Short range forecasting methods of fog, visibility and low clouds

Short range forecasting methods of fog,

visibility and low clouds

Second Midterm Workshop Langen, Germany, 20 October 2005

Scientific Committee

Wilfried Jacobs Silas Chr. Michaelides

Vesa Nietosvaara Andreas Bott Jörg Bendix Jan Cermak

COST- the acronym for European COoperation in the field of Scientific and Technical Research- is the oldest and widest European intergovernmental network for cooperation in research. Established by the Ministerial Conference in November 1971, COST is presently used by the scientific communities of 35 European countries to cooperate in common research projects supported by national funds. The funds provided by COST - less than 1% of the total value of the projects - support the COST cooperation networks (COST Actions) through which, with only around €20 million per year, more than 30.000 European scientists are involved in research having a total value which exceeds €2 billion per year. This is the financial worth of the European added value which COST achieves. A “bottom up approach” (the initiative of launching a COST Action comes from the European scientists themselves), “à la carte participation” (only countries interested in the Action participate), “equality of access” (participation is open also to the scientific communities of countries not belonging to the European Union) and “flexible structure” (easy implementation and light management of the research initiatives ) are the main characteristics of COST. As precursor of advanced multidisciplinary research COST has a very important role for the realisation of the European Research Area (ERA) anticipating and complementing the activities of the Framework Programmes, constituting a “bridge” towards the scientific communities of emerging countries, increasing the mobility of researchers across Europe and fostering the establishment of “Networks of Excellence” in many key scientific domains such as: Physics, Chemistry, Telecommunications and Information Science, Nanotechnologies, Meteorology, Environment, Medicine and Health, Forests, Agriculture and Social Sciences. It covers basic and more applied research and also addresses issues of pre-normative nature or of societal importance.

TABLE OF CONTENTS

Editorial .................................................................................. 1

Microphysical data of fog observed in Clermont-Ferrand and corresponding satellite images..................................... 2

Abstract .................................................................................. 2

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

2. Geographical characteristics of the fog event .......... 3

3. Microphysical data....................................................... 6

4. Meteorological situation and satellite data................ 8

5. Conclusion and further work ...................................... 9

References ........................................................................... 11

How to improve the fog and low clouds forecasting at Casablanca-Nouasseur Airport? ........................................ 12

Abstract ................................................................................ 12

1. Introduction ................................................................ 12

2. Fog Climatology......................................................... 13 2.1. Methodology.....................................................................13 2.2. LVP conditions .................................................................15 2.3. Meteorological fog (visibility lower than 1000m)...............16

3. The COBEL-ISBA numerical prediction system at Casablanca-Nouasseur ....................................................... 19

3.1. COBEL-ISBA prediction system.......................................19 3.2. Adaptation of the COBEL-ISBA prediction system for Casablanca-Nouasseur airport.....................................................20

4. Quality assessment of the COBEL-ISBA numerical forecast system : ................................................................. 21

4.1. Case study .......................................................................21 4.2. Further study ....................................................................22

5. Conclusions ............................................................... 23

References ........................................................................... 23

Development of a satellite based fog detection scheme using model screen temperature........................................ 25

Abstract ................................................................................ 25

1. Introduction ................................................................ 26

2. Data ............................................................................. 26 2.1. GOES-12 satellite observations .......................................27 2.2. Operational model data ....................................................27 2.3. Surface Observations.......................................................28

3. Method ........................................................................ 30

4. Results ........................................................................ 34 4.1. Case studies.....................................................................34 4.2. Statistical evaluations.......................................................40

5. Discussions................................................................ 42

6. Conclusions ............................................................... 45

References ........................................................................... 46

Fog / Low Stratus Detection and Discrimination Using Satellite Data ........................................................................ 50

1. Motivation and Context ............................................. 50 1.1. Background ......................................................................50 1.2. Operational Framework....................................................50 1.3. Concept and Approach.....................................................51

2. Low Stratus Delineation ............................................ 51

3. Ground Fog Detection ............................................... 54

4. Validation / Outlook ................................................... 56

References ........................................................................... 57

Discrimination between fog and low clouds using a combination of satellite data and ground observations... 59

References ........................................................................... 62

Cost 722 - Short range forecasting methods of fog, visibility and low clouds

1

Editorial Action COST-722 is a consortium of 14 European states

and Canada operating within the framework of the European Co-operation in the field of Scientific and Technical Research. The main objective of the Action is to develop advanced methods for very short-range forecasts of fog, visibility and low clouds, adapted to characteristic areas and to user requirements. This overall objective includes: The development of pre-processing methods of the necessary input data the development of the appropriate forecast models and methods and the development of adaptable application software for the production of the forecasts.

COST-722 started at the end of November 2001. After an inventory phase (Two working groups: “Existing forecast methods“ and “Requirements from the forecasters and from the customers“), the scientific research started in October 2003 (Three Working groups: “Initial data“, “Models“ and “Statistical methods“). Eight meetings of the Management Committee in combination with meetings of the working groups, several Short Term Scientific Missions and Expert Meetings took place. Details can be found in Action’s website will run till end of May 2007.

These proceedings have been prepared from presentations during the second midterm Workshop of COST-722 that was held in Langen, Germany, on 20 October 2005.

We wish to thank all the contributors and the members of the Scientific Committee for their collaborations. We thank also the European Science Foundation and the COST Office in Brussels that have supported this publication. We would like to acknowledge the support of the local host of the Workshop, namely the Meteorological Training and Conference Centre of the DWD and the Meteorological Service of Cyprus for the final formatting of the proceedings.

Wilfried Jacobs and Silas Chr. Michaelides Editors


2

Microphysical data of fog observed in Clermont-Ferrand and corresponding

satellite images

Michèle Colomb (1) and Diane Tzanos (2)

(1) Laboratoire Régional des Ponts et Chaussées, 8-10 rue B. Palissy, 63017 Clermont-Ferrand, France

(2) Météo-France, Direction des Systèmes d’Observation, 42 avenue G. Coriolis, 31057 Toulouse, France

Abstract

Microphysical characteristics provide useful data for fog description, visibility calculation and also modelling of satellite observation. The French road and bridges laboratory at Clermont-Ferrand is equipped with fog sensors and an optical granulometer for its research activities in the field of visibility in fog for road safety studies. Most of the experiments are carried out in an artificial fog experimentation centre, but when natural fog occurs outside the laboratory, measurements are undertaken for comparative studies. Last winter, 2004-2005, microphysical data were recorded in natural fog. Measurements taken over several days were selected and compared with visibility measurements from the nearest airport weather station. The corresponding satellite images were analysed by the French weather service in order to verify how fog can be distinguished from low clouds in this particular situation where ground observation data are available.

1. Introduction Transport is often disrupted in wintertime by fog

occurrence, which is responsible for dramatic accidents causing injuries and fatalities. In addition to this impact on traffic safety, it also has an economic impact, by causing major delays in the transport of goods and passengers. Airline


3

companies or traffic managers require more efficient fog occurrence and visibility level forecasts. Fog can be defined as when visibility is less than 1000 m. However, for road traffic, it is only when visibility becomes less than 200m that fog is considered a traffic hazard for road transport. Forecasts of visibility levels would therefore assist traffic authorities in making decisions relating to driver information and traffic control measures.

In order to improve forecasting methods, input data are required and satellite data have to be connected to ground observation measurements. This paper will present ground observations based upon microphysical measurements. Visibility is calculated from these data. One particular set of measurements is presented in comparison to corresponding satellite images. The Méteo France fog analysis method is also tested in this particular situation.

2. Geographical characteristics of the fog event

An analysis of the geographical characteristics of the area of the city of Clermont-Ferrand helps to understand specific situations leading to the formation of radiation fog. The city is located at an altitude of about 300 to 400m, in the plain, along the valley of the Allier river, at the border of the “Massif Central” mountain chain (about 1000m in altitude) cf. Figure 1, Figure 2. Most situations were radiation fog (Figure 3). Data from fog that occurred on 1st of December 2004 were documented (Figure 4). Measurements are carried out next to the experimentation fog chamber (Colomb et al. 2004).


4

Figure 1. Location of the city of Clermont-Ferrand.

Figure 2. The city of Clermont-Ferrand in Auvergne region.


5

Figure 3. View from the mountain of the top of the fog layer above the valley.

Figure 4. Fog at the laboratory on 1st of December 2004, visibility about 200 m.


6

3. Microphysical data Microphysical data are obtained with a “Palas” optical

granulometer. It is a single particle counter measuring light diffusion at 90°. The measuring range is from 0.4 to 40 µm for droplet diameter, and from 1 to 105 particles/cm3 for total concentration. A calibration curve is given for water droplets. The sensor is specifically designed for measurements in wet environments: the optical head is waterproof and connected to the control unit with optic fibres of 15 or 30 metres in length. This allows the sensor to be set up in fog, a dozen or so metres from the control unit, which is located in a dry room.

The data can be given as a droplet distribution characterized by total concentration, droplet mean diameter, Sauter diameter (either effective, or surface diameter), and mass (or volume) mean diameter (Hind 1999). It is also possible to calculate an estimated visibility, cf. Figure 5, from the extinction coefficient K, using Koschmieder’s law. Liquid water content is also determined from the droplet distribution. Measurements taken over one hour are presented, during which the fog dissipates. Estimated visibility increases from about 50 m to 8000m. Four selected periods are analysed in detail. Droplet distribution from period number 2 is presented.

Figure 5. Visibility calculated from droplet distribution measured with Palas Sensor, from 8h43 to 9h43.

010002000300040005000600070008000

00:00 00:05 00:10 00:15 00:20 00:25 00:30 00:35 00:40 00:45 00:50 00:55 01:00

Time (h:m)

Visibility (m)

1 2 3 4


7

Figure 6. Droplet distribution from period number 2, corresponding to low visibility.

