Search

The transition from foraging to the domestication of plants and animals in the Atacama Desert (ca. 3,000 BP) seemed to spark the emergence of social complexity, sedentarism, and the end for Hunter-Gatherer lifeways. Nonetheless, ethnohistorical evidence suggests the opposite. To challenge current narratives this work proposes an alternative perspective where the key is to ignore simple versus complex dichotomy and instead look deeply at the sources of evidence available. To test this proposed new narrative the basin of the Camarones Valley was the sampling area; a river valley located at the core of the Atacama that covers latitudinally from the Andean mountains to the Pacific. Therefore, an assessment of historical sources, the interpretation of freely available satellite imagery, and a sample of the new sites remotely sensed were surveyed. As a result, approximately 2,000 new archaeological sites in the basin of Camarones shows that agriculture was very limited, small-scale sites constitutes 90% of the record, and that pre-Hispanic people settled and moved around the landscape as a whole deriving in a high-mobility landscape showing limited evidence of sedentarism, disproving the current narrative of 'social complexity' and its close relation with sedentism. With this new information, a reorganization of the current multi-ethnic paradigm was also proposed for the different periods after the Formative. This change reflects that at least two distinct groups of foragers inhabited the western Chilean valleys during the Late Intermediate Period, and possibly prior to this. Finally, together with the proposal of the paradigmatic enhancement, it was possible to identify two types of archaeological sites unpublished for the Camarones Valley, the stoneworks for hunting called chaku and caycu. These have great implications for future research that will allow us to know in more detail the role of hunters in pre-Hispanic times.

Remote sensing provides image and LiDAR data that can be useful for a number of tasks such as disaster mapping and surveying. Deep learning (DL) has been shown to provide good results in extracting knowledge from input data sources by the means of learning intermediate representation features. However, popular DL methods require large scaled datasets for training which are costly and time-consuming to obtain. This thesis investigates semantic knowledge extraction from remote sensing data using DL methods in regimes with limited labelled data. Firstly, semantic segmentation methods are compared and analysed on the task of aerial image segmentation. It is shown that pretraining on ImageNet improves the segmentation results despite the domain shift between ImageNet images and aerial images. A framework for mapping road networks in disaster struck areas is proposed. It uses pre and post disaster imagery and labels from OpenStreetMaps (OSM), forgoing the need for costly manually labelled data. Graph-based methods are used to update the pre-existing road maps from OSM. Experiments on a disaster dataset from Palu, Indonesia show the efficacy of the proposed method. A method for semantic feature extraction from aerial imagery is proposed which is shown to work well for multitemporal high resolution image registration. These feature are able to deal with temporal variations caused by seasonal changes. Methods for tree identification in LiDAR data have been proposed to overcome the need for manually labelled data. The first method works on high density point clouds and uses certain LiDAR data attributes for tree identification, achieving almost 90% accuracy. The second uses a voxel based 3D Convolutional Neural Network on low density LiDAR datasets and is able to identify most large trees. The third method is a scaled version of PointNet++ and achieves an F_score of 82.1 on the ISPRS benchmark, comparable to the state of the art methods but with increased efficiency. Finally, saliency methods used for explainability in image analysis are extended to work on 3D point clouds and voxel-based networks to help aid explainability in this area. It is shown that edge and corner features are deemed important by these networks for classification. These features are also demonstrated to be inherently sparse and pruned easily.

