Model Sensitivity, Performance and Evaluation Techniques for The Air Pollution Model in Southeast Queensland Natalie Joanne Leishman Bachelor of Applied Science In partial fulfilment of the requirements for the degree of Master of Applied Science School of Natural Resource Sciences 2005
128
Embed
Model Sensitivity, Performance and Evaluation Techniques ...eprints.qut.edu.au/16148/1/Natalie_Leishman.pdf · performance in predicting natural processes is no easy task. The problem
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Model Sensitivity, Performance
and Evaluation Techniques
for
The Air Pollution Model
in Southeast Queensland
Natalie Joanne Leishman
Bachelor of Applied Science
In partial fulfilment of the requirements for the degree of
* Percent chance = (Number of events with ozone>6pphm or NOx>15 pphm /number of cluster events)
24
Table 3.4: Cluster definitions for significant cluster types for pollution events
Cluster 1
EAG AMB
Cold light southwesterly winds in the morning changing to warm light northerly winds by the afternoon followed by very light west-northwesterly winds overnight. Cold light west-northwesterly winds in the morning changing to warm light northwesterly winds in the afternoon.
Cluster 3
EAG AMB
Warm light southwesterly winds in the morning changing to light northeasterly winds in the afternoon. Winds remaining from the northeasterly direction in the evening. Warm light westerly winds in the morning becoming slightly warmer in the afternoon.
Cluster 7
EAG AMB
Cold moderate west-southwesterly winds in the morning warming slightly during the day before becoming cooler light overnight winds. Cool light west-northwesterly winds in the morning changing to moderate westerly winds in the afternoon. High atmospheric pressure, no rain.
Cluster 14
EAG AMB
Cool moderate southwesterly winds changing to warm moderate east-northeasterly winds in the afternoon before cooling and becoming still overnight. Cool very light west-southwesterly winds changing to warm light northeasterly winds. High pressure system
Cluster 15
EAG AMB
Cold moderate southwesterly winds changing to easterly winds in the afternoon becoming cold light southerly winds in the evening. Cool light west-southwesterly winds changing to warm very light easterly winds in the afternoon. High pressure system
Cluster 18
EAG AMB
Warm moderate northerly winds becoming warmer in the afternoon with strong north-northeasterly winds before cooling slightly in the evening, as winds are moderate northerly. Warn light north-northwesterly winds in the morning changing to hot moderate easterly winds in the afternoon.
Cluster 22
EAG AMB
Very light warm easterly winds in the morning becoming warmer during the day with moderate east-northeasterly winds in the afternoon. Still warm in the evening with winds light and from the northerly direction. Warm light northeasterly winds becoming hot moderate east-northeasterly winds in the afternoon.
Cluster 23
EAG AMB
Warm very light west-southwesterly winds in the morning becoming warmer moderate easterly winds in the afternoon before easing in the evening with north-northeasterly winds. Warm light northwesterly winds becoming hot easterly winds in the afternoon.
Cluster 27
EAG AMB
Warm light north-northwesterly wind changing to very warm moderate north-northeasterly winds in the afternoon. Lighter winds in the evening. Very warm light northwesterly winds, changing to hot moderate west-northwesterly winds. Low pressure system
25
3.3 Meteorological data sets
Having categorised the regional weather types, the performance of TAPM
meteorological simulator can be tested against a wider array of spatially distributed
meteorological monitoring stations in the region. There is a network of 17 at least
surface stations across Southeast Queensland. Eight Bureau of Meteorology stations
that measure wind direction, wind speed, temperature, gust, dewpoint, humidity,
pressure and rain at the surface. The measurements are taken for a 10-minute period
on the hour, 24 hours a day. Data are also available from nine EPA monitoring
stations that measure wind direction, wind speed, temperature, and rain, as well as
various pollutants such as ozone, oxides of nitrogen, sulfur dioxide, carbon monoxide
and coarse particles (PM10). These measurements are recorded continuously and a 30-
minute average reported.
For this analysis, five sites (mainly from the EPA network) were chosen. The sites are
Rocklea, Eagle Farm, Flinders View, Deception Bay and Moreton Island. Rocklea and
Eagle Farm represent urban locations, while Flinders View is located on the urban
fringe. Deception Bay is also located on the urban fringe but is very close to the coast.
Each of these sites also measures air pollutant concentrations. The Moreton Island
monitoring station is run by the Bureau of Meteorology and is located up on a cliff at
least 80 m above sea level and is dominated by marine influences and the sea breeze.
The location and pictures of the sites are shown in Figure 3.1. A description of each
site is found in Table 3.5.
26
Figure 3.1: Location of monitoring sites used for model validation
30 km N Is a coastal site that has been erected in a residential area. It began in 1994 and records wind speed and wind direction (10 m) as well as ozone and nitrogen dioxide (4 m). The site is compliant with the Australian Standards for Siting at a Station with the exception of trees within 20 m of the site. It is open to sea breezes but screened to the southwest of the site.
Rocklea (ROC) 6 km SW Established as a regional monitoring station, has been relocated in 1994 in an open area surrounded by residential and commercial areas. It began in 1978 and records wind speed, wind direction relative humidity and temperature (10 m) as well as ozone, nitrogen dioxide, visibility reducing particles and particulates (4 m). It is compliant with the Australian Standards for Siting at a Station.
Eagle Farm (EAG) 10 km NE Currently located in an industrial area the site was established to monitor the local air quality particularly due to the industrial activities at the mouth of the Brisbane River. It began in 1978 and records wind speed, wind direction relative humidity and temperature (10 m) as well as ozone, nitrogen dioxide and particulates (4 m).
Flinders View (RFV)
30 km SW Located in the Swanbank Exchange Grounds, the monitoring site is surrounded by a residential area. Monitoring commenced in 1995 and records wind speed, wind direction, temperature and relative humidity (10 m) as well as ozone, nitrogen dioxide, sulfur dioxide, visibility reducing particles and particulates (4 m). It is compliant with the Australian Standards for Siting at a Station except for a tree which is located within 20 m of the instrumentation. The height of the tree is kept below that of the inlet.
Moreton (MOR) Located on the northern tip of Moreton Island, the Cape Moreton site is located approximately 80 m above sea level. The site measures wind speed, wind direction and temperature.
28
3.3.1 Surface characteristics
The expected surface roughness length based on inspections of the site and those
included in the model are listed in Table 3.6.
Table 3.6: Surface roughness and soil type characteristics for each monitoring site.
