Nowcasting of Fog Using AI (NAI) Techniques Using Satellite Observations

Lliso, Llorenç⁽¹⁾. Peinado-Galan, Niobe ⁽¹⁾. Pellicer-Valero, Oscar J ⁽²⁾. Camps-Valls, Gustau ⁽²⁾. Lopez, Román ⁽¹⁾ . Calbet, Xavier ⁽¹⁾. Rípodas, Pilar ^{(1) )}. Gultepe, Ismail ^{(3) (4)}

(1) Agencia Estatal de Meteorología (AEMET/NWC SAF), Spain

⁽²⁾ Image Processing Laboratory (IPL), Universitat de València, Spain

(3) Ontario Tech University, Oshawa, Canada

(4) University of Notre Dame, South Bend, Indiana, USA.

The NWC SAF (Nowcasting Satellite Application Facilities) belongs to the SAF network of EUMETSAT. The NWC SAF provide software to derive nowcasting products from satellite data. The key objective is the nowcasting of the weather phenomena using observations from satellite combined with other secondary data sources, providing EUMETSAT member states and worldwide customers with an advanced, robust, and reliable forecasting system to support both operational and research activities.

At NWC SAF, a toolbox is developed for nowcasting called NowcastIng AI (NAI) which is a software package in the format of Jupyter notebooks, the software is released under the umbrella of the EUMETSAT training portal. The toolbox allows to combine GEO Clouds products (for MSG and GOES) with METAR data and leads to develop ML models for fog nowcasting. Using the previous experiences on NWC SAF products in machine learning algorithms, the presented work proposes NWC SAF cloud physical products as predictors to nowcast the fog. The fog forecasting is a cutting-edge problem with high economic and safety implications on aviation operations. Furthermore, the fog forecasting is an elusive problem, which represents a challenge for operational meteorologists.

The NAI tool allows to extract data cubes (volumes) centered at the airports identified by their ICAO code. The data is comprised of time series of stamps of NWC SAF clouds products, surface observations (temperatures, relative humidity, pressure, wind speed and direction as well as precipitation and visibility derived from METAR reports). The data cubes are stored as netcdf files and exposed as xarray datasets. Along with the data cube “builder”, the collection allows users to test their own models against the proposed model baselines. The study seeks to address fundamental questions concerning the efficacy of machine learning models, and their capacity to utilize satellite data effectively. This will lead to the feasibility of their operational usage for practical scenarios.

This presentation aims to introduce this new data source to the community and encourages researchers its usage and that can lead to the development of high-performance numerical forecasting models deployable in operational settings.