Abstract

Daily, KNMI processes a huge amount of heterogeneous spatial data involving models, observations, radar and - increasingly - satellite imagery. In order to link up with internal and external user needs, existing tools appear to be too dedicated and, hence, fail to allow for flexible data integration, exchange and visualisation. When KNMI started its METEOGIS project in 1997, experience with GIS at meteorological offices was limited. Since then, a number of studies has been initiated to demonstrate the value of GIS technology for the KNMI. Besides the introduction and implementation of GIS software in the existing technological infrastructure, the METEOGIS project has focused on e.g. the conversion of data stored in formats prescribed by the World Meteorological Organization (WMO) and integration (a fog sensitivity analysis).

Background

Founded in 1854, the Royal Netherlands Meteorological Institute (KNMI) is the national expert centre for weather, climate and seismology. Besides providing daily weather forecasts to the public and professional users (e.g. aviation and shipping), KNMI unfolds initiatives for (international) research in the above-mentioned areas of interest. The huge amount of data that has to be processed each day to meet the operational and scientific requirements is impressive, and acts as an incentive for the continuous improvement of the data infrastructure. The Observations & Modelling department is the "data pillar" of the institute and is, as such, responsible for the acquisition, processing, validation,storage and distribution of meteorological data. These data can be ordered in the following categories:

• Operational observations are acquired at ground level by observing stations equipped with instruments such as rain gauges or thermometers as well as in the upper-air by means of radiosondes. Moreover, weather radar provides information about rainfall whereas satellites add observations made over an extended area of the earth's atmosphere. Observations are important to assess the current state of the atmosphere (the "initial conditions") and are subsequently assimilated in the numerical models that are used to predict the weather. By means of assimilation, a meteorologist pursues a model representation that has to be consistent with the considered observations.
• Satellite data, which are in fact observations as well, deserve a separate treatment, partly because of their contribution to the understanding of long-term climatological processes (e.g. the hole in the ozone layer). The amount of meteorological satellite observations is increasing rapidly and their role as input to the numerical weather prediction systems is gaining weight accordingly. In addition, visual inspection of satellite images contributes to the analysis of the current state of the atmosphere - usually by simultaneously interpreting charts, satellite images and weather radar images. Satellite meteorology is becoming even more important in the near future, judging by the intended launch of a new generation of Eumetsat's Meteosat satellite by the year 2000. Their equipment will produce a data flow of one image per 15 minutes, recorded in 12 spectral channels!
• Model data are the result of solving a set of equations describing the physical processes in the atmosphere, starting from observations defining the initial conditions at a certain time. At KNMI, the HIRLAM model (High Resolution Limited Area Model) is used to forecast atmospheric conditions up to 48 hours.
• Climatological observations can be described as "historical" data because they refer to non-topical meteorological measurements, e.g. monthly weather reports.

A dedicated data and computer system infrastructure safeguards the operational tasks, i.e. the continuous monitoring of the weather pattern. Most of the data used in these processes are stored in binary formats such as BUFR (Binary Universal Form for the Representation of Meteorological Data) and GRIB (Gridded Binary), in accordance with the guidelines issued by the World Meteorological Organisation (WMO) .

The heterogeneousness of the described data sources and the desire to combine, analyse, visualise and otherwise manipulate the diverse data - beyond operational applications - has aroused the interest in GIS. Meteorological data are in essence spatio-temporal but this fact notwithstanding, the majority of meteorological institutes hasn't adopted GIS as a main part of their data processing infrastructure. The reasons why KNMI decided to invest in these tools are multiple and can be summarized as follows: to

• facilitate the integration of meteorological and other spatial data, e.g. combining up-to-date land cover and height maps with wind flow models;
• allow geometric transformations in order to project data sets easily according to the varied needs of users;
• improve and increase the exchange of meteorological data sets by offering possibilities to directly convert from WMO file formats to geographic standard formats;
• complement present visualisation packages (e.g. PV Wave) by providing easily accessible cartographic tools that enable quick but well-considered presentations;
• link up with existing applications and development tools by presenting a high level of "openness", for example expressed by ODE (Open Development Environment) enabled GIS software;
• introduce innovative techniques to approach the dynamics of the meteorological realm as close as possible, embodied by the idea that GIS will eventually handle all kinds of spatio-temporal data in the short term.

The ideas have taken shape with the start of the Meteogis project in April 1997. Over a period of 2 years, efforts are directed to the embedding of GIS technology in the existing IT-infrastructure. Being halfway through the project, this paper will report on the first results of a case study. In addition, it will clearify the considerations that have played a decisive role during the introduction of GIS "from scratch".

