forecasting_dashboard
The forecasting dashboard integrates the station monitoring service with calibrated forecasts computed using the EMOS and gEMOS routines. This service is designed to function as a simple workstation for the operational duties of the Agency for Hydrometeorology (TJHM) in Tajikistan. On the left side of the dashboard, a map displays the stations. In this example, the map includes both automatic weather stations and manual stations from TJHM, represented by square markers instead of round ones. Data from the stations are displayed in the lower-right section, where users can select different variables to view. These variables can also be visualized on the map. Additionally, key metadata for the stations can be accessed directly from the dashboard.
In the upper-right section, a meteogram displays the weather forecast for the next ten days. This meteogram shows calibrated forecasts that blend station data with global ECMWF forecasts (referred to as WWCS) and are represented in red. In these forecasts, the solid lines denote the mean estimates derived from calibrating an ensemble of ECMWF simulations. The shaded areas indicate the 66th and 95th percentile ranges of forecast uncertainty, reflecting the inherent variability in weather predictions. Observations, shown as a black line, are expected to fall within these ranges in at least 95% of cases (assuming the 95th percentile range). The spatially calibrated forecast is depicted as a color-coded raster in the background of the map. Users can click any point on the map to view location-specific forecasts based on this spatial calibration by selecting the "Map Forecast" tab in the meteogram.
The meteogram also includes precipitation forecasts, represented as blue bars showing six-hourly precipitation sums. Currently, these forecasts are derived directly from raw ECMWF data and are not yet calibrated with station data, as longer datasets are required for precipitation calibration. When measured precipitation data is available, observations will appear as additional, darker blue bars, allowing for forecast verification.
The meteogram displays a weather pictogram for each day, summarizing the average weather conditions in the forecast. The identification of these pictograms is based on a series of routines detailed in the service file process_pictos.R. Using various physical variables and decision trees, a pictogram is selected for each hour from a set of 42 possible pictograms (see image). These hourly pictograms can then be aggregated into other temporal resolutions, such as six-hourly or daily pictograms, through additional decision-tree routines. The current implementation supports both six-hourly and daily aggregations.
The full documentation of the decision trees is an internal MeteoSwiss document. However, the processes can be reproduced using the functions described in the code. It is important to note that the weather pictograms are proprietary files owned by MeteoSwiss. To ensure full compatibility with external systems, these icons would need to be replaced with an open-source set of weather icons, and the decision trees reconstructed accordingly.
The dashboard allows also a comparison of the forecasts with the past observations. By changing in the forecast reference time in the sidebar the past forecasts are shown in the meteogram jointly with the observations in black and precipitation in dark blue if the station includes any. This allows for a visual verification of the forecasts in the past two months and gives an estimate of the average forecast error. In the side bar you additionally also change the displayed period of the station observations or the lead time of spatial forecasts from the gEMOS, as well as the administrative areas of the country.
Figure: The dashboard is used operationally by TJHM (5th December 2024) as well as by the Agency for Civil Protection on a daily basis. Its serves as a daily workstation to assess local forecasts which have been unprecedented in the country.
Figure:The forecast information is also shared through an API to personal stations deployed in villages ("Kiosks") which show the forecasts also for remote locations which low mobile penetration.