ACCESS Sensitivity Experiments for fog cases at Perth Airport...experiments. Perth Airport depicted...

Bureau Research Report - 033

ACCESS Sensitivity Experiments for fog cases at

Perth Airport Belinda Roux December 2017

ACCESS SENSITIVITY EXPERIMENTS FOR FOG CASES AT PERTH AIRPORT

i

ACCESS Sensitivity Experiments for fog cases at Perth Airport

Belinda Roux

Bureau Research Report No. 033

December 2017

National Library of Australia Cataloguing-in-Publication entry

Author: Belinda Roux

Title: ACCESS Sensitivity Experiments for fog cases at Perth Airport

ISBN: 978-1-925738-04-9

Series: Bureau Research Report – BRR033


Enquiries should be addressed to: Contact Name: Belinda Roux Bureau of Meteorology GPO Box 1289, Melbourne Victoria 3001, Australia Contact Email: [email protected]

Copyright and Disclaimer

© 2016 Bureau of Meteorology. To the extent permitted by law, all rights are reserved and no part of

this publication covered by copyright may be reproduced or copied in any form or by any means

except with the written permission of the Bureau of Meteorology.

The Bureau of Meteorology advise that the information contained in this publication comprises

general statements based on scientific research. The reader is advised and needs to be aware that such

information may be incomplete or unable to be used in any specific situation. No reliance or actions

must therefore be made on that information without seeking prior expert professional, scientific and

technical advice. To the extent permitted by law and the Bureau of Meteorology (including each of its

employees and consultants) excludes all liability to any person for any consequences, including but

not limited to all losses, damages, costs, expenses and any other compensation, arising directly or

indirectly from using this publication (in part or in whole) and any information or material contained

in it.


iii

Contents

Executive summary ...................................................................................................... 1

1. Introduction .......................................................................................................... 2

2. Region description and observations ............................................................... 3

3. NWP model setup ................................................................................................ 5

4. Sensitivity experiments ....................................................................................... 6

5. Case study: 21 August 2016 ............................................................................. 10

6. Case Study: 3 October 2015 ............................................................................. 18

7. Case study: 4 July 2015 .................................................................................... 26

8. Case study: 9 June 2013 ................................................................................... 34

9. Conclusions ....................................................................................................... 38

10. Acknowledgements ........................................................................................... 39

11. References ......................................................................................................... 39


iv

List of Tables

Table 1: Description of experiments performed to test ACCESS-C2 model sensitivity ................ 7

Table 2: Average statistics (best=blue and worst=orange) of time series for different experiments at Perth Airport on 21 August 2016. ................................................................ 17

Table 3: Average statistics of the time series for different experiments at Perth Airport for 0100 UTC, 3 October to 0300 UTC, 4 October 2015 .................................................................... 21

Table 4: Average statistics of time series for different experiments at Perth Airport on 9 June 2013...................................................................................................................................... 35

List of Figures

Figure 1: Google image showing Perth airport and the surrounding region 3

Figure 2: Map of locations of automatic weather stations in the greater Perth region and topography (m), generated from the 3 second (~90m) Shuttle Radar Topography Mission (SRTM) Smoothed Digital Elevation Model (DEM-S) data (http://www.ga.gov.au/metadata-gateway/metadata/record/gcat_72760) 4

Figure 3: Land surface fractions used by the experimental ACCESS-C2 model over the Perth area. Perth Airport depicted by the black box. 6

Figure 4: Changed surface-type fractions used in the sensitivity experiments. Perth Airport depicted by the black box and the AWS used for verification is located at the centre of the 10 x 10 km changed area (near the top of the black box). 8

Figure 5: Standard (top) and changed (bottom) canopy heights used for the sensitivity experiments. Perth Airport depicted by the black box and the AWS used for verification is located at the centre of the changed area (near the top of the box). 9

Figure 6: MSLP analysis for 0600 UTC (left) on 21 Aug and 0000 UTC (right) on 22 Aug 2016. 10

Figure 7: Time series of observations from Perth airport AWS (YPPH) for 21 August 2016. (Sunset is at 0955 UTC and sunrise at 2255 UTC.) 11

Figure 8: ACCESS-C2 time series at Perth Airport for 21 August 2016. The model base date is 0000 UTC. 11

Figure 9: ACCESS-C2 dewpoint depression (shades) and 10 m winds over the Perth region for 2300 UTC on 21 August 2016. AWS observations are plotted in green. 12

Figure 10: Observed temperature, dewpoint temperature and winds at Perth airport for 1100 UTC (top) and 2300 UTC (bottom) on 21 August 2016. 13

Figure 11: Time series of the ACCESS-C2 vertical profile of relative humidity (shades), horizontal winds (barbs) and vertical winds in pa/s (red contours negative, blue contours positive) at Perth airport for 21 August 2016. RH above 80 % is shaded in light green, and above 90% in dark green. 13

Figure 12: Time series of the ACCESS-C1 (top) and ACCESS-C2 (bottom) temperature (red), dewpoint temperature (blue), visibility (orange) and winds (light green) against observations (dashed lines and dark green wind barbs) at Perth airport for 21 August 2016. 14


Figure 13: Himawari 8 night microphysics (left) and LIFR probabilities (right) for 21 August 2016 at 2200 UTC 15

Figure 14: ACCESS-C1 (left) and C2 (right) visibility (shades, km) and fog fraction (contours) for 22 UTC on 21 August 2016. Observations of visibility (in km) are plotted in dark green. 15

Figure 15: ACCESS-C1 (left) and C2 (right) dewpoint depression (shades) and10 m winds (barbs) for 22 UTC on 21 August 2016. Observations are plotted in dark green. 15

Figure 16: Bias (model-observations) time series of different experiments at Perth airport for temperature (solid lines) and dewpoint temperature (dashed lines) for 21 August 2016. 16

Figure 17: Visibility (km) when calc_prob_of_vis=0.6 (left) and calc_prob_of_vis=0.3 (right) for 22 UTC on 21 August 2016. Visibility observations are plotted in dark green and the contours are for fog fraction. 17

Figure 18: MSLP analysis for 0600 UTC on 3 October 2015 (left) and 0000 UTC on 4 October 2015 (right) 18

Figure 19: Observed temperature, dewpoint temperature and winds at Perth airport for 2300 UTC on 3 October 2015 18

Figure 20: Time series of observations from Perth airport AWS (YPPH) for 3 October 2015 19

Figure 21: MTSAT satellite image of fog and low cloud over SWWA on 3 October at 1824 UTC. Fog and low cloud is denoted by the blue shades. 20

Figure 22: Bias (model-observations) time series of different experiments at Perth airport for temperature (solid lines) and dewpoint temperature (dashed lines) for 10 October 2015. 20

Figure 23: Time series of the modelled temperature (red lines), dewpoint temperature (blue lines), visibility (orange lines) and winds (light green barbs) against observations (dashed lines and dark green wind barbs) at Perth airport for the operational ACCESS-C1 (top); standard ACCESS-C2 experiment (middle); and experiment "R" (bottom). 22

Figure 24: DPD and winds (left) and visibility and fog fraction (right) for ACCESS-C1 nested in R1 (top), ACCESS-C2 nested in R1 (middle) and experiment 'R' (C2 nested in R2)(bottom) for 2000 UTC on 3 October 2015. Black contours for fog fraction at 0.01, 0.3 and 0.7 and observations are plotted in green. 23

Figure 25: Time series of the ACCESS-C2 experiment "R" vertical profile of relative humidity (shades), horizontal winds (barbs) and vertical winds in pa/s (red contours negative, blue contours positive) at Perth airport for 3 October 2015. 24

Figure 26: Experiment R east-west (left) and north-south (right) cross sections of winds (barbs), temperature (black contours) and vertical motion (red is up and green is down) for 2000 UTC on 3 October 2015. Perth Airport AWS is denoted by the red dot. 25

Figure 27: Experiment R winds (barbs) and temperatures (black contours) at 33 m (left) and 360 m (right) for 2000 UTC on 3 October 2015. Shades are for topography and Perth airport AWS is denoted by the black cross. 25

Figure 28: MSLP analysis for 0600 UTC on 4 July 2015 (left) and 0000 UTC on 5 July 2015 (right) 26

Figure 29: Satellite image of fog and low cloud over SWWA on 4 July at 1753 UTC and 2253 UTC. Fog and low cloud is denoted by the blue shades. 26

Figure 30: Observed temperature, dewpoint temperature and winds at Perth airport for 1100 UTC (top) and 2300 UTC (bottom) on 4 July 2015 27

