MetGIS™ is an innovative Java-based, combined Meteorological and Geographic Information System, with a specific emphasis on snow and mountain weather. This constantly upgraded prediction scheme has been developed within the framework of a number of interdisciplinary international research projects. A principal focus of the system is the automated production of high-resolution, downscaled forecast maps of meteorological parameters such as precipitation, fresh snow amounts, the snow limit, the form of precipitation, wind and air temperature.
The geographic part of the system includes topographies relying on data bases such as SRTM (Shuttle Radar Topographic Mission) and representations of roads, rivers, railway lines, political borders and cities. On top of these, partly linked to terrain features, down-scaled meteorological information can be visualized in a variety of display styles. Meteorological forecast data of any numerical model with common output data formats can be used as a starting point for the downscaling procedures. Currently, the real-time output of the GFS (Global Forecast System of the US National Weather Service) is used as a base for MetGIS™ forecasts. Verification results are quite encouraging so far. Mean absolute errors are in the range of 1.3–3 °C for 36 h temperature forecasts, and around 80% of the 24 h forecasts predicted correctly, if the precipitation will be below or above 1 mm.
High-resolution precise meteorological forecasts are of great importance for all people and institutions whose activities depend on mountain weather. The base of a modern weather forecast is the output of a global Numerical Weather Prediction (NWP) model, for example the Integrated Forecast System (IFS, ECMWF, 2008) operated at the European Centre for Medium Range Weather Forecasts (ECMWF), or the Global Forecast System (GFS, Environmental Modelling Center, 2003), provided by the National Center for Environmental Prediction (NCEP, USA). The results of the computations of these global models (and of higher-resolution, derived models) can be used by meteorologists to forecast the weather and are often viewed as contour plots of meteorological fields projected onto a map, with no relation between the model topography and the topography eventually displayed in the map. Another common way of displaying the output of global NWP models is by showing the time evolution of meteorological parameters at specific gridpoints. However, since the model topography is smoother than the true topography, a gridpoint's height above sea level may differ between model and reality. Since the gridpoint distance of global NWP models is usually between 20 and 50 km, they are not able to capture small scale variations caused by the complex topographic features of alpine regions, but are capable of predicting large scale weather evolution for several days into the future.
The forecast information delivered by global NWP models can be refined in a process referred to as ‘downscaling’, in order to produce weather forecasts for mountainous regions. Presently used downscaling methods are commonly divided in to statistical and dynamical methods, and are often used to improve the results of climate models (Matulla et al., 2009), for the production of real time forecasts (e.g. Kang et al., 2009) and to introduce further information to analyses of observational data (Steinacker et al., 2006).
Dynamic downscaling methods use regional NWP models. These generally work with spatial resolutions below 15 km, to refine the output of global NWP models (see e.g. Kanamaru et al., 2008; Mailhot et al., 2010). Figure 1 shows a forecast of the temperature, 2 m above ground, calculated by the Weather Research and Forecast model (WRF, Skamarock et al., 2006) for a part of the Eastern Alps with a gridpoint distance of 12 km. Although this spatial resolution does improve the forecast of the global NWP, it is still far from resolving small scales, for example variations of temperature in narrow valleys. For this, a further increase in resolution would be needed, but even with the most powerful computers horizontal gridpoint distances smaller than 2 km can only be computed within a reasonable time for small areas and short forecast ranges. Results of these regional NWP models are usually visualized in the same way as results of global NWP models, but one of their drawbacks is the fact that they still work with a model topography that is quite smooth compared to reality, and it still takes a skilful forecaster to convert the NWP output to a forecast for a specific point in a complex orography.
Statistical downscaling uses global and regional NWP forecasts and observations to determine equations describing the links of the large scale weather pattern to the regional variations. Techniques using multilinear regression models to modify the output of NWP models in order to forecast the weather are commonly referred to as Model Output Statistics (MOS, e.g. Zokol, 2005), and were invented in the 1960s/1970s. Glahn et al. (2009) developed a procedure to create MOS for points on a regular grid, but this approach still needs observations within a specific radius to the gridpoints to yield respectable results. A variety of methods following more sophisticated approaches, for example using empirical orthogonal functions of meteorological fields as predictands for meteorological parameters, were developed and they are often used to downscale the results of global climate models (Schmidli et al., 2007). The New Digital Forecast Database (NDFD, Glahn and Ruth, 2003) of the US National Oceanic and Atmospheric Administration (NOAA) is a model/operational forecaster blend which can be displayed readily in a variety of GIS related manners (for example, Google Earth). Other prominent downscaling approaches are represented by LEAD (LEAD, 2011), PRISM (Di Luzio et al., 2008) and INCA (Haiden et al., 2011), the last focusing on nowcasting.
