SASUF Sustainability Forum 2024

Drastic shifts in climate-weather patterns, with the abundant rainfall and resulting floods have
caused damages all over the world and put many populations in precarious position. Challenges
posed by related crisis have pushed governments, NGOs, and the private sector to increase efforts
to leverage the power of Artificial Intelligence to mitigate climate change effects and provide
response to natural disasters. Flood mapping is a critical component of disaster response and
mitigation efforts. Conventional flood mapping techniques often rely on remote sensing data and
satellite imagery, for instance provided by Maxar in a collaboration with LiU, but they may fall short
in accurately capturing the three-dimensional geometry of flooded areas as well as the capacity of
different types of vegetation to bind water. This workshop aim to discuss the feasibility of exploring
the development of deep learning models that leverage advanced geometric estimation techniques
to enhance the spatial accuracy of flood maps as well as to accurately predict the drivability of roads
in the case of floods and the capacity of areas to bind water depending on the classification of
vegetation.

The detailed agenda is as follows:

10.00 - 10.05 Opening
10.05 - 10:35 Amy Loutfi, AI enabled and Semantic Image Recgonition for Flood Analysis and Decision Support
10:35 - 10.40 Questions
10.40 - 11.10 Njoya Ngetar, Remote Sensing and AI for Mitigation of Climate Change Effects: An Environmentalist's Perspective
11.10 - 11.15 Question
11.15 - 11.35 Jules Tapamo, Exploring Deep Representation Geometric Learning for Road Networks Drivability Prediction from Remote Sensing Images
11.35 - 11.55 Michael Felsberg, Explainability, Uncertainty, and Bias in Physics Informed Machine Learning on Remote Sensing Data
11.55 - 12.00 Questions and Closing

It is envisaged that the following activities will be carried out:

Activity 1: AI and mitigation of Climate change effects
Discussion on the state of readiness of Sweden and South Africa to leverage AI technology to mitigate climate change effects and appropriately respond to natural disasters.

Activity 2: Flooding Area Prediction by Geometry Estimation Models
Discuss current machine-learning-based geometry estimation models, such as PointNet and Graph
Convolutional Networks, to extract three-dimensional geometric features from the input data while
exploiting geometric inductive biases.

Activity 3: Estimating Water Binding Capacity of Vegetation from Multi-source Data
Discuss the relevance of multi-source data, including high-resolution satellite imagery, LiDAR data,
and hyper-spectral images, to provide a basis for estimating the water binding capacity of
vegetation.

Activity 4: Road network drivability prediction from Earth Observation Data
Explore avenues to develop deep learning models, incorporating geometric features obtained from
the geometry estimation (activity 2) and satellite-based classification of roads for the prediction of
drivability and accessibility of read networks, e.g. by means of attention mechanisms and
transformers.

Activity 5: Formulation of joint research proposal
Formulate research proposal to investigate the integration of geometry- and vegetation-aware deep
learning models to help improve mitigating the effects of floods, as well as providing more detailed
and reliable information on flooded areas.

This workshop is a follow up, on an initiative that started in in 2017. In fact, Prof Felsberg organised
the 17th International Conference on Computer Analysis of Images and Patterns, 2017, which was
attended by Prof Tapamo. It was an opportunity to know more about research projects in the
Computer Vision Laboratory at LiU, which to a great extend could be very helpful to many disciplines
at UKZN, such as Computer Science, Computer engineering and Environmental Science, were there
are various research interests in computer vision. In 2020, Prof Felsberg was invited at UKZN where
he gave talks on Computer vision. Subsequent to that an MOU was signed between UKZN and LiU
and Prof Felsberg was appointed Honorary Professor at the School of Engineering at UKZN. This
workshop will be used to initiate joint work between the Computer Vision Laboratory at LiU and The
Image Processing, Computer Vision and Data Mining Research at UKZN. We will want the start a
project that will investigate the integration of advanced geometry estimation techniques with deep
learning models to improve the accuracy and detail of flood mapping.

Activity 1: AI and mitigation of Climate change effects (15min)
Activity 2: Flooding Area Prediction by Geometry Estimation Models (30 min)
Activity 3: Estimating Water Binding Capacity of Vegetation from Multi-source Data (30 min)
Activity 4: Road network drivability prediction from Earth Observation Data (30 min)
Activity 5: Formulation of joint research proposal (15 min)

Keywords: Climate Change effects, floods mapping, road network classification,
Computer Vision, Deep Learning