Figure 31: Time series of observations from Perth airport AWS (YPPH) for 4 July 2015. (Sunset was at 0910 UTC, sunrise was at 2310 UTC. 28


vi

Figure 32: Bias (model-observations) time series of different experiments at Perth airport for temperature (solid lines) and dewpoint temperature (dashed lines) for 4 July 2015 28

Figure 33: Time series of the modelled temperature (red lines), dewpoint temperature (blue lines), visibility (orange lines) and winds (light green barbs) against observations (dashed lines and dark green wind barbs) at Perth airport for the operational ACCESS-C1 (top); standard ACCESS-C2 experiment (middle); and experiment "R" (bottom). 30

Figure 34: ACCESS-C1 (left) and ACCESS-C2 (right) 10 m winds and DPD for 2100 UTC on 4 July 2015. Observations are plotted in green. 31

Figure 35: ACCESS-C1 (left) and ACCESS-C2 (right) visibility (shades in km) and fog fraction (contoured at 0.01, 0.3 and 0.7) for 2100 UTC on 4 July 2015. Visibility observations are plotted in green. 31

Figure 36: Time series of the ACCESS-C2 vertical profile of relative humidity (shades), horizontal winds (barbs) and vertical winds in pa/s (red contours negative, blue contours positive) at Perth airport for 4 July 2015. 32

Figure 37: ACCESS-C2 east-west (left) and north-south (right) cross sections of winds (barbs), temperature (black contours) and vertical motion (red is up and green is down) for 1700 UTC on 4 July 2015. Perth Airport AWS is denoted by the red dot. 33

Figure 38: ACCESS-C2 10 m winds and DPD for 1700 UTC on 4 July 2015. Observations are plotted in green. 33

Figure 39: Bias (model-observations) time series of different experiments at Perth airport for temperature (solid lines) and dewpoint temperature (dashed lines) for 9 June 2013 34

Figure 40: Urban fractions from different resolution land use ancillaries and land-sea mask combinations. Perth Airport is denoted by the black rectangle. The top left picture was used in the standard case studies (top right for Experiment E) and the bottom right will be used in future model runs (with the addition of inland water bodies when available). 35

Figure 41: Bias (model-observations) time series of different masking experiments at Perth airport for temperature (solid lines) and dewpoint temperature (dashed lines) for 9 June 2013 36

Figure 42: Time series of the modelled temperature (red lines), dewpoint temperature (blue lines), visibility (orange lines) and winds (light green barbs) against observations (dashed lines and dark green wind barbs) at Perth airport for ACCESS-C2 (top), Experiment "F" (middle) and Experiment "E" (bottom). 37


Page 1

EXECUTIVE SUMMARY

Numerical modelling case studies using the ACCESS-City Perth model have been conducted to better understand the dynamics leading to fog at Perth Airport, to compare the operational APS2 ACCESS-C model (ACCESS-C2) to the previous operational APS1 ACCESS-C model (ACCESS-C1) and to assess the model sensitivity to surface characteristics, initial and boundary conditions and visibility parameterisations. Case studies of four fog events at Perth Airport were performed using the ACCESS-C2 model. Consistent with earlier studies, it was found that the model represented the mesoscale dynamics well and the topography in the region plays an important role in the formation of fog. In addition to the initial hypothesis that fog can form near the boundary of the moist environmental westerlies and katabatic easterlies off the escarpment (Golding, 1993), the possible role of blocked flow as a mechanism for fog formation was revealed. Environmental flow from the Indian Ocean to the west can be blocked by the escarpment, resulting in northerly winds on the coastal plain which can either advect fog to the south or fog can form through the mixing of the two air masses. In the case of moist southerly environmental flow, cold air drainage down the Helena valley can result in a stagnation point and fog can be formed (or enhanced) when the cool drainage flow mixes with the southerlies. Additional experiments were conducted for each fog event to test the model sensitivities to different surface and input data. Results of the sensitivity studies varied between cases. It was found that results are sensitive to the choice of land surface characteristics in the model as well as the initial and boundary conditions. Comparing to the operational model, it was found that ACCESS-C2 generally gives better results than ACCESS-C1, with lower dewpoint depressions and more accurate visibilities at the airport during the fog events. In three out of the four cases the rate of temperature increase after sunrise is slower than observed. At the airport, the model tends to simulate its lowest visibilities a few hours earlier than observations, and the visibility has not been simulated to be as low as observed for any of the cases.


1. INTRODUCTION

Fog is defined as the suspension of water droplets near the ground, reducing the visibility to below 1000 m (Gultepe et al., 2007). The aviation industry is one of the industries most influenced by the detrimental effects of fog, both for economic and safety reasons. Accurate forecasts of fog are particularly important at remote locations such as Perth Airport, where the closest alternative for large aircraft is more than two hours flying time away. The airport has an average of 12 fog events per year, the majority of which occur in the cool season (April to October).

Perth Airport is located on the coastal plain in south-west Western Australia (SWWA). The weather at Perth airport is often influenced by the interactions between weather systems coming from the Indian Ocean to the west and the Darling Escarpment which runs parallel to the coast and is located to the east of the airport. Perth Airport experiences around 12 fog events per year, mainly in the cool season (1 April to 31 October) during the night and early morning.

The Bureau of Meteorology uses the Australian Community Climate and Earth-System Simulator (ACCESS) for its operational numerical weather prediction (NWP) modelling. The system is based on the Unified Model (UM) from the UK Met Office, coupled with the Joint UK Land Environment Simulator (JULES) land surface model, and a four-dimensional variational data assimilation scheme (Bureau of Meteorology, 2010; Puri et al., 2013). The model suite, also known as the Australian Parallel Suite (APS) consists of global, regional and city scale models. Numerical weather prediction models are continually being improved and the Bureau of Meteorology updates its operational models every few years to stay up to date with the latest science and improvements. ACCESS has recently been upgraded from APS1 to APS2, improving model physics as well as resolution. More detail on the model setup is given in Section 3.

In case studies of fog over Perth Airport, Potts & Roux (2016) found that an experimental version of the APS2 ACCESS city model generally gave a good representation of the mesoscale dynamics but that the model did not produce low enough values of visibility. While the model was able to produce a good representation of the time series of wind, temperature and dew point, values of the dew point depression (DPD) were not low enough during periods when fog was observed, which directly affected the corresponding forecasts of visibility. A comparison of observed and model-predicted values of hourly visibility and DPD at eight Australian airports over the period 2011-2015 (Dare, 2017) also found that low DPDs and visibility during fog events at Perth and other locations were not well predicted by the operational APS0 and APS1 models.

The aim of the current work is to better understand the dynamics leading to fog at Perth, and to investigate possible reasons for errors in forecasts of visibility and DPD. To achieve this, four case studies are used as a basis to perform a range of sensitivity experiments. Three of the case studies are for recent 2015/16 fog events and the fourth is a reference case study of the 2013 fog event described by Potts and Roux (2016), with extension to the different sensitivity experiments. The sensitivity experiments help to determine the influence of land surface characteristics (surface types, vegetation heights and soil moisture), the choice of initial/boundary conditions, and the tuning of the visibility scheme. Along with these tests, the important influence of the mesoscale dynamics of fog in the vicinity of Perth Airport is considered in each case study. Results from a 1.5 km resolution experimental APS2 ACCESS city model (ACCESS-C2) are compared to the recent 4 km operational APS1 ACCESS city model (ACCESS-C1).


Page 3

In Section 2 the terrain surrounding Perth airport is described together with the observations used in the case studies. In Section 3. the numerical weather prediction (NWP) model setup at the Bureau of Meteorology is laid out and Section 4. provides details of the sensitivity experiments. Results from the four case studies, including the associated sensitivity experiments, are discussed in Sections 5 to 8. Conclusions are presented in Section 9.

2. REGION DESCRIPTION AND OBSERVATIONS

A satellite image from Google Earth shows the airport and its surroundings (Figure 1). Perth airport (demarcated with a yellow box) is located to the east of the city centre (marked as "Perth") on the coastal plain about 20 km east of the Indian Ocean. The Darling Scarp runs parallel to the coast and is about 10 km to the east of the airport. The Swan River passes north of the airport and runs southwest out to sea. The Canning River joins the Swan River around 10 km south-west of the airport with an extended area of open water where the rivers join. This area of open water, known as Melville Water, has the potential to impact fog formation, especially since synoptic scale winds are often south-westerly in the cool season.

Figure 1: Google image showing Perth airport and the surrounding region

The land use to the west of the airport is mainly residential (houses with trees and some parks), with bigger areas of farmland to the north and some light industry directly to the south. There


are mainly woodlands to the east of the escarpment. The airport itself consists of tarmac runways with grasslands in between and a few relatively low-rise buildings. There are also some natural patches with scrubs and grass directly to the north-east and south-east of the airport.