Most of the droplet distributions show an initial maximum between 1 and 2 µm and a second between 5 and 20µm (Figure 6). Even if the first part of the distribution composed of small droplets has a high concentration, its water content is very low as the droplets are small. However, droplet characteristics vary when visibility changes. For dense fog (visibility of 55m), the distribution is characterized by a total concentration of 737 particles/cm3, a mean diameter of 4.4 µm, a Sauter diameter of 15 µm, a mass diameter of 17 µm, and a liquid water content of 0.3 g/m3. When it dissipates to light fog with a visibility of 820m, it has a total concentration of 229 particles/cm3, a mean diameter of 1.9 µm, a Sauter diameter of 10.3 µm, a mass diameter of 12.6 µm, and a liquid water content of 0.01 g/m3 (see Table 1 ).

As a future development, when a larger set of microphysical data will be collected, it could contribute to test

0

100

200

300

400

500

600

700

0.10 1.00 10.00 100.00Diameter (µm)

N (cm-3.µm-1)

---N= 594 cm-3, LWC=0.1 g/m3

V=127m

---N= 737 cm-3, LWC=0.3 g/m3

V=55 m

--- N= 484 cm-3 LWC=0.1 g/m3

V=111m


8

some parameterisation scheme for visibility forecasting (Gultepe 2005). Table1. Droplet distribution characteristics.

Total Concentration

(n/cm3)

Mean Diameter

(µm)

Sauter Diameter

(µm)

Mass Diameter

(µm)

Liquid Water Content (g/m3)

Visibility (m)

737 4.4 15.0 16.9 0.30 55 594 3.1 13.5 16.0 0.10 127 532 2.5 11.2 13.6 0.05 230 601 2.2 9.3 12.6 0.03 297 229 1.9 10.3 12.6 0.01 820 306 1.5 5.2 7.7 0.003 1576

4. Meteorological situation and satellite data An analysis of the meteorological situation shows a

large area of low pressure set over the western part of Europe. A cold cut-off set over the southern part of France moves towards Germany and another cut-off is expanding over Portugal, on the tail of an upper-air trough. Early in the morning, this situation generates many areas of fog, which extend from the centre of France to Poland.

During daytime, cloud type images from satellite MSG clearly identify an area of very low cloud located in the valley and an area of snow, which covers the mountain (Figure 7). By night, the cloud type algorithm cannot detect snow and snow-covered areas are classified as "land" (i.e. clear sky area). At dawn (08 UTC), the very low cloud area vanishes from the cloud type image because the sun is too low for the visible channels to be useful, but sufficiently high to contaminate the 3.9 µm brightness temperature.

The Météo-France fog analysis (Guidard and Tzanos 2005), based on the above-described cloud type images, identifies this very low cloud area as a fog area. Very low visibilities are observed at the local airport of Clermont-Ferrand (200m or less from 06 to 09 UTC) which confirm this diagnosis. During this time, fog is quite perceptible. At 09 UTC, the fog density decreases and the visibility is 900 m at 10 UTC. The


9

area of fog analysed decreases too and at 11 UTC, there is no more fog in Clermont-Ferrand.

Figure 7. Cloud type identified at 9h00 UTC, from satellite MSG (by SAF-NWC).

The calculated value of the visibility for the analysed

period, cf. Figure 4, has the same range as the measured value at the local airport of Clermont-Ferrand (at less than 2 km from the laboratory): 0.2 Km (Figure 8). The fog analysis method is in agreement with these ground observations; it gives a high risk and medium risk level in the area of Clermont-Ferrand at 09 UTC.

5. Conclusion and further work When developing forecasting methods, satellite images

are compared to ground data. Information on visibility is usually provided by airport instrumentation, transmissometers or diffusiometers. A specific instrumentation for fog characterization was used in this study: an optical granulometer allowing the droplet distribution, the diameter and concentration of droplets to be determined. Visibility is calculated from these parameters.


10

Figure 8. The Météo-France fog analysis at 09 UTC and measured visibility in km.

Measurements taken over several days were selected

and compared with visibility measurements from the nearest airport weather station. The fog that occurred on 1st December is investigated in this paper. The analysis of the microphysical data and the measurements at the airport shows the same range of visibility levels. The fog analysis method concords with the ground observations for the period studied and predicts a high or medium risk of fog in the Clermont-Ferrand area.

This study is the first step of the research undertaken. The measurements carried out on other days are also being analyzed and a new series of measurements will be organized next winter in order to obtain more data for comparison with the model to diagnose fog occurrences.

In the future, a larger set of microphysical data will be collected and may be used to test a parameterization scheme for visibility forecasting. These data will also contribute towards an analysis of the microphysics of artificial fog produced in the


11

experimentation center for traffic safety studies, in comparison to natural fog.

References

Colomb M., Hirech K., Morange P., Boreux J.J., Lacôte P., Dufour J., 2004: Innovative artificial fog production device - A technical facility for research activities. Proceedings of The third International Conference on Fog, Fog collection and Dew, 11-15 Oct., Cape Town, South Africa.

Guidard V., Tzanos D., 2005: Discrimination between fog and low clouds using a combination of satellite data and ground observations. Proceedings of COST 722 Mid-Term Workshop on Short-range forecasting methods of fog, visibility and low clouds, 20 October 2005, Langen, Germany.

Gultepe I., 2005: A new warm fog microphysical parameterisation developed using aircraft observations. Proceedings of COST 722 Workshop on Short-range forecasting methods of fog, visibility and low clouds, 20 May 2005, Larnaca, Cyprus.

Hinds W. C., 1999: Aerosol technology: properties, behavior, and measurement of airborne particles - Second edition., John Wiley & Sons, Inc.


12

How to improve the fog and low clouds forecasting at Casablanca-Nouasseur

Airport?

Driss Bari (1) and Thierry Bergot (2)

(1)Direction de la Météorologie Marocaine (DMM), B.P.: 8106 Casa Oasis Maroc.

(2) Météo France, N° 42 Avenue Coriolis Gaspard, 31057 Toulouse.

Abstract

To improve the forecast of fog and low clouds at Casablanca Nouasseur airport, the COBEL 1D model has been used. The local assimilation scheme has been adapted in order to use hourly standard meteorological data from METAR and SYNOP, and vertical profiles of temperature and specific humidity from Radiosounding, but only at 00 UTC.

Some investigations of the local fog climatology were undertaken. It was shown that the predominant type of fog is the radiation one. The performance of the model has been tested by comparing the model results with observations for a case study on 02/01/2005.

1. Introduction The occurrence of fog and low-level stratus clouds plays

an important role for the safety and efficiency of ground and air transportation systems. In order to study fog mechanisms and to improve fog forecasting, a research field program has taken place, involving collaboration between the meteorological services of Morocco and France. In fact, this program concerns especially the Casablanca-Nouasseur international Airport.

Firstly, some investigation of the local fog climatology has been undertaken. This study demonstrates that the predominant fog at Casablanca-Nouasseur is associated with a surface cooling before the onset and light wind. These characteristics are typical of radiation fog. Consequently, it has


13

been decided to implement the COBEL-ISBA numerical prediction system at Casablanca-Nouasseur airport.

The first step of this collaboration between D.M.M. and C.N.R.M., has been the implementation and modification of the COBEL-ISBA numerical prediction system at Nouasseur Airport. The main modification concerns the assimilation scheme. In fact, the available observations at Casablanca-Nouasseur are extracted from hourly SYNOP, METAR and Radiosounding.

Next, we will evaluate the improvement of fog forecasting at the airport in comparison with observation during a winter season. The portability of the COBEL-ISBA system is also currently done for other sites, like JF Kennedy airport.

2. Fog Climatology

2.1. Methodology An increased understanding of local climatological

parameters associated with a specific fog type appears to carry some benefit. This would presumably allow better recognition and more accurate forecasting of the situations most suitable for a certain fog type to occur. The Nouasseur airport is located in a valley inside an open plain (Cf. Figure 1), at 33°14’ N and 07°15’ W, influenced by the Bouskoura forest from the north. The occurrence of fog at Casablanca-Nouasseur is 105 events per year and this airport gives a suitable environment for fog formation.

The data used to establish fog climatology cover the period from January 2000 to June 2005. Hourly surface observations of visibility, temperature, wind, precipitation, cloud cover and ceiling height are used in a simple algorithm to assign a fog type to every identified event at Nouasseur airport. In the following, the classification algorithm used in this study is outlined.

Fog events are first divided in precipitation and non-precipitation events. An event induced by precipitation is defined when some type of precipitation is observed at the onset of the event and/or during the two previous hours. The


14

working hypothesis is that the moistening and cooling of the lower atmosphere by the evaporation of precipitation are the main factors leading to fog onset.

For the non-precipitation events, observations at the time of the onset and previous five hours are used in order to identify the possible mechanisms leading to onset.

A radiation fog event is classically defined as being associated with surface radiational cooling under light wind conditions and under clear skies. Here, radiation fog is identified whenever fog onset is characterized by cooling of the 2m temperature during the 3 previous hours, with wind speed below 2.5 m/s and absence of ceiling the hour before. In order not to be too restrictive, a radiation fog event is also identified if cooling is observed as the ceiling height is increasing with total cloudiness below 4 octas, again under light wind conditions.

An advection fog event refers to scenarios where moist air is carried by the wind (i.e. advected) over land. In terms of surface observations, this would translate into a sudden reduction in visibility in association with flow coming from a large water body at a significant speed. In the classification scheme used here, the wind direction must be from the west to north before the onset of the fog, with a speed great than 3 m/s during the five previous hours and becoming weaker during the 2 previous hours.

