IAP-24-036

Aiding emergency services by detecting floods in urban areas using radar satellite data and machine learning

In this novel and timely research, we aim to improve the management and protection of floodplains and wetlands by using cutting-edge satellite images and drone data to develop field-tested methodologies to monitor flood level and extent on a weekly basis to inform flood risk management strategies. You will be performing experiments with a ground radar where we will take measurements of real life flooding as well as simulating flooding in the lab.

Building urban areas in floodplains offers fertile soil for agriculture, it provides flat, buildable land, making construction more straightforward and cost-effective. Additionally, proximity to rivers supports transportation and trade. However, it requires careful flood management to mitigate risks of seasonal flooding.

The work in this project is in the framework of the Scotland’s International Environment Centre (SIEC) [SIEC 2021], a £23M government investment to help Scotland (but also more broadly the UK and the World) to improve climate change resilience and to help deliver net zero carbon.

You will be working on 2 main test sites with different challenges. (i) The Forth Valley region (Scotland) is a thriving region in Scotland, but it is heavily affected by flooding which impacts local businesses and infrastructure. (ii) Colombo, in Sri Lanka, supports a high terrestrial and freshwater biodiversity, which is important for conservation but also supplies local people with a range of livelihood activities. However, Colombo is tormented by annual devastating floods.

In this project we will use algorithms to demonstrate internationally the benefits of satellite monitoring of flooding. We will use satellite Synthetic Aperture Radar (SAR), which is able to obtain images of the environment from space using microwaves. It allows us to acquire images independent of weather conditions and solar illumination, which is very valuable in areas with frequent cloud cover. We will also use a cutting edge radar technology called polarimetry interferometric (Pol-InSAR) [ESA-PolSAR] combined with state of the art Machine Learning and Deep Learning methodologies. In addition to using satellite data, we will be carrying out extensive experiments to ground-truth the data using a ground radar which can simulate the images obtained from satellites. Additionally, we will use ground sensors (water level probes, soil moisture probes) installed in key areas by SIEC’s infrastructure. Finally, we will make use of the emerging technology of UAV (“drone”) based observation for field validation and rapid local assessment using low cost aircraft.

A strong motivation for using satellite images is that we have entered a new era of freely available satellite data (e.g. the ESA Sentinel constellation missions [ESA-Sentinel]). We are experiencing a rapid growth of activities in the Space industry and the Earth Observation sector. When paired to the exponentially growing sector of unmanned aerial monitoring, this opportunity not only supports businesses activities but also provides many state of the art tools to the environmental management community.

The development work will be accompanied by significant fieldwork in Scotland (and possibly one trip to Colombo). In Scotland we will make use of the ground radar where the polarimetric radar signature of flooded areas will be analysed. We will also design an experiment with simulated floods in the lab. If successful, the processing stacks produced in this project will be incorporated in SIEC and feed into the Scottish Environmental Protection Agency (SEPA) flood management strategy.

Deliverables: In this project, we will set up a series of methodologies that will be able to provide weekly updates of flood and urban wetland depth and extent starting from images acquired from space. Among other products we are interested in monitoring drastic changes in water conditions which could help rapid intervention.

Novelty: Pol-InSAR is a cutting edge technology and is very useful to retrieve biophysical parameters of vegetation and soil [ESA-PolSAR]. However, we are in urgent need of controlled experiments on the ground, which will allow a much better understanding of the satellite signal over flooded areas (especially when this covers vegetation). Additionally, the use of Deep Learning for Pol-InSAR it is still at its infancy and more development is needed.

Data (satellite): Archived Pol-InSAR data are already available. Future acquisitions will be carried out synchronised to fieldwork. The datasets used will include at least the following satellite missions: ALOS-2 (Japanese Space Agency); SAOCOM (Argentinian Space Agency), NISAR (NASA and ISRO), BIOMASS (ESA) and Sentinel-1 (European Space Agency).

Data (ground): We will be using a ground radar built in the Stirling radar lab (based on a VNA architecture) to acquire images that emulate satellites. This can be tuned at different frequencies and acquire quad-polarimetric and interferometric data. It can be easily transported and installed on a tripod. In one of the novel experiments we will reproduce a flood around concrete where we will progressively add water to a basin that can be closed containing concrete structures until the water level is increased.

Data (drone): We will be using our DJI Phantom-4 drone to collect aerial images of the flooded areas during fieldwork.

Algorithm development: In this project we will develop algorithms that exploit weekly available Pol-InSAR images combined with sparse ground measurements to monitor water quantity especially when this is surrounded by concrete walls or under vegetation.
1) We will monitor changes in water quantity, by applying scattering models and change detectors. One of the methodologies will be based on the use of optimisations of polarimetric data [Marino et al 2014].
2) Analysis of time series. This will allow the evaluation of trends in flood frequency and extent.
3) Deep Learning with special interest for Convolutional Neural Networks will be developed to extract information from Pol-InSAR data

Year 1

Preparing a literature review on the topics: SAR, drone imaging, floods. Fieldwork: Start working on ground measurements and monitoring of flood with Pol-InSAR. Start of lab experiment with simulated flood. Attending international training events. Expected submission of a journal paper on monitoring urban floods and wetlands with Pol-InSAR.

Year 2

Monitor multi-year changes in water level in urban areas. Finilise results of lab experiment. Expected submission of a journal paper on using Deep Learning with Pol-InSAR and ancillary data.

Year 3

Use models to evaluate the sustainability of human activities based on temporal trends observed. Writing up of thesis chapters. Expected submission of journal paper on sustainability assessment.

Year 3.5

Complete thesis, submission and viva.

This is a multi-disciplinary project including topics related to (a) satellite Earth Observation; (b) drone surveys; (c) physical models (electromagnetic scattering); (d) data analysis; (e) floods, peatlands and grasslands; (f) programming.
The successful candidate will have the opportunity to gain valuable skills in the context of: (a) analysing and processing satellite and drone images using Python; (b) planning and accomplishing ground radar and drone campaigns; (c) developing analytical and empirical models to measure biophysical parameters of the environment; (d) using Geographical Information Systems (GIS) software.
The training will also include the attendance of major international training events such as the training on polarimetric SAR data, provide by ESA in Italy.