A map of the topography and locations of the Bureau's automatic weather stations (AWS) is shown in Figure 2, with the Perth Airport AWS identified as "YPPH". This map clearly illuminates the sharp rise in elevation of the Darling scarp to the east and the location of the valleys. The valley directly to the east of the airport is called the Helena Valley (circled in blue), and the influence of variations in the Scarp like the Helena Valley on the local dynamics at the airport and its contribution to fog formation/dissipation is investigated.

Figure 2: Map of locations of automatic weather stations in the greater Perth region and topography (m), generated from the 3 second (~90m) Shuttle Radar Topography Mission (SRTM) Smoothed Digital Elevation Model (DEM-S) data (http://www.ga.gov.au/metadata-gateway/metadata/record/gcat_72760)

The Perth airport AWS is located to the north-east of the runways and is used in the time series analysis in the case studies. All the stations with available data are used in the spatial verifications. Most stations record at least temperature, dewpoint temperature and winds with the exception of the Melville Water AWS (DOPY) which is located in the Swan River and Ocean Reef (OCEA), both of which record winds only. Apart from YPPH, the AWS at Pearce (YPEA) and Jandakot (YPJT) airports are the only stations in the vicinity which report visibility.


Page 5

Other information used in the case studies include vertical profiles of temperature, dewpoint temperature and winds from radiosonde balloon flights at Perth airport taken at 1100 UTC (1900 WST) and 2300 UTC (0700 WST); the Bureau's national 6 hourly mean sea level pressure (MSLP) analysis over Australia, and available satellite products.

From 2005 to early 2016 multispectral satellite data were received from the Multi-functional Transport Satellites (MTSAT) which is the geostationary weather satellite series operated by the Japan Meteorological Agency (JMA). Satellite fog and low cloud products were developed by Weymouth (2006) using the brightness temperature difference of the infrared and near infrared channels from MTSAT (Barras & Miao, 2011; Boneh et al., 2015). In late 2015 images from Himawari-8, JMA's new geostationary satellite became available (Bessho et al., 2016). Himawari 8 has a higher spatial and temporal resolution as well as more channels than MTSAT. The night microphysics product combines multiple satellite channels to monitor the evolution of night-time fog and stratus. More detail on this product is available from Zeschke (2015). The Geostationary Cloud Algorithm Testbed (GEOCAT) fog and low cloud products (FLS) (Calvert and Pavolonis, 2010) were also useful. The GEOCAT Low Instrument Flight Rules (LIFR) products are used in this study. These present the probability the cloud ceiling is below 500 ft and/or the visibility is less than 1 mile (<1.6 km). Details on the accuracy of the fog and low stratus satellite products are given in a report by Lucas (2017). Whenever available, the different satellite products will be used concurrently to indicate the spatial extent of the fog.

3. NWP MODEL SETUP

The ACCESS NWP modelling suite at the Bureau of Meteorology consists of a global model (ACCESS-G) with a horizontal resolution of about 25 km, a regional model covering Australia and the surrounding oceans (ACCESS-R) with a ~12 km resolution and higher resolution city scale models (ACCESS-C) covering the major populated regions. The global and regional models were updated to APS2 in 2016 and the city scale models were updated in 2017, increasing the horizontal resolution from 4 km to 1.5 km. The models run four times daily, at 0000 UTC, 0600 UTC, 1200 UTC and 1800 UTC but only the 0000 UTC runs were considered in this work.

The UM is coupled to the Joint UK Land Environment Simulator (JULES) land surface model (Best et al., 2011; D. B. Clark et al., 2011). JULES uses nine tiles to represent the fractional surface types covering each grid box, by defining five vegetation surface types and four non-vegetation types. The total surface flux for a grid box is computed by taking an area weighted average of the fluxes for each surface type present in that grid box. Figure 3 shows the current land surface fractions in the model over the Perth area, with the airport denoted by the box roughly in the middle of the map. With this dataset the airport is classed as having a 30-40 % urban fraction with 10-20% each for broadleaf trees, temperate grass and tropical grass fractions and 0-10% for scrubs and bare soil. The surface fractions used to date are relatively smooth and the land-sea mask is coarse, resulting in the Melville Water area being classed as ocean points. Both the surface type fractions and the land-sea mask are currently being updated. The effect of the basin in the model can be tested once the updated mask is available.

The visibility calculation in the UM is a diagnostic process and uses screen temperature, specific humidity, pressure, liquid water mixing ratio and the dry aerosol mass mixing ratio (P. A. Clark et al., 2008; Claxton, 2013). In the UK the Met Office has constructed a "MURK" fieldwhich serves as a proxy field for the total aerosols that can be used to specify the dry aerosolmass mixing ratio in the visibility scheme. MURK has been tailored for the UK and isconstrained by the assimilation of visibility observations (Claxton, 2013). The advantage of theMURK field is that it can have surface sources, be mixed by turbulence and advected by


dynamics, and it can be removed by precipitation. However, in Australia visibility is not assimilated and the use of a single climatological approximation for the dry aerosol mass mixing ratio was proposed instead of MURK. This followed an investigation in 2014 into the impact of the MURK filed in the APS1 model suite.

Figure 3: Land surface fractions used by the experimental ACCESS-C2 model over the Perth area. Perth Airport depicted by the black box.

In the APS1 suite MURK was only used for the visibility calculations but in the APS2 suite MURK also sets the cloud droplet number for the cloud microphysics and has been turned back on. It would be possible to change the code so that MURK is not used for the visibility calculations even if it is turned on in the rest of the model. This option could change the diagnosed visibility without influencing the prognostic model variables and was tested in 2017, with results to be presented in a separate report.

4. SENSITIVITY EXPERIMENTS

All case studies have been run with an experimental version of the APS2 ACCESS-C model (hereafter ACCESS-C2), which will be regarded as the standard run. The city models have not been running with land surface ancillary files consistent with model resolution, and updates to the operational ACCESS-C2 are being made accordingly. The original high resolution surface type fraction dataset was somewhat offset which resulted in the land use at the airport almost entirely being classed as urban fraction. In a single case study done with these erroneous surface type fractions the modelled screen temperature was much higher at night than other model runs for the same date with more reasonable fractions. These results are discussed with the case study in Section 8.

a b

d e f

c


Page 7

A set of experiments, all based on the same model code, have been run for the case studies and are compared to the standard run. A description of the experiments is given in Table 1.

Table 1: Description of experiments performed to test ACCESS-C2 model sensitivity

Description Experiment Settings/Changes

Standard run S UM8.2, forecast-only, Nested in APS1 ACCESS-R (APS2 for 21 Aug 2016), 00Z run, 1.5 km horizontal resolution, 70 vertical levels.

Changed surface type fractions F Fractions of surface types have been changed over the airport to increase grass fractions and decrease the urban fraction.

Changed vegetation heights G Height of vegetation over the airport has been halved so that trees are around 4m, grass around 0.25-0.5m and shrubs around 0.75m.

Updated soil moisture with consistent soil ancillaries

P New soil moisture content and consistent soil ancillaries were used

Standard run but nested in APS2 ACCESS-R

R Standard run but initial and boundary conditions taken from APS2 ACCESS-R instead of APS1

Using new land-sea mask T Standard run with ancillary files using new land sea mask (not final version yet)

Visibility parameter V Standard run but with adjustments to visibility parameter '"calc_prob_of_vis"

The standard run (experiment "S") is using the UM version 8.2 and all cases, except for the latest case on 21 Aug 2016, have been nested in APS1 ACCESS-R. The horizontal resolution of the ACCESS-C2 model is 1.5 km with 70 levels in the vertical up to a height of 40 km. The model is a forecast only model (no data assimilation at the city scale) and the 0000 UTC run has been used. Assimilating observations in the city scale model could potentially lead to improvements in the fog forecasts, but this would only be available with the next model upgrade (APS3).

In the next experiment ("F") the surface type fractions in a 10x10 km box around the airport have been changed so that the grass tiles have a higher fraction. The properties associated with the urban tile are more geared toward European cities with high density tall buildings which is clearly not the case for the airport. Figure 4 shows the changes to the surface type fractions around the airport.


Figure 4: Changed surface-type fractions used in the sensitivity experiments. Perth Airport depicted by the black box. The AWS used for verification is located at the centre of the 10 x 10 km changed area (near the top of the black box).

Dharssi et al. (2015) found that the tree heights can have a significant impact on forecasts of temperature and wind speed. Based on this, the heights of the vegetation in the box around the airport have been halved in experiment "G" to test the impact of vegetation heights at Perth Airport on the forecast. Figure 5 gives the vegetation heights for the standard (top) and changed (bottom) runs. An experiment was also performed with both the changed surface type fractions and vegetation heights together, but results did not provide much added information and are not presented for the sake of simplicity.

