1 of 35

The Met Office Urban-scale modelling programme��Humphrey Lean, Jon Shonk, Lewis Blunn,�Kirsty Hanley and many others.��Urban Scale Modelling Research�RMED, MetOffice@Reading��

Urban Fluid Mechanics Meeting, Reading, Sept 2022

www.metoffice.gov.uk

2 of 35

Plan of Talk

Introduction to Urban-scale modelling.

Issues with Urban-scale modelling.

Specific questions around urban representation and applications.

Preliminary results of using aircraft to look at urban boundary layer

Conclusions.

3 of 35

The path to high resolution

Urban Scale Modelling

Vision Statement:

Deliver an enhanced Urban-scale modelling capability (an atmospheric model with grid lengths in the range 25-300m) for application across timescales to exploit next-generation supercomputing.

Key part of Met Office Research and Innovation Strategy. “The path to High Resolution” theme

Comparison of London urban land use fraction for 100m (left) and 1.5km models.

100m

1.5km

4 of 35

The path to high resolution

Urban Scale Modelling

Scope and Terminology

Talking about NWP/climate models with gridlengths in range ~300-25m (c.f. current UK model 1.5km)

We use the term “Urban-scale” because the some key applications are likely to be urban related. Analagous to term “Convective scale” for ~1km models.

Don’t confuse “Urban-scale” with “Urban”: “Urban-scale” encompasses non-urban applications of these high resolution models.

5 of 35

Potential benefits of high resolution

Can capture urban/neighbourhood scale effects.

Potential for neighbourhood scale forecasts (within predictability constraints).

Good representation of boundary layer structure critical for AQ/Dispersion.

Better representation of underlying surface e.g. urban fraction, orography.

Benefits for orographic precipitation, cold-pooling in valleys, temperature distribution in cities.

Model dynamics will better resolve some atmospheric processes such as convergence lines, entrainment, turbulence.

Benefits for convection, sea breezes, tornadoes

The path to high resolution

Urban Scale Modelling

6 of 35

The path to high resolution

Urban-scale Modelling

Key activities.

300m London Model upgrade (PS49).

300m variable resolution, small ensemble.

Paris 2024 RDP - focus on 100m scale forecasting

Model Intercomparisons (Heat, convection).
Obs campaign, PANAME, 2022 (includes MO observational capabilities).

WesCon Wessex Convection Experiment 2023.

FAAM, OBR surface obs, Chilbolton, others..

Collaboration with city observational campaigns

FUTURE, ASSURE, Urbisphere etc..

Focus on Urban Heat climate services, NSWWS

Work towards 100m UK (or near UK) model for case study work.

7 of 35

Key problems which need to be addressed

Cost
Turbulence grey zone/representation of convection.
Predictability/ensembles
Observation sources
Data assimilation
Representation of urban surface (inc sources of urban data).

8 of 35

Cost of Urban-scale models.

These models are very expensive!

Emphasis of project is on developing capability and understanding resolution trade-offs for different applications.

Configuration	Tstep	levels	Npts factor	Npt facs*tstep fac
1.5km (UKV)	60s	70	1.0	1.0
300m LM	12s	70	25.0	125
100m	3s	140	450	9000
55m	1s	140	1487	87,480
25m	1s	140	7200	432,000

Approx. relative costs of models for same area and run length based on current research configurations.

9 of 35

When do we need Urban-scale models?

Example of 1.5m temperature over London on a sunny day in 100m model.

Much of this structure (river, large parks etc) is clearly coming from surface information.

Might be able to do as well using downscaling/ML techniques driven by coarser model.

However general NS streaky structure from BL rolls which are meteorological phenomenon reproduced by the model.

10 of 35

��COPE IOP2 5/07/2013 w comparison with aircraft flight level vertical velocity.

Kirsty Hanley

200m, 100m start to resolve turbulence.

Representation of convection/turbulence.

Transect across Cornwall peninsula and peninsula convergence line

11 of 35

Convective Boundary Layer

300m

100m

~1km deep BL 300m model struggling to resolve. Grid scale structure – grey zone artifacts.

12 of 35

Need to correctly handle turbulence grey zone

It has been noticed previously that 300m runs suffer from gridscale w structure in CBL situations.

Due to depth of mixed layer being similar to effective resolution of model.

Same is true in growing BL in 100m model if look early while mixed layer is shallow.

Answer should be that gridscale motions are parameterised (scale aware scheme).

13 of 35

Cumulus Convection

A-priori would hope convection would improve with increasing resolution.

Experiments show this is the case in some ways (e.g. light rain heavy rain balance) but not in others (e.g. extra small cells).

Expect this is due to compensating errors in low res models (which config is based on) and also due to inadequate (scale aware) parameterisations.

14 of 35

The path to high resolution

Urban-scale Modelling

100m scale convection issues.

