EO4NOWCAST Near Real-Time Soil Moisture Assessment and Pluvial Flood Nowcasting - Pipeline
- Generation of current Soil Moisture (SM) local maps based on the use of a ML model that processes ground-based precipitation measurements and Normalized Difference Moisture Index (NDMI) processed from the last available Copernicus S2 satellite tile.
- Pluvial flood nowcasting method based on the use of an ML model exploiting the information provided by updated soil moisture maps and common ground based atmospheric observation.
The exploitation of this asset is strictly limited for research/learning purposes.
The EO4NOWCAST sub-project of AI4Copernicus realised and demonstrateed an operational and replicable approach to assess severe weather events and related short-term hazards (nowcasting) built upon the synergy between EO and rainfall monitoring product. EO4NOWCAST is a generalized DSS platform that processes near real-time soil mositure maps and performs a pluvial flood nowcasting service based on the integration between satellite EO products and meteorological data through machine learning models.
Nowadays, flood prediction is mainly based through complex physical models that have some limitations, i.e., the missing estimation of the water content of the soil in their algorithms, which is a crucial parameter to estimate the risk of flood in the territory. Copernicus Sentinel satellites provides tiles of Normalized DIfference Water Index (NDMI) which is an indicator of the greenness of the biomes and can be processed to locally get approximated soil moisture information and with an update time of 5-10 days. To exploit this high spatial resolution information and, at the same time, to fill the need to have recent (near real-time) measurements, the Near Real-Time Soil Moisture Assessment AI model has been developed within the project to update the most recent NDMI map from Copernicus through rainfall map obtained by in situ monitoring network.
Within EO4NOWCAST project a novel approach is proposed to combine EOs and rainfall monitoring, i.e., accurate context awareness, to anticipate hazard conditions (from 20 to 120 minutes) and support decision makers. The Pluvial Flood Nowcasting method takes advantage of the information provided by updated soil moisture maps, which are pre-elaborated from the latest Copernicus imagery through the Near Real-Time Soil Moisture Assessment service developed by Artys and ground-based rainfall measurements.
This docker container contains the pipelines to process Near Real-Time Soil Moisture maps and to perform a Pluvial Flood Nowcasting in the area of interest of the Polcevera river basin (Genoa, Italy) with the methods developed in EO4NOWCAST project
The docker container allows to start an application that includes
- the machine learning pipeline for the hourly prediction of the soild moisture based on the NDMI maps from Copernicus S2 and precipitation monitoring products;
- the machine learning pipeline for the 5-min prediction of the flood risk based on the first service and precipitation monitoring products.
A user guide documentation for the execution of the docker container is provided in the annexed file.
The machine learning modules included in these pipelines have been trained for the relevant use case of the Polcevera river basin in Genoa (Italy).
The pipelines are scheduled, respectively, every hour and every 5 minutes. It is possible to monitor the status of each pipeline by means of a web console available once the application is up and running.
The attachment folder contains a README_1.md file describing all the instructions to perform the EO4NOWCAST Near Real-Time Soil Moisture Assessment and Pluvial Flood Nowcasting services.
The service does not violate the guidelines for Trustworthy AI.
The service processes and combines only a restricted type of data related to physical features of the area of interest. There is no storage of user information. Thus, the service is outside the scope of the GDPR.