where the influence from its ‘host’ model of 1.5km-RCM is strongest. The coarse resolution of the observations may contribute partly to such differences as well. Figure 11(d) displays model biases compared with observations, which were derived by regridding the 500m-RCM data to 5 km resolution (Figure 11(b)). This shows that the biases are generally lower than 25% and well within ±15 mm, apart from the corner in the northwest where the model has a dry bias of up to 30 mm. Results from the 1.5km-RCM (Figure 11(d)) show higher amounts of rainfall to the north of Dorchester and show much wetter climate over the study area as compared with observations and the 500m-RCM (Figure 11g and 11h), with wet biases reaching 30 mm. Table 5 shows that the 500m-RCM represents very well the frequencies of ranges of daily rainfall totals found in the observations at all three sites. The main difference in the frequency table is in the low daily amounts between 0.2 and 1 mm, which are more common in the model than in the observations. Table 5 also reveals that the model has c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

demonstrated very well its capability to capture the extreme rainfall totals that are found in the observations. 5. Conclusions The results presented in this article illustrate the added value of high-resolution modelling for gaining an insight into the climatology of a local area. Spatial variations caused by the unique shape of the orography and coastline can be seen clearly in the model results of 500m-RCM. With the analysis from the 1.5km-RCM and the 500mRCM, both models are able to resolve local features such as sea breezes but other local mesoscale features such as funnelling of winds through Portland Harbour and altitude and coastal effects are more evident in the 500m-RCM. These mesoscale features cannot to be resolved well by current RCMs, which normally have a grid length of about 25–50 km due to unresolved local orography and coastline details. The model outputs from the 500m-RCM also enable the diurnal cycle to be presented for a range of variables as well as the frequency distributions of these variables. In this study, the seasonal cycle is not analysed as the model was run only for the July–August period. The results in this article are based on a 10-year climatology of 1 month. Although this is shorter than the ideal period length of 30 years for climatology, it is long enough to cover a range of representative conditions and is a good compromise between statistical robustness and computing costs for such high-resolution climate modelling. Verification of the model results was carried out using hourly observations from two stations on Portland. The main purpose of the LCM is to generate long-term statistics, and overall the results were fit for purpose, with only a Q. J. R. Meteorol. Soc. 139: 1964–1976 (2013)

1976

C. Wang et al.

small bias in the mean for most of the climate variables. The error in hourly values was generally fairly high, but this was to be expected as it is not the purpose of the LCM to accurately model values on an hour-by-hour or day-by-day basis. The diurnal cycle was modelled very well for wind, but for other variables there was a tendency to underestimate the diurnal range. Extremes of wind speed and temperature were underrepresented in the frequency analysis, but not by a large amount. Extremes of daily rainfall were well modelled. Further work planned for the LCM includes optimization of parameters used in the three-dimensional Smagorinsky–Lilly subgrid turbulence scheme to improve and represent subgrid-scale mixing in the LCM model as well as using current 1.5 km variable resolution to replace the 4 km and the 1.5 km nesting to go with the 500 m. With further improvement of the model structure together with increased computing power, we are expecting to carry out continuous multidecadal climate simulations including considerations of simulating future local climate scenarios forced by CMIP5 (Coupled Model Intercomparison Project 5, Taylor et al., 2012) GCMs. Acknowledgement We gratefully acknowledge funding from the Met Office corporate investment fund to undertake this study. Thanks also go to the reviewers for their insightful comments and suggestions to shape up this research article. Richard Jones also gratefully acknowledges funding from the Joint Department of Energy and Climate Change (DECC) and Department for Environment Food and Rural Affairs (Defra) Met Office Hadley Centre Climate Programme – DECC/Defra (GA01101). References Ansumali S, Karlin IV, Succi S. 2004. Kinetic theory of turbulence modeling: smallness parameter, scaling and microscopic derivation of Smagorinsky model. Phys. A: Stat. Mech. Appl. 338: 379–394. Best MJ, Bornemann FJ, Chalcraft BV, Wilson CA. 2000. Mesoscale Model Upgrade – Introduction of the Land Surface Tile Sheme (MOSES 2). Forecast Research Technical Report No. 341, Met Office: Exeter, UK. Brown AR, Beare RJ, Edwards JM, Lock AP, Keogh SJ, Milton SF, Walters DN. 2008. Upgrades to the boundary-layer scheme in the Met Office numerical weather prediction model. Bound.-Lay. Meteorol. 128: 117–132. Clark PA, Harcourt SA, Macpherson B, Mathison CT, Cusack S, Naylor M. 2008. Prediction of visibility and aerosols within the operational Met Office Unified Model. 1: Model formulation and variational assimilation. Q. J. R. Meteorol. Soc. 134: 1801–1816. DOI: 10.1002/qj.318. Davies T, Cullen MJP, Malcolm AJ, Mawson MH, Staniforth A, White AA, Wood N. 2005. A new dynamical core for the Met Office’s global and regional modelling of the atmosphere. Q. J. R. Meteorol. Soc. 131: 1759–1782. DOI: 10.1256/qj.04.101. Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi S, Andrae U, Balmaseda MA, Balsamo G, Bauer P, Bechtold P, Beljaars ACM, van de Berg L, Bidlot J, Bormann N, Delsol C, Dragani R, ´ EV, Fuentes M, Geer AJ, Haimberger L, Healy SB, Hersbach H, Holm Isaksen L, K˚allberg P, K¨ohler M, Matricardi M, McNally AP, MongeSanz BM, Morcrette JJ, Park BK, Peubey C, de Rosnay P, Tavolato C, Th´epaut JN, Vitart F. 2011. The ERA-Interim reanalysis: configuration and performance of the data assimilation system. Q. J. R. Meteorol. Soc. 137: 553–597. DOI: 10.1002/qj.828. Denis B, Laprise D, Caya D, Cote J. 2002. Downscaling ability of one way