The disadvantage of statistical downscaling is the need for observation data to calibrate the equations, which limits its applicability. Results of statistical downscaling methods are usually shown as time evolution of meteorological parameters at specific locations, and are sometimes used for web based GIS, but since they rely on the output of global NWP models and empirical found and calibrated equations, their connection to the complex terrain of mountainous regions is still not given.
This paper presents an alternative downscaling method that connects NWP model forecasts with high-resolution terrain data bases, to describe the variations of meteorological parameters over complex orographic terrain. The approach has been implemented within the framework of MetGIS™, a combined Meteorological and Geographic Information System, and is suitable for operational application even over mountain areas which are scarce of meteorological observation data. This technique is similar to PRISM and INCA, which are also able to use topographic data with resolutions small enough to resolve mountain valleys and other small scale topographic features. This approach yields results that provide some value even for users without meteorological background. Various displaying methods have been developed for the results, including contour plots over maps that can feature by far more geographic information than the regular meteorological forecast maps (e.g. roads, railway lines, topography), and time evolution of meteorological parameters at specific locations.
MetGIS™ was constructed having the latest techniques of software engineering (Dumke, 2003; Endres and Rombach, 2003) and basics of geographic information systems (Jones, 1997; Burrough and McDonnell, 2000) in mind. The code consists mostly of Java-based object-oriented approaches (Lorenz, 1993; Naughton and Schildt, 1999) and some graphics libraries that have already been employed in the construction of the successful snow cover visualization software SN_GUI (Spreitzhofer et al., 2004, 2004).
Some basic ideas of MetGIS™ were inspired from the Integrated Data Viewer (IDV, 2011) and the now outdated, but methodically interesting, PC-based WeatherPro (formerly WELS) weather prediction scheme (Teixeira and Reiter, 1995; Spreitzhofer, 1997, 2000). The system is special in a way that meteorological mesoscale forecast data are downscaled to the points of high-resolution topographic databases, and subsequently stored in a format exactly the same as that of the terrain data. This allows performing complex transactions in which both meteorological and topographic data are involved.
From the start of the system development process, collaboration with international meteorological organizations and atmospheric research institutes has been established (see Table 1). This was to take advantage of the specific expertise of these institutions, to tune the emerging system with different sorts of geographical and meteorological data, and to facilitate a future international, wide-spread application of the system. Prototypes of MetGIS™ have successfully been operated with test data sets for specifically interesting meteorological situations (for example severe snowstorms) over Japan and South America (Spreitzhofer and Norte, 2006). For more information about the early development stage of MetGIS™, see Spreitzhofer and Steinacker (2006, 2008).
Table 1. Contributions of international research institutions to the development of the system
Alden/WELS (Alden Electronics, Inc./WELS Research Corp.)
USA (Boulder, Colorado)
Some basic ideas about the combination between geographic information systems and meteorological forecasts
WSL/SLF (Swiss Federal Inst. for Forest, Snow and Landscape Res., Swiss Federal Inst. for Snow and Avalanche Research)
Java technology for GUI programming. Visualization of the output of snowpack models.
SENAMHI (Servicio Nacional de Meteorología e Hidrología)
Start programming Java-based GIS. Tests of the prototype with a complete set of country-wide geographic vector data.
NIED/NISIS (National Research Institute for Earth Science and Disaster Prevention, Nagaoka Inst. for Snow and Ice Studies)
Continue GIS programming. Start programming interface for meteorological forecast models, using NHM model.
CRICYT/IANIGLA (Centro Regional de Invest. Científicas y Technológicas, Inst. Argentino de Nivelogía y Glaciología)
Inclusion of high resolution terrain data (SRTM, Shuttle Radar Topographic Mission).