The month of June was one of the worst modelled months based on mean wind speed
and temperature as well as using the index of agreement. Investigation of the
meteorological conditions in June shows that although winter is typically the driest
season, winter rainfall for 1999 was higher than usual. In particular for the month of
June in 1999, 198.2 mm of rainfall was measured at Brisbane compared with the 25
year average of 57.3 mm.
TAPM was rerun for the month of June. The deep soil moisture content was changed
to 0.4 (compared with initial selection of 0.15) and rain processes were included.
The changes to temperature and wind speed at Rocklea and Flinders View due to the
changes in soil moisture and rain selection are illustrated in Figure 5.15 to 5.18. The
timeseries plot of temperature would suggest that the correct levels of cloud cover
(comparison of observational data with TAPM with no rain) were not predicted by
TAPM for the period from June 17 – 25 and June 27 – 30 when the predicted
minimum temperatures were much warmer than the observed temperatures.
Interestingly, the model predicts the maximums quite well for this same period at
Flinders View, but not at Rocklea.
The incorrect cloud cover may result in difficulties calculating the correct heat
balances and temperatures over night. With the inclusion of rain processes, there was
an improvement in wind speed and temperature predictions, depending on the site.
For days that were sunny and cloudless, there were no changes in the predictions. For
the end of June at Rocklea, the predicted minimum temperatures with the rain
processes included were similar to the observed temperatures as was the case with
Flinders View. However, the wind speed predictions did not necessarily improve.
The wind speed at Flinders View was significantly overestimated in June. Table 5.4
illustrates the predicted and observed means for the surface meteorology parameters
of temperature, wind speed and wind speed components u and v. Overall there is no
52
significant improvement in the IOA’s or the predicted measurements for the monthly
data set.
Table 5.4: Statistics for (a) temperature (oC), (b) wind speed, (c) wind speed component u and (d) wind speed component v, for Flinders View and Rocklea, with and without rain processes.
Changes to the rain and soil moisture improved model predictions on days that had
rain. To improve the predictions of the meteorological parameters at Flinders View a
future step may be the investigation of surface roughness and soil type.
From this preliminary analysis there does not appear to be much gain in changing the
deep soil moisture content for the initial boundary conditions when modelling the
Southeast Queensland region, but the selection of rain processes may benefit the
analysis for days when rain occurred. The heat fluxes in the model changed
significantly for the days that had rain when the rain processes were included, which
resulted in changes to the temperature predictions.
53
The high wind speed observed on the 14th of June was observed at all sites, but it was
not predicted by the model at any of them. (While the wind speed was predicted at
Flinders View it was only due to the model continually overpredicting at this site.) On
closer investigation this was because this phenomenon was not included in the
synoptic information that is fed into TAPM as an input. Therefore, the model cannot
be expected to be able to pick this event if the information is not available to it.
54
Figure 5.15: Comparison of observed temperature with predicted temperature (with and without rain and change in soil moisture) for Rocklea
Figure 5.16: Comparison of observed wind speed with predicted wind speed (with and without rain and change in soil moisture) for Rocklea
55
Figure 5.17: Comparison of observed temperature with predicted temperature (with and without rain and change in soil moisture) for Flinders View
Figure 5.18: Comparison of observed wind speed with predicted wind speed (with and without rain and change in soil moisture) for Flinders View
56
5.2.2 Selection of roughness length
Closer inspection of the vegetation category in TAPM for the area surrounding
Flinders View, identified grassland as the default land use type. The canopy height
for this category is 0.6 m, resulting in a roughness length of approximately 0.06 m.
Visual inspection of the monitoring site showed that the roughness length was much
greater. Since the monitoring site is located within a residential area, with forest
within a few kilometers, the urban land use was selected (as the canopy height of 10
m gives an approximate surface roughness of 1.0 m) and the month of June was re-
modelled (with rain processes on).
The results (Figure 5.19-20) illustrated a significant improvement in the prediction of
wind speed throughout the month of June even though the wind speed remained
overestimated. The mean observed wind speed for June was 1.3 ms-1 compared with
2.3 ms-1 for urban landuse and 3.6 ms-1 for the original TAPM run. The temperature
predictions did not improve, in fact generally for the month of June the results were
worse with the predicted minimum temperatures significantly higher (2-4oC) than the
observations. This is due to Flinders View not being an urban site. The heat fluxes
such as sensible heat and evaporative heat were investigated to check whether
temperature was dependent on these values. If it was, one would expect little change
in the fluxes between model runs, however there was a significant difference in the
heat fluxes for the different land use categories. It can be concluded that temperature
is very dependent on the heat fluxes.
The selection of urban land use with TAPM selects an alternate group of algorithms in
the model that deal with reproducing the wind parameters over urban areas (with the
model assuming that the land surface has the properties of concrete). This would
explain why the temperatures are much warmer and a better option would have been
to select a vegetation category with parameters similar to the vegetation at Flinders
View as it is on the edge of the urban area, not within it.
The wind direction is not improved by changes in the roughness length.
57
Figure 5.19: Timeseries of temperature and wind speed at Flinders View
(a) temperature
(b) wind speed
NOTE: Observations denoted by blue line, original TAPM run (vegetation = grassland) denoted by pink line and new TAPM run (vegetation = urban) denoted by green dashes).
58
Figure 5.20: Timeseries of (a) evaporative heat flux and (b) sensible heat flux at Flinders View for vegetation of grassland and vegetation of urban
(a) evaporative heat flux
(b) sensible heat flux
NOTE: Original TAPM run (vegetation = grassland) denoted by pink line and new TAPM run (vegetation =
urban) denoted by green dashes).
59
5.2.3 Sensitivity to soil type
Weather forecasting work conducted by Makar et al (2005) indicted that changes to
the heat fluxes had a significant impact on wind speed and temperature at the surface.
Soil type is another variable in TAPM that the user can select that can have a
significant effect on evaporative heat flux and sensible heat flux. A brief comparison
between two different soil types (with all other defaults kept the same was
conducted). For Flinders View the default soil type was sandy clay loam. The grid
point next to this one within the model was designated as clay. Figure 5.21 shows the
differences in wind speed and temperature between the two sites as well as the heat
fluxes. There are significant differences in the wind speed particularly and the
temperature and consideration should be given to the soil type chosen.
Figure 5.21: Timeseries of (a) temperature, (b) wind speed, (c) evaporative heat flux and (d) sensible heat flux for sandy clay loam soil and clay soil
(a) temperature
60
(b) wind speed
(c) evaporative heat flux
61
(d) sensible heat flux
5.2.4 Sensitivity to data assimilation
TAPM v 2.0 includes data assimilation as an option to help force the surface winds to
be modelled correctly. The month of June was rerun with the inclusion of Amberley
wind speed and wind direction data for June 1999. The weighting of the input data
was given as 1.0 and the radius of influence was 10 km. As expected there were
improvements in wind speed predictions at Flinders View due to its close proximity to
Amberley.
