Skip to main content

A Pre-operational Product for Cloud Screening and Snow Extent Retrieval using Copernicus Data in Finland

Point of contact
Ali Nadir Arslan
Finnish Meteorological Institute
Erik Palménin aukio 1
FI-00560 Helsinki
Phone: +358-50-3203386

The aim of this action is to develop a super high resolution snow extent product from Sentinel data using machine learning techniques to provide a pre-operational snow product service over Finland.

Snow plays a vital role in climate monitoring, thus this issue is continuously being researched. Some Snow cover products already exist but do not offer sub hundred meters resolution, and do not make use of fusion of multiple sensor data with in-situ measurements. At present, machine learning techniques are trending in image classification. This action will use machine learning to create self-learning algorithms for detecting clouds and snow, thus creating new ways to predict cloud and snow extent. It will take advantage of Sentinel data disseminated by the Finnish National Data Hub FINHUB. FINBUB is hosted by the National Satellite Data Centre NSDC which is the main dissemination point for Copernicus Sentinel data in Finland.

The outcome of this action will differ from existing products because we plan

  • To use machine learning techniques
  • To produce very high resolution products aimed at sub hundred meters
  • Combine different data, the sentinels and in-situ observations
  • Since the input data is available immediately after downlink, processing will begin immediately thus cutting down on time for searching and downloading data

Accurate and high resolution snow extent information has significant use in several fields such as meteorology and hydrology. Finland has high cloud coverage through the year, and working with optical data such as Sentinel-2 will require applying cloud screening. Other countries such as Sweden, Norway will also benefit from the methods as these could be applied to other areas who have similar cloud problems as well as mountainous areas where the resolutions is crucial.

As the amount of earth observation data in the Copernicus project increases and advances in machine learning techniques, implementing alternative classification methods and approaches on this immense amount of data is a self-proposing topic. We are aiming to develop this snow extent product by using in-situ snow depth measurements to train our model and produce super high resolution snow detection. Apart from targeting the scientific community, through this action the general public will also have a chance to use Copernicus data through an application that will provide snow extent. For instance: 1) for logistics purposes in traffic e.g. cleaning of roads, determining of routes in case of heavy snow situations; 2) Timber harvesting from the woods; 3) for recreational use e.g. skiing, summer cottage maintenance, when people want to visit the summer cottages they need to check the snow situation; 4) Professional and authority use, since there are a lot of floods in Finland during the snow melting season, flood prediction is very important. Therefore hydrologists would benefit from snow extent products with better resolution in order to improve their prediction accuracy. In Agriculture, farmers need to check the snow situations to make decisions for relating to the snow melting season.

Outputs and Results

  • Pre-operational high resolution snow extent product over Finland
  • Improved cloud screening technique using machine learning
  • Minimum of 1 scientific paper
  • Attention to the use of modern ML techniques on EO data
  • A service for users to be able to view sentinel imagery and snow extent with respect to their area of interest with or without cloud screening
  • End user oriented workshop
  • A research oriented workshop