1. D.S.Hammond, L.Chapman & J.E.Thornes
Improving estimates of roughness length (Z0) in a road weather prediction model using airborne LIDAR data
D.S.Hammond, L.Chapman & J.E.Thornes
School of Geography, Earth and Environmental Sciences, University of Birmingham, Edgbaston, Birmingham, B15 2TT, UK
2010 Standing International Road Weather Commission Biennial Conference, Quebec City, Canada, February 5th -7th 2010.

2. Geographical & Infrastructure Parameters used to drive the road weather prediction model
Geographical Parameters
Infrastructure
Sky View Factor (ψs)
Road Type
Altitude
Roughness Length (Z0)
Slope
Traffic Density
Aspect
Emissivity
Latitude
Albedo
Longitude
Some parameters can be easily obtained e.g. Altitude and slope from DEM, sky view from fisheye photography or by proxy techniques
Other parameters are less well defined in the model – we’ve been looking at ways of better parameterising some of these key variables in the model
Today I’m going to focus on one of these less well defined variables, looking at work we’ve recently undertaken to improve parameterisation of roughness length in the route based forecast model

3. Roughness length (Z0)
Air flow in boundary layer largely controlled by frictional drag imposed on flow by the underlying surface
Source: Oke (1992)
So first of all, what do we mean by roughness length?
Well, air flow in the boundary layer is largely controlled by frictional drag imposed on flow by the underlying rigid surface
This frictional drag retards the air flow close to the ground, causing a sharp decrease in wind speed near the surface
Z0 is a measure used for quantifying the aerodynamic roughness of a surface that causes this frictional drag
Z0 - measure of the aerodynamic roughness of a surface
“Height at which the neutral wind profile extrapolates to a zero wind speed.” (Oke, 1992)

4. How is Z0 currently parameterised in the road weather prediction model
Simple look-up table of Z0 values assimilated from scientific literature
Ordinal dataset of Z0 – major oversimplification
Rural
Semi-rural
Suburban
Urban
City Centre
Motorway
0.50
0.75
1.00
2.00
A-road
0.25
B-road
C-road
Parameterised with respect to ordinal land use and road type classifications at each forecast point in the model using a simple look-up table of Z0 values assimilated from scientific literature
What we end up with is an ordinal classification of Z0 which is a major oversimplification
Fails to account for variations in the surface elements within land use classes, and takes no account of wind direction and the surface elements within the upwind fetch
Modified from Chapman (2002)

5. Example of Z0 classification
If we visualise this in a GIS – we can see from this example that all the forecast points located on the motorway or a C classified road passing through a semi-rural land use area are assigned a Z0 value of 0.5m.
Whilst all the points located on a C classified road that passes through a rural land use area are auto assigned a Z0 value of 0.25m
Wanted a method of parameterising Z0 which gets us away from this simple ordinal classification
Looking for a method which can provide realistic estimates of surface roughness at each forecast point along our route, taking into account the surface elements within the upwind fetch – this would be a vast improvement on the current parameterisation

6. Use LIDAR data to obtain surface elements heights
(Oke, 1992; Grimmond & Oke, 1999)
Simple rule of thumb:
(Garratt, 1992; Hanna & Chang, 1992; Grimmond & Oke, 1999)
Effective roughness length (Zeff):
(Vihma & Savijärvi, 1999)
The technique which is proposed in the paper uses height data from high-resolution LIDAR data coupled with spatial processing techniques to obtain an estimate of Z0eff at each forecast point
Technique relies on the well-known simple rule of thumb whereby Z0, to a first order, is related to the height of the surface elements by an empirical coefficient derived from observation
Numerous studies have estimated this coefficient to be approx equal to 1/10th of the height of the surface elements
But this in itself doesn’t account for variations in surface elements within the upwind fetch, and whilst Z0 is well defined for homogeneous terrain, in a road environment where we have much more heterogeneous terrain, its much more appropriate to calculate an Z0eff value based on the distribution of local Z0 values
The simplest way of calculating Z0eff is to take the areal average of the available local Z0 values within a defined upwind area
LIDAR data © 2009 Landmap

