Zhang, Caiyun

This study examines the impact of uncertainty associated with Light Detection and Ranging (LiDAR) derived Digital Elevation Models (DEMs) on flood risk mapping in the North Biscayne Bay sub-watershed. A comparison of flood extent and generation of the probability of flooding was carried out using the bathtub and probabilistic approaches respectively. The water level was computed separately for original and refined DEM using Cascade 2001 hydrological model. Using land cover based corrected DEMs reveals a 12% reduction in flooded areas in contrast to original DEM, considering uncertainties associated with land cover. Probabilistic flood modeling via Gaussian Geostatistical Simulation accounts for DEM uncertainty, yielding nuanced probability flood risk maps (0-100%). Findings emphasize DEM refinement before conducting flood mapping to address uncertainties. Future research should explore other mediums of correction incorporating effects of point density of LiDAR, methods of DEM generation, use of diverse scenarios, and kriging techniques for flood modeling and mapping while using LiDAR derived DEM.

Gentrification describes rapid infrastructure development and investment in areas with lower income classes. It may cause potential erasure of the original neighborhood's unique culture and the displacement of residents. Due to rising sea levels and the increase in the frequency and intensity of storms, the inundation of Florida will increase as time passes. This creates an ironic relationship where historical coastal areas inhabited by an affluent population will move inland to historically lower-income populations. This thesis developed a Climate Gentrification Index (CGI) to identify areas at risk of gentrification caused by inundation of storm scenarios in Tampa City, Florida. Socioeconomic data and inundation data produced from a hydrological model were combined to define CGI and areas with high risk were mapped and discussed.

Much of Interior Alaska contains permafrost, which is a permanently frozen layer found within or at the surface of the Earth. Historically, this permafrost has experienced relative stability, with limited thaw during warmer summer months and fire events. However, largely due to the impact of a warming climate, among other factors, permafrost that would typically experience limited thawing during the summer season has recently been thawing at an unprecedented rate. Trapped by this layer of permafrost is a large quantity of carbon (C), which could be released into the atmosphere as greenhouse gases such as carbon dioxide (CO2) and methane (CH4). Due to the remoteness of the Arctic, there is a lack of yearly recorded permafrost thaw depth and snow depth values across much of the region. As such, the focus of this research was to establish a framework to identify how permafrost thaw depth and snow depth can be predicted across both a 1 km2 local scale and a 100 km2 regional scale in Interior Alaska by a combination of 1 m2 field data, airborne and spaceborne remote sensing products, and object-based machine learning techniques from 2014 – 2022. Machine learning techniques Random Forest, Support Vector Machine, k-Nearest Neighbor, Multiple Linear Regression, and Ensemble Analysis were applied to predict the permafrost thaw depth and snow depth. Results indicated that this methodology was able to successfully upscale both the 1 m2 field permafrost thaw depth and snow depth data to a 1 km2 local scale before successfully further upscaling the estimated results to a 100 km2 regional scale, while also linking the estimated values with ecotypes. The best results were produced by Ensemble Analysis, which tended to have the highest Pearson’s Correlation Coefficient, alongside the lowest Mean Absolute Error and Root Mean Square Error. Both Random Forest and k-Nearest Neighbor also provided encouraging results. The presence or absence of a thick canopy cover was strongly connected with thaw depth and snow depth estimates. Image resolution was an important factor when upscaling field data to the local scale, however it was overall less critical for further upscaling to the regional scale.

Salt marshes are highly dynamic ecosystems that rely on multiple environmental and physical drivers that determine species distribution and soil property distribution. However, climate change and human interference are threatening the delicate ecosystem. One of the easiest ways to monitor marsh dynamics is through remote sensing. Traditional methods may not handle the large, non-parametric datasets well and often do not spatially determine areas of uncertainty. This dissertation research developed a framework to map marsh species and predict ground soil properties using multiple remote sensing data sources by integrating modern Object-based Image Analysis (OBIA), machine learning, data fusion, and band indices techniques. It also sought to determine areas of uncertainty in the final outputs and differences between different spectral resolutions. Five machine learning classifiers were examined including Support Vector Machine (SVM) and Random Forest (RF) to map marsh species. Overall results illustrated that RF and SVM typically performed best, especially when using hyperspectral data combined with DEM information. Seven regressors were assessed to map three different soil properties. Again, RF and SVM performed the best no matter the dataset used, or soil property mapped. Soil salinity had r as high as 0.93, soil moisture had r as high as 0.91, and soil organic an r as high as 0.74 when using hyperspectral data.