Data assimilation was not included in this work for the sites used for model
comparison. Good model evaluation should include comparison of modelled
information with field data unrelated to the inputs within the model.
5.2.5 Sensitivity to grid resolution
An analysis of TAPM’s sensitivity to the grid resolution was conducted to determine
whether local features such as topography in each nested grid were influencing air
movements. This was important due to the complexity of the topography in Southeast
Queensland, particularly with the Range to the north and south. Figure 5.22 illustrates
62
the similarity between the 1 km, 3 km and 6 km grid resolutions for wind speed.
Table 5.5 shows there is not much difference in the annual mean temperatures and
wind speeds for modelled data.
Table 5.5: Predicted annual temperature and wind speed for various grid resolutions at Flinders View.
Temperature Wind Speed
Observed mean 19.1 1.45
Predicted mean 1 km 19.23 3.55
Predicted mean 3 km 19.18 3.71
Predicted mean 6 km 19.21 3.51
Figure 5.22: Predicted wind speed at Flinders View for the 1 km, 3 km and 6 km resolution grids.
5.3 Performance of model based on cluster types
To seek more detailed information on performance, the index of agreement was
calculated for each cluster type to determine whether particular day types were
modelled more accurately by the model than others.
63
Figure 5.23 illustrates a box and whiskers plot of the index of agreement for each
cluster type. The cluster types with high and low IOAs are listed in Table 3.4.
Figure 5.23: Box and whisker plot of IOA for each cluster type for wind speed each site.
Table 5.6: Best (IOA>0.8) and worst (IOA<0.5) cluster types based on IOA based on predictions of the model (wspd and wdir)
Sites Good agreement Poor agreement
Rocklea 24, 21 1, 3, 6, 14
Eagle Farm 24, 22, 23, 20, 21, 2 3, 4, 6
Flinders View 22, 23, 20 6, 29, 11, 8, 15
Deception Bay 23, 22, 24, 20 1, 29, 3
Moreton 5, 16 23, 18, 26
Assessment of the best and worst cluster types using the index of agreement,
suggested that the day types that were more typical of summer and spring days were
modelled more accurately than winter days. Clusters 24 and 22 were predominantly
summer days with hot moderate northeasterly winds in the afternoon, while clusters
14 and 15 are more typical of winter days with cool light to moderate winds in the
afternoon from the north to easterly sector.
64
Investigating model performance based on cluster type provides more detail than
looking at 1999 as a whole year. However, in appreciating which cluster types are
predicted better than others, better clarification can be sought through looking at
cluster types that are important for a particular application. Assuming, that dispersion
of pollution is the application of interest, assessing the accuracy of the model for
cluster types that are known to be pollution-conducive days may give substantially
more information than the general annual statistics. Based on information in Section
3.2, days with warm light north-northwesterly winds in the morning changing to hot
moderate north-northeasterly winds in the afternoon at Eagle Farm (BOM station),
can be conducive to high ozone events (Cluster 27) as can very warm light
southwesterly winds in the morning changing to moderate northeasterly winds in the
afternoon (Cluster 23). The winter days with cool light to moderate winds in the
afternoon from the north to easterly sector associated with high pressure systems
(Cluster 14 and 15) can also be conducive high pollution events, with elevated NOx
concentrations observed during the morning between 7 am and 9 am. Further analysis
of the diurnal profile for these days could elicit more information about the model’s
performance. In Appendix D the diurnal profiles for summer and winter cluster types
reviewed here are presented.
5.3.1 Diurnal profiles of wind speed and temperature
Previously in this work the diurnal profile for the mean wind speed was presented for
Deception Bay. The results showed that on average the predicted wind speed was
consistently 2 ms-1 higher than the observed wind speeds for each hour. The
differences between the predicted wind speed and observed wind speed for a summer
cluster type day and a winter cluster type day for Deception Bay are shown in Figure
5.24. The results illustrated very different profiles for wind speed. For the summer
day type the late afternoon breeze is underpredicted by the model (mean difference
approximately 2 ms-1), whereas for the winter day type for the same period the wind
speed was overpredicted by the model (mean difference 3 ms-1).
The mean difference between predicted and observed wind speeds during the day for
summer cluster day types is less than 0.5 m/s whereas for winter it is much higher.
This difference in wind speeds is due to the wind direction. For the summer cluster
65
days the wind direction during the day (7 am to 4pm), predicted by the model was
predominantly northwesterly, whereas for the winter days investigated the
predominant wind direction was southeasterly for the same time period. Analysis of
the overall wind speed for the year showed that the greatest difference in predicted
and observed wind speeds at Deception Bay resulted when winds were blowing from
the southeast, therefore it is not surprising that the diurnal profile for the winter days
demonstrates a different profile than for summer.
The diurnal profiles of the difference between predicted and observed temperatures at
Flinders View for the two synoptic cluster types is shown in Figure 5.25. The annual
predicted mean temperature shown previously showed very little difference to the
observed mean temperature. The mean difference in temperature at Flinders View for
the winter cluster type illustrates that the temperatures before midday are over
predicted compared with the afternoon, night temperatures. The frequency of winds
tends to be from the southeast in the afternoon (better prediction), where as the wind
direction tends to be predominantly south to southwesterly in the morning.
The profiles of wind speed at Flinders View illustrate that the sea breeze on the
summer days tends to be overpredicted by the model, whereas the winter wind speed
seems to be consistently overpredicted.
Each profile provides a little more detail about every site. For Rocklea the wind
speed profiles are very similar for the summer and winter cluster type. The afternoon
wind speeds are underpredicted for each season by the model. This appears to be
independent of wind direction (for this example) as the main wind direction in the
afternoon for winter is southeasterly and northeasterly for summer, yet the same
profiles are seen. Whereas TAPM overpredicts the afternoon wind speed for winter
day types and underpredicts for the summer day types at Eagle Farm.
Figure 5.24: Wind speed difference profiles for Deception Bay (a) summer and (b) winter pollution conducive days (a) summer pollution conducive day (b) winter pollution conducive day
Figure 5.25: Temperature difference profiles for Flinders View (a) summer and (b) winter pollution conducive days
(a) summer pollution conducive day (b) winter pollution conducive day
67
5.3.2 Case day 30 January 1999
January 30, 1999 was selected for further investigation into model performance. It is
an example of a typical summer day that is conducive to high pollution episodes.
