What if marine pollution could be curbed at its urban source?
What if we could predict the probability of presence of urban marine litter?
Urban Ocean Marine Litter Assessment
Interact with the dashboard below ↓
The Challenge
Litter accumulation in marine ecologies poses significant cyclical health, biodiversity, and climate risks. It is estimated that 11 million metric tons of plastic waste enter the ocean every year, and this number expected to triple by 2040. Originating in urban areas, this trash accumulation poses a systemic issue in developing countries with lacking waste disposal systems and recycling programs. This often disproportionately impacts vulnerable coastal dwellers and fisherman who heavily rely on the marine ecosystem.
Urban Ocean, a program by Ocean Conservancy has been developing initiatives that mitigate marine pollution, assess waste management, and enable cities to address ocean plastics and resilience. Site selection of pilot sites is made difficult due to lacking information around litter which makes it difficult for Urban Ocean’s zero-waste pilot to continue to make strides in reducing marine litter.
The Solution
This project aims to navigate zero-waste siting in areas facing bleak marine litter conditions through geospatial machine learning. We present, for Urban Ocean managers, a web-based dashboard with litter data metrics, demographic metrics, and a litter accumulation risk site-selection tool serving the 12 urban ocean cities in siting zero-waste pilots.
Data analysis
While conducting our analysis it was difficult to unify datasets across geographies. We used global datasets and used counts of datasets like population but proxies like proximity to restaurant and shops in lieu of social behavior datasets. We then used a fishnet grid (described above) to model risk and aggregate and visualize data in ewach cell across all cities. For the proxy datsets, we implemented a K-Nearest Neighbor analysis to determine the proximity of each data point to one another and incorporate that as an additional variable to our model. K-Nearest Neighbor (KNN) helped to find the proximity of another variant to determine high activity spots in the area. The nearest neighbor factor was repeated for land-use, roads, restaurants, water, and waste facilities.
Ultimately created three regression models: a random forest model, a linear regression model, and a “mixed” model that combined the two. We found our mixed model the best for our use case with regards to residual error when comparing predicted to observed results.
We the replicated this analysis across the 12 Urban Ocean cities: Pune, Salvador, Santa Fe, Santiago, Semarang, Surat, Chennai, Bangkok, Can Tho, Melaka, Mumbai and Panama City.
Finally, we created a dashboard (see image below) via Javascript and HTML/CSS for this model for easier access and usability for an Urban Ocean project manager. You can explore it here under Marine Litter Assessment Dashboard.