7. Calculate Z0eff using areal area average of local Z0 values
Methodology
Process LIDAR data
DSM – DTM = ZH
Apply f0 = 0.1 rule of thumb
Z0 = 0.1 x (ZH)
Calculate Z0eff using areal area average of local Z0 values
Z0eff = <Z0>

8. Methodology ArcMap Focal Mean neighbourhood functionZ0
247.5°
292.5°
Upwind Fetch (m)
5 distances of upwind fetch = 5 Z0eff datasets
Prevailing westerly wind
This was done in a GIS by running a focal mean neighbourhood function on the roughness modified LIDAR data at each forecast point
The neighbourhood function was modified to account for 5 different distances of upwind fetch, from 100m out to 500m, for a prevailing westerly wind direction
Z0eff roughness values were calculated for each forecast point based on the average of the local Z0 values within the defined upwind areas

9. Distribution of Z0eff values
If we take a look at the percentage distribution of Z0eff values around the route, we see that the roughness values are positively skewed towards the lower end of the roughness scale, as we might expect given the predominantly rural to suburban nature of the route
Maximum Z0eff values occur with an upwind fetch of just 100m, and are mainly located in the urbanised city centre as you’ll see in the next slide, where values up to 3.1 metres are found
As the distance of upwind fetch increases, the range of roughness values around the route decreases, most likely due to the dampening of average surface element heights by an increasing proportion of lower-rise surface elements within the defined neighbourhood area over larger fetches

10. Landscape Description
Distribution of Z0eff values
Davenport classification of effective terrain roughness
Z0 (m)
Landscape Description
“Sea”
Open sea or lake (irrespective of wave size), tidal flat, snow-covered flat plain, featureless desert, tarmac and concrete, with a free fetch of several kilometres.
“Smooth”
Featureless land surface without any noticeable obstacles and with negligible vegetation; e.g. beaches, pack ice without large ridges, marsh and snow-covered or fallow open country.
“Open”
Level country with low vegetation (e.g. grass) and isolated obstacles with separations of at least 50 obstacle heights; e.g. grazing land without wind breaks, heather, moor and tundra, runway area of airports. Ice with ridges across-wind.
“Roughly Open”
Cultivated or natural area with low crops or plant covers, or moderately open country with occasional obstacles (e.g. low hedges, isolated low buildings or trees) at relative horizontal distances of at least 20 obstacle heights.
“Rough”
Cultivated or natural area with high crops or crops of varying height, and scattered obstacles at relative distances of 12 to 15 obstacle heights for porous objects (e.g. shelterbelts) or 8 to 12 obstacle heights for low solid objects (e.g. buildings).
6. 0.5
“Very Rough”
Intensively cultivated landscape with many rather large obstacle groups (large farms, clumps of forest) separated by open spaces of about 8 obstacle heights. Low densely-planted major vegetation like bush land, orchards, young forest. Also, area moderately covered by low buildings with interspaces of 3 to 7 building heights and no high trees.
7. 1.0
“Skimming”
Landscape regularly covered with similar-size large obstacles, with open spaces of the same order of magnitude as obstacle heights; e.g. mature regular forests, densely built-up area without much building height variation.
8. ≥ 2.0
“Chaotic”
City centres with mixture of low-rise and high-rise buildings, or large forests of irregular height with many clearings.
Here we have our new Z0eff values (for 100m fetch) plotted along our route, categorised using the Davenport classification of effective terrain roughness, which is seen in the literature as probably the best field validated roughness classification to date
If we look at the map we can see straight away that the distribution of roughness values are what we’d expect to see, with the highest values in the city centre and the lowest values towards the more rural westerly end of the route
Also the range of roughness values is typical of the values we would expect based on the Davenport classification
In the city centre we have values above 2 metres that fall within the “chaotic” category, and the vast majority of the forecast points located in the rural and semi-rural areas of the route have roughness values between 3 and 25cm, placing them within the “open”, “roughly open” or “rough” categories of Davenport classification
Likewise, most of the points located in the suburban and urban areas of the route have roughness values between 25cm and half a metre, which corresponds to the “rough” and “very rough” categories
So, the overall range of roughness values seems typical of the values we could expect based on the general land use classes around the route
LIDAR data © 2009 Landmap