The Louisiana coastal ecosystem is experiencing increasing threats from human flood control construction, sea-level rise (SLR), and subsidence. Louisiana lost about 4,833 km2 of coastal wetlands from 1932 to 2016, and concern exists whether remaining wetlands will persist while facing the highest rate of relative sea-level rise (RSLR) in the world. Restoration aimed at rehabilitating the ongoing and future disturbances is currently underway through the implementation of the Coastal Wetlands Planning Protection and Restoration Act of 1990 (CWPPRA). To effectively monitor the progress of projects in CWPPRA, the Coastwide Reference Monitoring System (CRMS) was established in 2006. To date, more than a decade of valuable coastal, environmental, and ground elevation data have been collected and archived. This dataset offers a unique opportunity to evaluate the wetland ground elevation dynamics by linking the Rod Surface Elevation Table (RSET) measurements with environmental variables like water salinity and biophysical variables like canopy coverage. This dissertation research examined the effects of the environmental and biophysical variables on wetland terrain elevation by developing innovative machine learning based models to quantify the contribution of each factor using the CRMS collected dataset. Three modern machine learning algorithms, including Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN), were assessed and cross-compared with the commonly used Multiple Linear Regression (MLR). The results showed that RF had the best performance in modeling ground elevation with Root Mean Square Error (RMSE) of 10.8 cm and coefficient of coefficient (r) = 0.74. The top four factors contributing to ground elevation are the distance from monitoring station to closest water source, water salinity, water elevation, and dominant vegetation height.

Traditional sand analysis is labor and cost-intensive, entailing specialized equipment and operators trained in geological analysis. Even a small step to automate part of the traditional geological methods could substantially improve the speed of such research while removing chances of human error. Digital image analysis techniques and computer vision have been well developed and applied in various fields but rarely explored for sand analysis. This research explores capabilities of remote sensing digital image analysis techniques, such as object-based image analysis (OBIA), machine learning, digital image analysis, and photogrammetry to automate or semi-automate the traditional sand analysis procedure. Here presented is a framework combining OBIA and machine learning classification of microscope imagery for use with unconsolidated terrigenous beach sand samples. Five machine learning classifiers (RF, DT, SVM, k-NN, and ANN) are used to model mineral composition from images of ten terrigenous beach sand samples. Digital image analysis and photogrammetric techniques are applied and evaluated for use to characterize sand grain size and grain circularity (given as a digital proxy for traditional grain sphericity). A new segmentation process is also introduced, where pixel-level SLICO superpixel segmentation is followed by spectral difference segmentation and further levels of superpixel segmentation at the object-level. Previous methods of multi-resolution and superpixel segmentation at the object level do not provide the level of detail necessary to yield optimal sand grain-sized segments. In this proposed framework, the DT and RF classifiers provide the best estimations of mineral content of all classifiers tested compared to traditional compositional analysis. Average grain size approximated from photogrammetric procedures is comparable to traditional sieving methods, having an RMSE below 0.05%. The framework proposed here reduces the number of trained personnel needed to perform sand-related research. It requires minimal sand sample preparation and minimizes user-error that is typically introduced during traditional sand analysis.

The Florida Everglades ecosystem is experiencing increasing threats from anthropogenic modification of water flow, spread of invasive species, sea level rise (SLR), and more frequent and/or intense hurricanes. Restoration efforts aimed at rehabilitating these ongoing and future disturbances are currently underway through the implementation of the Comprehensive Everglades Restoration Plan (CERP). Efficacy of these restoration activities can be further improved with accurate and site-specific information on the current state of the coastal wetland habitats. In order to produce such assessments, digital datasets of the appropriate accuracy and scale are needed. These datasets include orthoimagery to delineate wetland areas and map vegetation cover as well as accurate 3-dimensional (3-D) models to characterize hydrology, physiochemistry, and habitat vulnerability.