The Meteogis project

In view of the presence of large amounts of spatio-temporal data, the absence of commercial GIS packages at meteorological institutes is striking. From an exploratory study, preceding the introduction of GIS at KNMI, it appeared that only a small group of national services in Europe had acquainted themselves with this particular software (e.g. Météo France , DNMI from Norway).

Two important short comments can be given on this fact. In the first place, climatological research institutes are known to be more familiar with GIS than operational services - the exploration of the "historical" databases lends itself particularly well for a geographic approach. Secondly, there exist numerous "GIS-like" applications, especially developed for meteorological purposes and establishing seamless links with current operational tools by the ability to process WMO data formats directly. The ECMWF for example (European Centre for Medium-Range Weather Forecasts) distributes its Metview software freely among its member institutes, including KNMI. The advantages of such dedicated software notwithstanding, there remains a need for indeed less customized but therefore more flexible and open GIS tools, as pointed out in the previous section. (Another example of dedicated meteorological software is Zebra).

In the framework of the Meteogis project, a user requirement analysis has been conducted. Five different classes of requirements were taken into account, to wit:

1. Functional. The contents of the GIS "tool box" should be well-defined;
2. Organizational. The organization must rely on sufficient and adequately trained GIS staff while the gradual implementation of the technology should anticipate the interaction with existing data flows. Furthermore, the compatibility with the external GIS community is embraced as a principle (no exotic and poorly supported package!);
3. Technical - organizational. Embedding GIS in the existing IT-infrastructure means adopting stiff requirements with respect to platforms, operating systems, database connectivity, security, networking and metadata handling;
4. Technical - operational. The configuration of the GIS environment is decisive for its success in a particular organization. Moreover, limited processing time, userfriendliness and openness encourage the acceptance of GIS as a supporting tool.
5. Institutional. As always when introducing new technology, success can only be achieved if an organization is prepared to invest a critical amount of money. But perhaps more important is the willingness to provide "back-up" in order to help germinate the GIS seeds.

After an extended weighing of all the contributed criteria and an even more elaborated study of commercial GIS packages, KNMI decided to buy Esri and Erdas products. As a start, both for Unix and Windows95, 2 licenses Arcview, including 1 Spatial Analyst, were purchased - together with one license ArcInfo (with GRID and TIN). After an implementation and training stage, involving a limited but enthusiastic group of "pioneers", the Meteogis project started with its first case study to inform interested persons on the possibilities of GIS. Because the results of such a "pilot" are far-reaching as far as the receptivity of potential users is concerned, a topical and meteorologically interesting subject has been selected: the fog sensitivity of highway sections, directly relating to severe traffic accidents.

Case study: a fog sensitivity analysis for western North-Brabant

Objectives

The case study focuses on the derivation of a fog sensitivity map, thereby illustrating the potential of GIS when dealing with meteorological and climatological issues. The usefulness of these tools may be beyond dispute in geo-disciplines, in meteorology it is only starting to find its niche. As an example, consider the fact that a search on the Esri Internet site resulted in only slightly more then 10 hits! Eventually, the fog project could arouse the interest of different authorities in order to start a study to identify possible risky road sections on a more structural basis. The application of such information in the development of well-balanced fog detection and traffic control systems is easily imaginable, given the fact that dangerous and suddenly looming patches of fog are held responsible for tragic accidents involving many road casualties .

Study area

The area on which the case study is focused is part of the province of North-Brabant (figure 1). A closer look is taken at the surroundings of Breda, because of a serious and tragic accident in 1990, as a direct consequence of treacherous fog reducing visibility values dramatically. Since then, visibility sensors have been installed along that particular road section thus providing valuable reference data for the case study.

Fog

Fog often manifests itself as a "cloud near the earth's surface", thereby seriously hampering visibility values (less than 1000 meter) by the presence of small drops of water in the atmosphere (figure 2). Meteorologists make a distinction between several types of fog: advection fog, frontal fog, hill fog and radiation fog, for example (Musk, 1988). In this particular paper, the latter is considered because it is held responsible for the suddenly looming and dense fog banks in which motorists are caught (mostly between October and March) causing severe traffic accidents.

Radiation fog is caused by an interaction of different processes:

• the nights preceding misty mornings are calm and practically cloudless;
• as a consequence, the earth's surface can emit its heat thereby causing a considerable drop in temperature near the ground;
• a process of condensation occurs because moist, slightly over-saturated air near the surface is cooling down to dew-point temperature;
• the resulting patches of fog could be very local and difficult to anticipate.