11. Ordinal v Ratio Dataset
Existing Ordinal Z0 Classification
New LIDAR based Z0eff Classification
If we zoom in to our map to look in a bit more detail, we see immediately the improvements in roughness parameterisation using the LIDAR data
From the figure on the left we see that the existing ordinal Z0 classification would have assigned all the points along the motorway a roughness value of 0.5m, and the points along the minor road a value of either 25cm or 0.5m depending on the land use classification at each point
With the new LIDAR based Z0eff values however, we can clearly see variations in surface roughness along both the motorway and the minor road resulting from variations in the height of the upwind surface elements
Roughness values along the motorway now range from 10cm up to approximately 40cm, and on the minor road we see values as high as 1.4 metres where the road passes through a small forested area
Such differences are impossible to identify with the ordinal based Z0 classification, so the new method of parameterisation using LIDAR data appears to be an improvement over the existing technique
LIDAR data © 2009 Landmap

12. OWEN Land Use (Owen et al, 2006)
Statistical Analysis
Are there significant differences in Z0eff values between land use categories?
2 land use datasets used in the comparison
Kruskal-Wallis rank-order statistical analysis
ENTICE Land Use
OWEN Land Use (Owen et al, 2006)
1. Rural
1. Villages/farms
6. Urban
2. Semi-Rural
2. Suburban
7. Light urban/open water
3. Suburban
3. Light suburban
8. Woodland/open land
4. Urban
4. Dense suburban
5. City Centre
5. Urban/transport
Roughness values were compared using 2 different land use datasets
Existing ENTICE proxy land use classes used in the model, derived via a spatial density analysis of vector road data to locate dense areas of the road network, based on the assumption that more heavily urbanised areas have a denser road network than suburban and rural areas
More comprehensive land-cover dataset of the West Midlands produced by Sue Owen at Lancaster university, derived from dimensionality reduction of 25 spatial land-cover attributes using principal components analysis

13. Kruskal-Wallis Analysis
Results of Kruskal-Wallis analyses were highly significant (p < 0.001) over all 5 distances of upwind fetch for both land use datasets
Significant differences do exist in the Z0eff values between at least two land use classes in each dataset, but it doesn’t reveal where these differences exist
Wilcoxon rank-sum Tests
Analysis performed on the Z0eff values within each independent land use class
Results of the Kruskal-Wallis analyses were highly significant over all 5 distances of upwind fetch for both the ENTICE and OWEN land use datasets
This tells us that significant differences do exist in the Z0eff values between at least two of the land use classes in each dataset, but it doesn’t reveal where these differences occur
But by examining the mean rank scores assigned to each land use category, we can get a good feel for where these differences are likely to occur
For example, with the ENTICE land use dataset the lowest roughness values a found in the rural category which has a much lower mean rank score than the city centre category
Overall the vast majority of the land use comparisons are statistically significant for both land use datasets
New method of roughness parameterisation does distinguish well between different land use categories around the route

14. Multiple Regression on Thermal Mapping data
20 nights Thermal Mapping data (dependent variable)
ENTICE GPD parameters (independent variables)
Sky View
Road Type
Altitude
Slope
Aspect
Z0eff
Thermal
Mapping
Data
As an initial indicator as to the potential influence of the new Z0eff values on road surface temperature prediction, regression analysis was performed on 20 nights thermal mapping data for the study route, using parameters from the ENTICE geographical parameter database as the independent variables in the statistical model, one of which was surface roughness
1st run - Existing Z0 classification
2nd run - New Z0eff dataset

15. Statistical Model Performance
As we can see from the plot of R2 values, statistical prediction of RST improved on all but 1 of the 20 nights which gives us confidence that the new roughness parameterisation will improve predictions of surface temperature in the numerical model

16. Potential Future Improvements
Distance of upwind fetch calculated for each individual forecast point as a function of obstacle height
Same technique could be used to assimilate a look-up table of Z0eff values for various directions of upwind fetch
Limitations
Technique assumes constant direction of upwind flow, with each portion of the upstream surface considered to be an equal contributor to the aerodynamic character at a given forecast point
Technique fails to account for moving surface elements, such as vehicle traffic

17. References