The onset of fog related to the lowering Stratus is identified whenever a reduction in visibility below 1 Km is observed to occur following a gradual lowering of ceiling heights over the 6 previous hours. Ceiling heights must be observed to be below 700m, with a total cloudiness great that 4 octas, during that time period in order for the event to be classified as a fog resulting from cloud base lowering. If a fog event does not meet all the criteria outlined above, or there is insufficient valid data before onset, it is classified as indeterminate.

Using this algorithm, based on the well known mechanisms of fog formation, a classification of all fog events is performed for Nouasseur airport. Results indicate that radiation fog events are generally the most common in the airport with 47.6% of 473 detected fog events, followed by the “lowering cloud base” fog with 30.2%. The advection one is


15

less frequent (3.4 %); 18.6% of fog events are indeterminate because of missing data in the hours leading to onset. It contains also the events which occur under conditions that do not fit any of the simple conceptual models used in the classification algorithm.

Figure 1. Location of the Nouasseur region and its environment.

2.2. LVP conditions The occurrence of low ceiling and/or poor visibility

conditions restricting the flow of air traffic into major airport terminals is one of the major causes of aircraft delay. The low visibility procedures (LVP) conditions at Nouasseur airport are defined by a visibility less than 200 m and/or ceiling is below 25 m (82 feet). Table 1 sums up the monthly distribution of “Fog_LVP” occurrence at Nouasseur Airport during the period of January 2000 – June 2005. We have noticed 109 fogs LVP among 473 fog events, which represent 23% of the total observed events. This fog type tends to occur frequently during late winter with a likelihood of 4.8 % in January and 3.8 % in February.


16

2.3. Meteorological fog (visibility lower than 1000m) The monthly frequency of fog events is shown in Figure

2 to illustrate more clearly the seasonal distribution of events. The frequency is defined as the total number of fog events par month divided by the total number of events for each fog type.

There is a marked tendency toward a minimum likelihood of fog occurrence during the summer months. This is likely the result of the shorter nights and warmer temperatures characterizing the summer months. In fact, the short nights, particularly in July and August, minimize the likelihood of radiation fog.

Table 1. Monthly Fog LVP occurrence (Visibility <= 200 m and/or Ceiling <= 25 m (82 feet) ) at the Nouasseur Airport over the period

of January 2000 – June 2005.

Month Fog LVP Occurence Frequency(%) January 23 4.8

February 18 3.8 March 7 1.5 April 7 1.5 May 4 0.8 June 7 1,5 July 3 0.6

August 0 0.0 September 6 1.3

October 16 3.4 November 7 1.5 December 11 2.3

Total 109 23.0 %

Overall, the maximum fog frequency occurs during the late winter season (especially in January, February) and early spring (March) in the region. Sufficient moisture and a long enough period of nocturnal radiational cooling provide an explanation for the maximum in radiation fog occurrence.


17

0 2 4 6 8 10 12 14 16 18

Janv

Mars

Mai

Juil.

Sept.

Nov.

With_Precip Radiation Advection Low_Stratus Indeterminate

Figure 2. Frequency of the Monthly distribution of fog occurrence at

Nouasseur airport over the Jan. 2000 – June 2005 period.

In terms of diurnal dependence, not surprisingly, it is observed that fog onset occurs mostly during the night time hours. In fact, fog onset occurs most often in the latter half of the night. This suggests the role of nocturnal radiational cooling in leading to fog formation and is consistent with the predominance of radiation fog at Nouasseur Airport.

Near surface flow and temperature conditions at the time of fog onset have been explored to further define specific conditions leading to an enhancement in the likelihood of fog formation.

An examination of the temperature distribution at fog onset shows that significant fog frequencies are observed over a range of temperatures between 10°C and 14°C. The distribution of fog events with temperature could be related to the characteristic water vapour content of the air mass over the region. This takes place due to the evapotranspiration of the forest of Bouskoura.

In terms of dependence on wind speed, the corresponding distribution at fog onset (Cf. Figure 3) shows that Nouasseur airport has high frequencies of fog formation under calm wind conditions (51.5%). These results are


18

consistent with the fact that this location was found to be mostly under a radiation fog regime.

A detailed analysis of the onset and dissipation times of the fog events at Nouasseur airport should allow us to build a preliminary idea on the persistence of the phenomenon. The duration of fog event does not depend only on the season when the fog is formed but also it depends on the onset time. It has seen that:

- The fog events which develop in the night (between 9 pm and 3 am) are the most frequent and are often long durations for all the seasons, until 8 hours.

- The morning fogs (between 3 am and 6 am) are generally of short duration (2 to 3 hours).

- The other fog events do not last generally more than 2 hours.

Figure 3. Rose wind at Nouasseur airport, at fog onset, over the period of January 2000 – June 2005 The frequency of calm wind

(wind speed low than 2 m/s) is equal to 51.5%.


19

3. The COBEL-ISBA numerical prediction system at Casablanca-Nouasseur

3.1. COBEL-ISBA prediction system The COBEL-ISBA local prediction system is currently

operationally used at Paris-CdG international airport (Bergot et al. 2005). This prediction system is based on an adaptive assimilation scheme dedicated to surface boundary layer and on the local Cobel-Isba numerical model. The high resolution COBEL-ISBA 1D-model has been developed in collaboration between METEO-FRANCE/C.N.R.M., the Laboratoire d’Aérologie (Paul Sabatier University / C.N.R.S.) and Université du.Québec à..Montréal (U.Q.A.M.). It is adapted for forecasting the formation and development of fog (Bergot 1993, Bergot and Guedalia 1994).

Figure 4. Schematic depiction of the physical processes included in

The COBEL-ISBA forecast method. This model is forced with mesoscale parameters derived

from a 3D limited-area model which is ALADIN-MAROC in our case.


20

This system includes an adaptive assimilation scheme at local scale, a detailed parameterization of soil-atmosphere exchanges and parameterization of turbulence in stable layers in order to correctly simulate the nocturnal atmospheric cooling (Cf. Figure 4). The model equations are solved on a high resolution vertical grid with 20 vertical levels below 200m. The first level is at 50 cm.

3.2. Adaptation of the COBEL-ISBA prediction system for Casablanca-Nouasseur airport

The local assimilation scheme has been adapted in order to use classical meteorological observations (METAR and SYNOP). The observed state contains only classical meteorological observation, in opposition to the application done in France at Paris-CdG, Paris-Orly and Lyon. The 2m observations of temperature and humidity and the observed present weather come from SYNOP messages. The information of cloud cover comes from METAR. Also, the vertical profiles of temperature and specific humidity originate from the Radiosounding of Casablanca Anfa station at 00UTC.

The estimation of the temperature and humidity initial profiles uses a variational assimilation approach. This approach has been described in a number of articles. The principle is to minimize a cost function which measures the degree of fit to the observations and to some a priori information called “background”. This function takes the form

( ) ( ) ( ) ( )1 11 1( ) ( ) ( )2 2

TTb bJ x x x B x x y H x R y H x− −= − − + − −

Where bx is the background state (previous Cobel-Isba forecast), B is its error covariance matrix, y is the observation vector, H is the observation operator, and R is the error covariance matrix of the observations.

The assimilation procedure used to construct the initial conditions is based on different steps: first, the atmospheric vertical profiles of temperature and humidity are estimated in a 1D-Var framework, then these atmospheric profiles are adjusted to introduce the cloud cover when low clouds or fog are detected. Finally, the vertical profiles of the temperature and water content within the soil are estimated from


21

atmospheric measurements (temperature, wind, precipitation) with an offline version of the surface-vegetation-atmosphere transfert model ISBA.

4. Quality assessment of the COBEL-ISBA numerical forecast system :

4.1. Case study In order to illustrate the first results, a fog that took place

during the night of 2 January 2005 is studied. The figure shows the plot of the Radiosounding data at Casablanca Anfa Meteorological Station. The fog layer appears on 1st January around 23H45 with a minimum of visibility of 600m, under calm wind conditions. We notice clearly the thermal inversion in the boundary layer.

The COBEL-ISBA model has been run, up to 21h forecasts, starting from 00UTC on 02/01/2005. The initial conditions are derived from the assimilation procedure previously described. The mesoscale forcing consists only of the geostrophic wind and the cloud cover. The horizontal advections and vertical velocity are not taken into account.

Figure 5b shows the simulated time evolution of visibility for the day 02/01/2005 beginning at 00 UTC. Also, included in the figure are the observed values (blue line). The error on the forecast of the cessation of fog is of about 45minutes, which is very encouraging. However, the forecast value of the horizontal visibility inside the fog layer is too small. This result shows the tendency for the model to forecast too dense fog. This problem has been also observed at Paris-CdG Airport and is currently being studied.


22

Time (UTC)

3:002:302:001:301:000:300:00

Visi

bilit

y (m

)

1000

900

800

700

600

500

400

300

200

100

0

(a) (b)

Figure 5. (a) Radiosounding at Casablanca Anfa at 00 UTC on 02/01/2005, (b) Forecasted visibility (red solid line) and observed

visibility (blue dashed line) at Nouasseur Airport in function of time on 02/01/2005.

4.2. Further study The COBEL-ISBA system will continue to run every 3

hours, up to 21h forecasts, during one year (or a winter season at least due to the high frequency occurrence of fog). Then, we will evaluate statistically the quality of the fog forecast during the studied period. In fact, an inspection of events will be beneficial to assess the improvement of model forecasts in very short-term forecasts. Subsequently, the results will be classified according to the climatology previously detailed. Therefore, some statistics based on Hit Ratio (HR), False Alarm rate (FAR), Frequency Bias Index (FBI) and critical success Index (CSI) will be established in order to assess the quality of the forecasts. Further more, we will evaluate the forecast quality in function of the type of fog (issued from the climatology study) in order to understand better the reasons of


23

the errors and consequently to explore some way of improvement.