Mangrove forests are important in providing a range of ecosystem services, including food provision to local communities and carbon storage, while being globally restricted to tropical coastlines. The conservation and sustainability of mangrove forests is thus a globally important topic. Mangrove forests in the Niger Delta are known to be under high pressure from urbanisation, development, logging and oil pollution, and invasive species such as nipa palm (Nypa fruticans). These mangrove forests are poorly understood as a result of difficulty of access, social unrest and security restrictions. For example, there is no data on the relationship between disturbance and mangrove structure in the Delta, current area extent and biomass stocks of mangrove forest, its rate of loss, or the rate of nipa palm colonisation in the Niger Delta. The overall objective of this thesis is to utilise a combination of field data and earth observation to resolve these knowledge gaps. This work will estimate area and biomass of mangrove forests in the Niger Delta, and their changes over recent years through disturbance and invasive species. I used an extensive field data collection in 2016-17 to establish 25 geo-referenced 0.25-ha plots across the Niger Delta and collected 567 ground control points. I estimated aboveground biomass (AGB) from a general allometric equation based on stem surveys. Leaf area index (LAI) was recorded using hemispherical photos. I performed and evaluated a land cover classification using a combination of Advanced Land Observatory Satellite Phased Array L-band SAR (ALOS PALSAR), Landsat ETM+ and the Shuttle Radar Topography Mission Digital Elevation Model (SRTM DEM) data. I also compared two supervised classification methods: Maximum Likelihood (ML) and Support Vector machine (SVM) classifiers. I established a relationship between field estimates of AGB and Advanced Land Observatory Satellite (ALOS) L-band radar backscatter. I also estimated the area of nipa palm and mangrove forests in the Niger Delta and generated the first mangrove biomass map for the region, for 2007 and 2017 to obtain change information. Plot estimated mean AGB was 83.7 Mg ha-1 and I found significantly higher plot biomass in close proximity to protected sites and tidal influence, and the lowest in the sites where urbanisation was actively taking place. The mean LAI was 1.45 and there was a significant positive correlation between AGB and LAI (R2= 0.28). Satellite observations of NDVI for the growing season correlated positively with in-situ LAI (R2= 0.63) and AGB (R2= 0.8). Lower stem sizes (5-15cm) accounted for 70% contribution to the total biomass in disturbed plots, while undisturbed plots had a more even contribution of different size classes to AGB. Nipa palm invasion was significantly correlated to plots with larger variations in LAI (i.e. more patchy cover) and proportion of basal area removed within plots. The classification results showed SVM (overall accuracy 99.9 %) performed better than ML (98.7%) across the Niger Delta. I estimated a 2017 mangrove area of 794 561 ha and nipa extent of 11,419 ha. I discovered a 12% decrease in mangrove area and 694 % increase in nipa palm between 2007 and 2017. The highest radar-AGB relationship was from the combination of HH: HV and HV bands (R2= 0.62, p-value < 0.001). Using this relationship, I estimated a mean and total AGB of 90.5 Mg ha-1 and 82 X 106 Mg in 2007; 83.4 Mg ha-1 and 65 X 106 Mg in 2017. Local wood exploitation is removing larger stems (> 15 cm DBH) preferentially from these mangroves and creates an avenue for nipa palm colonisation. I identified opportunities to use remote sensing to estimate biomass, based on the LAI-AGB-NDVI relationship I found, and can serve as a calibration dataset for radar data to provide effective monitoring of mangrove forest degradation. It is clear from these results that remote sensing can be used to map the extent and changes in these land cover types, and thus such mapping efforts should continue for policy targeting and monitoring. I was able to show that mangroves of the Niger Delta are at risk, from rapid clearance as well as from the invasive species nipa palm. I also provide evidence of mangrove cover loss of 11 000 ha yr-1 over a decade, resulting in biomass loss rate of 100 Mg ha-1 yr-1 while mangrove degradation rate of 56 Mg ha-1 yr-1 in the Niger Delta. Assessing carbon stock of mangrove forests in the Niger Delta can create a baseline for regional conservation and regeneration plans. These plans can create opportunities for generating carbon credits under reducing emissions from deforestation and forest degradation (REDD+).