In the current configuration for ACCESS-C the soil moisture content is downscaled from the global model. Separate work is currently underway to develop a high resolution soil moisture field for Australia to support fire-weather forecasting requirements. Updated ancillaries of soil moisture content and soil type from that project have been used in experiment "P" to investigate the influence of better soil moisture content on the formation of fog.

For the two case studies in 2015 both the APS1 and APS2 ACCESS-R model output were available. In experiment "R" the standard run was nested in the APS2 ACCESS-R and compared to the standard run nested in APS1 ACCESS-R to investigate the influence on initial and boundary conditions on the model run.

In experiment "T", one case study has been performed using a new mask (with high resolution land fractions) which is under development, but the Swan river basin has not yet been captured in this dataset so proper experimentation should be done once that is fixed.


Page 9

Figure 5: Standard (top) and changed (bottom) canopy heights used for the sensitivity experiments. Perth Airport depicted by the black box and the AWS used for verification is located at the centre of the changed area (near the top of the box).

Lastly in experiment "V" the visibility parameterisation has been tuned. Changes to the visibility parameterisation do not influence the state variables such as temperature, dewpoint temperature and winds so the only reason for this experiment was to see how much a single tuning parameter can improve the model's forecast of visibility.


5. CASE STUDY: 21 AUGUST 2016

In the first case study fog developed at around 1930 UTC, 21 Aug 2016 (0330 WST, 22 Aug 2016) and cleared around 0030 UTC, 22 Aug (0830 WST, 22 Aug). There was some rain observed early in the morning of the 21st, and again late afternoon and early evening with the cloud clearing about four hours before the onset of fog. The 10 m winds backed from west north-west in the afternoon to south-west during the night with calm conditions or light south easterlies after fog formation. Both temperature and dewpoint temperature dropped rapidly at the time of the fog formation over the airport.

This case was chosen because forecasters reported that the operational models did not capture the event and the Himawari 8 satellite observations could provide useful insights into the spatial extent of the fog. Fog formed in the early morning hours at Perth airport on 22 August 2016 and lasted about five hours. Figure 6 shows the Bureau's operational mean sea level pressure (MSLP) analysis over Western Australia for the afternoon of 21 August and the morning of 22 August 2016. A cold front passed across south-west Western Australia (SWWA), bringing some rain in the early evening. A high pressure system was situated to the west of the country with predominantly southerly winds over the area. The Perth Regional Forecasting Centre (RFC) has identified a number of synoptic situations favourable for the occurrence of fog and this event could be classified as a post-front diffluent southerly (PF-DS).

Figure 6: MSLP analysis for 0600 UTC (left) on 21 Aug and 0000 UTC (right) on 22 Aug 2016.

A time series of the observations from the Perth airport automatic weather station (AWS) is shown in Figure 7. The solid red and blue lines are for the screen level temperature and dewpoint temperature, respectively, and the 10 m wind is plotted as wind barbs in green. The brown solid line shows horizontal visibility (in km, on the left axis), while the orange dashes show the 30 min ceilometer mean (in meters above ground level on the right axis). The orange numbers show the 30 min ceilometer mean total cloud coverage in octas and the grey dashed line is the mean sea level pressure (MSLP) in hPa minus 1000 (on the left axis). 10-minute rainfall accumulations are shown with black bars in mm (on the left axis) and the hourly accumulations are noted in black below the wind barbs. The dashed brown vertical lines show the event start and end times (the first and last times when the visibility was below 1 km). Note that with the 10 minute visibility observations, 10 km is the maximum threshold reported and equates to clear air.


Page 11

The sudden observed drop in temperature and dewpoint coincides with the winds easing and turning south-easterly. This may indicate cold air drainage from the Helena valley mixed with the moist air from the onshore flow and prior precipitation, contributing to the initiation fog at Perth airport. The fog cleared with the air warming and drying and the winds backing to south-westerlies in the morning around 0900 WST (0100 UTC).

Figure 7: Time series of observations from Perth airport AWS (YPPH) for 21 August 2016. (Sunset is at 0955 UTC and sunrise at 2255 UTC.)

Figure 8: ACCESS-C2 time series at Perth Airport for 21 August 2016. The model base date is 0000 UTC.

Figure 8 shows the time series of the variables simulated by ACCESS-C2 at the airport. Overall the model gives a good representation of the meteorological variables observed at the airport. ACCESS-C2 produced more rain than observed, but the cloud amounts and heights as well as


the timing of the rainfall were fairly accurate. The simulated visibility was decreased although not to the extent of the observations and the light south-easterly wind observed after fog formation was not simulated. Although the screen level temperature and dewpoint temperatures were simulated well, the sudden drop and moistening that accompanied the observed fog at the airport were not simulated. There is evidence of a cold air drainage flow from the Helena Valley in the model (Figure 9), but it doesn't penetrate to the airport.

Figure 9: ACCESS-C2 dewpoint depression (shades) and 10 m winds over the Perth region for 2300 UTC on 21 August 2016. AWS observations are plotted in green.

The observed vertical profiles of temperature, dewpoint temperature and winds for 1100 UTC and 2300 UTC are shown in Figure 10. A moist layer with some cloud up to the 600 hPa pressure level and fairly strong winds were observed over the airport at 1100 UTC. By 2300 UTC there has been substantial warming above 650 hPa, and drying particularly above 750 hPa. A strong nocturnal inversion is also evident.

Figure 11 shows a time series of the vertical profile of the ACCESS-C2 relative humidity, winds and vertical motion for 21 August at Perth airport. The model simulates westerly winds in the morning (at least up to the 700hPa pressure level), which backed to south-westerly around 11 UTC and southerly shortly thereafter. The relative humidity is greater than 80% between 900hPa and 700hPa in the morning, which extends to the lower levels between 0830 and 1930 UTC. The winds were fairly strong throughout the lower part of the atmosphere, and the model seems to be in good agreement with the observed profiles at 11 and 23 UTC.


Page 13

Figure 10: Observed temperature, dewpoint temperature and winds at Perth airport for 1100 UTC (top) and 2300 UTC (bottom) on 21 August 2016.

Figure 11: Time series of the ACCESS-C2 vertical profile of relative humidity (shades), horizontal winds (barbs) and vertical winds in pa/s (red contours negative, blue contours positive) at Perth airport for 21 August 2016. RH above 80 % is shaded in light green and above 90% in dark green.

To compare ACCESS-C2 with ACCESS-C1, a time series of the observed and modelled screen level temperature and dewpoint, 10 m wind and horizontal visibility is given in Figure 12. The biggest difference between C1 (top) and C2 (bottom) is that C2 gave a fair bit more rain in the


late afternoon (not shown). ACCESS-C2 had a total of 8.8mm between 0700 UTC and 1000 UTC as opposed to 0.5 mm at around 1000 UTC from ACCESS-C1. There was about 2 mm of rain observed between 0830 and 1230 UTC. ACCESS-C2 had a more realistic temperature and dewpoint temperature during the night at the airport (C1 overestimated the dewpoint temperature in the afternoon and underestimated both temperature and dewpoint during the night). The sudden moistening and cooling observed with the fog onset and the south-easterly wind change were not simulated by either model.

Figure 12: Time series of the ACCESS-C1 (top) and ACCESS-C2 (bottom) temperature (red), dewpoint temperature (blue), visibility (orange) and winds (light green) against observations (dashed lines and dark green wind barbs) at Perth airport for 21 August 2016.

The availability of 10 minute data from the Himawari-8 geostationary satellite provides a useful tool to get an idea of the spatial extent of low cloud and fog. Figure 13 shows the Himawari-8 night microphysics (left) and the Low Instrument Flight Rules (LIFR) fog and low cloud probabilities at 2200 UTC on 21 August 2016. The simulated visibility and fog fractions from the ACCESS-C1 and ACCESS-C2 models for the same hour are given in Figure 14 and the corresponding 10 m winds and screen level dewpoint depression are given in Figure 15.


Page 15

Figure 13: Himawari 8 night microphysics (left) and LIFR probabilities (right) for 21 August 2016 at 2200 UTC

Figure 14: ACCESS-C1 (left) and C2 (right) visibility (shades, km) and fog fraction (contours) for 22 UTC on 21 August 2016. Observations of visibility (in km) are plotted in dark green.

Figure 15: ACCESS-C1 (left) and C2 (right) dewpoint depression (shades) and10 m winds (barbs) for 22 UTC on 21 August 2016. Observations are plotted in dark green.


ACCESS-C2 captured the low visibilities and fog to the west of the escarpment fairly well. ACCESS-C1 gives an indication of slightly lower visibilities to the west of the escarpment, but not to the same extent and the simulated visibilities was far above the values required for fog. Both models show similar wind patterns, but with its higher spatial resolution ACCESS-C2 shows much more variability in the dewpoint depression and lower dewpoint depressions, which corresponds better to the observations.