15 of 35

Location of convergence lines agrees well but no rain in reality.
Need to understand reason for spurious rain:

Hypothesis: too strong vertical velocity due to turbulence at edge of updrafts not being correctly represented. WesCon will shed light on whether this is correct.
Could also be microphysics issue.

Common issue with 100m model is small showers precipitating too easily

Kirsty Hanley

16 of 35

Predictability

As we move to forecasting on smaller scales unpredictable scales become larger compared to the areas you are interested in forecasting.
Example shows 12 hour forecast of a storm in the London model which could easily be elsewhere in the domain.
Depending on what model is required for (i.e. for anything other than locally forced effects) high res modelling will need to be in an ensemble context.
Need to understand optimal way to set up 100m scale ensembles.

17 of 35

Observations

Obs R&D in Met Office working on strategy for Urban-scale observations.
Includes obs for model development, verification and data assimilation.
Lot of interest in cloud sourced observations (WOW, netamo, Davis etc).
Many issues to think about regarding representivity, errors and more practical issues such as ownership, metadata, timeliness.

Netamo stations for London area for 2020.

Blue points for full 12 months.

These points have T and RH. Half have rain and quarter wind data of all points in UK (may be different in urban areas).

18 of 35

Data Assimilation

An obvious possible use for Urban-scale models is nowcasting. For this will need DA at these scales.
Very little work on this so far.
Step in right direction: Ed Pavelin (Met Office nowcasting group) developing OI technique to fit WOW observations to gridded analysis.

19 of 35

Summary of Challenges in Urban-scale NWP

Costs!
Stability and dynamics
Representation of turbulence and convection
Observations to use for verification
Data assimilation
Predictablity at these scales
Representation of the (urban) surface
Detailed land-use classification
Spin-up at the boundaries

Planned: Peer-reviewed position paper to review progress in 100m-scale modelling (not exclusively urban) and scope out R&D work needed.

2-3 Experts on each of above topics
Invite-only workshop in Dec 2022 @ KNMI, the Netherlands

Lead authors/organisers: Humphrey Lean (UKMO), Natalie Theeuwes (KNMI)

20 of 35

Motivation for urban NWP

Large proportion of the population live in cities

There are a number of meteorological hazards that we would like to forecast on weather and climate timescales.

Several involve other coupled models (e.g. air quality requires chemistry model, flooding requires hydrology) but:

Good representation of urban meteorology is fundamental

21 of 35

Urban Surface Model Development Priorities

improving representation of anthropogenic fluxes;
implementing a vertically distributed canopy;
enhancing land cover and morphology data;
including vegetation and water in the urban canopy;
making changes specific to high resolution running.

22 of 35

Anthropogenic Heat Emissions (AHEs)

Model in:	Input Data	Spatial Usage	Time Variation	Dynamic?
JULES trunk	1995 -2003 average UK monthly energy consumption	Global	Monthly, no diurnal variation	No
JULES branch	City/country level data used by MO partners	Regional	Diurnal variation, same every day	No
Future	?	Global	Annual, daily, and diurnal	Yes

Firstly, I am going to give a brief overview of the AHE models used currently in JULES which is the land surface model coupled to the UM, and a new anthropogenic heat emissions model that is currently being developed.
The AHE model in the JULES trunk is based on 1995 -2003 average UK monthly energy consumption. It was designed for use in UK regional forecast models but can be used globally in UM suites. The emissions data is monthly and there is no diurnal variation.
Since MO partners often use JULES for urban applications outside of the UK, a branch of JULES was developed to enable users to supply their own AHE ancillary, and a diurnal profile is applied. There are currently plans to get this into the JULES trunk.
Ideally there should be a AHE model that is applicable globally, and is dynamic i.e., responds to temporal changes in air temperature and population density etc.
Sylvia Bohnenstengel in collaboration with Sue Grimmond is developing a model of the form here….. The anthropogenic heat flux is a function of population density, and there are terms relating to building cooling and heating when air temperature goes above and below certain thresholds, and there is a term relating to temperature independent emissions such as from humans and transport. The values of the coefficients can be made to depend on socio, economic, and cultural conditions, e.g. in higher-income countries like Singapore A/C is more likely to be turned on than in lower-income countries like Sri Lanka when temperature exceeds a threshold.

(Rho is the population density. A0 is the temperature independent energy use parameter. The CDD and HDD are temperature relative to some threshold temperature required for air conditioning and heating to be turned on respectively.)

23 of 35

Steps to a vertically distributed canopy

Our final aim is a fully distributed canopy scheme with:

vertically distributed drag;
vertically distributed scalar fluxes;
vertically distributed radiative fluxes.

We have a plan to implement the vertically distributed drag as the first step.

D(z)

z

Drag will be applied, either diagnostically or prognostically, within the canopy as an extra term in the boundary layer solution.