Based on monitoring information of ozone concentrations at each of the four
monitoring locations it was selected as a high pollution day for 1999. Hourly ozone
concentrations at all sites were above 6 pphm with the highest concentrations
recorded at Deception Bay (9.15 pphm) and Flinders View (6.75 pphm). There were
no records of bush fires or dust events that may have contributed to the higher
concentrations.
Time series figures of wind speed and wind direction for the 30th January provide
important information about the northeasterly change in the afternoon. For Deception
Bay and Eagle Farm the highest ozone concentrations were recorded at 11 am in the
morning of the 30th January.
Figure 5.26a illustrates the arrival of a moderate sea breeze by 10 am which remains
for most of the day as observed at Deception Bay. The sea breeze is predicted to occur
for a similar time period, with slightly stronger winds in the afternoon. This may lead
to the dispersion of the pollution to be greater than in reality for the afternoon.
Figure 5.26c and d illustrate the early prediction of the sea breeze by the model at
Rocklea (3 hours early) and Flinders View (2 hours early). Of particular interest is
that when the sea breeze arrives at Flinders View as predicted by the model the wind
speed increases dramatically from 1 ms-1 to 7 ms-1 which is not mirrored in the
observational data. This could have a significant impact on the subsequent prediction
of pollution concentrations at this site, particularly as the elevated levels of ozone
were observed to occur at 4 pm (same time as the model predicted 7 ms-1 winds).
Figure 5.26: Timeseries plots of wind direction and wind speed for (a) Deception Bay, (b) Eagle Farm, (c) Rocklea and (d) Flinders View, 30 January 1999, ozone pollution conducive day
(a) Deception Bay
(b) Eagle Farm
(c) Rocklea
(d) Flinders View
70
5.3.3 Case day 5 March 1999
The other example of a day identified as significant for ozone pollution was the 5th of
March (Cluster type 23). This day is characteristic of the typical sea breeze afternoons
that are experienced in summer in Southeast Queensland. The arrival and strength of
this sea breeze is extremely important in the dispersion of pollutants from the City to
the more southwestern districts. Days that fit into this category exhibit abrupt changes
in wind direction in the afternoon with an onshore wind blowing all the way to
Amberley. The 5th of March was selected for closer examination as the IOA suggested
that the model predicted the wind direction, wind speed and temperature satisfactorily.
It is interesting that for this day the model was able to predict the change in the wind
direction at all four sites, on the hour, with the easterly winds remaining for up to five
hours. Closer inspection of the pollution data for this day, suggested that perhaps
dispersion could be modelled well as the change in wind direction to the east and the
arrival of the sea breeze at each of the sites coincided with the maximum ozone
concentrations observed.
The maximum wind speed is over predicted by the model for most of the sites.
Figure 5.27: Timeseries plots of wind direction and wind speed for (a) Deception Bay, (b) Eagle Farm, (c) Rocklea and (d) Flinders View, 5 March 1999, ozone pollution conducive day
(a) Deception Bay
(b) Eagle Farm
(c) Rocklea
(d) Flinders View
73
5.3.4 Case day 21 June 1999
The 21st June was identified as a high NOx pollution day, typical of winter time
pollution. For winter time NOx pollution events, elevated concentrations of NOx are
observed generally between 7 am to 9 am when the wind speeds are light, and the air
more stagnant leading to an accumulation of pollution until the wind speed picks up
and disperses the pollution. This accumulation of pollution may on occasions be
emissions from the previous day.
The indices of agreement have shown that TAPM generally predicts these day types
less satisfactorily than summer day types. In particular the wind speed in the early
hours of the morning is overpredicted on this day by 50 to 200 % before 10 am at all
locations. The wind direction is predicted quite well in the early hours of the morning
at the coastal sites of Deception Bay and Eagle Farm while there is quite different
wind directions predicted at the more inland sites of Rocklea and Flinders View. It is
interesting to note that on these winter pollution type days the days are more stable.
The overprediction of the nighttime winds by the model may lead to very significant
differences to any pollution modelling conducted and the pollution observed.
Overprediction of these morning winds could lead to early dispersion of pollution and
the morning peak concentrations will not be predicted.
Figure 5.28: Timeseries plots of wind direction and wind speed for (a) Deception Bay, (b) Eagle Farm, (c) Rocklea and (d) Flinders View, 21 June 1999, NOx pollution conducive day
(a) Deception Bay
(b) Eagle Farm
(c) Rocklea
(d) Flinders View
77
6. Discussion
Due to the growing use of TAPM as a meteorological processor for areas where there
is no or poor observational data the model was run for one year using the defaults for
soil moisture and sea surface temperatures as well as without rain processes on.
The model was compared to observational data at five monitoring sites across
Southeast Queensland. Each site was unique due to its location A preliminary
analysis of the mean wind speed and temperature showed that the meteorology at the
urban sites were modelled well, while the difference in predicted and observed data at
the urban fringe sites was slightly greater, with the general trend being that the wind
speed is overpredicted by the model. The wind speed was overpredicted by 12 % to
143% depending on the location of the site, which was quite similar to the results
found by Hibberd et al (2003), where for various sites studied in Western Australia
the wind speed was overpredicted between 31% to 105%. Jackson et al (2003) also
found that overall wind speeds were overpredicted at a variety of sites across
Australia, irrespective of their relative location to the coast. This then suggests that
how the surface rougheness is treated in the model, may not adequately represent
reality. For Moreton Island, a complex site for any model to adequately predict, the
temperature was significantly overpredicted but this was due to the model treating the
site as a water location. The seasonal months of summer and spring tended to be
modelled to better accuracy than the autumn and winter months. Investigation of the
frequency distributions of the wind direction suggested that there may be a few
problems with the wind direction.
Closer examination of the observational data for the month of June, showed that for
1999, it was an extremely wet month. Wetter than the average in the last 25 years.
TAPM was rerun for June with the inclusion of rain. (Inspection of the temperature
timeseries suggested that the model was not adequately predicting the cloud cover,
hence not correctly determining the heat fluxes and subsequent temperature as the
predicted minimum temperatures were much warmer than observed across the sites.)
The model predictions (with the inclusion of rain processes) provided better
agreement between observational and modelled data at Rocklea for days on which it
78
actually rained. From this it can be recommended that if there is knowledge of high
rain that the inclusion of rain processes within the model may give some
improvement. Therefore for all model runs rain processes should be included.
