Project Mission
The project aims to improve the algorithm for predicting drifting sea ice velocity components and provide helpful visualisations for sea ice motion in the Arctic, therefore taking an adaptation approach to climate change.
Problem Statement
The most reliable data used for sea ice movement tracking is collected by passively drifting buoys that were deployed in the Arctic over the last few decades. These are GPS-equipped instruments that provide precise measurements, but offer limited spatial coverage. This shortfall can be compensated by remote sensing & image processing on sequences of satellite imagery. However, this data may not be available all year long - which in turn has led researchers to use atmosphere/ wind data to estimate sea ice drift velocity. The challenge is to use AI to build the model that best reproduces the buoy drift, based on the wind fields using forty years data record of drifting Arctic buoys, and co-located winds, and additional environmental parameters (such as sea ice conditions: sea ice concentration, sea ice thickness).
Value Proposition
Industry: With the Northwest passage used to reduce journey times, the industrialisation of the Arctic is on the rise: studies estimate that shipping activity in the Arctic will increase more than 50% by 2050 [source]. The transport industry poses two risks to the Arctic ecosystem - further heating at the North Pole resulting in ice shrinkage & heavy fuel oil spill ecological hazards. Accurate sea ice motion predictions would enable better logistic planning for the shipping industry reducing the duration of trips, and hence carbon emissions and oil spill hazard, through the shortest path optimisation. Academic Research: The primary need is to better understand the dynamics of climate change in the Arctic Ocean to answer research questions such as the transition from older sea ice to a younger seasonal ice pack, the transport of ice-rafted sediments, or pollutants in the context of an oil spill, and the risk assessment associated with navigation and marine operations in the Arctic.
Development Process
Carry out initial data pre-processing, analysis and visualisation using boxplots, histogram and correlation map. Work on visualising & mapping buoy movement on EASE grid using basemap. Perform feature engineering and PCA. Research regression models and trial various options - linear regression, multi-output regression, random forest regressor, k-neighbours regressor, decision tree regressor.
Challenges We Ran Into The team having different local machine OS proved a challenge in the first day of hackathon, for example some of the key code was developed in Linux which was difficult to replicate on Mac OS causing delays. Looking back, we would have taken advantage of cloud computing from the beginning so that the entire team collaborates on the same OS.
Accomplishments that we’re proud of
Using new technologies to tackle the challenge!
What we learned
The part of the team less experienced with AI techniques got exposure to the model development process & data manipulation behind the scenes; also got to experiment with less familiar Python libraries such as pandas, plotly, scikit-learn & basemap and even installed VM for the first time to be able to run jupyter notebook with all requirements.
What’s Next for Vel-Ice
Develop sea ice thickness tracking visualisation more granularly at daily or weekly levels from the current yearly MVP submission.
NOTE Submission file of predictions (predictions.csv) added in google drive below
Built With
- jupyter
- python
Log in or sign up for Devpost to join the conversation.