Accurate, fast detection of clouds in satellite imagery has many applications, for example Numerical Weather Prediction (NWP) and climate studies of both the atmosphere and of the Earth’s surface temperature. Most operational techniques for cloud detection rely on the differences between observations of cloud and of clear-sky being more or less constant in space and in time. In reality, this is not the case - different clouds have different spectral properties, and different cloud types are more or less likely in different places and at different times, depending on atmospheric conditions and on the Earth’s surface properties. Observations of clear sky also vary in space and time, depending on atmospheric and surface conditions, and on the presence or absence of aerosol particles. The Bayesian approach adopted in this project allows pixel-specific physical information (for example from NWP) to be used to predict pixel-specific observations of clear sky. A physically-based, spatially- and temporally-specific probability that each pixel contains a cloud observation is then calculated. An advantage of this approach is that identification of ambiguously classed pixels from a probabilistic result is straightforward, in contrast to the binary result generally produced by operational techniques. This project has developed and validated the Bayesian approach to cloud detection, and has extended the range of applications for which it is suitable, achieving skills scores that match or exceed those achieved by operational methods in every case. High temperature gradients can make observations of clear sky around ocean fronts, particularly at thermal wavelengths, appear similar to cloud observations. To address this potential source of ambiguous cloud detection results, a region of imagery acquired by the AATSR sensor which was noted to contain some ocean fronts, was selected. Pixels in the region were clustered according to their spectral properties with the aim of separating pixels that correspond to different thermal regimes of the ocean. The mean spectral properties of pixels in each cluster were then processed using the Bayesian cloud detection technique and the resulting posterior probability of clear then assigned to individual pixels. Several clustering methods were investigated, and the most appropriate, which allowed pixels to be associated with multiple clusters, with a normalized vector of ‘membership strengths’, was used to conduct a case study. The distribution of final calculated probabilities of clear became markedly more bimodal when clustering was included, indicating fewer ambiguous classifications, but at the cost of some single pixel clouds being missed. While further investigations could provide a solution to this, the computational expense of the clustering method made this impractical to include in the work of this project. This new Bayesian approach to cloud detection has been successfully developed by this project to a point where it has been released under public license. Initially designed as a tool to aid retrieval of sea surface temperature from night-time imagery, this project has extended the Bayesian technique to be suitable for imagery acquired over land as well as sea, and for day-time as well as for night-time imagery. This was achieved using the land surface emissivity and surface reflectance parameter products available from the MODIS sensor. This project added a visible Radiative Transfer Model (RTM), developed at University of Edinburgh, and a kernel-based surface reflectance model, adapted here from that used by the MODIS sensor, to the cloud detection algorithm. In addition, the cloud detection algorithm was adapted to be more flexible, making its implementation for data from the SEVIRI sensor straightforward. A database of ‘difficult’ cloud and clear targets, in which a wide range of both spatial and temporal locations was represented, was provided by M´et´eo-France and used in this work to validate the extensions made to the cloud detection scheme and to compare the skill of the Bayesian approach with that of operational approaches. For night land and sea imagery, the Bayesian technique, with the improvements and extensions developed by this project, achieved skills scores 10% and 13% higher than M´et´eo-France respectively. For daytime sea imagery, the skills scores were within 1% of each other for both approaches, while for land imagery the Bayesian method achieved a 2% higher skills score. The main strength of the Bayesian technique is the physical basis of the differentiation between clear and cloud observations. Using NWP information to predict pixel-specific observations for clear-sky is relatively straightforward, but making such predictions for cloud observations is more complicated. The technique therefore relies on an empirical distribution rather than a pixel-specific prediction for cloud observations. To try and address this, this project developed a means of predicting cloudy observations through the fast forward-modelling of pixel-specific NWP information. All cloud fields in the pixel-specific NWP data were set to 0, and clouds were added to the profile at discrete intervals through the atmosphere, with cloud water- and ice- path (cwp, cip) also set to values spaced exponentially at discrete intervals up to saturation, and with cloud pixel fraction set to 25%, 50%, 75% and 100%. Only single-level, single-phase clouds were modelled, with the justification that the resulting distribution of predicted observations, once smoothed through considerations of uncertainties, is likely to include observations that would correspond to multi-phase and multi-level clouds. A fast RTM was run on the profile information for each of these individual clouds and cloud altitude-, cloud pixel fraction- and channel-specific relationships between cwp (and similarly cip) and predicted observations were calculated from the results of the RTM. These relationships were used to infer predicted observations for clouds with cwp/cip values other than those explicitly forward modelled. The parameters used to define the relationships were interpolated to define relationships for predicted observations of cloud at 10m vertical intervals through the atmosphere, with pixel coverage ranging from 25% to 100% in increments of 1%. A distribution of predicted cloud observations is then achieved without explicit forward-modelling of an impractical number of atmospheric states. Weights are applied to the representation of individual clouds within the final Probability Density Function (PDF) in order to make the distribution of predicted observations realistic, according to the pixel-specific NWP data, and to distributions seen in a global reference dataset of NWP profiles from the European Centre for Medium Range Weather Forecasting (ECMWF). The distribution is then convolved with uncertainties in forward-modelling, in the NWP data, and with sensor noise to create the final PDF in observation space, from which the conditional probability that the pixel observation corresponds to a cloud observation can be read. Although the relatively fast computational implementation of the technique was achieved, the results are disappointingly poor for the SEVIRI-acquired dataset, provided by M´et´eo-France, against which validation was carried out. This is thought to be explained by both the uncertainties in the NWP data, and the forward-modelling dependence on those uncertainties, being poorly understood, and treated too optimistically in the algorithm. Including more errors in the convolution introduces the problem of quantifying those errors (a non-trivial task), and would increase the processing time, making implementation impractical. In addition, if the uncertianties considered are too high then a PDF flatter than the empirical distribution currently used would be produced, making the technique less useful.