Over drainage due to water management practices, abundance of native and rare
species, and low-lying topography makes the coastal Everglades especially vulnerable to
Sea-Level Rise (SLR). Water depths have shown to have a significant relationship to
vegetation community composition and organization while also playing a crucial role in
vegetation health throughout the Everglades. Modeling potential habitat change and loss
caused by increased water depths due to SLR requires better vertical Root Mean Square
Error (RMSE) and resolution Digital Elevation Models (DEMs) and Water Table
Elevation Models (WTEMs). In this study, an object-based machine learning approach
was developed to correct LiDAR elevation data by integrating LiDAR point data, aerial
imagery, Real Time Kinematic (RTK)-Global Positioning Systems (GPS) and total
station survey data. Four machine learning modeling techniques were compared with the
commonly used bias-corrected technique, including Random Forest (RF), Support Vector
Machine (SVM), k-Nearest Neighbor (k-NN), and Artificial Neural Network (ANN). The k-NN and RF models produced the best predictions for the Nine Mile and Flamingo study
areas (RMSE = 0.08 m and 0.10 m, respectively). This study also examined four
interpolation-based methods along with the RF, SVM and k-NN machine learning
techniques for generating WTEMs. The RF models achieved the best results for the dry
season (RMSE = 0.06 m) and the wet season (RMSE = 0.07 m) WTEMs. Previous
research in Water Depth Model (WDM) generation in the Everglades focused on a
conventional-based approach where a DEM is subtracted from a WTEM. This study
extends the conventional-based WDM approach to a rigorous-based WDM technique
where Monte Carlo simulation is used to propagate probability distributions through the
proposed SLR depth model using uncertainties in the RF-based LiDAR DEM and
WTEMs, vertical datums and transformations, regional SLR and soil accretion rates. It is
concluded that a more rigorous-based WDM technique increases the integrity of derived
products used to support and guide coastal restoration managers and planners concerned
with habitat change under the challenge of SLR. Future research will be dedicated to the
extension of this technique to model both increased water depths and saltwater intrusion
due to SLR (saltwater inundation).

Global sea-level rise SLR is projected to accelerate over the next century, with research
indicating that global mean sea level may rise 18–48 cm by 2050, and 50–140 cm by 2100.
Decision-makers, faced with the problem of adapting to SLR, utilize elevation data to identify
assets that are vulnerable to inundation. This paper reviews techniques and challenges stemming
from the use of Light Detection and Ranging LiDAR Digital Elevation Models DEMs in support
of SLR decision-making. A significant shortcoming in the methodology is the lack of
comprehensive standards for estimating LiDAR error, which causes inconsistent and sometimes
misleading calculations of uncertainty. Workers typically aim to reduce uncertainty by analyzing
the difference between LiDAR error and the target SLR chosen for decision-making. The
practice of mapping vulnerability to SLR is based on the assumption that LiDAR errors follow a
normal distribution with zero bias, which is intermittently violated. Approaches to correcting
discrepancies between vertical reference systems for land and tidal datums may incorporate tidal
benchmarks and a vertical datum transformation tool provided by the National Ocean Service
VDatum. Mapping a minimum statistically significant SLR increment of 32 cm is difficult to
achieve based on current LiDAR and VDatum errors. LiDAR DEMs derived from ‘ground’
returns are essential, yet LiDAR providers may fail to remove returns over vegetated areas
successfully. LiDAR DEMs integrated into a GIS can be used to identify areas that are
vulnerable to direct marine inundation and groundwater inundation reduced drainage coupled
with higher water tables. Spatial analysis can identify potentially vulnerable ecosystems as well
as developed assets. A standardized mapping uncertainty needs to be developed given that SLR
vulnerability mapping requires absolute precision for use as a decision-making tool.

Human activities in the past century have caused a variety of environmental
problems in South Florida. In 2000, Congress authorized the Comprehensive Everglades
Restoration Plan (CERP), a $10.5-billion mission to restore the South Florida ecosystem.
Environmental projects in CERP require salinity monitoring in Florida Bay to provide
measures of the effects of restoration on the Everglades ecosystem. However current
salinity monitoring cannot cover large areas and is costly, time-consuming, and laborintensive.
The purpose of this dissertation is to model salinity, detect salinity changes, and
evaluate the impact of salinity in Florida Bay using remote sensing and geospatial
information sciences (GIS) techniques. The specific objectives are to: 1) examine the
capability of Landsat multispectral imagery for salinity modeling and monitoring; 2)
detect salinity changes by building a series of salinity maps using archived Landsat images; and 3) assess the capability of spectroscopy techniques in characterizing plant
stress / canopy water content (CWC) with varying salinity, sea level rise (SLR), and
nutrient levels.
Geographic weighted regression (GWR) models created using the first three
imagery components with atmospheric and sun glint corrections proved to be more
correlated (R^2 = 0.458) to salinity data versus ordinary least squares (OLS) regression
models (R^2 = 0.158) and therefore GWR was the ideal regression model for continued
Florida Bay salinity assessment. J. roemerianus was also examined to assess the coastal
Everglades where salinity modeling is important to the water-land interface. Multivariate
greenhouse studies determined the impact of nutrients to be inconsequential but increases
in salinity and sea level rise both negatively affected J. roemerianus. Field spectroscopic
data was then used to ascertain correlations between CWC and reflectance spectra using
spectral indices and derivative analysis. It was determined that established spectral
indices (max R^2 = 0.195) and continuum removal (max R^2= 0.331) were not significantly
correlated to CWC but derivative analysis showed a higher correlation (R^2 = 0.515 using
the first derivative at 948.5 nm). These models can be input into future imagery to
predict the salinity of the South Florida water ecosystem.