Date of Degree

9-2020

Document Type

Dissertation

Degree Name

Ph.D.

Program

Earth & Environmental Sciences

Advisor

Maria Tzortziou

Advisor

Kyle McDonald

Committee Members

Andrew Reinmann

Dorothy Peteet

Subject Categories

Environmental Monitoring | Hydrology

Keywords

remote sensing, tidal marshes, wetlands, radar, radiometric modeling, tidal inundation

Abstract

This dissertation focuses on the monitoring and characterization of tidal marshes using remote sensing-based approaches. Chapter 1 introduces the topics of wetland ecology and remote sensing. Chapters 2-4 are the main research chapters of the dissertation covering the topics of tidal marsh mapping, tidal marsh vegetation characterization, and assessment of tidal marsh inundation patterns. Chapter 5 summarizes the preceding chapters and highlights future research directions.

The primary research objective of Chapter 2 is the mapping and study of tidal marshes of the Chesapeake and Delaware Bays. This chapter also features a thematic focus on the evaluation of various forms of satellite imagery for wetlands studies in general. In this chapter, Chesapeake and Delaware Bay tidal marshes were mapped with producer’s accuracy greater than 88% and user’s accuracy greater than 83% using a random forest classification. A second classification effort focused on the mapping of wetlands vegetation at the Jug Bay wetlands complex in the Patuxent River. This classification, which also utilized the random forest technique, yielded accuracies of greater than 90% for all mapped wetlands vegetation types. These two classifications made use of SAR and optical/IR satellite imagery as input layers. Post-classification analysis demonstrated that optical/IR images were most useful for providing accurate classifications for Chesapeake Bay tidal marshes while the SAR images were most useful for classifying different wetlands vegetation at Jug Bay. This highlights the importance of SAR-optical/IR fusion for providing flexibility in achieving accurate classifications when mapping diverse wetlands types with a single classifier.

Chapter 3 builds on many of the findings from the evaluation portion of Chapter 2. In Chapter 3, a mapping methodology utilizing SAR-optical/IR fusion is used to expand tidal marsh mapping beyond the Chesapeake and Delaware Bays and into the entire Mid-Atlantic region. Chapter 3 also seeks to map not only tidal marshes, and distinguish them from freshwater marshes, but also map marshes that have been invaded by Phragmites australis. These three types of marshes were all mapped with greater than 80% user’s and producer’s accuracy in the Mid-Atlantic region. This marsh mapping effort was also carried out in the Gulf Coast region, where marshes were mapped with greater than 91% user’s accuracy and 95% producer’s accuracy, although the three individual marsh classes were often confused with one another. In addition to effectively mapping Phragmites australis with supervised classification approaches in the Mid-Atlantic region, a decision tree-based approach was developed to map invasive aquatic water chestnut (Trapa natans) with greater than 96% accuracy. This decision tree approach utilized multitemporal Sentinel-1 C-band SAR imagery. The same decision tree classification was utilized to map non-persistent emergent vegetation, an indicator of tidal freshwater wetlands, with greater than 93% accuracy.

Chapter 4 seeks to assess tidal marsh inundation patterns. In this chapter, numerous satellite image-based inundation products were developed and validated using in-marsh water level observations from study sites. In this chapter, optical/IR, C-band SAR, and L-band SAR inundation products are validated with water level sensors and assessed for overall performance. This validation suggest that L-band inundation products offered the best performance for mapping tidal marsh inundated area with 90% accuracy. L-band inundation products were developed using backscatter intensity thresholding-based approaches which were derived empirically. Radiometric modeling efforts were also utilized to elucidate changes in scattering mechanisms in support of empirical image analyses. The radiometric modeling efforts and empirical image analyses demonstrate that C-band and L-band SAR backscatter tends to decrease in response to inundation in tidal marshes. However, an important distinction that the radiometric modeling efforts revealed was that C-band signals interact much more strongly with vegetation, while L-band signals respond more strongly with the surface underlying vegetated canopies. Further, L-band signals decrease to a greater magnitude in response to inundation making L-band imagery much more effective for assessing tidal marsh inundation state than C-band imagery. These findings are also supported by image-based polarimetric decompositions which capture similar scattering shifts in response to inundation as those demonstrated by radiometric modeling efforts at L-band frequencies.

Chapter 5 summarizes the previous chapters and discusses the launch of three satellites that are anticipated to advance the study of tidal marsh systems. The most relevant of which is the NISAR satellite which will operate at an L-band frequency with an anticipated 12- day revisit time. The findings of this dissertation demonstrate the utility of L-band SAR for characterizing tidal marsh inundation state, combined with a 12-day revisit, NISAR imagery should greatly improve the characterization of tidal marsh inundation patterns.

This work is embargoed and will be available for download on Thursday, September 30, 2021

Graduate Center users:
To read this work, log in to your GC ILL account and place a thesis request.

Non-GC Users:
See the GC’s lending policies to learn more.

Share

COinS