As mentioned before, the case studies were also used to test various model sensitivities. Figure 16 shows a time series of the difference (model – observations) of screen level temperature (solid lines) and dewpoint temperature (dashed lines) for four of the experiments performed on 21 August. All the experiments overestimate the dewpoint temperature in the early afternoon and underestimate the screen temperature in the late afternoon but the night time (1000 UTC – 2300 UTC) temperatures and dewpoint temperatures were simulated fairly well by the model. The evolution of the temperatures and dewpoint temperatures were similar for the different experiments, indicating that the temperature sensitivity for this case was moderate at most.

Figure 16: Bias (model-observations) time series of different experiments at Perth airport for temperature (solid lines) and dewpoint temperature (dashed lines) for 21 August 2016.

Average statistics for the experiments are given in

Table 2. In this case changing the land fractions produce slightly worse results overall while the standard experiment is generally better, but there were not significant and consistent differences between the different runs. Altering the land fractions and vegetation heights across a larger area would have produced greater differences among experiments, especially considering the dynamic influences on fog at Perth airport. Such tests should be more easily achieved in the future with later versions of the model code from the UK.


Page 17

Table 2: Average statistics (best=blue and worst=orange) of time series for different experiments at Perth Airport on 21 August 2016.

21 Aug 2016 S (standard)

F (surf fractions)

G (veg heights)

P (soil moisture)

Correlation coefficient

Temp 0.965 0.959 0.971 0.966

Dewpt 0.874 0.862 0.859 0.844

Mean difference

Temp ‐0.321 ‐0.616 ‐0.325 ‐0.368

Dewpt ‐0.203 ‐0.096 ‐0.417 ‐0.053

RMSE Temp 1.119 1.298 1.056 1.176

Dewpt 1.047 1.106 1.138 1.112

In considering the tuning of parameters in the visibility scheme, Figure 17 shows the difference in visibility if the calc_prob_of_vis is set to the standard 0.6 (left) or 0.3 (right). This change mainly affects the visibilities in the lower ranges (below 10 km). The fog fraction doesn't change but with a parameter value of 0.3 the simulated visibilities of 1 km have a similar pattern to the 0.3 fog fraction contour. Similarly, if calc_prob_of_vis is set to 0.1, visibilities of 1 km or lower follows the 0.1 fog fraction contour.

Figure 17: Visibility (km) when calc_prob_of_vis=0.6 (left) and calc_prob_of_vis=0.3 (right) for 22 UTC on 21 August 2016. Visibility observations are plotted in dark green and the contours are for fog fraction.

This experiment indicates that the calc_prob_of_vis parameter can be useful for better calibrating the visibility without affecting the rest of the model. This will be tested for a possible upgrade to operations in the near future.


6. CASE STUDY: 3 OCTOBER 2015

The second case study is for 3 October 2015. Fog formed around 1945 UTC 3 Oct (0345 WST, 4 Oct) and cleared just over 3.5 hours later around 2320 UTC, 3 Oct (0720 WST, 4 Oct). Figure 18 gives the MSLP analysis over Australia for 0600 UTC on 3 October and 0000 UTC on 4 October. Using the synoptic situations favourable for the occurrence of fog identified by the Perth RFC, the synoptic situation could be classified as a post front ridging between fronts (PF-RI-BF), where fronts are typically 10 to 30 hours apart with a moderate ridge north of Perth.

Figure 18: MSLP analysis for 0600 UTC on 3 October 2015 (left) and 0000 UTC on 4 October 2015 (right)

Figure 19 shows the observed vertical profiles of temperature, dewpoint temperature and wind at Perth airport for 2300 UTC when fog was present. There was a moist layer below 890 hPa, capped by subsidence inversion and nocturnal inversion below 1000 hPa, believed to cap the fog layer. There are light northerly winds at the surface and stronger north-westerlies aloft.

Figure 19: Observed temperature, dewpoint temperature and winds at Perth airport for 2300 UTC on 3 October 2015 (0700 WST, 4 Oct)


Page 19

A time series of the observations from the Perth AWS is given in Figure 20. The dewpoint temperature stayed above 12 degrees throughout the period. Winds were westerly to north-westerly throughout the afternoon and early evening, turning to light northerlies during the night before backing to westerlies midmorning, around two hours after the fog cleared. The screen temperatures dropped during the night and the air moistened, becoming saturated around 2000UTC before warming and drying around 2330 UTC, 3 Oct (0730 WST, 4 Oct), shortly after the fog cleared. There was some low cloud observed during the night before the fog event. It did not show a gradual lowering to fog as one might expect, but the fog did lift to low cloud in the morning.

Figure 20: Time series of observations from Perth airport AWS (YPPH) for 3 October 2015. Sunrise is at 2150 UTC.

A satellite image of the fog and low cloud product (derived from MTSAT) for 1824 UTC 3 October (0224 WST, 4 Oct) is given in Figure 21. Unfortunately this is the latest MTSAT image available for the event, but there is an area of fog and/or low cloud visible to the north of the airport which has likely been advected to the south with the low level northerly or further developed to the south as the conditions became more favourable for fog.

Figure 22 shows a time series of the difference between observed and modelled temperature and dewpoint temperature from the different experiments. Most experiments show similar biases with a large warm bias for screen temperature during the night and early morning and an even larger cold bias for dewpoint temperature in the early morning hours.

The "R" Experiment (using APS2 ACCESS-R input as initial and boundary conditions) is the only experiment which shows a significant difference with a much reduced temperature bias, especially during the night and early morning. The dewpoint temperature bias is higher than the other experiments in general, but the early morning negative bias (between 0700 and 0900 LST) is also reduced in this experiment.


Figure 21: MTSAT satellite image of fog and low cloud over SWWA on 3 October at 1824 UTC. Fog and low cloud is denoted by the blue shades.

Figure 22: Bias (model-observations) time series of different experiments at Perth airport for temperature (solid lines) and dewpoint temperature (dashed lines) for 10 October 2015.


Page 21

A comparison of the average statistics over the time series for the different experiments is given in Table 3. This shows that Experiment R is clearly better when it comes to screen temperature, and even though the mean difference in dewpoint temperature is the largest due to its mostly positive bias, the root mean square error and correlation coefficient of the dewpoint depression is comparable to the other experiments.

Table 3: Average statistics of the time series for different experiments at Perth Airport for 0100 UTC, 3 October to 0300 UTC, 4 October 2015

3 Oct 2015 S (standard)

F (surf fractions)

G (veg heights)

P (soil moisture)

R

(C2 in R2)

Correlation

coefficient

Temp 0.948 0.943 0.948 0.947 0.980

Dewpt 0.632 0.629 0.654 0.605 0.621

Mean

difference

Temp 1.243 1.001 1.179 0.818 0.329

Dewpt ‐0.276 ‐0.227 ‐0.260 ‐0.134 0.788

RMSE Temp 1.886 1.679 1.822 1.768 0.859

Dewpt 1.251 1.240 1.205 1.097 1.197

Time series of the observed and modelled temperature, dewpoint temperature, 10 m winds and visibility are given in Figure 23 for the operational run (ACCESS-C1 nested in ACCESS-R1) as well as the ACCESS-C2 standard ("S") and "R" experiments. Experiment "R" shows much improved DPD during the night with visibilities below 4 km, although the period of low visibilities is about 4 hours too early. This experiment is also the only experiment to simulate the observed northerlies during the night and early morning. The dewpoint temperatures were a bit warmer than observed but overall this model run produced a much more realistic simulation.

At closer inspection of the spatial fields from the model simulations, it is clear that experiment "R" also captured the mesoscale circulations of the area much better than the other experiments. Figure 24 shows the spatial DPD, winds, visibility and fog fraction of the three model runs for 2000 UTC. In the top panel, the operational run (ACCESS-C1) has some indications of northerlies to the west of the escarpment, but it does not extend towards the coast as the observations suggest and the visibility over Perth airport is simulated to be more than 20 km. The standard experiment (ACCESS-C2 nested in APS1 ACCESS-R) did not simulate the northerly winds on the coastal plain and also were much drier over the central part of the domain in contrast to the observations (middle panel). In Experiment R, where ACCESS-C2 is nested in APS2 ACCESS-R (bottom panel), the winds and dewpoint depressions coincide well with the observations. The model simulates reduced visibilities, even though the simulated visibility is still generally higher than observed. The broad area of low visibilities and fog to the north of the airport has also been simulated by the model.

Based on the more realistic simulation of Experiment R when compared to observations, this run was used to look at the mesoscale dynamics of the fog event. It should be noted that APS2 ACCESS-R became operational in July 2016 so Experiment R should reflect the operational performance of the model in future events.