5. Conclusions

The main objective of the EU Cost action 722 is to develop advanced methods for very short-range forecasts of fog, visibility and low clouds, adapted to characteristic areas and to user requirement. This includes the development of adaptable application software for the production of the forecasts. This collaboration between D.M.M. and C.N.R.M. is in totally agreement with this goal. Moreover, the same kind of approach is currently done for Warsaw international airport.

The methodology consists of exploiting the characteristic of the COBEL-ISBA numerical forecast system to the possible extent. The portability of COBEL-ISBA assimilation-forecast numerical system allow an easy deployment of this kind of study and it seems to be beneficial for accurate forecast of fog events at Nouasseur Airport.

This methodology will be tested over a winter season at least, and will be compared with other methods (Hirlam1D, statistical methods). Further more, we will evaluate the forecast quality as a function of the type of fog (issued from the climatology study), in order to understand better the reasons of the errors, and consequently to explore some way of improvement.

References

Bergot T., Carrer D., Noilhan J., Bougeault P., 2005: Improved Site-Specific Numerical Prediction of Fog and Low Clouds: A Feasibility Study, Weather and Forecasting, 20, 627-64.

Bergot T., Guedalia D., 1994: Numerical forecasting of radiation fog. Part I: Numerical model and sensitivity tests. Mon. Wea. Rev., 122, 1218-1230.


24

Bergot T., 1993 : Modélisation du brouillard à l’aide d’un modèle 1D forcé par des champs méso-échelle : application à la prévision. PhD Thesis – Université Paul Sabatier, Toulouse, France.


25

Development of a satellite based fog detection scheme using model screen

temperature

Ismail Gultepe Science and Technology Branch

Cloud Physics and Severe Weather Research Section 4905 Dufferin St. Toronto, ONT. M3H-5T4, Canada

Abstract

A warm fog detection (air temperature>0°C) algorithm using a combination of GOES-12 satellite observations and screen temperature data based on an operational numerical model has been developed. This algorithm was tested on a large number of daytime cases during the spring and summer of 2004. Results from the scheme were compared with surface observations from four manned Canadian weather stations in Ontario, including Ottawa (YOW), Windsor (YQG), Sudbury (YSB), and Toronto (YYZ). Initially, when all cases are included, fog detection (hit rate) by the satellite scheme ranged between 0.50 and 0.60 when fog was also reported at the surface. It is suggested that mid- or high-level clouds within the satellite imagery during the observed foggy periods affected the scheme’s performance in detecting surface level fog for the majority of the cases. When cases with mid and high-level clouds are removed, the hit rate improves up to 0.92. Average false alarm rate about 37% suggested that inclusion of model based sounding values could improve the results from the satellite based algorithms up to 92%. Average differences between the model-based screen temperature (Tsc) and the surface observed air temperature (Tso) were found to be up to 2ºC and this can likely account for some discrepancies in detecting fog. Finally, averaging GOES and model data to scales, representing single data point observations, likely resulted in some of the failure of the fog algorithm.


26

1. Introduction Fog formation is related to thermodynamical, dynamical,

radiative, aerosol, and microphysical processes, and surface conditions (Gultepe et al. 2006, Bott and Trautmann 2002). Within fog, the extinction of light at visible ranges results in low visibilities that can affect low-level flight conditions, marine traveling, shipping, and transportation. The high frequency of fog occurrence, experienced greater than 10% of the time in some regions of Canada (Whiffen 2001) requires improvements in fog nowcasting and/or forecasting models. Particularly, visibility (Vis) which is related to fog intensity should be more accurately simulated to reduce the costs of fog-related accidents and delays at airports and in marine environments. Pagowski et al. (2004) studied a fog episode related to a major traffic accident in Canada and suggested that the forecasting of fog can be very difficult because of its high variability and because the physical processes occurring in the boundary layer are not simulated adequately, e.g. microphysics and turbulent fluxes. In their analysis, they emphasized the importance of land-surface schemes to obtain screen temperatures at the surface that were used to predict fog formation and development.

In this study, the operational regional Global Environmental Multi-scale (GEM) model (Belair et al. 2003, Cote et al. 1998) is used to obtain the screen temperatures (air temperature that is given at 1 m height) to be included in the analysis of a GOES-based day-time fog algorithm. The objective is to develop an algorithm that uses both operational model data and GOES observations to detect fog regions. Presently, the GOES based algorithms use only channel differencing between ch2 and ch4 observed at the cloud (fog) top. The results are verified using surface observations collected over the four stations in the Ontario region from March 1 to September 30 2004. The results are then evaluated using forecasting statistics such as hit and false alarm rates.

2. Data Data extracted from 1) GOES-12 observations, 2) the

GEM regional operational model, and 3) surface station


27

observations, were used in the analysis and summarized below.

2.1. GOES-12 satellite observations Satellite imagery was obtained from GOES-12

observations. Reflected radiances measured at five channels are available from the multispectral imager instrument aboard the GOES-12. The time resolution of the imagery is approximately every 15 minutes. Channels 1 (visible 0.5-0.7 um), 2 (near IR 3.8-4.0 um) and 4 (thermal IR 10.2-11.2 um) albedos or black body temperatures (BBT) are used here, with native ground resolutions of 1 km, 4 km, and 4 km, respectively. Details on GOES-12 observations can be found in Schmit et al. (2005).

This comparison uses a satellite sector covering south-central Ontario and Quebec centered at (45ºN, 77ºW). During post-processing procedures, channel data are registered to approximate 5 km square pixels, creating an image over the study area of 1270x1000 km2 (254 x 200 pixels).

2.2. Operational model data The operational numerical model data were obtained

from the Canadian Meteorological Centre (CMC). In this analysis, forecast screen-level temperature (Tsc) and relative humidity with respect to water (RHw) data from the 0000 UTC run of the regional Global Environmental Multiscale (GEM) model were used. Prior to May 18 2004, regional GEM had a horizontal resolution of 24 km. After this date, upgrades to the operational model by CMC increased the resolution to 15 km. The time resolution was 450 seconds (Mailhot et al. 2005, Belair et al. 2003) and the model output was obtained every 8 time steps (instantaneous, not averaged).

Screen-level temperature (Tsc) was obtained from an interpolation between the surface skin temperature estimated using the surface-land schemes (Belair et al. 2003) and the air temperature at the first atmospheric model level around 50 m. The interpolation is done using surface layer stability functions. Surface screen temperature comparisons between 2 model runs with 15 and 24 km resolutions have been studied and


28

shown in www.msc.ec.gc.ca/cmc/op_systems/doc_opchanges, indicating that differences in Tsc were negligible. For this work, numerical model data were interpolated to the same ground resolution as the post-processed satellite data (5 km).

Figure 1. Map of the surface station locations. The map covers the extent of the satellite sector used in this work.

2.3. Surface Observations Hourly surface observations were retrieved from

Environment Canada climate archives for four weather stations in south and central Ontario. These stations are listed in Table 1. A map of their locations is shown in Figure 1. Figure 2 is a digital picture of fog conditions at approximately 1140 UTC taken in the Greater Toronto Area (GTA), just north of the city limits. The main observations used in the analysis from the surface observation are temperature (Tso), relative humidity with respect to water (RHw), and visibility (Vis). Visibility is obtained based on human observers that report “prevailing visibility” which is an "average" value of the visibility around the airport determined by identifying specified targets at known


29

distances as defined in the Observers Manual (MANOBS 1977).

Figure 2. Picture of fog conditions in the GTA on Sept 03 at 1140 UTC taken by author.

Table 1. Surface stations used in the analysis.

Station Name Station ID WMO ID Location

Ottawa, ON YOW 71628 45.32 N, 75.67 W

Windsor, ON YQG 71538 42.27 N, 82.97 W

Sudbury, ON YSB 71730 46.62 N, 80.80 W

Toronto, ON YYZ 71624 43.67 N, 79.60 W


30

3. Method Hourly observations from each of the four weather

stations were examined between the dates of March 1 and September 30, 2004 for warm fog events (T > -5ºC). Off-hour “special” reports were not included. For this comparison, only daytime events were selected, i.e. solar zenith angles in the satellite images were < 70º. Observed fog events at the surface were categorized as those that had at minimum three consecutive hours of fog (listed under the “Weather” column of the station reports) with at least some of those hours in the daytime. After matching these times with the available satellite and NWP data, 19, 25, 31 and 24 cases of fog (total of 99 cases) were identified at YOW (Ottawa), YQG (Windsor), YSB (Sudbury) and YYZ (Toronto), respectively. The same number of random non-fog hours for 198 cases was also selected at each station to balance the comparison statistics.

These cases were then processed using the warm fog satellite detection scheme. Model data from the closest forecast hour were used in the algorithm. For example, when a satellite image at 1615 UTC is processed, model data from the 16-h forecast are chosen and so forth. A 3 x 3 pixel area (15 km x 15 km) centered at each of the four weather stations was extracted for comparison against the surface observations for the corresponding hour. Results from the satellite scheme were considered “foggy” if at least 1 out of 9 pixels were classified as foggy by the detection scheme. The comparisons between surface observations and model values are made by using the model output, representing 15 km scale, over the weather station.