However, the issue of inadequate cloud cover calculations (resulting in much warmer
predicted minimum temperatures) is still unresolved and further work is needed.
The model performance at the urban fringe site of Flinders View was not satisfactory
for the prediction of wind speed but this may be a consequence of the monitoring site
location rather than model performance. The high wind speeds predicted from the
model suggested that the monitoring site may not be adequately represented through
the selection of vegetation, similar to results discussed in Hibberd et al (2003). The
high wind speeds would suggest that there was not enough surface roughness to slow
the winds down. The vegetation was changed from grassland to urban (as the actual
monitoring location is in a residential area surrounded by forest) for a radius of 6 km
around the site. Re-running the model showed a significant improvement in the wind
speeds. The effect of soil type selection was also investigated. Changes to the soil
type affected the heat fluxes, but the full extent of the impact this has on the
modelling is yet to be evaluated. These studies confirmed that the user must carefully
prescribe the vegetation and soil type parameters within the model.
Data assimilation was used for a site 6 km from Flinders View. While this improved
the wind speed performance yet again, there was no improvement to the wind
direction or temperature profiles at Flinders View due to the distance away from the
site. Luhar et al (2004) found there was significant improvement in all surface
meteorological parameters (temperature, wind speed and wind speed components U
and V when data assimilation was included, however the sites investigated in that
study where much closer to the actual location of the observed data included for data
assimilation.
Using the index of agreement as a statistical measure allows for a quick evaluation of
model performance, however can be limiting in providing information where a model
may or may not perform well.
79
Cluster analysis was used to categorise days into smaller subgroups based on synoptic
characteristics. This allowed for more detailed interest as one could look at days of
interest for a particular application. It enabled the identification of the types of days
that the model predicted well and the ones the model didn't. Days that were typical of
summer conditions were modelled better than days typical of winter conditions.
Cluster types were then grouped further through determining pollution conducive
days, that is the cluster types that are typical for ozone pollution conducive days and
cluster types for winter pollution days.
Cluster 27, which was extremely similar to the day types previously investigated by
Coffey (1993), Physick (1993). These days in the morning had light north-
northwesterly winds which changed to very warm moderate north-northeastlery winds
in the afternoon at Eagle Farm. The model over predicted parameters at some sites
and underpredicted at others. Further investigation of the diurnal profiles determined
that for that particular day it is the onshore winds in the afternoon that are important
for polllution dispersion. The 30th January was investigated further Sea breeze was
predicted early in some cases and at twice the strength.
Another ozone conducive day type was evaluated, Cluster 23. These days typically
had light very light west-southwesterly winds in the morning before coming warm
moderate easterly winds in the afternoon at Eagle Farm. At Amberley Airport light
northwesterly morning winds were observed to change to hot easterly winds in the
afternoon. The daily wind direction and wind speed profiles for the 5th March were
evaluated. On this occasion the model predicted the sea breeze very well. Further
investigation is required to determine why this day the afternoon breeze was predicted
well, while there was a delay for the 30th January.
Winter pollution conducive days (elevated NOx concentrations) were investigated
(Cluster 14 and 15). These days were typically associated with cool, light to moderate
winds in the afternoon from the north to easterly sector combined with a high pressure
system. The results illustrated very different profiles for wind speed at each of the
sites compared with the summer profiles. In the early hours of the morning wind
speeds were generally overpredicted for most sites in winter.
80
Viewing the diurnal profiles of significant cluster types provided more information
about the performance of the model, than the mean profiles for wind speed and wind
direction. It showed that across sites, the differences in wind speed and temperature
can vary, and understanding why these variations occur is very important for correct
analysis of model outcomes.
Depending on the application for the modelling work being undertaken the additional
methods of evaluating the model performance such as the fractional bias will assist in
assessing the parameters that are significant for the particular application.
81
7. Conclusions
TAPM is a powerful prognostic meteorological modelling tool. Databases for terrain,
synoptic meteorology, land use and vegetation, and soil types are included with the
package on purchase eliminating the need for additional site-specific information to
be included in the model inputs. It appears very simple to use if the defaults and
databases are used. TAPM is being increasingly used to generate surface
meteorological statistics for dispersion modelling assessments, particularly (but not
exclusively) where there are not field data/observations.
The results showed that TAPM predicted the surface meteorology satisfactorily. The
use of synoptic clusters helps in data validation because it helps to understand what
wind speed and wind direction phenomenon is predicted or not predicted in the
model. This may provide useful feedback for ongoing model fine-tuning. It is
essential to include the appropriate land use, vegetation and soil type to maximise
model performance.
Overall TAPM adequately represented the air flows in the Southeast Queensland
airshed. The following were found:
• Annual average temperature and wind speed statistics were modelled well. At
most sites the annual average temperature was overpredicted by 1.3 oC and the
annual average wind speed by 0.5 m/s.
• The distribution of wind direction frequency remains a concern, however if the
regional wind flows (which were modelled satisfactorily) are the winds of
interest rather than the local winds then perhaps the prediction of local flows
around each station is not as important.
• TAPM satisfactorily predicts the sea breeze on synoptic day types
representative of sea breeze days. On some days in particular the arrival of the
sea breeze in the afternoon is predicted within an hour of what is observed at
the monitoring stations. On other days it may be a few hours late.
• The strength of the sea breeze is over predicted at the sites such as Deception
Bay and Flinders View. The wind speed was underpredicted at the urban sites.
82
• The wind direction on winter mornings (that may see recirculation) does not
appear to be modelled well.
• The wind speed on winter pollution conducive days was modelled
satisfactorily. The accuracy of the wind speed predictions tended to be reliant
on the direction of the wind flows. Windspeeds from the northeast and
northwest were predicted better than winds from the southeast.
• The differences in modelled and observed temperatures also seemed to be
dependent on wind direction.
The method of evaluation used in this work provided useful information about the
model performance and reliability on observational data.
• The standard statistical methods of mean, standard deviation and index of
agreement allowed for assessment of annual averages of wind speed and
temperature.
• Subsets of the annual data into seasonal and diurnal variation provided insight
into when TAPM predicted the observations well and when it didn't.
• Synoptic clustering and determination of significant clusters based on summer
ozone and winter pollution conducive days allowed easy evalution of the
performance of the model for days which were deemed important for
dispersion. This allowed another level of detail that the previous statistical
methods didn't. It provided a good first step, especially if diurnal behaviour of
residuals is investigated.
• Selection of observational data sets is important. In identifying the surface
boundary layer conditions of each of the monitoring stations it was easy to
ascertain which sites the model should not be expected to predict accurately.