Figure 23: Time series of the modelled temperature (red lines), dewpoint temperature (blue lines), visibility (orange lines) and winds (light green barbs) against observations (dashed lines and dark green wind barbs) at Perth airport for the operational ACCESS-C1 (top); standard ACCESS-C2 experiment (middle); and experiment "R" (bottom).


Page 23

Figure 24: DPD and winds (left) and visibility and fog fraction (right) for ACCESS-C1 nested in R1 (top), ACCESS-C2 nested in R1 (middle) and experiment 'R' (C2 nested in R2)(bottom) for 2000 UTC on 3 October 2015. Black contours for fog fraction at 0.01, 0.3 and 0.7 and observations are plotted in green.


The difference in results between the standard run and experiment R highlights the importance of initial and boundary conditions in some cases. Getting the dynamics right it often critical to fog formation. Ensemble modelling would be a good way to account for this sensitivity through simulating multiple scenarios. There are plans to include a city scale ensemble in the next model upgrade at the Bureau (APS3). It is recommended that the application for fog forecasting be investigated once an experimental version of the ensemble model becomes available to better position aviation forecasters to utilise this functionality once it becomes operational.

Figure 25 gives a time series of the vertical profile of the relative humidity and winds from Experiment "R" at Perth airport. Similar to the observations at 2300 UTC (Figure 19), the model simulated a moist layer up to the 900 hPa pressure level with the light northerlies near the surface and stronger north-westerlies aloft.

Figure 25: Time series of the ACCESS-C2 experiment "R" vertical profile of relative humidity (shades), horizontal winds (barbs) and vertical winds in pa/s (red contours negative, blue contours positive) at Perth airport for 3 October 2015.

Cross sections through the latitude and longitude of Perth airport reveal that the northerlies of the coastal plain could be a result of the environmental flow being blocked by the escarpment. This phenomenon is also observed in the Adelaide region and is described in more detail by Tepper & Watson (1990). Both the east-west and north-south cross sections from Experiment "R" over Perth airport for 2000 UTC are given in Figure 26, with the y axis showing geopotential height in meters.


Page 25

Figure 26: Experiment R east-west (left) and north-south (right) cross sections of winds (barbs), temperature (black contours) and vertical motion (red is up and green is down) for 2000 UTC on 3 October 2015. Perth Airport AWS is denoted by the red dot.

Looking at the east-west cross section (left), it is clear that the northerly winds are confined to the coastal plain below the height of the escarpment while the winds above are north-westerly. The north-south cross section (right) indicates that a similar situation is observed in both directions along the coastal plain. This blocking flow of the coastal plain can also be seen when looking at maps of the model winds below and above the height of the escarpment (Figure 27).

Figure 27: Experiment R winds (barbs) and temperatures (black contours) at 33 m (left) and 360 m (right) for 2000 UTC on 3 October 2015. Shades are for topography and Perth airport AWS is denoted by the black cross.

The relatively warm and moist onshore flow mixing with the cooler air from the blocked flow likely contributed to the fog formation over Perth Airport on 3 October 2015 and the model captured it very well.


7. CASE STUDY: 4 JULY 2015

Fog formed around 1830 UTC, 4 Jul 2015 (0230 WST, 5 Jul) and cleared about 0045 UTC (0845 WST) on 5 July. There was rain observed prior to the fog in the afternoon and evening and the visibility gradually decreased from 1500 UTC. This was the second day in a four day period of foggy nights, and this day in particular had the lowest visibility for the longest period in the four days. The synoptic situation can be described by looking at the afternoon and morning MSLP map analysis over Australia (Figure 28). A cold front passed over SWWA in the afternoon with fog forming in the early morning hours. Similar to the previous case study, the synoptic type was classed as the long wavelength variant of the post front ridging between fronts by the Perth RFC.

Figure 28: MSLP analysis for 0600 UTC on 4 July 2015 (left) and 0000 UTC on 5 July 2015 (right)

The MTSAT satellite images of fog and low cloud over SWWA suggest that fog was primarily observed on the plateau to the east of Perth airport. Figure 29 shows the satellite image at 1753 UTC (shortly before fog onset at the airport) and 2253 UTC.

Figure 29: Satellite image of fog and low cloud over SWWA on 4 July at 1753 UTC and 2253 UTC. Fog and low cloud is denoted by the blue shades.


Page 27

The observed vertical profiles of temperature, dewpoint temperature and winds at the airport at 1100 UTC and 2300 UTC are given in Figure 30. At 1100 UTC there was a subsidence inversion at 800hPa with moist air below and a surface nocturnal inversion. The winds were light northerlies at the surface with westerly winds aloft. By the next morning (2300 UTC) the airport was in fog with a marked nocturnal inversion that capped the fog layer. The lower level winds (up to about 700hPa) are more variable than the previous case study, with light winds just above the moist layer at 2300 UTC.

Figure 30: Observed temperature, dewpoint temperature and winds at Perth airport for 1100 UTC (top) and 2300 UTC (bottom) on 4 July 2015

From the time series of observations from the Perth Airport AWS (Figure 31) one can see that the 10 m winds were westerly in the afternoon shifting to relatively light northerly winds during the night, with clouds and frequent periods of light rain. The operational model (ACCESS-C1) did not forecast any rain at the airport but ACCESS-C2 did forecast some rain in the afternoon and early evening.


Figure 31: Time series of observations from Perth airport AWS (YPPH) for 4 July 2015. (Sunset was at 0910 UTC, sunrise was at 2310 UTC.

Figure 32 shows the time series of the model minus observed temperature and dewpoint temperature at Perth Airport for the different sensitivity experiments. Experiment R shows the largest dry bias around sunset (with too warm temperatures and too cold dewpoint temperatures) but all the experiments show similar biases during the night and morning hours.

Figure 32: Bias (model-observations) time series of different experiments at Perth airport for temperature (solid lines) and dewpoint temperature (dashed lines) for 4 July 2015


Page 29

The time series of the temperature, dewpoint temperature, winds and visibility for the operational model, standard experiment and experiment R are given in Figure 33. The modelled 10 m winds are fairly consistent between the different model runs and are in general agreement with the observations. The two ACCESS-C2 experiments have lower visibilities at the airport than the operational run but the models again reduce the visibility earlier than the observations. The temperature and dewpoint temperatures were forecast to be lower than observed for a few hours before the event, especially the operational model, but all the models cleared the low visibilities too early and simulated temperatures and dewpoint temperatures around sunrise that were too high.

Even though the models show less variation in the wind compared to the observations, the general evolution of the winds throughout the period were captured. There were northerly winds in the morning of 4 July, which shifted westerly in the afternoon and evening before becoming more consistently north to north-easterly in the early morning hours during the fog event. It should be noted that light winds tend to be more variable and the wind direction is often difficult to forecast.

Figure 34 shows the surface maps of the 10 m wind and dewpoint depression at 2100 UTC for ACCESS-C1 and ACCESS-C2. ACCESS-C2 has more spatial variability in both fields and a lower DPD overall. The model simulated a drier area on the coast around the Swanbourne AWS (31.96°S; 115.76°E) which extended towards Perth airport in the afternoon and early evening. The wind strength was simulated well and in general the direction was reasonable, although there are times and locations where the wind direction were in the opposite direction to the observations. This can partly be attributed to the variable nature of the observed light winds and partly to the timing of wind direction changes.

The visibility and fog fraction for ACCESS-C1 and ACCESS-C2 for 2100 UTC over the bigger Perth domain are given in Figure 35. ACCESS-C2 has lower visibilities across the coastal plain, although the area of fog fraction above 0.3 in ACCESS-C1 is bigger. ACCESS-C1 has a more extensive area of low (<1 km) visibilities on the plateau to the east of the airport and ACCESS-C2 has more spatial variability. Looking at the observations at Pearce and Perth airport, the visibilities of both models are too high in general, at least on the coastal plain. To the east of the escarpment the models have low visibilities (and a higher fog fraction). Unfortunately observations over these areas are limited, but compared to the satellite images from MTSAT (Figure 29) these look reasonable. With the recent addition of Himawari-8 satellite data better spatial validation should be possible for future fog events.


Figure 33: Time series of the modelled temperature (red lines), dewpoint temperature (blue lines), visibility (orange lines) and winds (light green barbs) against observations (dashed lines and dark green wind barbs) at Perth airport for the operational ACCESS-C1 (top); standard ACCESS-C2 experiment (middle); and experiment "R" (bottom) for 4 July 2016.


Page 31

Figure 34: ACCESS-C1 (left) and ACCESS-C2 (right) 10 m winds and DPD for 2100 UTC on 4 July 2015. Observations are plotted in green.