Warm daytime fog regions were identified in GOES-12 satellite imagery using the criteria given in Table 2. On a pixel-by-pixel basis, each satellite pixel was individually checked for the conditions specified in Table 2. In order for the scheme to detect the presence of clouds/fog, the first four conditions were satisfied. The first test is an indicator of reflectance from cloud tops. The second test is an indicator of mid and high level clouds. The third test is a condition to specifically check for liquid water clouds. For this criterion, daytime channel 2 data is first separated into its SW albedo and IR components. Then, the SW albedo component is used to identify liquid water phase clouds. For the fourth criterion, the channel 2 – channel


31

4 IR temperature difference test is used an indicator of liquid water phase.

Table 2. Criteria used in the fog algorithm. The model-based screen temperature is shown as Tsc. A represents albedo and SZA is defined as solar zenith angle. RHw is the relative humidity with respect to water based on model runs. Ch1, ch2, and ch4 represent channel 1, 2, and 4 from GOES. BL is defined as the boundary layer.

# Criteria Explanation 1 Ach1 ≥ 20% Cloud detection 2 | Tch4 – Tsc | ≥ 10ºC Mid/high-level cloud

detection 3 Ach2 ≥ 8% Liquid cloud detection 4 Tch2 – Tch4 > 10ºC Liquid cloud detection 5 SZA < 70º Daytime condition 6 | Tch4 – Tsc | < 10ºC BL fog/cloud detection 7 Tch4 > -5ºC Warm fog detection 8 RHw> 80% BL fog/cloud detection

Following the cloud mask tests, all criteria 5-8 in Table 2

need to be satisfied for the satellite pixel to be considered foggy. The SZA condition simply confirms daytime conditions. In test 6, it is assumed that |Tch4-Tsc|<10°C approximately represents a maximum height of 1000 m if the dry adiabatic lapse rate is assumed. This assumption is done to increase the chance of detecting fog formation. It is clear that if the air saturated and well mixed, the fog layer thickness can become more than 1000 m. The Tch4 > -5°C condition is used for representing the liquid phase of fog because droplets are very commonly found at temperatures warmer than –5°C (Gultepe and Isaac 2004). Finally, the occurrence of fog over small scales is possible when RHw (~80%) from a large-scale model (e.g. < 50 km) is less than 100% as shown in Figure 3. This value (80%) can be increased when high-resolution models (with grid size less than 1 km) are applied for fog forecasting.


32

Figure 3. Surface observations and GEM model data for Sept 03 2004 at 1315 UTC for Toronto Pearson Airport (YYZ). Model points are averaged values from the 3 x 3 pixel grid over the station. Top figure shows observed and model surface temperatures and observed visibility. Bottom figure shows observed and model surface relative humidity. Yellow box indicates time period when fog was reported at the surface station (during the 11, 12 and 13 UTC hours).

The statistical descriptions (Tapp et al. 1986, Woodcook 1976, Saseendran et al. 2002), including hit rate (HR), false alarm rate (FAR), and true skill statistics (TSS), are used to evaluate the success of the fog algorithm. HR is a measure of the accuracy of the validation. FAR is the fraction of the


33

predictions that were not actually observed. TSS is the measure of relative forecasting accuracy.

Figure 4. GOES satellite observations for the south-central Ontario study area for Sept 03 2004 at 1315 UTC (Boxes a-d) and GEM-based screen temperature (Tsc) and RHw are shown for 1300 UTC in boxes e and f, respectively. Boxes a and b show Tch2 and Tch4, respectively. Box c shows the Tch2-Tch4 difference. Box d shows the channel 2 albedo (Ach2).


34

4. Results In this section, two examples, representing a good and a

bad case, are described. Then, an overall evaluation of the results for all of the 198 cases based on the statistical concepts is provided and discussed.

Figure 5. This figure shows the RBG image of ch1 albedo, ch4 IR temperature, and the fog mask for September 3 2004. The light yellowish color over Toronto area indicates the foggy area.

4.1. Case studies For this case, the average model surface temperature

was 19.6ºC and the observed dry bulb temperature was 17.1ºC at 1300 UTC at YYZ. Figure 3a shows observed surface temperature, visibility, and averaged GEM operational model-based screen surface temperature. The observed visibility was 1.6 km and fog was recorded in the hourly surface report at 1300 UTC. Averaged model-based RHw during fog episode was 82% (Figure 3b) and the observed RHw was 100%. Based on this result, a low RHw value criterion for fog algorithm was chosen but needs to be tested for future studies. Figure 4 depicts a series of GOES satellite-related images.


35

Figure 6. Surface observations and GEM model data for July 31 2004 at 1315 UTC for YYZ. Top figure shows observed and model surface temperatures and observed visibility. Bottom figure shows observed and model surface relative humidity. Model data are for 1300 UTC. Yellow box indicates time period when fog was reported at the surface station (during the hours of 7-14 UTC).

The time of the satellite data is 1315 UTC, while the

model data are for the 1300 UTC forecast hour. Figure 4a shows that the channel 2 IR cloud top temperature (Tt) indicates some clouds nearby Toronto. The IR image from channel 4 (Figure 4b) did not indicate any fog/low clouds over the Lake Ontario region, specifically at YYZ. When the Tch2-Tch4 difference is greater than 10°C (Figure 4c), the pixel is accepted as containing liquid water. For this case, this test was able to identify liquid water clouds in southern Ontario. The fog regions are also identified in the Ach2 image (albedo ≥ 0.08 indicates likelihood of liquid water) in areas of yellow color


36

(Figure 4d). The operational GEM regional model output of Tsc and RHw are shown in Figures. 4e and 4f, respectively. The warm surfaces (> 18°C) are seen nearby YYZ station where RHw is approximately greater than 0.80, indicating consistency between model based fog definition and GOES fog algorithm indication.

Application of the satellite fog algorithm provides the final fog detection output in Figure 5 which shows the foggy region over YYZ. This result is verified by station observations and separately by an external human observer (Figure 3).

In case of mid- and high-level clouds, satellite based algorithms do not work but summarizing this condition here gives us a chance to show applicability of an operational model. For this case, the results are summarized similar to the September 3 case. Time series of surface observations and GEM model data at YYZ are shown in Figure 6. The RHw and T values from GEM model and observations were found comparable. Reported visibility was 2.4 km, and fog and rain showers were recorded in the hourly surface report at YYZ at 1300 UTC. Figure 7 depicts a series of GOES satellite-related images as in Figure 4. The satellite data was collected at 1315 UTC, while the model data are for the 1300 UTC. A swath of cold cloud is seen across the image covering Lake Ontario and the GTA. This is indicated by the white colored clouds in channels 2 and 4 (Figures 7a and 7b). The average cloud top temperature (Tt) over YYZ from GOES channel 4 was -30.8ºC. Figure 7c shows Tch2-Tch4 > 10°C areas of liquid water cloud or fog in the vicinity of YYZ. Liquid water cloud regions are also identified in the Ach2 image in areas west of the GTA, but not at YYZ (Figure 7d). High level clouds with dark blue color with smaller albedo values are see in and around the GTA. The model surface screen temperature was about 20ºC at YYZ (Figure 7e) and the observed dry bulb temperature was 19.6ºC at 1300 UTC at YYZ (Figure 6). The model RHw was 95% (Figure 7f) and the observed RHw was 98%.


37

Figure 7. This figure is obtained similar to Figure 4. The GOES satellite observations and GEM model data are for July 31 2004 at 1315 UTC and 1300 UTC, respectively.


38

Figure 8. This figure is obtained similar to Figure 5 except for July 31 2004 case. The light blue color indicates the fog regions.

The derived GOES cloud mask indicates cloudy regions over most of the image (Figure 8). Fog was not detected at YYZ over where the cold clouds were seen overhead. In this case, high-level clouds obscured the fog conditions below, making fog undetectable by the satellite algorithm. Some fog areas are seen far west, indicating that foggy layers were likely exist below the low level clouds over YYZ area, and this is verified with low visibility values observed at YYZ (Figure 6) and high RHw values from the GEM model. If high- and mid-level clouds exist, then a high resolution GEM model together with surface based observations (e.g. microwave radiometers) should be used for fog forecasting rather than a satellite based algorithm.


39

Table 3. True skill score statistics for the satellite fog scheme compared to surface observations.

Station name GOES

Algorithm Total Statistics YOW (Ottawa) Yes No HR FAR TSS Surface Obs Fog Yes 5 14 19 .61 .39 .21 Surface Obs Fog No 1 18 19

GOES

Algorithm YQG (Windsor) Yes No HR FAR TSS Surface Obs Fog Yes 8 17 25 .56 .44 .12 Surface Obs Fog No 5 20 25

GOES

Algorithm YSB (Sudbury) Yes No HR FAR TSS Surface Obs Fog Yes 8 23 31 .60 .40 .20 Surface Obs Fog No 2 29 31

GOES

Algorithm YYZ (Toronto) Yes No HR FAR TSS Surface Obs Fog Yes 7 17 24 .58 .32 .16 Surface Obs Fog No 3 21 24

GOES

Algorithm Total Yes No HR FAR TSS Surface Obs Fog Yes 28 71 99 .59 .41 .17 Surface Obs Fog No 11 88 99


40

4.2. Statistical evaluations The results were analyzed by computing hit rates (HR), false alarm rates (FAR) and true skill statistics (TSS). HR is computed as the fraction of GOES foggy samples compared to the total number of foggy observations reported at the surface. FAR is the fraction of GOES non-foggy samples compared to the total number of reported non-foggy observations. The true skill score (TSS) is HR–FAR. These are summarized in Table 3. The hit rates for the four stations ranged between 0.56 and 0.61 and for all cases combined was 0.59. The false alarm rates varied between 0.32 and 0.44.

When all cases with |Tt-Tsc|>10ºC implying were removed from the initial dataset, a total of 63 (out of 198) cases remained. Similar skill statistics as in Table 3 was generated and shown in Table 4. When cases with suspected mid or high-level clouds are removed, the hit rates at each of the stations improve significantly from those in Table 3. Hit rates ranged from 0.55 in Windsor to 1.0 in Ottawa. False alarm rates decreased to 0 at both Windsor and Toronto. Overall true skill scores ranged between 0.55 and 0.86; 0.59 for all four stations combined.

