Case study of an urban heat island in London, UK: Comparison between observations and a high resolution numerical weather prediction model Siân Lane, Janet Barlow, Humphrey Lean With thanks to Christos Halios, Aurore Porson, Sylvia Bohnenstengel, and John Lally

Why is this interesting? Lots of people rely on forecasts of urban weather, but cities can be difficult to represent in forecast models. Urban schemes must be compared with observations to assess their accuracy and impact. Over 80% of the UK population live in urban areas (Office for National Statistics, 2010). Urban areas have their own microclimate (urban heat island, increased pollution etc.) With a grid-spacing of ~1 km, NWP models like the UK Met Office’s UKV model are able to resolve some features of cities, but many features which have a significant effect on the boundary layer are still too small to be resolved, so must be parameterized. Note the animation covers up a more ‘sciencey’ list of motivations

Met Office 1.5 km model grid Regent’s Park The urban surface is very varied (parks, river, buildings etc.) Tiled surface scheme can deal with change in land-use, but still leaves a lot of variation to be represented by a single urban tile (different heights/arrangements of buildings, building materials, etc.) Point out location of obs. sites Marylebone Road Hyde Park

Representation of urban surfaces in the UKV Separate Roof and canyon tiles Accounts for in-canyon radiation exchange Roof thickness and canyon dimensions can be altered to suit local morphology Single ‘slab’ with defined roughness and heat capacity Radiatively coupled with soil (like a vegetation canopy) Simpler to implement - MORUSES is more physically realistic, does it produce better results? MORUSES (new) (Porson et al. 2010) Best scheme (operational) (Best 2005)

Science questions Can the UKV (1.5 km version of the Unified Model) accurately represent the structure of the urban boundary layer? Can the UKV reproduce the diurnal behaviour of temperature, heat fluxes etc. in London? Does the new heat flux scheme of MORUSES produce more physically realistic behaviour? What is causing any differences noted between the UKV and observations?

Case study (30/09 – 01/10 2011) Chosen because no clouds or fronts affected the area – simple(ish) situation. Conditions would be expected to produce a strong UHI, and remained the same for several days. The UBL would be strongly influenced by the urban surface energy balance in these conditions Chilbolton (~100 km WSW of London) used as a rural reference site wind from south not strong enough for advection to be a problem Point out location of high pressure system Wind

London, Chilbolton – model = pale lines Day time BL depth is underestimated in both rural and urban areas Height (m) Time (UTC) London, Chilbolton – model = pale lines

Using MORUSES increases sensible heat flux and reduces model time lag. Rooftop BT Tower - still underestimate @ BT during day Model not –ve at night after scaling -> BT is in residual layer so fluxes set to zero H @ BT is larger than at roof -> possibly due to location of roof obs in RSL – analagous to momentum flux profiles? Profiles in canyon & RSL would be useful (Model pre-scaling denoted by plain line)

Improvement in H doesn’t translate to an improvement in temperature Rooftop (21 m) BT Tower (190 m) - scaled Suppressed diurnal cycle at rooftop site is similar to modelled rural behaviour. Delayed warming and cooling seems to be an urban effect.

Rural site shows similar suppressed diurnal cycle, but no lag Chilbolton Temperature (K) Time (UTC) Observations, operational UKV, UKV + MORUSES

Model captures structure of the wind field well, but tends to underestimate speed slightly 6 12 18 Time (UTC) Time (UTC)

Long-term comparison between model and observations doesn’t show a clear difference Temperature (150 days) Underestimate during afternoon? TUKV – TBT (K) Do you have rural station comparisons to put alongside these urban ones? That will be a question folks ask, how well was the model doing outside urban areas? **JFB: remind folks what the tendency was in the case study Time (UTC) 12

Hourly average wind speed (150 days) Long-term comparison between model and observations doesn’t show a clear difference Hourly average wind speed (150 days) The case study results may only apply in similar conditions – the longer dataset requires filtering to determine the affects of different meteorological conditions WindUKV – WindBT (ms-1) I guess the point here is that the spread is large – be clear what the whiskers and outlier points mean – is this all hourly averages? How many datapoints? Time (UTC) 13

www.actual.ac.uk – s.e.lane@pgr.reading.ac.uk Conclusions Some features of the UBL are reproduced by the model (higher BL top in urban area, wind field structure), but some are not (BL growth rate and maximum depth). This is influenced by the background state of the model. The operational UKV underestimates heat flux in London, and produces a phase lag of ~ 1-2 hours in the diurnal cycles of T and H. (heat capacity too large?) Using MORUSES improves the modelled values and timing of H, but only gives a slight improvement in T. These conclusions cannot be generalised to the whole dataset – filtering by different conditions is required www.actual.ac.uk – s.e.lane@pgr.reading.ac.uk