DGF (Departamento de Geofísica, Universidad de Chile)
Chile (Santiago de Chile)
Integrate visualization of the output of the MM5 model for two domains covering the Andes range.
IMG (Institute of Meteorology and Geophysics, University of Vienna)
Construction of an operational forecasting system, driven by the GFS model. Inclusion of European topographic data.
3. Components of MetGIS™
3.1. Structure review and technical remarks
The basic structure of MetGIS™ is depicted in Figure 2. Part of MetGIS™ is an independent Geographic Information System which essentially has two functions: to support the forecast visualization modules with geographical background information, and to deliver very detailed topographical input to the downscaling module. The latter is also fed by external mesoscale forecast models and produces high-resolution meteorological forecast fields. These refined predictions can be visualized by the stand-alone MetGIS™ Java GUI (Graphical User Interface) or the MetGIS™ Web Interface.
3.2. The Driving Meteorological Forecast Model
As mentioned above, MetGIS™ is driven by the gridpoint output of atmospheric mesoscale forecast models, but operating these is not part of MetGIS™ proper. In the past the NMH (Non-Hydrostatic Model) of the Japan Meteorological Agency and a version of the MM5 model, operated by the University of Chile, have been used to launch MetGIS™ forecasts. In principle, all numerical model output which is based on NetCDF, GRIB, GRIB2 and simple GrADS-compatible binary formats can be processed and refined by MetGIS™. However, real-time MetGIS™ forecasts are currently calculated based on the output of the GFS (Global Forecast System) of the United States National Oceanic and Atmospheric Administration (NOAA). Alternatively, MetGIS™ could also be driven by ECMWF forecasts, but this is currently not practised due to lack of free data access on the part of the European Centre for Medium Range Weather Forecasts. To obtain a general idea about the performance of global NWPs, visit http://www.emc.ncep.noaa.gov/gmb/STATS/STATS.html
Using global forecast models as a starting point for MetGIS™ bears the advantage that a quick first guess forecast for any mountain region in the world can be calculated without major adjustments of the system software setup. Nevertheless, operating MetGIS™ from the output of higher-resolution regional forecast models is likely to have a positive impact on the overall forecast quality. Therefore, work is currently in progress, enabling the output of the WRF model (Skamarock et al., 2006) to be integrated in the MetGIS™ forecasting process.
3.3. The Geographic Information Module
This module is independent of commercial software such as ArcGIS and structured in the form of layers that can be selected for display independently. This permits the individual or combined visualization of city positions, vector data information (road systems, rivers, railways and political borders) and of various terrain characteristics (elevation, slope and azimuth) in different resolutions and colour scales.
Data are currently available for most of the mountainous regions of Asia, Europe, Australia, New Zealand, North and South America and the Antarctic. However, the geographical databases are constantly upgraded in agreement with external demand.
Geographical data used by MetGIS™ stem from a variety of sources. Concerning topographic properties, the system relies on data of the Japanese Geographic Survey Institute (GSI) and the US Geological Survey. Data bases used from the latter include GTOPO30 (Global Topographic Data) and SRTM (Shuttle Radar Topographic Mission). The highest terrain resolutions used are less than 100 m horizontally (see Figure 3).
3.4. The downscaling module
The downscaling module uses the high-resolution geographic information included in MetGIS™ to assess meteorological information for scales much smaller than those resolved by operational mesoscale models. It relies on a combination of algorithms developed by Spreitzhofer and Raderschall (2004) and ‘VERA-style’ techniques. VERA (Vienna Enhanced Resolution Analysis, Steinacker et al., 2000, 2006) incorporates an objective, automated downscaling and analysis approach for meteorological data over complex topography. The method includes the influence of the high-resolution topography on specific meteorological parameters in the form of so-called ‘fingerprints’. This fingerprint technique uses knowledge about the horizontal and vertical distributions of meteorological parameters in conjunction with geographical conditions, to refine the NWP output. Since the NWP model is using a smoothed topo-graphy for its calculations, information about the ‘real’ height of the considered area can be used to gain information about the horizontal distribution of various meteorological parameters within the downscaling process. An example of the benefits of this procedure is the forecast of areas above and below the snowline. Information about the height of the snowline is estimated from the NWP output, but to determine the exact areas affected by snow, additional information about the detailed distribution of the surface height in the area of interest is needed. This information is provided by the Geographic Information Module and referred to as a ‘fingerprint’.