Because surface temperatures play an important part in the satellite detection scheme, the ability of the numerical model to forecast this parameter plays an important factor in the satellite algorithm’s overall performance. For this reason, a comparison was made between the model screen temperatures (Tsc) against the values actually observed (Tso) at the stations. Average model screen temperatures across the 3x3 pixel grid were computed for each of the cases at each station and then differenced from the reported dry bulb temperature at that hour. Absolute average, minimum and maximum differences were computed and shown across the 198 cases shown in Table 5.

The average absolute temperature difference (ΔTdif) between the model and surface observations was between 1.4 and 1.8ºC. Note that some of these differences are likely in part due to comparisons made between point observations and scaled averaged values from the model runs.


41

Table 4. True skill statistics for the satellite fog scheme compared to surface observations when mid- and high-level cloud cases are removed. Station Name YOW (Ottawa)

GOES Algorithm Total Statistics

Surface Obs for Yes Yes No HR FAR TSS Surface Obs Fog No 5 0 5 0.92 .08 .86 YQG (Windsor) 1 6 7 Surface Obs Fog Yes Yes No HR FAR TSS Surface Obs Fog No 6 5 11 .69 31 .55 YSB (Sudbury) 0 5 5 Surface Obs Fog Yes Yes No HR FAR TSS Surface Obs Fog No 6 2 8 .80 .20 .58 YYZ (Toronto) 2 10 12 Surface Obs Fog Yes Yes No HR FAR TSS Surface Obs Fog No 5 3 8 .80 20 .63 Total 0 7 7 Surface Obs Fog Yes Yes No HR FAR TSS Surface Obs Fog No 22 10 32 .79 .21 .59 3 28 31


42

Table 5. Absolute average, minimum and maximum temperature differences between model based screen temperatures (Tsc) and surface reported values (Tso) for the 198 cases.

Differences [ºC] YOW YQG YSB YYZ

Average 1.5 1.8 1.6 1.4 Minimum 0.0 0.0 0.1 0.0 Maximum 5.2 7.2 5.6 7.3

5. Discussions Fog detection using satellite observations have been

studied in previous works by Ellrod (1995), Bendix (2002) and Lee et al. (1997). Ellrod (1995) showed that cloud physical thickness can be obtained using temperature differences from channels 2 and 4 from GOES. This method will not work when mid- and high-level clouds are present. Their method can be compared with the method given here that is based on both model runs and satellite observations. This work will be performed in future. It is clear that when satellite observations cannot be applicable to a case with mid- and high- clouds, model based LWC values together with surface observations should be used to estimate visibility at various levels as described in K84 (Kunkel 1984).

Using a high-resolution model run for a given point, fog physical thickness (Tsc-Tt) can easily obtained from the results of the present work and that can be used for calculating the dissipation time of the fog.

A simple test for mid- and high-level cloud identification is differencing the temperature between the surface and the cloud tops (test 2 in Table 2). If the temperature difference is relatively large (e.g. > 10ºC) then there may be evidence of mid- and high-level clouds. For cases when the stations reported fog and GOES produced non-foggy results, absolute Tt-Tsc differences were examined. Averages of these differences for the 3 x 3 pixel grid were calculated in Table 6 for all four stations. The calculations show an absolute average temperature difference of between 31ºC and 35ºC across the four stations for such cases.


43

It should be noted that a night-time fog detection can be least accurate when some of information from satellites cannot be obtained such as ch1 albedo. The ch2 values at night represent only IR component due to non-existence of solar radiation; therefore, it can be used directly in the fog detection schemes as described in Ellrod (1995).

Using operational models, fog regions and visibility in the BL can be determined. The RUC model presently uses physical parameterizations to calculate visibility (Benjamin et al. 2004, Smirnova et al. 2000). Their visibility calculations are based on LWC values calculated using the parameterized equations given in Stoelinga and Warner (1999). They also calculate visibility using RHw values when clouds did not occur. These parameterizations were not available in GEM simulations that can be used for further validations for fog microphysics.

The relationships between visibility and Nd were studied by Gultepe (2005) who showed that Nd should be included in visibility parameterizations that includes only LWC. These studies suggested that visibility can be obtained as a function of both LWC and Nd, and therefore the new parameterizations together with satellite-based algorithms can be used to obtain visibility directly from satellite observations.

The cold bias at the surface from operational GEM runs is usually indicated and needs to be improved because Tsc-Tso in the present work reaches up to 2°C. This may cause some fog detecting problems related to RHw and T. More information on this topic can be found on http://www.msc.ec.gc.ca/cmc/op_systems/doc_opchanges.


44

Table 6. A listing of GOES cloud top temperature (Tt) and model screen temperature (Tsc) absolute differences for cases when the satellite method returned non-foggy results compared to a reported fog event by the surface station. Average, minimum and maximum differences are provided in the last 3 rows from the top to the bottom, respectively.

|Tt-Tsc| for individual cases [°C]

Station name

YOW (14

cases)

YQG (17

cases)

YSB (23

cases)

YYZ (17

cases) |Tt-Tsc| 56.1 23.3 16.4 18.0 |Tt-Tsc| 41.8 57.4 5.5 26.2 |Tt-Tsc| 59.3 21.0 38.5 19.4 |Tt-Tsc| 45.7 23.8 4.2 23.0 |Tt-Tsc| 21.7 78.8 33.4 52.9 |Tt-Tsc| 20.1 11.3 31.5 26.4 |Tt-Tsc| 13.2 20.8 35.5 42.6 |Tt-Tsc| 20.9 44.7 59.2 25.5 |Tt-Tsc| 38.9 20.9 49.6 23.4 |Tt-Tsc| 61.2 58.3 30.5 12.7 |Tt-Tsc| 2.8 15.8 39.4 51.5 |Tt-Tsc| 30.7 16.6 15.3 31.0 |Tt-Tsc| 18.9 25.4 23.0 25.5 |Tt-Tsc| 55.4 46.7 65.3 28.7 |Tt-Tsc| 31.0 35.6 15.7 |Tt-Tsc| 44.9 19.8 51.3 |Tt-Tsc| 12.9 59.5 56.0 |Tt-Tsc| 10.9 |Tt-Tsc| 30.4 |Tt-Tsc| 14.6 |Tt-Tsc| 43.9 |Tt-Tsc| 37.8 |Tt-Tsc| 10.0

Average 34.8 32.6 30.9 31.2 Minimum 2.8 11.3 4.2 12.7 Maximum 61.2 78.8 65.3 56.0

A new version of GEM called GEM-LAM (Limited Area

Model) has the horizontal and vertical resolutions set at 0.0225


45

deg (~2.5 km) and 58 levels, respectively. The lowest vertical level for the GEM is approximately at 50 m. The output of the operational GEM-15 run at 0000 UTC is being used to provide the initial and boundary conditions for the two runs. They are being initiated at 1200 UTC and provide hourly forecasts out to 24 hours. The aim of the GEM-LAM project is to develop a high-resolution model that offers a better representation of local conditions (orography, vegetations, etc), physical processes (cloud microphysics, radiation, etc) and dynamical organization of weather systems at all scales. Therefore, it will be better suitable for fog studies. The GEM-LAM operational/research- oriented versions will be used for fog and visibility analysis. Clearly, state-of-the-art high-resolution model development will be very helpful for fog nowcasting/forecasting. In the future studies, use of such a LAM model for short time forecasting would be very useful to plan field projects such as Fog Remote sensing And Modeling (FRAM) (Gultepe et al., 2005).

6. Conclusions The case studies analyzed from the spring/summer

2004 showed that a satellite-based warm fog detection scheme using model derived screen temperatures was initially able to verify fog conditions for 59% of cases reported at four surface observation stations in Ontario. A potential reason for the disparity, namely interference from overlying cloud layers, is suggested. Also, when optically thin fog occurs, significant contribution from the surface to reflected radiance at visible and near IR channels can complicate the algorithm.

The following conclusions are drawn from this work:

♦When cases with suspected mid or high-level clouds are removed, the satellite scheme’s mean hit rate increased to 0.79 overall, with individual success rates of up to 0.69 to 0.92.

♦ The accuracy of the GEM model screen temperatures is very crucial to increase hit rates. The comparisons between Tso and Tsc indicate that differences between them can be up to 2°C.

♦ It is possible that using criteria of RHw>80% based on model can result in higher occurrences of fog conditions and


46

needs to be improved if no observations at the surface are made.

♦The GOES satellite algorithm based on only channel differencing cannot work accurately when an elevated fog layer (cloud) occurs. In this case, high-resolution model runs should be integrated into a decision making system

At times when fog can be properly deduced from satellite and model based data, microphysical parameters, including liquid water path (LWP), mean droplet diameter (Dm), and visibility, can also be determined using traditional look up tables that uses radiative transfer equations. Satellite data can also be used to derive such fog-related information using parameterizations that have been developed using aircraft measurements (Gultepe and Isaac 2004, Gultepe 2005).

Acknowledgments Funding for this work was provided by the Canadian

National Search and Rescue Secretariat and Environment Canada. Some additional funding was also provided by the European COST-722 fog initiative project office. Technical support for the data collection was provided by Cloud Physics and Severe Weather Research Section of Science and Technology Branch, Environment Canada, Toronto, Ontario. Author thanks to G. Isaac, S. Cober, and J. Reid of Environment Canada for discussions related to the data analysis. Authors are also thankful to Milbrandt and S. Belair of RPN, Environment Canada, Montreal, Quebec, and Dr. W. Jacobs, Langen, Germany, for discussions during the course of this work.