Satellite images have been used as a complementary information for geological studies. In order to realise the maximum potential of satillite imagery, then improvements are needed, both in the visual presentation of such images, and in their automatic classification , in order to reveal the rock differences. Methods of processing imagery, were evaluated (band ratio, principal components, decorrelated stretch and maximum likelihood) and new (canonical regression, hue-saturation-intensity HSI transform, with modified manipulation, and watershed) were evaluated with respect to their ability to reveal rock differences. It was found that the HSI method gave the best results, both for visual presentation and automatic classifcation. This method has the ability to enhance both spectral and spatial information simultaneously without any data loss which is not the case in the other image enhancement methods (band ratio, principal components or decorrelation stretch). For automatic classification, the 'hue' images produced by the HSI transformation typically gave accurate (91%) classification of all the major rock types. Further, it was shown that the watershed method of classification was superior error rate = 9% to the maximum likelihood method (error rate 14%) for the same inputs. The new method of canonical regression was evaluated and although it was not very successful, the results were encouraging and it was concluded that this method may enable the estimation of the chemical composition of exposed rocks directly from satellite imagery.

A combination of factors including land use change and fire suppression has resulted in the loss of pine savanna habitats across the southeastern U.S., affecting many avian species dependent on these habitats. However, due to the ephemeral nature of the habitat requirements of many pine savanna species (e.g., habitat is only present for a couple of years after a fire), targeted management of such habitats can be challenging. Moreover, the growing numbers of imperiled pine savanna species can make prioritizing management difficult. One potential tool to better inform management of pine savanna species is satellite imagery. Sentinel-2 satellite imagery data provides an instantaneous snapshot of habitat quality at a high resolution and across a large geographic area, which may make it more efficient than traditional, ground-based vegetation surveying. Thus, the objectives of my research were to 1) evaluate the use of remote sensing technology to predict habitat occupancy for pine savanna species, and 2) use satellite imagery-based models to inform multispecies management in a pine savanna habitat. To meet my objectives, I conducted point count surveys and built predictive models for three pine savanna bird species: Bachman's Sparrow (BACS; Peuacea aestivalis), Northern Bobwhite (NOBO; Colinus virginianus), and Red-Cockaded Woodpecker (RCW; Dryobates borealis) across Georgia. I assessed the performance of satellite imagery in predicting habitat occupancy of these pine savanna species and its potential for multispecies management. I found that models created using satellite imagery habitat metric data performed well at predicting the occupancy of all three species as measured by the Area Under the Receiver Operating Characteristic Curve: BACS=0.84, NOBO=0.87, RCW=0.76 (with values between 0.7-1 defined as acceptable or good predictive capacity). For BACS and NOBO, I was able to compare these satellite imagery models to field-based models, and satellite models performed better than those using traditional vegetation survey data (BACS=0.80, NOBO=0.79). Moreover, I found that satellite imagery data provided useful insights into the potential for multispecies management within the pine savanna habitats of Georgia. Finally, I found differences in the habitat selected by BACS, NOBO, and RCW, and that BACS may exhibit spatial variations in habitat use. The results of this study have significant implications for the conservation of pine savanna species, demonstrating that satellite imagery can allow users to build reliable occupancy models and inform multispecies management without intensive vegetation surveying.

The Scottish Office Department of Agriculture and Fisheries Department has been concerned to know the area of bracken (Pteridium aquilinum ) in Scotland and how rapidly it is spreading. It offers a potential danger to animal health and a devaluation of land for agricultural practice. Bracken is present in patches which are often difficult to delimit because of variable density and appearance throughout the year and is not consistently identifiable at a single minimum spatial resolution. Bracken provided a candidate vegetation type for prototyping a knowledge-based system (KBS) which synthesized environmental data expressed spatially, with knowledge on vegetation dynamics and vegetation succession in particular. Temporal models of vegetation types, which have different degrees of reliability with analysis of static spatial data were incorporated. The knowledge-based system permits the user access to knowledge about botany, ecology, mapping, image processing and spatial data analysis for an informed and informative answer to their query. Bracken was mapped at three experimental sites using land survey, photogrammetric techniques and satellite image processing to quantify the extent of bracken and bracken change. An existing data model was enhanced by building-in the hierarchy of survey observations and associated error bands. The model was then implemented within an expert system shell with KBS routines written in POP-11 and Fortran. The coupling of the GIS and KBS routines was of the form of: 1. organisation of the transfer of descriptive values as parameters between the data-models of the KBS and the GIS routines and 2. read/write of the spatial data into the data-models for analyses by KBS routines.