Figure 35: ACCESS-C1 (left) and ACCESS-C2 (right) visibility (shades in km) and fog fraction (contoured at 0.01, 0.3 and 0.7) for 2100 UTC on 4 July 2015. Visibility observations are plotted in green.

The vertical time series of the ACCESS-C2 winds at Perth Airport (Figure 36) shows fairly good agreement with the observations at 1100 UTC and 2300 UTC (Figure 30). The remnants of the fog from the previous day can be seen in the morning of 4 July with the relative humidity close to the ground above 90% for the first three hours of simulation and also between 0600 and 0800 UTC where the model simulated some rainfall. Westerly winds were simulated above the 950hPa pressure level during the day and evening, turning more south-westerly in the early morning hours. Winds in the lowest levels were light and more variable.


Figure 36: Time series of the ACCESS-C2 vertical profile of relative humidity (shades), horizontal winds (barbs) and vertical winds in pa/s (red contours negative, blue contours positive) at Perth airport for 4 July 2015.

Figure 37 gives the east-west and north-south vertical cross sections through Perth airport at 1700 UTC, when the model visibility was at its lowest. The model again simulated blocked flow, with north-easterlies below the height of the escarpment and south-westerlies aloft. At the base of the scarp the winds turn more easterly, which indicates some drainage flow down the Helena valley. The drainage flow down the valleys is also visible in a spatial map of the 10 m winds and DPD at 1700 UTC (Figure 38).


Page 33

Figure 37: ACCESS-C2 east-west (left) and north-south (right) cross sections of winds (barbs), temperature (black contours) and vertical motion (red is up and green is down) for 1700 UTC on 4 July 2015. Perth Airport AWS is denoted by the red dot.

Figure 38: ACCESS-C2 10 m winds and DPD for 1700 UTC on 4 July 2015. Observations are plotted in green.

Overall, ACCESS-C2 produces a reasonably accurate representation of the mesoscale flows. Experiment R did not have the same dramatic improvements as in the previous case, highlighting how case dependent results can be. The observed low level flows were generally light and a bit more variable in this current case.


8. CASE STUDY: 9 JUNE 2013

In the fourth case study, fog formed around 1300 UTC on 9 June 2013 and lasted until 2300 UTC. This 10 hour fog event was an unusual case and although it has already been described in detail by Potts and Roux (2016), it is used here as a reference date for the sensitivity studies. As such the focus of this section will be on the sensitivity studies performed. Figure 39 gives the time series of the temperature and dewpoint temperature difference (modelled-observed) for the standard experiments for 9 June 2013. For the screen temperature all the experiments had a cold bias in the morning and a warm bias at night. Experiment F (where the surface type fractions around the airport are changed) had the lowest temperature bias during the night. In the early morning hours experiments G and P both had a few hours where the warm bias was greater than the other experiments.

Figure 39: Bias (model-observations) time series of different experiments at Perth airport for temperature (solid lines) and dewpoint temperature (dashed lines) for 9 June 2013

A breakdown of the average statistics from the bias time series is given in Table 4. The best result is coloured blue and the worst is orange for each of the statistics. From this it is clear that the changed land surface experiment (F) provided the best results for this case study. The only statistic where the experiment did not have the best score was the mean temperature difference, but the higher warm bias at night in the other experiments compensated for the cold daytime bias so the results can be misleading if evaluated on their own.

The original high resolution land surface ancillaries had a slight spatial offset which resulted in the land use at the airport being classified almost entirely as urban fraction. This has since been rectified and this reference case study was used to test the influence of the different land surface ancillaries and land-sea masks available. Figure 40 shows the urban fraction in different combinations of land-sea mask and land fraction ancillaries. Perth Airport is demarcated by the black rectangle. When using the low resolution land surface ancillary files, the airport has a surface fraction of about 0.3. With the high resolution land fractions the airport has an urban fraction of 0.9 for the old (slightly offset) ancillaries or less than 0.1 (for the new ancillaries). The high resolution land fractions with the new land-sea mask are used in the operational


Page 35

ACCESS-C2 model, with an update planned to include the inland water bodies (such as the Swan river basin) following as soon as it has been tested and cleared for operations.

Table 4: Average statistics of the time series for different experiments at Perth Airport for the period of 0100 UTC 9 June to 0300 UTC 10 June 2013.

9 Jun 2013 Ex s f g p

Correlation coefficient

Temp 0.855 0.880 0.825 0.823

Dewpt 0.960 0.971 0.965 0.965

Mean difference

Temp ‐0.182 ‐0.508 ‐0.057 0.020

Dewpt ‐0.002 0.001 0.047 0.349

RMSE Temp 1.691 1.628 1.834 1.879

Dewpt 0.678 0.607 0.631 0.742

Figure 40: Urban fractions from different resolution land use ancillaries and land-sea mask combinations. Perth Airport is denoted by the black rectangle. The top left picture was used in the standard case studies (top right for Experiment E) and the bottom right will be used in future model runs (with the addition of inland water bodies when available).

Old mask, high res. frac. Old mask, low res. frac.

New mask, high res. frac. New mask, low res. frac.


Figure 41 shows the time series bias of the temperature and dewpoint temperatures for the four experiments using different combinations of the land-sea mask and land surface ancillaries. The one experiment that clearly stood out was when the old land-sea mask was used together with the high resolution surface ancillaries, resulting in the airport having an urban fraction of 0.9. This is called Experiment E and the night time temperatures were up to two degrees higher than the other experiments during the night, probably as a result of the urban heat island effect.

Figure 41: Bias (model-observations) time series of different masking experiments at Perth airport for temperature (solid lines) and dewpoint temperature (dashed lines) for 9 June 2013

The time series of the temperature, dewpoint temperature, wind and visibility for the standard ACCESS-C2 together with experiments F and E are given in Figure 42. As mentioned before all the experiments have a warm bias during the night for this case but the bias in Experiment E is much more extensive and the dewpoint temperature for this experiment also has a warm bias in the early morning hours. The visibility was similar for the different experiments and this was the one case study where the model visibility did not decrease before the observations. This could be explained by the fact that fog formed unusually early in the night for this case. There were only small differences between the runs for the 10 m winds at the airport.

Old mask, low res fracs Old mask, high res fracs New mask, low res fracs New mask, high res fracs


Page 37

Figure 42: Time series of the modelled temperature (red lines), dewpoint temperature (blue lines), visibility (orange lines) and winds (light green barbs) against observations (dashed lines and dark green wind barbs) at Perth airport for ACCESS-C2 (top), Experiment "F" (middle) and Experiment "E" (bottom).


9. CONCLUSIONS

The purpose of the case studies was to better understand the mesoscale dynamics leading to fog at Perth Airport, to compare results from the ACCESS-C2 model to the operational ACCESS-C1 model and to assess the model sensitivity to surface characteristics, initial and boundary conditions and visibility parameterisation.

Four case studies of fog events at Perth Airport were performed using the ACCESS-C2 model, with additional experiments for each case study to test the model sensitivities to different surface and input data. Results of the sensitivity studies showed that the model is sensitive to the choice of land surface characteristics in the model as well as the initial and boundary conditions. Compared to ACCESS-C1, it was found that ACCESS-C2 generally gives better results, with lower DPD and visibilities at the airport during the fog events. In three out of the four cases the rate of temperature increase after sunrise is slower than observed (for 21 August 2016 it is comparable to the observations). The model tends to simulate its lowest visibilities at the airport a few hours earlier than observations, and the visibility has not been simulated to be as low as observed for any of the cases.

The first case was for a fog event on 21 August 2016. Fog formed around 1930 UTC, 21 Aug 2016 (0330 WST, 22 Aug 2016) and cleared at 0030 UTC, 21 Aug (0830 WST, 22 Aug). The environmental flow was southerly, and fog formed with a temperature decrease of more than 2 degrees Celsius and a calming in the wind followed by a south-easterly change. This suggests that the cold air drainage down the Helena Valley in combination with the southerly environmental flow might contribute to the fog formation and enhancement at Perth airport. In previous case studies (Potts and Roux 2016) there were indications that the drainage flow may also lead to the dissipation of fog through increased turbulence and mixing of dry air. Overall ACCESS-C2 simulated the mesoscale dynamics well for this case and the simulated areas of low visibility on the coastal plain coincided well with observations from the Himawari-8 satellite imagery. There were no significant differences between the standard ACCESS-C2 simulations and the sensitivity experiments, although ACCESS-C2 showed much more detail in the temperature and dewpoint temperature fields and the low visibility simulations were much improved over ACCESS-C1.