References

Bélair S., Crevier L.-P., Mailhot J., Bilodeau B., Delage Y., 2003: Operational implementation of the ISBA land surface scheme in the Canadian regional weather forecast model. Part I: Warm season results. J. Hydromet., 4, 352-370.


47

Bendix J., 2002 : A satellite-based climatology of fog and low-level stratus in Germany and adjacent areas. Atmos. Res., 64, 3-18.

Benjamin S. Dezsö Dévényi G., Weygandt S.S., Brundage K.J. , Brown J.M. , Grell G.A. , Kim D., Schwartz B.E. , Smirnova T.G. , Smith T.L., Manikin G.S., 2004: An Hourly Assimilation–Forecast Cycle: The RUC. Mon. Wea. Rev., 132, 495–518.

Bott A., Trautmann T., 2002: PAFOG-a new efficient forecast model of radiation fog and low-level stratiform clouds. Atmos. Res., 64, 191-203.

Côté J., Desmarais J.G., Gravel S., Méthot A., Patoine A., Roch M., Staniforth A., 1998: The operational CMC-MRB Global Environmental Multiscale (GEM) Model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 1373-1395.

Ellrod G. P., 1995: Advances in the detection and analysis of fog at night using GOES multi-spectral infrared imagery. Weather and Forecasting, 10, 606-619.

Gultepe I., Isaac G.A., 2004: An analysis of cloud droplet number concentration (Nd) for climate studies: Emphasis on constant Nd. Q. J. Royal Met. Soc., 130, Part A, No. 602, 2377.

Gultepe M. Müller D., Boybeyi Z., 2006: A new warm fog parameterization scheme for numerical weather prediction models. J. Appl. Meteor., Accepted.

Gultepe I., Cober S., Isaac G.A., King P., Reid J., Gordon M., Taylor P., Hudak D., Rodriguez P., 2005: Canadian Testbeds and Observations During The FRAM Fog Field Project. COST722 Workshop on fog, Larnaka, Cyprus, May 15-17, 2005. 8 pp.

Gultepe I., 2005: A new warm fog microphysical parameterization developed using aircraft observations. COST722 Workshop on fog, Larnaka, Cyprus, May 15-17, 2005. 10 pp.


48

Kunkel B. A., 1984: Parameterization of droplet terminal velocity and extinction coefficient in fog models. J. Appl. Meteor., 23, 34-41.

Lee T. F., Turk F. J., Richardson K., 1997: Stratus and Fog Products Using GOES-8–9 3.9-μm Data, Weather and Forecasting, 12, 664–677.

MANOBS Seventh Edition, 1977: Atmospheric Environment Service. Downsview, Ont. Canada, M4H5T4.

Mailhot J., Bélair S., Lefaivre L., Bilodeau B., Desgagné M., Girard C., Glazer A., Méthot A., Patoine A., Plante A., Talbot D., Tremblay A., Vaillancourt P., Zadra A., 2005: The 15-km version of the Canadian regional forecast system. Accepted for publication in Atmos.-Ocean.

Pagowski M., Gultepe I., King P., 2004: Analysis and Modeling of an Extremely Dense Fog Event in Southern Ontario. J. of Appl. Meteor., 43, 3–16.

Saseendran S.A., Singh S.V. , Rathore L.S., Das S., 2002: Characterization of Weekly Cumulative Rainfall Forecasts over Meteorological Subdivisions of India Using a GCM. Weather and Forecasting, 17, 832–844.

Schmit T.J., Gunshor M.M. , Menzel W.P, Gurka J.J. , Li J., Bachmeier A.S., 2005: Introducing the next-generation advanced baseline imager on GOES-R. Bull. of the Amer. Met. Soc., 86, 1079–1096.

Smirnova T.G., Benjamin S.G., Brown J.M., 2000: Case study verification of RUC/MAPS fog and visibility forecasts. Preprints, 9th Conf. on Aviation, Range, and Aerospace Meteorology, AMS Orlando, FL., Paper # 2.3. 6 pp.

Stoelinga M.T., Warner T.T., 1999: Nonhydrostatic, Mesobeta-scale model simulations of cloud ceiling and visibility for an east coast winter precipitation event. J. Appl. Meteor., 38, 385-404.


49

Tapp R.G., Woodcock F., Mills G.A., 1986: The Application of Model Output Statistics to Precipitation Prediction in Australia. Monthly Weather Review, 114, 50–61.

Whiffen B., 2001: Fog: Impact on aviation and goals for meteorological prediction. 2nd Conference on Fog and Fog Collection, St. John’s Canada, 15-20 July 2001. 525-528.

Woodcook F., 1976: The Evaluation of Yes/No Forecasts for Scientific and Administrative Purpose. Monthly Weather Review, 104, 1209–1214.


50

Fog / Low Stratus Detection and Discrimination Using Satellite Data

J. Cermak and J. Bendix

Laboratory for Climatology and Remote Sensing (LCRS) University of Marburg

Germany

1. Motivation and Context

1.1. Background Fog is a meteorological phenomenon with distinct

spatio-temporal ramifications. The interpolation of point measurements of cloud cover and visibility entails significant uncertainties; satellite retrievals of fog and low stratus presence on the other hand are a useful alternative.

In the past, methods for fog detection have been developed and implemented on polar orbiting systems such as Terra / Aqua MODIS and NOAA AVHRR at the Laboratory for Climatology and Remote Sensing (LCRS). Until recently, geostationary satellite sensors did not have the spectral and spatial potential for fog detection. With the advent of Spinning Enhanced Visible and Infrared Imager (SEVIRI) aboard the Meteosat 8 satellite, operational fog detection and monitoring have become possible. Meteosat 8 is the first of a total three planned Meteosat Second Generation (MSG) satellites and has been in operation since 2003. The SEVIRI has a total 12 spectral channels; 11 of these have a sampling distance of 3 km at nadir and cover visible and infrared wavelengths. The high resolution visible (HRV) channel reaches a spatial resolution of 1 km at sub-satellite point. The system's repeat rate for full disk scans is 15 minutes (Schmetz et al. 2002).

1.2. Operational Framework MSG SEVIRI data is received at the Marburg Satellite

Station (MSS) along with other satellite systems (Figure 1). It is then operationally processed in the FMET framework. FMET


51

includes image calibration, geolocation and subsetting, and also provides a framework for the operational creation of level 2 products (such as the fog / low stratus mask). Calibrated data and products for a European domain are presented on the LCRS web page and archived.

1.3. Concept and Approach The delineation of fog and low stratus is addressed as a

two-step problem: In a first step, low stratiform clouds need to be separated from other surfaces, in the 2D perspective innate to satellite imagery. Once this is accomplished, some information on cloud geometry is needed in order to establish whether or not ground visibility impairment is likely in a given location. Figure 2 visualizes this problem.

2. Low Stratus Delineation Satellite-based fog detection has been addressed in a

wide range of papers. Most of these deal with night-time fog delineation. Most commonly, fog is subsumed with low stratiform clouds, not considering whether or not visibility is impaired at ground level. In this paper, 'fog / low stratus' will be used to address both features. A common approach to nighttime fog / low stratus detection makes use of the emissivity differences the small droplets in fog and low stratus display in the thermal (10.8 µm) and middle infrared (3.9 µm). This behaviour distinguishes fog and low stratus from clouds with larger droplets, so that they can easily be identified. This technique has been implemented on a number of platforms by various authors (e.g. Eyre et al. 1984, Bendix and Bachmann 1991, Ellrod 1995, Dybbroe 1993, Bendix 2002). More recently, the same algorithm was successfully implemented on MSG (Cermak et al. 2004).

Daytime fog detection is more critical, as a solar component mixes into the 3.9 µm signal and the IR emissivity differences used at nighttime can no longer be established. A newly developed scheme therefore applies a combination of spectral and spatial tests for assumed fog / low stratus properties to each satellite scene. This scheme is summarized


52

in Figure 3 and described in detail in Cermak and Bendix (forthcoming).

Figure 1. Marburg Satellite Station. The figure gives an overview of the satellite and ground-based systems in operation.

Figure 2. The concept of ground fog retrieval from satellite imagery. In order to establish whether or not a cloud is likely to impair visibility in a given location, information on cloud and surface geometry is required.


53

Figure 3. Overview of the low stratus detection scheme. A series of tests is applied sequentially; any unsuited pixels are rejected in the process.

Figure 4. A histogram of brightness temperature differences T10.8-T3.9 for the MSG scene 1100 UTC, 5 November 2003. Locations of the clear and cloudy peaks as well as the threshold are indicated. Figure from CITATION.


54

The basic properties assumed of fog / low stratus in the algorithm are

• cloud • water phase • small droplets • low altitude • homogeneous cloud top surface ('stratiform')

Initially, cloud-contaminated pixels are separated from clear pixels in a gross cloud check. This test makes use of the 10.8 and 3.9 µm channels. At daytime, the latter contains a solar component, as stated above. Clouds have a larger reflectance at this wavelength, so that the radiance difference R(10.8) – R(3.9) is considerable smaller for clouds than for land. A threshold is extracted automatically from the histogram of radiance temperature differences (cf. Figure 4).

Following on the gross cloud check are snow exclusion and water phase determination tests, using SEVIRI channels at 0.6, 1.6, 0.8, 10.8, 8.7 and 12.0 µm. The water clouds left are then tested for the small droplet presence. This test again uses the 3.9 µm radiances. Small droplets are more reflective at this wavelength than larger drops (Figure 5), so that small droplet cloud pixels can be singled out.

Cloud altitude and homogeneity are tested on the basis of cloud clusters rather than individual pixels. Spatially coherent and discrete entities of cloudy pixels are identified in the given scene. For each of these entities, a rough estimate as to whether the cloud cluster could be low is made based on surface elevation and temperature. For stratiformity, the variation of cloud top temperature within each entity is tested against a threshold value.

