## Intro
Coastal areas host a considerable part of human life and activities, providing societal, economical and ecological benefits. The health of these ecosystems, however, highly depends on the fragile balance between climate influence and local anthropogenic constraint. Among coastal areas, low lying sandy beaches are disproportionately affected by the impacts of climate change and their vulnerability has become increasingly evident worldwide, especially in the light of rising sea levels and increased extreme weather patterns. Erosion, coastal flooding, damaging land use changes are now putting the ecosystems and seafront population at greater risk. Therefore, understanding and predicting the sandy coastlines evolution is of paramount importance for coastal zone management, to anticipate potential threats with sufficient lead-time, as well as to provide mitigation strategies and more adapted resilience procedures. 

## Study dataset
For our study we selected a set of cross shore transects every 50km over the world shoreline and keep only the transects of sandy coasts. On those transect we computed monthly shoreline anomaly (difference to a reference of worldwide shoreline position) using LANDSAT missions between 1st January 2000 to 31th December 2012 and then interpolated daily. To predict shoreline evolution, three variables are necessary : Sea Level Anomaly, River Discharge and Wave Energy. Therefore, we created a dataset containing those 3 variables with data retrieved  from drivers over the same period. This process gives us with a dataset containing 8857 spatial positions and 4749 days. We processed the dataset to map the spatial points into a 2D matrix. The idea is that we want to represent the world atlas as a smaller scale to keep the spatial information (not really present in a vector). The 8857 spatial points are placed in a 100x100 matrix representing the world from -180 to 180 degrees of longitude (x-axis) and -60 to 60 degrees of latitude (y-axis). This mapping is done by looking to each point of the input vector and find the corresponding closest point in the 2D output matrix “without repetitions”. This means that a point placed in the matrix can’t be overwrite. Therefore we have all input point spatially placed and empty area representing the biggest gap between data (oceans).



## Acknowledgements

We thank Dennis Wilson for organizing this Hackathon, Grégoire Thoumyre, Jennifer Montaño, Marcan Graffin, Grégoire Abessolo, Rafael Almar, Erwin Bergsma, Julien Boucharel for providing this dataset.