The above description of the development of radiation fog is simplified. In fact, the process of emitting long-wave radiation from ground to atmosphere (loss of heat) is complicated and affected by several factors. Especially the very first steps in the development of radiation fog are hardly understood (Wessels, 1993). In general, the most favourable meteorological conditions for the occurrence of radiation fog can be summarized as follows, according to Musk (1988): clear night sky (loss of radiation), moist air at sunset (automns and winter, after rain or near open water) and a weak wind (vertical mixing of misty layer to spread cooling upwards).

Nowadays it has become clear that not only meteorological factors play a role in the development of radiation fog. The availability of sufficient moist near the surface is dependent on the climatological history of an area (e.g. the amount of rainfall during the last 24 hours), but the characteristics of the soil (e.g. sandy or clay-soil) and the ability of vegetation to hold moisture are decisive as well. The cooling of the earth's surface is determined by the coverage of the soil and the soil type itself, although Wessels (1993) correctly states that this factor is less leading because the emission is stronger above sandy (and thus relatively dry!) grounds.

The assessment of fog sensitivity for the study area has adopted two different methodologies, to wit the soil moisture approach and the Fog Potential Index (Dixon, 1989; Musk, 1988). Both methods consider non-meteorological factors, such as soil type and topography, and are in general well-documented. For the considered case study, concepts have been translated in GIS fuctionality such that an arbitrary road section can be subjected to either of two approaches. It is emphasized that relating actual meteorological parameters with physiographic characteristics in a GIS environment is beyond the scope of this paper, although it is perceived as a logical follow-up in the research project.

Methods

The derivation of a fog sensitivity map is achieved according to the following two approaches:

1. Soil Moisture;
2. Fog Potential Index (FPI).

The former results in absolute estimates of fog sensitivity, expressed as the possibility that visibility values decrease to less than 200 meter. The latter method compares the differences between road sections by assigning an index number between 0 and 100, with higher values corresponding to potentially more (fog-)sensitive locations. Terpstra et al. (1997) note that high values do not necessarily mean risky road locations as this may depend also on road conditions (windings, congestion...).

GIS tools

For the fog sensitivity analysis, both Arcview and ArcInfo are used. The latter has been applied mainly for preprocessing tasks (e.g. joining and editing map sheets) whereas Arcview has been selected as the environment in which the actual demonstration has been developed. Obviously, the learning curve of Arcview is far more steeper than ArcInfo's (remember that KNMI started from scratch!) - an advantage for the newly entered GIS operators who had to work under stiff time limitations! The Spatial Analyst has proved to be of the utmost importance for KNMI-applications, especially when extending its functionality with Avenue scripts.

Approach 1: soil moisture

Originally, soil type is taken as a measure for the amount of moisture that is present near the surface. As stated previously, soil type is important as it is related to the humidity of the surface and the capacity of that surface to radiate heat - and thus contributes to the prerequisites for the development of radiation fog. In this study, however, soil type is only indirectly applied - here more objective ground water levels (as indicated on the soil map of the Netherlands 1:50000) are choosen as a starting point (see figure 3). Table 1 shows the classification of ground water levels that has been adopted during the study. Four soil moisture classes have been discerned, after discussions with meteorologists of KNMI.

In theory, the class values can be inserted in a formula with which permillage fog is derived, that is, the permillage of time that is characterized by a visibility of less than 200 meters (dense fog) :

The above formula is the result of a regression (Wessels, 1993), based on visibility measurements recorded by fog detectors along the A-16, a highway intersecting the study area. The fog sensitivity map resulting from the soil moisture method is not converted to these quantitative classes, however, because it is felt that more statistical material (prolonged visibility measurements along highway sections) are needed for a reliable estimation. Moreover, the soil mositure classes which are applied in the regression are based on soil types, not on ground water levels as is the case in the present case study! The soil moisture map as derived from the ground water level map is shown in figure 4.

The resulting map is shown in figure 5. Some topography is added for reasons of orientation. The highways are shown in a colour that corresponds with a moisture class that is representative for the direct surroundings of these roads. Moist spots are noticed there where the highways cross meadow-land whereas dry sections are identified near the dry sandy grounds. The general impression of the map complies with the existing ideas based on extended field-work. As a reference (but given the limitations mentioned earlier) a number of mean annual fog detector observations are provided (as a promille). It becomes clear that in the open area northward and southward of the city of Breda, higher values are more likely to occur. But again, it is hardly possibly to draw any conclusions from these data! At the same time it is admitted that there are probably more criteria to reckon with in order to obtain a more reliable image. Consider for example the proximity of factories emitting large quantities of water vapour or the presence of a pine forest that more or less dissolves the patches of fog!