3. Ground Fog Detection After the identification of fog and low stratus areas, the

pixels where the cloud is likely to impair ground visibility need to be determined. This decision requires some knowledge of cloud geometry: ground contact is assumed where zb ≤ zs, with


55

zs height of the ground surface below the cloud and zb the computed cloud base height. The latter is obtained from

zb = zt – Δz with zt cloud top height and Δz cloud geometrical thickness.

Figure 5. The dependence of VIS and MIR reflectance on droplet effective radius and cloud optical depth.

Cloud top height is computed on the basis of the cloud

entities introduced above. Where the cloud is limited by terrain (tested on the basis of a digital elevation model), cloud margin height is extracted from the same. Otherwise, an assumed environmental temperature lapse rate is applied. Margin height is then interpolated into the cloud surface.

Cloud geometrical thickness is obtained from radiative transfer. Photon penetration depth into the cloud varies with wavelength in the middle infrared. Based on this feature, a combination of multiple wavelengths in this range can be utilized to extract vertical cloud information. Lookup tables computed in radiative transfer simulations are fitted with observed radiances in several wavelengths and cloud


56

geometrical thickness obtained in this way. The method is described in full in Bendix et al. (2005).

4. Validation / Outlook Since no comparable and reliable spatial data set exists

for validation, point measurements have been used to this end. Station measurements of cloud height and ground visibility provided by the Deutscher Wetterdienst (DWD) were compared to satellite classification results in contingency tests. For low cloud detection, a false alarm ratio of 0.021 was reached, which means that almost all cases identified as low cloud by the satellite algorithm are indeed covered by low clouds. The probability of detection was 0.701, indicating that 30% of low cloud cases were missed by the satellite algorithm in the cases considered. Two main problems were identified to cause these misses: cloud layer overlap and sub-pixel effects. Where a higher level cloud covers a low cloud situation, the latter cannot be detected by the satellite sensor. So, after exclusion of all such situations, the probability of detection rises to 0.834. The remaining misses are mostly sub-pixel clouds (Cermak and Bendix, forthcoming, Cermak et al. 2006).

The ground fog detection scheme has been validated in its MODIS implementation and compared to other cloud thickness retrieval algorithms. A quantitative evaluation based on SYNOP data yields a higher probability of detection for ground fog with the new scheme as compared to other published approaches. The false alarm ratio was reduced as well (Bendix et al. 2005).

The scheme is implemented in operational processing of satellite data received at LCRS and constantly being scrutinized. Operational products are available on the LCRS homepage at http://www.lcrs.de.

For the future it is planned to look at pathways for integrating satellite data with ground measurements and models for product generation. Also, with the new ground truth station to be established at LCRS, fog vertical profiles and the role of microphysical parameters will be investigated for further algorithm improvement.


57

Acknowledgements This research is generously funded by the German

Research Council DFG (BE 1780/8-1; 8-3) within the NEKAMM project. The study is part of the ESF COST action 722 (Short Range Forecasting Methods of Fog, Visibility and Low Clouds) and the ESA/EUMETSAT MSG RAO PI programme (ID 141). The authors wish to thank the DWD (P. Bissolli) for the provision of validation data.

References

Bendix J., Bachmann M., 1991: Ein operationell einsetzbares Verfahren zur Nebelerkennung auf der Basis von AVHRR-Daten der NOAA-Satelliten. Meteorol. Rdsch., 43, 169-178.

Bendix J., 2002: A satellite-based climatology of fog and low-level stratus in Germany and adjacent areas. Atmos. Res., 64, 3–18.

Bendix J., Thies B., Cermak J., Nauss T., 2005: Ground fog detection from space based on MODIS daytime data - A feasibility study. Weath. Forecasting, 20, 989-1005.

Cermak J., Thies B., Bendix J., 2004: A new approach to fog detection using SEVIRI and MODIS data. Proc. 2004 EUMETSAT Met. Sat. Conf.. Darmstadt, 130–136.

Cermak J., Schneebeli M., Nowak D., Vuilleumier L., Bendix J., 2006: Characterization of low clouds with satellite and ground-based remote sensing systems. Meteorol. Z., accepted.

Cermak J., Bendix J., A novel approach to fog / low stratus detection using Meteosat 8 data. Submitted to Atmos. Res.

Dybbroe A., 1993: Automatic detection of fog at night using AVHRR data. Proc. 6th AVHRR Data Users’ Meeting, 254–252.

Ellrod G. P., 1995: Advances in the detection and analysis of fog at night using GOES multispectral infrared imagery. Weath. Forecasting, 10, 606–619.


58

Eyre J. R., Brownscombe J. L., Allam R. J., 1984: Detection of fog at night using Advanced Very High Resolution Radiometer. Meteorol. Mag. 113, 266–271.

Schmetz J., Pili P., Tjemkes S., Just D., Kerkmann J., Rota S., Ratier A., 2002: An introduction to Meteosat Second Generation (MSG). Bull. Am. Meteorol. Soc., 83, 977–992.


59

Discrimination between fog and low clouds using a combination of satellite data and

ground observations

Vincent Guidard and Diane Tzanos Météo-France, Direction des Systèmes d’Observation, Toulouse,

France

Many users need a spatialized information on fog occurrence (needs of local information, typically airports, are well satisfied by human and instrumental observation). In order to elaborate such a product, we decided to rely on a satellite-based detection of low clouds, which is an information available over the whole domain of interest (France in our case). Then, we apply an algorithm which uses spatialized ground observations. Eventually the product is evaluated with a one-year set of data.

Low clouds are identified thanks to the Cloud Type product of the SAF-Nowcasting of EUMETSAT, which is generated from MeteoSat8 data (Le Gléau and Derrien 2000). The cloud classes in which fog is expected to occur are “low cloud” and “very low cloud”. This product has a resolution of about 3 km x 6 km over France. Nevertheless, the product is projected using a polar stereographic projection, leading to a grid size of about 3 km. In order to avoid non-detections due to the spatial discretization of the cloud type, the occurrence of low or very cloud is identified on a group of 3 x 3 of those new pixels centered on the point of interest. From a one-year set of hourly data, we can evaluate that 66% of fog occurrences happen under low or very low clouds, 19% under higher clouds and 14% of fog events occur in satellite clear sky areas. The latter percentage is especially due to the non-detection of low clouds at twilight (dawn, in most of cases), because of the contamination of the 3.9 µm channel by the visible radiation.

As not all low and very low clouds correspond to fog, we have to use ground observation in order to discriminate between low clouds and fog. The parameters which are used in the described study are: relative humidity at 2 m, wind strength


60

at 10 m and precipitation. For relative humidity and wind strength, we use hourly analyses routinely produced at Météo-France. These analyses are based on an optimal interpolation scheme and they are specifically designed to fit the observations (Calas 1997), thus they are not suitable for NWP model initialisation. Radar reflectivities are used as precipitation analyses.

Figure 1. Risk of fog analysis, valid at 12 UTC on 03 December

2004. Observed visibilities in km: less than 1 km are plotted in red, between 1 and 5 km in cyan, and greater than 5 km in black.

The algorithm is rather simple. First, we evaluate

whether fog occurrence is possible or not. Fog is expected not to occur if one of the following criteria is fulfilled: relative humidity is below 90%, or the cloud type product indicates a clear sky area, or the wind strength is greater than 7 m/s, or precipitation are higher than 0.2 mm/h. Then, in regions where fog is possible, if the cloud class is “low” or “very low”, we estimate a level of risk of fog occurrence. Thresholds on relative humidity lead to 3 levels of risk: low-level if humidity is between 90% and 95%, medium-level if humidity is between


61

95% and 97%, and high-level if humidity is greater than 97%. If the cloud class indicates higher clouds, then the risk is not estimated. An example of a risk of fog analysis is given on Figure 1.

The evaluation of these levels of risk is performed versus a one-year set of hourly SYNOP data, from March 2004 to February 2005. Many objective scores have been computed, but only the values for the potential of detection (POD) and the false alarm rate (FAR) are discussed here (Table 1). The product is to be evaluated from two points of view: the discrimination between low clouds and fog, and the risk of fog analysis. There is a high POD (93%) for the discrimination, whereas the FAR is not negligible (63%). The potential of detection for the whole analysis is smaller (73%), according to the satellite non-detections, and the FAR is obviously the same. Table 1. Objective Scores. POD FAR

Discrimination between fog and low clouds Humidity threshold of 90 % 0.93 0.87 Humidity threshold of 97 % 0.57 0.63

Risk of fog analysis Humidity threshold of 90 % 0.73 0.87 Humidity threshold of 97 % 0.41 0.63

Case studies show that these analyses give a relevant

information on the horizontal extension of the fog area. The three levels of risk provide useful hints on the severity and the evolution of the fog events. Nevertheless, areas where the risk is not estimated may be a limit of this product.

To conclude, these analyses are helpful and they are a big step forward, as nothing similar was available at Météo-France before. Some improvements are expected after a period of evaluation by French nowcasters during the winter 2005-2006. The next step will be the implementation of a quantification of the level of horizontal visibility within areas where the fog is possible.


62

References

Calas D., 1997: Apport d'un analyse à mésoéchelle pour la prévision immediate des systèmes convectifs. Ph. D. dissertation, Université Paul Sabatier, 240 pp. [Available from Université Paul Sabatier, 118 route de Narbonne, 31062 Toulouse cedex, France].

Le Gléau H., Derrien M., 2000: Nowcasting and very short range forecasting SAF. Prototype scientific description, CMS Météo-France. [Available from http://www.meteorologie.eu.org/safnwc].

Short range forecasting methods of fog, visibility and low clouds

Documents

Transcript of Short range forecasting methods of fog, visibility and low clouds