Tracer concentration mapping in a stream with hyperspectral images from unoccupied aerial systems

Abstract

The release of anthropogenic chemicals to streams, stemming from contaminated sites, agriculture, urban sources, or accidental input, represents a significant threat to water resources and thus the health of humans and aquatic ecosystems. Predicting the transport and fate of chemicals is critical to quantifying contaminant concentrations and developing environmental quality standards (EQS). Tracer tests are a well-established tool for such hydrological investigations in various aquatic systems. In stream settings, the experiments have predominantly investigated longitudinal mixing and flow velocities by measuring the tracer concentration in a few discrete locations; few studies have focused on the transverse mixing properties. Recent progress in hyperspectral remote sensing from Unoccupied Aerial Systems (UAS) allows advancing the two-dimensional monitoring of tracer tests, by mapping the tracer concentration with a high spatial resolution in narrow streams with difficult accessibility. So far, such methods have mainly been demonstrated in controlled settings or in ocean waters, but rarely in optically complex streams. In this study, we evaluated the performance of a miniaturized hyperspectral imaging system and a consumer grade camera onboard a UAS, to map the concentration of the fluorescent tracer Rhodamine WT in a stream impacted by a contaminated site. In order to estimate tracer concentrations from the remotely sensed data, a ratio of the red and blue band was used for the RGB camera, while a vector-based method, estimating the spectral angle in regards to a reference spectrum, was applied for the continuum-removed hyperspectral data. The RGB camera performed well only in sections of the stream exposed to direct sunlight (R2: 0.83; nRMSE: 10.2%), failing to map the concentration in all locations, which included areas where direct sunlight was blocked by riparian trees (R2: 0.17; nRMSE: 26.19%). In contrast, the advanced spectral information allowed the hyperspectral- system to map the concentration well in all sections of the stream (R2: 0.78; nRMSE: 13.35%), regardless of illumination changes. This demonstrated the advantage of hyperspectral imaging systems for measuring water-leaving irradiance in hundreds of contiguous narrow spectral bands that also allow detecting finer absorption and emission features. The approach described here could help to improve the knowledge of contaminants mixing in streams, i.e. to predict the location of fully transverse mixing for contaminated sites discharging to streams via groundwater-surface water interactions, as well as general assumptions behind mixing and dilution models.