High-resolution satellite imagery has become an essential data source for various geospatial applications, ranging from 3D reconstruction and topographic mapping to environmental monitoring and agricultural management. Moreover, multi-temporal satellite image sequences provide unprecedented opportunities to explore and monitor the dynamics of natural processes, like changes in land cover phenology. However, processing and extracting useful information from satellite imagery poses significant challenges. It requires technology capable of efficiently and automatically processing vast amounts of data while ensuring robustness against restrictive imaging conditions such as varying illumination, atmospheric influences, and occlusions due to clouds and shadows. In this thesis, novel methodologies have been developed that integrate state-of-the-art deep learning techniques into the processing workflow of satellite images, with a focus on enhancing the quality of two standard Earth observation products: digital surface models (DSMs) and temporally ordered, spatially co-registered sequences of ortho-images depicting the same geographic location. First, a data-driven DSM refinement method is developed based on an explicit 2D scene representation. The core of the method is a deep convolutional neural network that acts as a smart, learned post-processing filter and can seamlessly complement any stereo matching pipeline. The network learns to refine an imperfect input DSM by regressing a per-pixel correction to the height, using both the DSM and (ortho-rectified) stereo images as input. It removes artefacts of the preceding stereo algorithm, restores shape regularities such as straight and crisp crease edges, and completes missing surface details. Notably, the prior encoded in the network weights remains valid for DSM refinement under varying imaging conditions and for previously unseen geographic locations. Second, a novel method is introduced that enables the generation of high-quality DSMs directly from raw photogrammetric point clouds. The method leverages a neural representation of the 3D scene as an implicit, continuous occupancy field. Compared to learned DSM refinement schemes, the proposed implicit surface reconstruction approach excels at reconstructing fine-grained shape details such as dormers and produces exceptionally crisp and straight building edges. Third, a sequence-to-sequence model is designed that reconstructs coherent, complete, and gap-free Sentinel-2 image time series from a cloudy input sequence. The model combines 2D convolutions across the spatial and spectral dimensions and 1D self-attention across time to effectively capture an expressive prior over multi-spectral and multi-temporal reflectance patterns.

Tidal marshes are unique communities that are subjected to environmental stressors including sea level rise, salinity change, and drought, resulting in constant change. It is important to monitor these changing areas because of the ecosystem services they provide to us, such as protection from storms and carbon sequestration. The proposed work for this thesis project is focused on the study of tidal marshes and the dynamics between the vegetation species within them. The aim of this project is to use geospatial technology and analyses, along with machine learning classification methods, to monitor change in these valuable ecosystems. The Georgia coast is home to a large section of marsh on the Atlantic coast of the United States, and this project will take advantage of the benefits provided by previous work in this area. The two objectives of this study are to 1) examine multiple machine learning algorithms to determine the best supervised classification method for the Georgia coast, and 2) quantify the relationship between species-specific aboveground biomass of vegetation with ecotone movement between the three tidal marsh domains. Objective one of this study will compare two different supervised classification methods, Random Forest and Artificial Neural Networks, to determine which supervised classification performs best in mapping vegetation species and ground cover within the study area. In objective 2, the most accurate classifier will be used to examine ecotone movement over time and quantify the relationship between aboveground biomass of vegetation and ecotone movement.

Globally, tidal marshes cover about 90,800 km. Within the state of Georgia tidal marshes are primarily located behind the barrier islands and total 1,619 km2. The combination of high salinity environments and daily inundation, and being dependent on river output, make these dynamic systems. Tidal marshes provide numerous ecosystem services such as carbon and nitrogen sequestration, flood control, coastal protection, and numerous biogeochemical processes. Due to their unique position, tidal marshes are under threat from sea level rise, drought, coastal development, and large-scale disturbance events. Tidal freshwater marshes are especially susceptible to these threats due to their geographic location and small extent which have been historically understudied. By mapping tidal marshes, species composition is better understood and can be used to scale up ecosystem services, biogeochemical processes, and above ground biomass using remote sensing imagery. This study uses aerial orthoimagery along with a digital elevation model, National Wetland Inventory, and vegetation indices to map salt, brackish, and tidal freshwater marshes along the entire coast of Georgia. Higher spectral and spatial resolution PlanetScope 4- and 8-band satellite imagery was also used to map salt, brackish, and tidal freshwater marshes of the three main watersheds in coastal Georgia which include the Ogeechee, Altamaha, and Satilla Rivers. The aerial orthoimagery classification had an accuracy of 86.3% with salt marshes making up 67.8%, brackish 28.7%, and tidal freshwater 3.5% of the classified image and showed the importance of using a DEM and NWI for tidal marsh mapping. The PlanetScope classifications were comparable to the aerial classification with an accuracy of 86.5% (Ogeechee), 88.1% (Altamaha), and 75.9% (Satilla). Differences between the 4-band and 8-band PlanetScope imagery proved to be minimal. Due to the vulnerability of salt marshes to climate change, this study aims to contribute and expand upon current remote sensing studies on tidal marsh mapping.

