Objectives of the service
At the current state of technology, river flow monitoring still relies on in-situ stations installation on selected points of interest. This approach is subject to many limitations like the continuous maintenance and the difficulty to expand the observations to other sites because of the cost of installation and potential lack of power supply and data connection, especially in remote areas. As far as the forecast approach is concerned, traditional prediction methods are often based on a “past-to-future” approach, i.e., historical data are simply extrapolated to the future. This approach is generally subject to poor accuracy as it does not take into account the current status of the catchment and the predicted weather forecast for the next forecast horizon.
WatAspace will tackle these limitations by using a multi-model multi-data approach. Satellite images from different sensors (Sentinel-1, 2 and 3) are used in order to enrich the information content and improve the update frequency. At the same time, meteorological data are collected to provide valuable ancillary information to be later used in prediction mode. Then, tailored machine learning algorithms and physical models are calibrated against historical water observations to estimate water discharge in rivers and water level in reservoirs.
Finally, WatAspace may be used to predict water flow in the next hours, days or months, according to meteorological forecasts.
Users and their needs
Many public and private actors are interested in river monitoring for various reasons:
Water utilities and Irrigation agencies: they use reservoirs to stock water for civil or agricultural use. They need to monitor the current water level and forecast the water evolution in the next days and weeks to comply with extreme events like floods and drought.
Hydropower companies: like Water Utilities, they need to manage the level of the reservoir during the extreme events and to optimise water withdrawal according to the energy price.
Public governments: river authorities and regional agencies need to monitor river discharge in given outlets for civil protection during the extremes and for hydrological balances.
The challenges to fulfil these needs are related to the spatial size of the water body to be monitored (the smaller, the more difficult to detect from satellite) and the temporal aggregation of the results (the smaller the interval, say hourly, the more difficult). To solve the spatial issue, high resolution satellite images will be used, whereas to solve the temporal issue, weather forecast and physical models will be tested to integrate the satellite revisit time.
The experimentation is currently ongoing on test sites in Italy, but WatAspace can be used to monitor any water body all over the world.
Service/ system concept
WatAspace aims to develop two services:
monitoring service: river and reservoir water level monitoring based on satellite data and machine learning without the need of in-situ installation;
prediction service: water inflow forecast for the next few days or weeks based on numerical weather prediction and calibrated on historical remote sensed data.
The satellite images and meteorological observation will be collected from Copernicus and local Open Data portals respectively. WatAspace will be calibrated on nearby locations where historical water observations are available. Then, the system will be exported to the target location, by downloading historical data and creating a final hybrid monitoring and prediction model putting together the WatAspace technology with the hydrological model of Waterjade.
Finally, the service will be deployed via a dashboard, where the user can login and see the water data in a convenient and graphical way, or simply via APIs for a faster and easy-to-integrate manner.
Space Added Value
The novelty of WatAspace is the joint use of different satellite images in order to exploit the different information content about the status of a waterbody:
Radar altimeters (e.g., Sentinel-3) provide a direct measurement of water level in rivers or reservoirs;
Multispectral sensors (MS, e.g., Sentinel-2) provide multi-band images containing different information about the soil (e.g., presence of water-covered areas, vegetation, snow);
Synthetic Aperture Radar (SAR, e.g., Sentinel-1) provides images in which water bodies can be easily discriminated against in dry areas.
The joint use of the three sensors can take advantage of the pros of each one, mitigating its weak points at the same time. In particular, Sentinel-3 is used in its altimetry mode to detect the water level, whereas Sentinel-1 and Sentinel-2 are used to detect the presence of water, providing an estimate of the cross-section of the target.
By using a dedicated Artificial Intelligence infrastructure, the three different data will be finally merged to obtain a robust and comprehensive dataset.
At present, WatAspace is being tested on pilot sites in Italy with the idea to start from the big rivers/reservoirs where we have in situ observations. The first analysis concentrated on the Po river (300-500 m width) and Lake Bilancino (5 km2) using Sentinel-1 and 2. Preliminary results are encouraging. The next step will be to decrease the size of the water body and to use Sentinel-3 and high-resolution images.