For Flinders View and Deception Bay the locations of trees within a few metres
of the sites, caused concern that the monitoring instruments may be sheltered
and therefore deviations of wind speed, wind direction and temperature
paramters may result. The complexity of both of these sites due to surrounding
terrain and coastline could cause local effects that would not be expected to be
predicted by the model.
Varying the inputs into the model showed the following when the model is used for
the Southeast Queensland airshed:
83
• The selection of vegetation type/land use is very important. Changes in the heat
fluxes, tempertures and wind speed predictions occurred. Care should be taken
when determining whether an area is to be classified as urban as it appears that
the workings within the model may cause overestimation of surface
temperature to be greater than what is observed.
• The selection of soil type affected the surface heat fluxes significantly and
subsequently affected ground-level temperature and wind speed predictions
within the model.
• The effect of the selection of rain processes varied for each site, however, this
variation at each site was probably more to do with the way the site was
represented rather than the selection of the rain processes. The inclusion of rain
processes is not expected to hamper the performance of TAPM, therefore
should be selected as on.
• Data assimilation did improve wind speed at Flinders View when it was
included for a site nearby. Should reliable wind speed and wind direction data
exist within 10 km of the site, this can be included within the model to assist in
predicting the correct wind flows. At a distance further than this it is unlikely
that data assimilation will improve performance because it will beyond the
radius of influence.
• The various grid resolutions used for this modelling showed little variation
between annual average predictions of wind speed and wind direction.
However, this may not be the case for all model set ups. Verifying that terrain
information is adequately represented is important as the averaging of terrain of
1km2 or greater may cause a loss of features such as valleys and mountains.
Going to much finer resolution, may have showed quite different results.
• The selection of deep soil moisture content alone did not improve the overall
performance of TAPM significantly.
84
85
8. References
ASTM (2000) Standard guide for statistical evaluation of atmospheric dispersion
model performance, Designation: D6589-00. American Society for Testing and
Materials, West Conohocken, PA.
Azzi M., Hyde R., and Duc H. (2002) Comparison of predicted and observed
temperature and wind profiles on a high ozone event in Sydney. Proceedings of the
16th International Clean Air and Environment Conference. Christchurch, New
Zealand, 2002, Clean Air Society of Australia & New Zealand.
Coffey Partners International Pty. Ltd., (1993) Brisbane Wind-field Study: Final
report to Department of Environment and Heritage. Report E184/1 May 1993.
Department of Environment (Qld), 1997 Air Quality in the southeast Queensland air
shed – State of Knowledge Report.
Elbir T. (2003) Comparison of model predictions with the data of an urban air quality
monitoring network in Izmir, Turkey. Atmospheric Environment 37, 2149-2157.
Graham L. and Bridgman H. (2002) Applying The Air Pollution Model (TAPM), in
the Lake Macquarie airshed. Proceedings of the 16th International Clean Air and
Environment Conference. Christchurch, New Zealand, 2002, Clean Air Society of
Australia & New Zealand. pp 270-273
Hibberd M, Physick, B and Park G. (2003) Verification of several aspects of TAPM
against multi-year monitoring data at Collie. Proceedings of the 17th International
Clean Air Conference. Newcastle, New South Wales, 2003, Clean Air Society of
Australia.
86
Hurley P. (2000) Verification of TAPM meteorological predictions in the Melbourne
region for a winter and summer month. Aust. Met. Mag 49, 97-107.
Hurley P. (2001) The Air Pollution Model (TAPM) Version 2. Part 1: Technical
Description. CSIRO Atmospheric Research Technical Paper No.55
Ischtwan and Cope (1996) Modelling of photochemical smog in the South East
Physick B., Blockley A., Farrar D., Rayner K, and Mountford P. (2002a) Application
of three air quality models to the Pilbara region. Proceedings of the 16th International
Clean Air and Environment Conference. Christchurch, New Zealand, 2002, Clean Air
Society of Australia & New Zealand.629-634.
Physick B.L., Hurley P.J, Blockley A., Rayner K.N. and Mountford P. (2002b)
Verification of the air quality models TAPM and DISPMOD in coastal regions.
Presented at 4th International Conference on Environmental Problems in Coastal
Regions. 16-18 September 2002. Rhodes, Greece.
Rao S.T., Zurbenco I., Neagu R., Porter P., Ku J.Y., Henry R. (1999) Space and time
scales in ambient ozone data. Bulletin of American Meteo. Soc., 78, 2153-2166.
Schlünzen K.H., Baechlin W., Brünger H., Eichhorn J., Grawe D., Schenk R., and
Christof Winkler (2004) 9th Int. Conf. on Harmonisation within Atmospheric
Dispersion Modelling for Regulatory Purposes, 147-150
88
Simpson RW and Auliciems A, (1989) Air pollution in Brisbane. Institute of Applied
Environmental Research, Griffith University, Nathan, Q4111, 109 pp.
Skjøth CA., Brandt J. and Christensen JH (2005) Validation methods in
meteorological and air pollution modelling. Paul Scherrer Institute www.psi.ch
Stone RC (1989) Weather types at Brisbane, Queensland: an example of the use of
principal components and cluster analysis. International Journal of Climatology, 9:3-
32.
Tustison, Foufoula-Georgiou E, and Harris D. (2003) Scale-recursive estimation for
multisensor Quantitative Precipitation Forecast verification: A preliminary
assessment. J Geophys Res. 107, D8, 2-1 – 2-14
Wilmott C.J. (1981) On the Validation of Models. Phys. Geography 2 184-94.
89
Appendix A - Meteorological component of TAPM
The mean horizontal wind components u and v are derived from the momentum equations (Hurley, 2001).