In the second case, fog formed around 1945 UTC 3 Oct 2015 (0345 WST, 4 Oct) and cleared just over 3.5 hours later around 2320 UTC, 3 Oct (0720 WST, 4 Oct). There was fog to the north of the airport earlier in the night and it advected south with light northerly surface winds which existed throughout much of the night. In this case Experiment R, where the APS2 ACCESS-R data was used for initial and boundary conditions, provided a much improved simulation over both the ACCESS-C1 run and the other ACCESS-C2 experiments. The other model runs did not have the light northerly winds before the fog and had a large warm bias during the night. Further evaluation of Experiment R together with surface and upper air observations revealed that the northerly flow came from north-westerlies being blocked by the Darling escarpment. In previous case studies Potts and Roux (2016) suggested that the blocked flow may precondition the environment before fog formed. The current case studies suggest that it can also have a more direct impact by transporting fog from the north and/or mixing cooler air from the blocked flow with the moist environmental onshore flow. ACCESS-C2 simulated the mesoscale dynamics very well when nested in ACCESS-R2, highlighting the importance that boundary conditions can have on the simulation of fog.

In the third case study, fog was observed between 0300 and 0900 WST on the morning of 5 July 2015. Surface winds were more variable in this study and the model did not capture the full variability but overall ACCESS-C2 provided a reasonably good representation of the mesoscale


Page 39

flows. Experiment R did not have the same dramatic improvements as in the previous case, highlighting how case dependent results can be.

The last case study date is 9 June 2013. Fog was observed for much of the night from 2100 WST, 9 June, to 0700 WST, 10 June. This case study was already described by Potts and Roux (2016) and has been used as a reference. Results from the sensitivity experiments show that the choice of land fractions had a big influence on the simulation of night time temperature, for this case in particular.

Overall the model represented the mesoscale dynamics at Perth Airport well. The topography in the region plays an important role in the formation of fog. The case studies of 2015 suggested that a possible mechanism for fog is the blocking of the moist environmental flow from the Indian Ocean to the west by the escarpment, resulting in northerly winds on the coastal plain which can propagate fog to the south by advection or through the mixing of two air masses. In moist southerly environmental flow (like the case of 21 August 2016) cold air drainage down the Helena Valley may contribute to the development of fog when the cool drainage flow mixes with the moist southerlies.

The case studies indicated that the boundary conditions and initial conditions, as well as the accurate representation of the land surface and coast, are important in forecasting fog at Perth airport. Uncertainty in the initial and boundary conditions as well as in the model parameterisations could be addressed through the use of ensemble modelling.

10. ACKNOWLEDGEMENTS

Thanks to all the people in the Bureau who contributed to the work. Kathleen Hirst provided the DEM topography data and Chris Lucas generated the Himawari-8 and GEOCAT fog analysis products. Imtiaz Dharssi has been instrumental in setting up the surface fractions and soil moisture for the sensitivity studies, as well as providing information on land surface modelling at the Bureau. Wenming Lu was a tremendous help with setting up the modelling experiments and resolving technical issues. Rod Potts, Beth Ebert and Richard Dare provided valuable feedback. Thanks also to Gary Dietachmayer for supporting the work on upgrading the model visibility. The collaboration with Adrian Lock and Ian Boutle from the UK Met Office is also very much appreciated.

11. REFERENCES

Barras, V. J. I., & Miao, Y. (2011). GENESIS: forming an ACCESS‐SCM testbed from NWP analyses. CAWCR Research Letters, (6), 11–15.

Bessho, K. et al. (2016), An Introduction to Himawari‐8/9 ‐ Japan’s New‐Generation Geostationary Meteorological Satellites, J. Meteorol. Soc. Japan. Ser. II, 94(2), 151–183, doi:10.2151/jmsj.2016‐009.

Best, M. J., Pryor, M., Clark, D. B., Rooney, G. G., Essery, R. L. H., Ménard, C. B., … Harding, R. J. (2011). The Joint UK Land Environment Simulator (JULES), model description – Part 1: Energy and water fluxes. Geoscientific Model Development, 4(3), 677–699. https://doi.org/10.5194/gmd‐4‐701‐2011

Boneh, T., Weymouth, G. T., Newham, P., Potts, R., Bally, J., Nicholson, A. E., & Korb, K. B. (2015). Fog Forecasting for Melbourne Airport Using a Bayesian Decision Network. Weather and Forecasting, 30(5), 1218–1233. https://doi.org/10.1175/waf‐d‐15‐0005.1

Bureau of Meteorology. (2010). Operational implementation of the ACCESS Numerical


Weather Prediction systems. Operations Bulletin, (83), 1–35. Retrieved from http://web.bom.gov.au/nmoc/stan/opsbull/apob83.pdf

Clark, D. B., Mercado, L. M., Sitch, S., Jones, C. D., Gedney, N., Best, M. J., … Cox, P. M. (2011). The Joint UK Land Environment Simulator (JULES), model description – Part 2: Carbon fluxes and vegetation dynamics. Geoscientific Model Development, 4(3), 701–722. https://doi.org/10.5194/gmd‐4‐701‐2011

Clark, P. A., Harcourt, S. A., Macpherson, B., Mathison, C. T., Cusack, S., & Naylor, M. (2008). Prediction of visibility and aerosol within the operational Met Office Unified Model. I: Model formulation and variational assimilation. Quarterly Journal of the Royal Meteorological Society, 134, 1801–1816.

Claxton, B. M. (2013). Parameters controlling the aerosol hydration scheme within the UKV visibility parametrization. Met Office, Cardington Technical Note No. 85, (85), 1–35.

Dare, R. A. (2017). Evaluation of numerical model forecasts of visibility and fog at Australian airports. Bureau Research Report No. 024.

Dharssi, I., Steinle, P., & Fernon, J. (2015). Improved numerical weather predictions by using optimised urban model parameter values and satellite derived tree heights. In 21st International Congress on Modelling and Simulation Gold Coast, Australia, 29 Nov to 4 Dec 2015 (pp. 1161–1167). Retrieved from www.mssanz.org.au/modsim2015/M4/dharssi.pdf

Golding, B. W. (1993). A Study of the Influence of Terrain on Fog Development. Monthly Weather Review.

Gultepe, I., Tardif, R., Michaelides, S. C. S., Cermak, J., Bott, A., Bendix, J., … Cober, S. G. (2007). Fog Research: A Review of Past Achievements and Future Perspectives. Pure Appl. Geophys., 164, 1121–1159. https://doi.org/10.1007/s00024‐007‐0211‐x

Lucas, C. (2017). Validation of a Satellite-based Fog and Low Stratus Detection Algorithm Using Himawari-8 over Australia. Bureau Research Report No. 025, (September), 82 pp

Potts, R., & Roux, B. (2016). A High Resolution NWP Modelling Study of Fog at Perth Airport. Melbourne.

Puri, K., Dietachmayer, G., Steinle, P., Dix, M., Rikus, L., Logan, L., … Engel, C. (2013). Implementation of the initial ACCESS numerical weather prediction system. Australian Meteorological and Oceanographic Journal, 63(2), 265–284. Retrieved from http://www.scopus.com/inward/record.url?eid=2‐s2.0‐84874408763&partnerID=tZOtx3y1

Tepper, G., & Watson, A. (1990). The wintertime nocturnal northeasterly wind of Adelaide, South Australia: an example of topographic blocking in a stably‐stratified air mass. Australian Meteorological Magazine. Retrieved from https://scholar.google.com.au/scholar?as_q=blocked+flow&as_epq=&as_oq=Adelaide&as_eq=&as_occt=any&as_sauthors=tepper+watson&as_publication=&as_ylo=1990&as_yhi=1999&btnG=&hl=en&as_sdt=0%2C5

Wilkinson, J. M., Porson, A. N. F., Bornemann, F. J., Weeks, M., Field, P. R., & Lock, A. P. (2013). Improved microphysical parametrization of drizzle and fog for operational forecasting using the Met Office Unified Model. Quarterly Journal of the Royal Meteorological Society, 139(671), 488–500. https://doi.org/10.1002/qj.1975

Weymouth, G. T., 2006: Fog and low cloud imagery from satellite. Asia–Pacific Satellite Applications Training Seminar, Melbourne,VIC, Australia, Australian Bureau of Meteorology.

Zeschke, B., 2015: The Night Microphysics RGB product, Australian Vlab Centre of Excellence National Himawari-8 Training Campaign resource. http://www.virtuallab.bom.gov.au/files/8314/4100/7579/NightMicrophysicsRGBTOTALoneslide_Compatibility_Mode.pdf

ACCESS Sensitivity Experiments for fog cases at Perth Airport...experiments. Perth Airport depicted...

Documents

Transcript of ACCESS Sensitivity Experiments for fog cases at Perth Airport...experiments. Perth Airport depicted...