c 2013 Royal Meteorological Society and  British Crown Copyright, the Met Office

nested regional climate models: the big-Brother Experiment. Climate Dyn. 18: 627–646. DOI: 10.1007/s00382-001-0201-0. Edwards JM, Slingo A. 1996. Studies with a flexible new radiation code. I: Choosing a configuration for a large-scale model. Q. J. R. Meteorol. Soc. 122: 689–719. Essery R, Best M, Cox P. 2001. MOSES 2.2 Technical Documentation. Hadley Centre Technical Report No. 30, Met Office Hadley Centre: Reading, UK. Feser F, Rockel B, von Storch H, Winterfeldt J, Zahn M. 2011. Regional climate models add value to global model data: A review and selected examples. Bull. A. Meteorol. Soc. 92: 1181–1192. DOI:10.1175/2011BAMS3061.1. Gregory D, Rowntree PR. 1990. A mass flux convection scheme with representation of cloud ensemble characteristics and stabilitydependent closure. Mon. Wea. Rev. 118: 1483–1506. Hardiman SC, Butchart N, Osprey SM, Gray LJ, Bushell AC, Hinton TJ. 2010. The climatology of the middle atmosphere in a vertically extended version of the met office’s climate model. Part I: Mean State. J. Atmos. Sci. 67: 1509–1525. Lean H, Clark PA, Dixon M, Roberts NR, Fitch A, Forbes R, Halliwell C. 2008. Characteristics of high resolution versions of the Met office Unified Model for forecasting convection over the UK. Mon. Weather Rev. 136: 3408–3424. Lock AP. 2001. The numerical representation of entrainment in parametrizations of boundary layer turbulent mixing. Mon. Weather Rev. 129: 1148–1163. Lock AP, Brown AR, Bush MR, Martin GM, Smith RNB. 2000. A new boundary layer mixing scheme. Part 1: Scheme description and single-column model tests. Mon. Weather Rev. 128: 3187–3199. Noguer M, Jones R, Murphy J. 1998. Source of systematic errors in the climatology of a regional climate model over Europe. Climate Dyn. 14: 691–712. Perry MC, Hollis DM, Elms M. 2009. The Generation of Daily Gridded Datasets of Temperature and Rainfall for the UK. National Climate Information Centre Climate Memorandum No. 24, Met Office: Exeter, UK. Roberts NM. 2003. The impact of a change to the use of the convection scheme to high-resolution simulations of convective events. Met Office Technical Report 407, 30 pp. Met Office: Exeter, UK. Rummukainen M. 2010. State-of-the-art with regional climate models. WIREs Clim. Change 1: 82–96. DOI: 10.1002/wcc.8. Salath´e EP, Steed R, Mass CF, Zahn P. 2008. A high-resolution climate model for the U.S. Pacific Northwest: Mesoscale feedbacks and local responses to climate change. J. Climate 21: 5708–5726. Taylor K, Stouffer R, Meehl G. 2012. An overview of CMIP5 and the experiment design. Bull. Am. Meteorol. Soc 93: 485–498. DOI: 10.1175/BAMS-D-11-00094.1. Uppala SM, K˚allberg PW, Simmons AJ, Andrae U, da Costa Bechtold V, Fiorino M, Gibson JK, Haseler J, Hernandez A, Kelly GA, Li X, Onogi K, Saarinen S, Sokka N, Allan RP, Andersson E, Arpe K, Balmaseda MA, Beljaars ACM, van de Berg L, Bidlot J, Bormann N, Caires S, Chevallier F, Dethof A, Dragosavac M, Fisher M, Fuentes M, ´ E, Hoskins BJ, Isaksen L, Janssen PAEM, Jenne R, Hagemann S, Holm McNally AP, Mahfouf J-F, Morcrette J-J, Rayner NA, Saunders RW, Simon P, Sterl A, Trenberth KE, Untch A, Vasiljevic D, Viterbo P, Woolen J. 2005. The ERA-40 re-analysis. Q. J. R. Meteorol. Soc. 131: 2961–3012. DOI:10.1256/qj.04.176. Walters DN, Best MJ, Bushell AC, Copsey D, Edwards JM, Falloon PD, Harris CM, Lock AP, Manners JC, Morcrette CJ, Roberts MJ, Stratton RA, Webster S, Wilkinson JM, Willett MR, Boutle IA, Earnshaw PD, Hill PG, MacLachlan C, Martin GM, MoufoumaOkia W, Palmer MD, Petch JC, Rooney GG, Scaife AA, Williams KD. 2011. The Met Office Unified Model Global Atmosphere 3.0/3.1 and JULES Global Land 3.0/3.1 configurations. Geosci. Model Dev. Discuss. 4: 1213–1271. Wang C, Clark PA, Haack T, Millington S. 2009. Mesoscale modelling for radar propagation prediction during the Wallops-2000 Experiment. In Proceedings of 23rd Conference on Weather Analysis and Forecasting/19th Conference on Numerical Weather Prediction, Omaha, NE 1–5 June 2009. American Meteorological Society: Boston, USA. Wang C, Wilson D, Haack T, Clark P, Lean H, Marshall R. 2012. Effects of initial and boundary conditions of mesoscale models on simulated atmospheric refractivity. J. Appl. Meteorol. Climatol. 51: 115–132. DOI: http://dx.doi.org/10.1175/JAMC-D-11-012.1. Wilson DR, Ballard SP. 1999. A microphysically based precipitation scheme for the UK Meteorological Office Unified Model. Q. J. R. Meteorol. Soc. 125: 1607–1636.

Q. J. R. Meteorol. Soc. 139: 1964–1976 (2013)