A basic feature of MetGIS™ downscaling is that the resolution level of the geographical databases used is determined automatically at run time, according to the requirements of the processed operations. The degree of flexibility to operate MetGIS™ in new geographical regions is quite high, and downscaling shows reasonable results on a variety of horizontal scales.
In the current setup, downscaling is just applied upon a number of parameters that have a strong dependence on altitude, such as temperature and the depth of fresh snow. The correct accurate description of the vertical structure of the atmosphere of the model driving MetGIS™ is crucial for the forecast quality of the MetGIS™ output. While at this time the terrain elevation is the most relevant parameter for downscaling, work is going on to include also the specific shape of valleys to derive even more accurate predictions of meteorological parameters, especially the height of the snow level.
3.5. The MetGIS™ Java GUI
The MetGIS™ Java GUI (see Figures 3, 4) is used to visualize and further manipulate the data delivered by the downscaling module. The Java GUI is the proper heart of MetGIS™ and a quite complex software package, including a wealth of display options. Using so-called ‘map generalization techniques’, the detail of geographic information displayed is automatically adjusted when zooming in or out or when switching between differently sized domains. These can cover all scales between entire mountain ranges stretching over hundreds of kilometres, and areas of a few square kilometres.
The current main task of the Java GUI is to feed the MetGIS™ Web Interface automatically with forecast maps in PNG format, but it can also be used to construct high-resolution weather prediction maps for external web pages.
3.6. The MetGIS™ web interface
The MetGIS™ web interface (http://univie.ac.at/amk/ metgis™/) is a highly simplified version of the MetGIS™ Java GUI. It has been designed as an easy-to-use tool for applied users of weather forecasts (such as traffic operation managers and avalanche control staff) who normally have a very limited knowledge of meteorological processes. It offers partly password-protected real-time access to MetGIS™ downscaled short-range forecasts. A number of predefined forecast domains for various countries is available. The forecast range is 2 days (on demand longer), forecasts are available for 3 h intervals, and the predictions are updated four times a day. Forecast fields can be displayed for specific times or in a time-lapse mode.
Very detailed forecast information about some important meteorological parameters is offered, e.g. the spatial distribution of temperature, the form of precipitation (snow, sleet, rain: Figure 5) and the depth of fresh snow (Figure 6). Forecast fields can be displayed for specific times or in a time-lapse mode, and display styles available are ‘Colour Areas’ (used in Figures 5 and 6) or ‘Numbers’ (numerical values on a regular grid). The detailed terrain representation included in MetGIS™ allows for an easy detection of areas above the snow line or the freezing level.
The MetGIS™ Web Interface also offers the possibility to display forecast histograms (time evolution of selected parameters), valid for specific locations that are clicked in the forecast maps.
4. Forecast quality
The forecast quality of MetGIS™ is constantly monitored, since the detection of weaknesses of the used downscaling approach may provide the base for further improvements of the applied techniques. A recent verification study (Sperka and Spreitzhofer, 2009), covering most of the year 2008 and comparing MetGIS™ point forecasts with the measurements at a number of SYNOP stations around the Alps and the Pyrenees yielded quite reasonable scores for temperature and precipitation forecasts. A short summary of these results is presented in the following paragraph.
Concerning temperature, high-altitude stations delivered in the case of 36 h forecasts a mean absolute error of around 1.5 °C (see Table 2). Slightly larger mean absolute errors of 2–3 °C were found at lowland stations, where temperature inversions in the winter are not uncommon. MetGIS™ can only display such temperature inversions if the underlying NWP forecast contains information about them, and inversions are only predicted by NWPs if the lowland is evident in their topography. Since the topography of a NWP is always a smoothed picture of reality, inversions in mountainous regions are more likely to be missed. Precipitation scores were screened mainly for 24 h periods (see Table 3), with four different classes included (<0.1 mm, 0.1–1 mm, 1.1–10 mm, > 10 mm). In around 90% of the predictions the observed and forecast precipitation amounts were in the same class or in adjacent classes. The verification of predicted fresh snow amounts is planned for the near future, but the main problem in this task will be the availability of reliable observation data.