)()(''
ssv uuNfvxxz
uwuFdtdu
−−+⎟⎠⎞
⎜⎝⎛
∂∂
∂∂
+∂∂
−∂∂
∂∂
+=σ
σππθσ
σ
)()(''
ssv vvNfuyyz
vwvFdtdv
−−+⎟⎟⎠
⎞⎜⎜⎝
⎛∂∂
∂∂
+∂∂
−∂∂
∂∂
+=σ
σππθσ
σ
⎟⎟⎠
⎞⎜⎜⎝
⎛∂∂
∂∂
+⎟⎠⎞
⎜⎝⎛∂∂
∂∂
+⎟⎟⎠
⎞⎜⎜⎝
⎛∂∂
+∂∂
−=∂∂
yv
xu
yv
xu σ
σσ
σσσ&
where:
x,y, σ are the components of the coordinate system (m) θv is the potential virtual temperature, π is 3.14159265, Ns is the large scale nudging coefficient (1/(24x3600)) us, vs and θs are the large scale synoptic winds and potential virtual temperature,
⎟⎟⎠
⎞⎜⎜⎝
⎛−−
=sT
sT zz
zzzσ , where z is the Cartesian vertical coordinate (m), zT is the
height of the model top (m) and zs is the terrain height (m). Scalar equations are solved for potential virtual temperature
)()(''
vsvsv
vv NS
zw
Fdt
dv
θθσσθ
θθ
θ −−+∂∂
∂∂
+=
1−
⎟⎠⎞
⎜⎝⎛∂∂
−=∂∂
zg
v
H σθσ
π
where g is the gravitational constant (8.91ms-2)
vv qpradiation
v Sct
TT
S λθθ −⎟
⎠⎞
⎜⎝⎛∂∂
=
where T is temperature (K) λ is the latent heat of vaporisation of water (2.5x106 Jkg-1) cp is the specific heat at constant pressure (1006 Jkg-1K-1)
90
91
Appendix B – Statistical Formulae
Statistics (based on Willmott, 1981 and Pielke, 1984) were calculated as follows:
Observed mean: ∑=
=N
iimean O
NO
1
1
Predicted mean: ∑=
=N
iimean P
NP
1
1
Observed Standard Deviation: ∑=
−−
=N
imeanistd OO
NO
1
2)(1
1
Predicted Standard Deviation: ∑=
−−
=N
imeanistd PP
NP
1
2)(1
1
Pearson Correlation Coefficient
⎥⎥⎦
⎤
⎢⎢⎣
⎡⎟⎠
⎞⎜⎝
⎛−⎟⎠
⎞⎜⎝
⎛
⎥⎥⎦
⎤
⎢⎢⎣
⎡⎟⎠
⎞⎜⎝
⎛−⎟⎠
⎞⎜⎝
⎛
⎟⎠
⎞⎜⎝
⎛⎟⎠
⎞⎜⎝
⎛−⎟⎠
⎞⎜⎝
⎛
=
∑∑∑∑
∑∑∑===
22
22
111
N
Ii
N
ii
N
Ii
N
ii
N
ii
N
ii
N
iii
PPNOON
POPONr
Root Mean Square Error: ( )∑=
−=N
iii OP
NRMSE
1
21
Unsystematic Root Mean Square Error: ( )∑=
−=N
iii PP
NURMSE
1
2ˆ1_
Systematic Root Mean Square Error: ( )∑=
−=N
iii OP
NSRMSE
1
2ˆ1_
Index of Agreement: ( )
( )∑
∑
=
=
−+−
−−= N
imeanimeani
N
iii
OOOP
OPIOA
1
2
1
2
1
92
P̂
Measures of Skill: stdP
URMSEESKILL __ =
std
std
OP
VSKILL =_
SKILL_R = stdP
SRMSEURMSE __ +
Where: N is the number of observations/predictions O is the Observed data P is the Predicted data
is the linear regression fitted formula with intercept a and slope b
93
Appendix C - Cluster definitions for cluster types
Cluster 1 EAG: Cold light southwesterly winds in the morning changing to warm light northerly winds by the afternoon followed by very light west-northwesterly winds overnight. AMB: Cold light west-northwesterly winds in the morning changing to warm light northwesterly winds in the afternoon.
Cluster 2 EAG: Warm light to moderate east-southeasterly winds in the afternoon changing to moderate northeasterly winds in the afternoon. AMB: Warm light easterly winds in the morning changing to moderate east-northeasterly winds in the afternoon High pressure system
Cluster 3 EAG: Warm light southwesterly winds in the morning changing to light northeasterly winds in the afternoon. Winds remaining from the northeasterly direction in the evening. AMB: Warm light westerly winds in the morning becoming slightly warmer in the afternoon.
Cluster 4 EAG: Very light cool winds in the morning, with north-northwesterly winds in the afternoon. Still very light conditions. AMB: Cool very light west-northwesterly winds in the morning, remaining light throughout the day
Cluster 5 EAG: Cool light to moderate westerly winds in the morning with winds strengthening in the afternoon. AMB: Cool moderate west-northwesterly winds strengthening to warm strong westerly winds.
Cluster 6 EAG: Warm light south-southeasterly winds. East-southeasterly winds in the afternoon. AMB: light south-southeasterly winds in the morning changing slightly to easterly in the afternoon. Warm day. Some rain
Cluster 7 EAG: Cold moderate west-southwesterly winds in the morning warming slightly during the day before becoming cooler light overnight winds. AMB: Cool light west-northwesterly winds in the morning changing to moderate westerly winds in the afternoon. High-pressure system, no rain.
Cluster 8 EAG: Warm moderate southerly winds becoming warmer throughout the day AMB: Warm moderate southerly winds becoming warmer throughout the day Low pressure system
Cluster 9 EAG: Cold moderate easterly winds, with a cool change in the afternoon.
94
AMB: Cool light southeasterly winds remaining light changing to easterly in the afternoon. Low pressure system, Very wet
Cluster 10 EAG: Warm moderate easterly winds in the morning, remaining the same throughout the day. AMB: Moderate easterly winds warming during the day but easing slightly Low-pressure system, Wet.
Cluster 11 EAG: Warm moderate to strong southerly winds becoming stronger southeasterly winds in the afternoon. AMB: Moderate warm south-southwesterly winds in the morning strengthening during the day.
Cluster 12 EAG: Warm light to moderate southerly winds changing to easterly in the afternoon. AMB: Very light southerly winds changing to light to moderate easterlies in the afternoon.
Cluster 13 EAG: Warm light to moderate south-southwesterly winds remaining the same throughout the day. AMB: Light to moderate southerly winds remaining the same throughout the day.
Cluster 14 EAG: Cool moderate southwesterly winds changing to warm moderate east-northeasterly winds in the afternoon before cooling and becoming still overnight. AMB: Cool very light west-southwesterly winds changing to warm light northeasterly winds.
Cluster 15 EAG: Cold moderate southwesterly winds changing to easterly winds in the afternoon becoming cold light southerly winds in the evening. AMB: Cool light west-southwesterly winds changing to warm very light easterly winds in the afternoon.
