Work Package 1 (WP1) Aim:
To integrate EO data into our DT as well as calibrate satellite observations using terrestrial sensors and subsequently use the data to improve the spatial resolution of soil moisture measurements, which is a key model parameter and driver of flood, drought and groundwater.
Specific tasks are:
Task 1.1 aims to derive near-real-time, catchment-wide soil moisture at high spatial resolutions (~20m) using in-situ, satellite, and model-derived datasets. Historic radar images from Sentinel-1 will be processed and integrated with data from Sentinel-2 and COSMOS ground data to establish correlations with soil moisture. A Deep Learning model (LSTM) will be used for future estimations. This approach leverages the dielectric properties contrast between water and dry soil. The data will be fed into a digital hub for assessment and calculation of catchment soil moisture, and radar backscatter will help determine historic flood extents. The observations will calibrate and validate the Digital Twin's components, driving forecasting models and alert systems, resulting in a near-real-time catchment soil moisture map. This task enhances soil moisture dynamics understanding, improving water resource management and flood forecasting.
Task 1.2 aims to enhance EO data with in-situ sensor networks. Digital twins offer a unique opportunity to improve Earth Observation (EO) data by integrating it with in-situ sensor data. Ground observations can directly calibrate EO data through bias-correction, spatial downscaling, and temporal interpolation. Advanced calibration methods can also use proxy data and model states, such as downscaling EO-based soil moisture observations using hydrological models. This integration will enable data assimilation and forecasting, with AI playing a key role. To test this, a network of soil moisture sensors from University of Hull’s SUDs lab will be used to complement the COSMOS network, enhancing the value of remotely sensed data. This data will be fed into groundwater and surface water models to improve the accuracy of soil moisture estimations.
The Long Short Term Memory predictions for soil moisture. The model has been trained on some COSMOS sites, we use this model to predict soil moisture at the Hull sites based purely on SAR satellite data as part of WP1.
WP1 Contributing partners:
Dr Alessandro Novellino, British Geological Survey:
Coordinating EO activities.
Dr Kathryn Leeming, British Geological Survey:
Developing of the LSTM model
Ms Erin Mills, British Geological Survey: Analysis and interpretation of optical satellite data
Ms Holly Hourston, British Geological Survey: Analysis and interpretation of radar satellite data
Prof. Wouter Buytaert, Imperial College London: Development of a data assimilation scheme to use observations from satellite and in-situ sensors to test the improvement of soil moisture estimations in a digital twin framework.
Dr Rike Becker, Imperial College London: Technical implementation of a data assimilation scheme to use observations from satellite and in-situ sensors to test the improvement of soil moisture estimations in a digital twin framework.