Landsat and Sentinel are earth observation programs run by the United States and European Union respectively, and both serve similar goals of creating multispectral imagery of the planet for monitoring and mapping purposes. The Landsat program, in operation since the 1970s, is currently on its seventh successful mission with the Landsat 8 satellite, and has demonstrated its value for earth observation many times over. Sentinel 2 is a newer platform in operation since 2015, and with its more frequent image collection and higher resolution it has quickly proved its value. Each Landsat and Sentinel satellite revisits the same spot on earth every 16 days and 10 days respectively, each platform has two satellites in an orbit that is offset in such a way that the revisit time is cut in half, resulting in eight and five day revisits respectively. These orbital differences allow the satellites to observe the same area of the planet the same day several times a year. This research analyses a pair of datasets from the two satellite platforms collected three minutes apart on June 23, 2019, and compares the results of each platform on common classification and analysis tools employed by researchers who use the two platforms. In addition to comparing results of each independent platform in classification and indexing analysis methods, the power of data fusion of these short time difference datasets in classification schemas will be demonstrated.

Feature detection is a popular subject of remote sensing, and satellite imagery has been classified extensively with neural networks. Neural networks are computer algorithms that aim to mimic processes in the human brain, and are used for classification in many fields. Classification of LiDAR data in literature has been limited to low density autonomous vehicle LiDAR units, and very few studies have looked at classification with neural networks. In my research I am applying the popular backpropagation neural network because of its widespread use and documented success, and the ARTMAP neural network because it is able to avoid ‘catastrophic forgetting’ that is sometimes a problem with backpropagation. There are two parts to my study, first I will look at aerial LiDAR, where the neural networks will be classifying aerial LiDAR data derivatives combined with satellite imagery, and the second where I will look at terrestrial laser scanning (TLS) LiDAR returns, RGB color values, and geometric calculations. The neural networks will be evaluated for their ability to detect shell midden land from classification signatures in data collected from Charlotte Harbor Preserve State Park at the mouth of the Myakka River and Cockroach Bay Preserve State Park in Tampa Bay. The neural networks are implemented through the Neural Network code in MATLAB, and the data is preprocessed for MATLAB use using a Python script. As LiDAR is a dimensional increase from 2D satellite imagery, the neural networks are expected to perform just as well as they do with classifying satellite imagery . Results yielding the ability to accurately detect shell midden signature from remotely sensed LiDAR data will improve archaeological survey and management capabilities.

On July 14, 2017, a cover-collapse sinkhole formed in the front yard of a home in Pasco County, FL. Starting as a depression, the initial collapse occurred rapidly (120 minutes) with subsequent slumping over the course of three days. The sinkhole is oval and cone-shaped with a northeast-southwest long axis and ridges on the northwest and west slopes. A combination of remote sensing, geophysics, and soil borings are used to characterize the temporospatial surface changes and subsurface structures at this sinkhole. Repeat surveys started four days post-collapse and concluded 10 months post-collapse. Surface changes over time are computed using terrestrial laser scanning (TLS) and drone-based structure-from-motion (SfM) photogrammetry, with the Multiscale Model to Model Cloud Comparison algorithm. The initial collapse area measured 1,395 square meters (day 4) and grew to a maximum of 1,626 square meters (day 32) before stabilization efforts partially the sinkhole and built up the perimeter walls. Post-collapse and pre-stabilization activity in the form of perimeter growth occurred on the northeast and southwest edges. Ground-penetrating radar detected a semicontinuous horizon within sands and silts that appears to correspond to the historic ground surface present before portions of the nearby Saxon Lake were filled in as part of agricultural and housing development modifications to the area in the 1960s and 70s. The direction of the collapse’s long axis, post-collapse activity, and the orientation and depth of a semicontinuous subsurface horizon all suggest a northeast-southwest trending linear or elongated karst feature contributed to the collapse and subsidence.

A sinkhole collapse on 14 July 2017 destroyed two homes and required evacuation of 9 additional residences in Land O’Lakes, Florida. The sinkhole formed rapidly and within 10 hours reached most of its final size of approximately 40 by 50 m and a maximum depth of about 15 m. According to Pasco County officials, this is the largest sinkhole to have formed in the county during the last 30 years. The site is located near two natural lakes and occurs within an area of well-developed karst. The county allowed USF geoscientists to study the sinkhole resulting in an extensive collection of LiDAR, GPR, and lake-bottom profiling data. The seismology group installed one broadband 3-component seismometer on 25 July ~20 m W of the sinkhole edge and a second one on 27 September about 10 m E of the sinkhole edge. Data are recorded continuously at 200 Hz. Drilling to understand soil structure integrity and remediation work to stabilize the sinkhole led to extreme noise levels during daytime operations through most of August 2017. Nighttime noise is lower, but due to nearby houses, sensors pick up strong 60 Hz noise caused by AC units. The only signal unequivocally attributable to the sinkhole occurred on 5 August just after stabilization work, consisting of partial backfilling with limestone rocks, began that caused the sinkhole to widen by about 3 m along its entire western edge. After completion of stabilization work, no further signals have been detected from the sinkhole. Besides sinkhole monitoring, we recorded several large global earthquakes, e.g., the Mw=8.2 and Mw=7.1 Mexico earthquakes on September 8 and 17, respectively, as well as Hurricane Irma as it passed within 20 miles as a quickly weakening category 1 hurricane. In addition to outreach and sinkhole documentation as part of a multi-sensor approach, we use the site to train seismology graduate students in field practices and conducted a class field trip.

The Saxon Lake sinkhole collapse of July 14, 2017 in Land O Lakes, Florida, caused the destruction of two homes and the evacuation of nine additional residences. The sinkhole is slightly oval with dimensions of approximately 51 meters east-west and 42 meters north-south, and it is reportedly 15 meters deep. This is presumably the largest sinkhole to form in Pasco County during the last 30 years. The surface collapse happened rapidly and continued over three days, with slumping and erosion increasing the size. The site is located near two natural lakes in a housing development from the late 1960s. This occurrence is within an area of well-developed karst, with a number of natural lakes. We present preliminary analysis of the sequence of deformation, sinkhole geometry, surrounding subsurface structures, and seismic activity. Data are assembled from terrestrial and aerial LiDAR, UAS survey and PhoDAR modeling, aerial imagery, ground penetrating radar, lake-bottom profiling, and seismic monitoring. Additionally, multi-sensor data were brought together in a Geographic Information Systems (GIS) and included an analysis of georeferenced historic imagery and maps. These spatial data indicate historic land use change and development alterations that included lake shore reconfiguration, canal construction, and connection of lake water systems in the area of impact. Three subsidence reports from the 1980s are also recorded within 500 meters of the collapse