Riverine litter monitoring from multispectral fine pixel satellite images

Abstract

Mismanaged litter or debris in aquatic systems can pose threats to water quality and blue economic activities. Monitoring strategies like remote sensing can complement and support the gathering of relevant descriptors useful for understanding challenges related to leakage litter. We present a robust technique for detecting and quantifying floating riverine litter, a soup of natural and anthropogenic materials. Spectral information of GeoEye, PlanetScope and Skysat fine resolution satellite imagery was statistically transformed into spatial anomalies correlated to fractional-pixel riverine litter abundance. Algorithm development also involved techniques that accounted for variation in satellite data characteristics. The detected fractional abundance was converted into a unit surface area in a region-of-interest. Intercomparison of derived area coverage from matching images captured by the various unique sensors in the PlanetScope and Skysat constellation were consistent (R² = 0.98). Likewise, the litter surface area derived manually had a very strong linear relationship to the algorithm estimates (R² > 0.99). The prospect of time series observations over several years at sub-daily to near daily intervals were also demonstrated using the cloud-free PlanetScope and Skysat imagery. Transferability as well as easy adaptation of the algorithm was further showcased by application over water bodies in Guatemala and Slovakia.