Approach 2: Fog Potential Index

Musk (1988) describes a method to index the susceptibility of a particular area to thick radiation fog. He expresses this Fog Potential Index as a function of four predictors:

with

d_w = a measure for the vicinity of water (brooks, rivers, lakes) and the extensiveness of
these water surfaces;
t_p = a derivative of local topography - height, slopes, curvature of the terrain;
s_p = a function of the position of the road itself (height, e.g. an embankment);
e_p = a factor describing environmental characteristics that either help or hinder the development of radiation fog (e.g. the proximity of urban areas and industry).

Musk (1988) has applied his method for study areas in Great-Britain and the exact weighing of the above-mentioned parameters is still not completely understood. Weights of 10, 10, 2 and 3 respectively are used in the original study, while the value of the parameters themselves varies between 0 and 4. The literature fails to provide rules for the assignment of these 5 classes to each of the predictors; without sufficient clues, comparisons between different FPI-studies remain hazardous operations! In consultation with experts of KNMI, the following grouping has been assessed (table 2). Notice that we have initially distinguished between small and large water surfaces - eventually, the value for d_w is determined by picking the maximum predictor value for either d_{w small} or d_{w large}.

From table 2, a number of observations can be made. In the first place, a higher value for the predictor means a higher potential contribution to the development of radiation fog. From this, it becomes clear that the environmental characteristics only consider the vicinity of urban areas and forests ("dissolving" fog) and exclude the presence of steaming industrial complexes (factories that are seriously producing water vapour are not identified in the study area). Next, information about the position of the road, relative to its surroundings, is lacking completely. The results of the present study will hopefully lead to a situation in which these valuable data (administered by the Dutch Directorate General of Public Works and Water Management - Rijkswaterstaat) come at the disposal of meteorological and climatological institutes. Finally, a remark can be made with respect to t_p. The susceptibility for radiation fog is severe in those areas that are concave and characterized by a certain minimum slope angle. These values appear to be critical for cold air masses that descend from higher spots and subsequently flow to lower situated road sections.

Figures 6 to 11 give an visual impression of the predictors. As can be seen from figure 7, the influence of the terrain is negligible - at least when using height data at the considered resolution! Only the red spots would contribute to the susceptibility for radiation fog, and they are at a considerable distance of highways.The height map (figure 6) is obtained from interpolating point data to a 50 meter grid (using ArcInfo, thereby reducing CPU time considerably), but more detailed data would possibly reveal interesting micro-rellief.

The vicinity of water, especially lakes, rivers, brooks and ditches seems to be of more importance for the fog susceptibility of the study area. It is interesting to note that small but characteristic ponds, e.g. as digged in the neighbourhoud of cloverleafs, are emerging in a very pronounced way from the map (figure 8, small and dark spots).

The environmental predictor (figure 9) can be interpreted according to the same legend as the previous one. Clearly, the proximity of urban area has a counteracting ("positive") effect on the sensitivity of the area with respect to radiation fog. Here, values are low and thus hardly contribute to a high index number.

The resulting FPI-map, summarizing all three predictors (remember that the road topography could not be evaluated, unfortunately) and weighing all contributions according to the earlier mentioned weights, shows a pattern that reveals a high correlation with the "water map" (figure 10).

References

Dixon, J.C. (1989): Current techniques for assessing (indirectly) the localized incidence of fog on roads. Meteorological Magazine, 118. pp. 99-104.

Musk, L.F. (1988): The assessment of local fog climatology for new motorway and major road schemes. Proceedings of the International Conference on Meteorology and road Safety. Florence, November 1988. pp. 779-797.

Terpstra, J.M., D. Blaauboer & P.A.T. van Es (1997): Mistonderzoek A-58 ten zuiden van Breda (in Dutch).

Wessels, H.R.A. (1993): Meteorologische evaluatie van de zichtmetingen langs de A-16. KNMI Technische Rapporten: TR-157 (in Dutch).


About the Author...

Frans J.M. van der Wel
Royal Netherlands Meteorological Institute
Observations & Modelling Department - Satellite Data Division
P.O. Box 201
3730 AE De Bilt
The Netherlands
Telephone: +31 30 2206822
Fax: +31 30 2210407
welvdf@knmi.nl