Thumbnail Image


Publication or External Link





Environmental remote sensing is the science of monitoring physical, chemical, and biological characteristics of the Earth through space and time, and from a distance, by measuring how these environments interact with electromagnetic energy, or more simply through changes in color. This dissertation leverages in situ, satellite, and unoccupied aircraft system (UAS, drones) data to enhance the efficacy of environmental remote sensing in Chesapeake Bay. Satellite data consists of distinct contributions of the surface under observation and the intervening atmosphere. Atmospheric correction (AC) processors seek to isolate the surface signal, and while several variants exist, their accuracy varies widely in optically complex coastal waters. Chapter 2 is a statistical evaluation of four common AC variants applied to data collected by the most recent operational ocean color sensor, the Ocean Land Color Instrument (OLCI) onboard Copernicus Sentinel-3A and -3B satellites. Remote sensing reflectance (Rrs), the product of AC processors from which a suite of water quality metrics is then derived, was obtained from each AC variant and matched in space and time with in situ Rrs data collected in the Chesapeake Bay. AC results varied widely, and the most statistically robust was a neural-net based algorithm (Case 2 Regional Coast Color, C2RCC). The resultant shape and magnitude of Rrs (e.g. color) is governed by the type and concentration of optically active constituents (OACs), namely phytoplankton pigments, chromophoric dissolved organic matter, inorganic sediment, and water itself. In coastal waters where OACs are dynamic and vary independently from each other, deriving accurate water quality metrics remains an open challenge. Chapter 3 applies a spectral clustering classification of OLCI Rrs data (2016-2022) and identifies the fifteen most dominant optical water types (OWTs) of Chesapeake Bay. OWTs were matched in space and time with Chesapeake Bay water quality monitoring data, and a statistical evaluation demonstrates how water quality data are constrained within and across OWTs. In contrast to earth-observing satellites, UAS equipped with optical sensors offer on-demand, highly resolved data. Aquatic UAS applications are in their infancy, and the critical removal of light reflected directly off the skin of water has received little attention in the literature. Chapter 4 proposes four different approaches to remove direct surface reflectance from UAS imagery and evaluates each against in situ Rrs data. The most accurate method is a simple empirical model that exploits measurements in the infrared where water strongly absorbs light; applying this model permits high resolution water quality retrievals with only modest uncertainty. Chapter 5 uses UAS imagery to monitor a wetland restoration site in the Chesapeake Bay across seasons and years. A supervised random forest model is developed with UAS data and used to classify species-specific marsh vegetation with 97-99% accuracy. Vegetation classification maps were compared to as-built planting plans to delineate instances of significant marsh migration. Chapter 6 summarizes how the environmental remote sensing methods used in this dissertation can contribute to a better understanding of coastal research, monitoring, and management by addressing challenges, gaps, and potential solutions at various scales.