Jon Shonk

My work has been towards a vertically distributed canopy. At present, the surface scheme only interacts with the boundary layer scheme via a single interface at the bottom model layer using a roughness length approach to define properties at the first level. This means that buildings cannot interact with the atmosphere throughout their height – the entirety of the drag they impart is through the bottom level of the model. Our grand plan here is to create a vertically distributed canopy that allows interactions throughout the height of the buildings. This is far from a trivial task – indeed, three large tasks are presented here, and we will address them in the coming years. The first part of the work is to focus on vertically distributed drag – the MORUSES approach here is simplest, so we can make improvements more easily.

24 of 35

Raising the blending height

The first step towards vertically distributed drag is to couple the atmosphere and surface schemes at a realistic blending height – above level one of the atmosphere scheme.

Blending height needs to be diagnosed.

1

2

3

4

5

6

z

X

blending height

2.0 m

5.3 m

10.0 m

16.0 m

23.3 m

32.0 m

(typical level heights)

(model level #)

Jon Shonk

25 of 35

Prognostic vs diagnostic profiles

1

2

3

4

5

6

(model level #)

U(z)

z

U(z)

z

1

2

3

4

5

6

(model level #)

The profile under the blending height can be either prognostically or diagnostically determined.

Jon Shonk

26 of 35

Summary

Vertically distributed urban canopy

Anthropogenic heat flux

JULES branch 🡪 Trunk
New dynamic parameterisation

Land cover and building morphology

Working towards new global land cover and morphology in ANTS
Forward plans to include urban vegetation and water in the canopy scheme

Strong collaboration with UM partners will be key

Raised blending height

(Mar 2023)

Prognostic drag

(2023-)

Diagnostic profiles

(Mar 2023)

Fully vertically distributed scheme

(2024-)

So to summarise our urban-scale parametrisation progress and plans:
A vertically distributed urban canopy scheme is currently being worked on
The AHE model currently on the JULES branch is being incorporated into trunk
A new dynamic model is being developed by Sylvia
We are working on including new improved land cover and building morphology data, and I will hopefully include that in ANTS, and we are thinking about how we might better represent green and blue land cover in surface scheme
We welcome collaboration from UM partners. Working with Mat Lipson at the BoM has helped us start making some really good progress on the land cover front, and it will be great to get feedback on new parametrisations when UM partners test them.
Okay thank you for listening. We will be happy to take any questions.

27 of 35

Questions about Urban Applications

100m model clearly isn’t resolving buildings. Rely on parameterising building effects.

Question is what can be done for urban applications with these models and what will still require CFD modelling of buildings/streets. NB work (e.g. NIWA) to couple CFD models to UM.

Advantage of 100m models is ability to take into account larger scale features.

100m grid superimposed on city of London

28 of 35

Example of convective overturning over London in 100m model.

Note sea breeze fronts at the end of animation.

25^th July 2012

29 of 35

Spatial structure of urban boundary layer

Good representation of boundary layer structure critical for air quality.

AQ/Dispersion often talked about as potential application of urban scale atmosphere models.

Look at comparing modelled spatial variations with observations.

From Oke 1987

30 of 35

100m model South-North transect

Mixing height (large domain)

Mixing height (small domain)

Urban fraction

Have previously validated clear convective boundary layer overturning against LIDAR observations.

Spatial variation across urban area looks sensible. Max mixing height appears to be capped by larger scale more stable region.

No validation of spatial structure.

Should be possible in near future due to field campaigns with many sites across cities.

0km

80km

Lean et al 2019 http://doi.wiley.com/10.1002/qj.3519

31 of 35

Met Office Atmospheric Survey Aircraft (MOASA).

Cessna-421 aircraft flown at about 480m over London.
Funded by Clean Air SPF project.
For boundary layer structure use upward pointing backscatter LIDAR and flight level winds (particularly vertical wind) from AIMMS probe.

Image courtesy Debbie O’Sullivan, Met Office, 2021

32 of 35

Urban Boundary Layer Spatial dependence

Upward pointing aircraft lidar

100m Model aerosol

South – North leg

Technique clearly shows potential for comparing spatial structure of BL to model.
Lacking a case with clear urban signal due to unfavourable meteorology (E wind).

22 Jul 2021

33 of 35

24/03/2022 – London poor AQ case

Aircraft level vertical velocity comparison.

Not much sign of urban effect in boundary layer.

Model did correctly predict higher aerosol concentrations over city due to higher emissions

34 of 35

Conclusions

Met Office has a project to develop the capability for “Urban-scale” 100m models.

Although these models are promising in research there are a number of challenges to enabling their use for practical applications.

For urban applications development is required to improve the representation of the urban surface along with research to understand the utility of these models for practical applications.

Have presented a recent attempt to evaluate Urban-scale models representation of the spatial structure of the urban boundary layer.

35 of 35

Questions?