Cluster 16 EAG: Warm strong westerly winds strengthening throughout the day. Temperatures reaching the high 20s. AMB: Warm moderate westerly winds in the morning strengthening to very strong westerlies Low pressure system
Cluster 17 EAG: Warm very strong southeasterlies becoming stronger during the day. Temperatures reaching the high 20s. AMB: Warm moderate east-southeasterly winds in the morning. Winds in the afternoon are very strong easterly winds. High pressure system
Cluster 18 EAG: Warm moderate northerly winds becoming warmer in the afternoon with strong north-northeasterly winds before cooling slightly in the evening as winds are moderate northerly.
95
AMB: Warm light north-northwesterly winds in the morning changing to hot moderate easterly winds in the afternoon.
Cluster 19 EAG: Warm moderate southerly winds in the morning changing slightly to southeasterly in the afternoon. AMB: light to moderate south-southeasterly winds in the morning changing to east-southeasterly in the afternoon. High pressure system
Cluster 20 EAG: Warm light to moderate northerly winds, changing to very strong north-northeasterlies throughout the day. AMB: Temperatures in the low 20s. Light northerly winds becoming hotter in the day as the winds become northeasterly.
Cluster 21 EAG: Warm moderate south-southeasterlies in the morning changing to easterlies in the afternoon. AMB: Temperatures in the mid 20s. Light southeasterly winds becoming moderate easterly winds in the afternoon.
Cluster 22 EAG: Very light warm easterly winds in the morning becoming warmer during the day with moderate east-northeasterly winds in the afternoon. Still warm in the evening with winds light and from the northerly direction. Temperatures reaching the high 20s. AMB: Warm light northeasterly winds becoming hot moderate east-northeasterly winds in the afternoon.
Cluster 23 EAG: Warm very light west-southwesterly winds in the morning becoming warmer moderate easterly winds in the afternoon before easing in the evening with north-northeasterly winds. Temperatures reaching the high 20s. AMB: Warm light northwesterly winds becoming hot easterly winds in the afternoon.
Cluster 24 EAG: Temperatures in the mid 20s, moderate easterly winds in the morning becoming warmer during the day. AMB: Warm light easterly winds with temperatures reaching high 20s as the moderate east-northeasterly winds arrive in the afternoon.
Cluster 25 EAG: Very strong easterlies throughout the entire day. AMB: warm moderate easterly winds in the morning becoming stronger during the day. Rain
Cluster 26 EAG: Cold strong west-southwesterly winds AMB: Temperatures in the low teens, with strong westerly winds. These winds strengthen throughout the day.
Cluster 27 EAG: Warm light north-northwesterly wind changing to very warm moderate north-northeasterly winds in the afternoon. Lighter winds in the evening. Temperatures reaching the high 20s. AMB: Very warm light northwesterly winds, changing to hot moderate west-
96
northwesterly winds. Cluster 28 EAG: Temperatures in the low twenties. Still morning with light to moderate
easterly wins in the afternoon. AMB: Very light northerly wind in the morning changing to easterly in the afternoons. Moderate rain
Cluster 29 EAG: Cold moderate south-southwesterly winds in the morning, warmer moderate southerly winds in the afternoon. AMB: Cold moderate southerly winds warming slightly during the day. Light Rain
Cluster 30 EAG: Temperatures in the mid teens before reaching twenties in the afternoon. Light to moderate south-southwesterly winds changing to easterly in the afternoon. AMB: very light southerly winds, becoming stronger slightly and easterly in the afternoon.
97
Appendix D Summer and winter/autumn pollution conducive days
Difference between predicted and observed data for autumn day type (Cluster 14); wind speed and wind direction for Deception Bay (a) wind speed
(b) wind direction
98
Difference between predicted and observed data for autumn day type (Cluster 14); temperature, wind speed and wind direction for Eagle Farm (a) temperature
(b) wind speed
(c) wind direction
99
Difference between predicted and observed data for autumn day type (Cluster 14); temperature, wind speed and wind direction for Rocklea (a) temperature
(b) wind speed
(c) wind direction
100
Difference between predicted and observed data for autumn day type (Cluster 14); temperature, wind speed and wind direction for Flinders View (a) temperature
(b) wind speed
(c) wind direction
101
Difference between predicted and observed data for autumn day type (Cluster 14); temperature, wind speed and wind direction for Moreton Island (a) temperature
(b) wind speed
(c) wind direction
102
Difference between predicted and observed data winter day type (Cluster 15); wind speed and wind direction for Deception Bay (a) wind speed
(b) wind direction
103
Difference between predicted and observed data for winter day type (Cluster 15); temperature, wind speed and wind direction for Eagle Farm (a) temperature
(b) wind speed
(c) wind direction
104
Difference between predicted and observed data for winter day type (Cluster 15); temperature, wind speed and wind direction for Rocklea (a) temperature
(b) wind speed
(c) wind direction
105
Difference between predicted and observed data for winter day type (Cluster 15); temperature, wind speed and wind direction for Flinders View (a) temperature
(b) wind speed
(c) wind direction
106
Difference between predicted and observed data for winter day type (Cluster 15); temperature, wind speed and wind direction for Moreton Island (a) temperature
(b) wind speed
(c) wind direction
107
Difference between predicted and observed data for summer day type (Cluster 23); wind speed and wind direction for Deception Bay (a) wind speed
(b) wind direction
108
Difference between predicted and observed data summer day type (Cluster 23); temperature, wind speed and wind direction for Eagle Farm (a) temperature
(b) wind speed
(c) wind direction
109
Difference between predicted and observed data summer day type (Cluster 23); temperature, wind speed and wind direction for Rocklea (a) temperature
(b) wind speed
(c) wind direction
110
Difference between predicted and observed data summer day type (Cluster 23); temperature, wind speed and wind direction for Flinders View (a) temperature
(b) wind speed
(c) wind direction
111
Difference between predicted and observed summer day type (Cluster 23); temperature, wind speed and wind direction for Moreton Island (a) temperature
(b) wind speed
(c) wind direction
112
Difference between predicted and observed data for summer day type (Cluster 27); wind speed and wind direction for Deception Bay (a) wind speed
(b) wind direction
113
Difference between predicted and observed data for summer day type (Cluster 27); temperature, wind speed and wind direction for Eagle Farm (a) temperature
(b) wind speed
(c) wind direction
114
Difference between predicted and observed data for summer day type (Cluster 27); temperature, wind speed and wind direction for Rocklea (a) temperature
(b) wind speed
(c) wind direction
115
Difference between predicted and observed for summer day type (Cluster 27); temperature, wind speed and wind direction for Flinders View (a) temperature
(b) wind speed
(c) wind direction
116
Difference between predicted and observed data for summer day type (Cluster 27); temperature, wind speed and wind direction for Moreton Island (a) temperature