Table 2. Temperature forecast verification for Rax (WMO station 11180, 1546 m), January 2008 to May 2009
The quantities are derived by comparing the observed temperatures to the respective MetGIS™ forecast. The temperature at noon on a specific day, for example, was compared to the 24 h MetGIS™ forecast issued at noon on the previous day. The first column referring to the 0 h forecast range values depicts a comparison between the observation at the time the forecast was issued and the respective MetGIS™ temperature at the forecast initial time. The following columns represent different forecast ranges. Table row ‘Checked Forecasts’ displays the number of observation-prediction pairs used to gain the statistic for the respective forecast range. The rows 2 to 4 show hit rates in percentage for the displayed accuracy, while the other rows show various error statistic parameters.
Mean abs. error (°)
RMS error (°)
Table 3. Contingency table of 24 h precipitation amounts for Rax, January to October 2008
Observed 24-h precipitation (in mm)
In this table predicted 24 h precipitation is compared to observations of the same time span. Example: The number 70 in the third row and second column means, that for 70 24 h intervals a precipitation greater than 1 mm and smaller than 10 mm was forecast, but more than 0.1 mm and less than 1 mm of precipitation was observed. All available 24 h intervals were considered.
Forecast 24-h precip.
The smallest horizontal scale, for which the use of MetGIS™ and its currently implemented downscaling algorithms can be recommended, strongly depends on the depicted parameter. For temperature forecasts, resolutions of around 100 m make sense as long as thermal inversions are reasonably described by the numerical model used to drive MetGIS™. The reasonable minimum scale for fresh snow forecasts is slightly larger. It has to be kept in mind that MetGIS™ forecasts the height of snow which a storm may drop on the ground, but it still does not include the effect of redistribution of once fallen snow by the wind.
5. Conclusions and outlook
MetGIS™ is a promising interface between geographical and meteorological information systems which produces terrain-adjusted meteorological forecasts with a specific focus on liquid and frozen precipitation in resolutions still unknown in present real-time weather prediction systems. The forecast quality of MetGIS™ will permanently increase, since gradually more VERA-style downscaling techniques will be implemented. Due to the global coverage of GFS forecasts and the sort of geographical data used, operational MetGIS™ predictions can be produced for any mountain region of the world. If higher-resolution limited area models are available for specific regions, these can easily be integrated with MetGIS™, since the meteorological model interface of MetGIS™ is quite flexible.
Further upgrades of MetGIS™ may include the usage of meteorological observation data for the purpose of forecast adjustment and fine-tuning. Another future option that can increase the power of MetGIS™ over Alpine terrain is the inclusion of snow cover characteristics via the application of snow cover models such as SNOWPACK (Bartelt and Lehning, 2002; Lehning et al., 2002a, 2002b). Even without this, conditional display modes of the type ‘Show all gridpoints above 2000 m, with a slope exceeding 30°, predicted fresh snow accumulation of more than 30 cm and a wind speed exceeding 20 m s−1’ will soon be possible, aiming at specific applications such as the estimation of the avalanche danger. Traffic operation centres may benefit from the introduction of line forecasts (predictions along the course of a highway) which are feasible through the availability of high resolution forecast data bases produced by MetGIS™. Finally, information about the shape of valleys, derived from the internal geographic information system, might be used to improve the forecast of the snow line.
Thanks for scientific, computational, conceptual and/or financial contributions to successful advances in the development of MetGIS™ go to M. Ristic, M. Blaschek and C. Pehsl (all Institute of Meteorology, University of Vienna), to Prof. emer. E. R. Reiter and L. Teixeira (formerly WELS Research Corp./Alden Electronics), to WSL/SLF, SENAMHI, NIED, CRICYT/IANIGLA and to DGF (see Table 1 for full names of these institutions). We are also grateful to IfGR (Institut für Geographie und Regionalforschung, University of Vienna), and we say thanks to ASFINAG (Autobahnen- und Schnellstraßenfinanzierungs-AG, Austria), to BMVIT (Austrian Ministry for Traffic, Innovation and Technology, project ‘Strassenforschung 3.237’) and to the Austrian Science Funds (FWF, grant L 696-N15) for sponsoring parts of the presented R&D work.