Accurate sugarcane yield prediction is important for the sugar industry in serving the demands for decision-making systems such as harvest timing, product handling, and forward sales. Accurate yield modelling offers sugarcane farmers with a deeper knowledge of spatial and temporal crop variability, enhancing the quality and quantity of sugarcane yields while minimizing production costs and alleviating adverse environmental consequences. High-performance Machine Learning (ML) algorithms were applied to Remote Sensing (RS) images so that the timely acquired data with both spatial and temporal resolutions could be processed efficiently to interpret the complexity and variability of sugarcane yield. In this context, we tested advanced ML algorithms on diverse RS datasets such as Unmanned Aerial Vehicle (UAV), Sentinel-2, and Landsat-8 images, validated the results using ground measurements such as wet and dry biomasses, and crop yields; and developed a model that predicts sugarcane crop yield at the earliest possible growth stage with the least amount of spectral data. To demonstrate the scalability of the proposed yield prediction model, its performance was assessed in two regions: an experimental site in Queensland, Australia, and some sugarcane fields in Khuzestan Province, Iran. The predictive model was expanded using freely accessible Sentinel-2 satellite data so that it could be applied to a variety of sugarcane yield studies in various crop systems. For example, the expanded prediction model is particularly useful if ground data collection is limited, or UAV data is not feasible due to surveying costs. This research is anticipated to benefit agricultural producers and farmers in their decision-making and agricultural operation planning and help establish their management practises for optimal productivity.

Contemporary American streets are dominated by the automobile, both its movement and storage, which took root in the early 20th century and has been supported by transportation planners in the decades since. Pedestrian fatalities remain stubbornly high – even in cities with “Vision Zero” goals – and cars generate significant levels of both local air pollution and globally-relevant greenhouse gasses. Moreover, the default use of curbside lanes as on-street parking precludes a number of other street designs, such as wider sidewalks, protected bike lanes, transit bulbs and islands, parklets, rain gardens, and street trees. This dissertation probes these long-standing and consequential street layouts with methods that are both highly granular (accurate to the block level), and at the same time, comprehensive to an entire municipality. San Francisco, California, a dense, high-income, and nominally “progressive” American city serves as the case for three separate analyses: an in-person census of roughly 3,000 municipal bus stops, and review of satellite-imagery of nearly 6,400 intersections for the presence or absence of marked crosswalks, and the provision of on-street parking and its obstruction of a growing bike-lane network. Chapter 1 demonstrates that crosswalk provision varies dramatically across San Francisco, from Census tracts with 100% of intersections exhibiting marked crosswalks, to others with less than 10% coverage. Chapter 2 entails the largest bus-stop census of its kind, covering the entire SFMTA system. This approach finds that one third of all bus stops are blocked by on-street parking, and two thirds lack seating of any kind, as well as spatial clusters of stop amenities. Indeed, both high-quality bus stops and marked crosswalks are present to a far greater extent in the city’s northern half, unexplained fully by transit service, density, or pedestrian volumes. Lastly, Chapter 3 identifies fifty miles of angled parking in San Francisco, a format that roughly doubles on-street parking capacity, and decreases the opportunity for high-quality bicycle lanes. These studies provide both salient datasets to guide San Francisco’s street improvements (of which some progress at City Hall has already occurred), and serve as proof of concepts for planners and researchers elsewhere to probe their own blocks granularly and at scale. Indeed, this dissertation demonstrates new frontiers in applied urban analytics, and broadens and challenges the academic literature on transportation equity. It does so by expanding the types of street elements to be spatialized across neighborhoods, and demonstrates how the provision of seemingly-mundane features such as bus stops, crosswalks, and on-street parking can mirror (and exacerbate) other long-running investment disparities.

Abstract: Wildland fires are natural occurrences in the woodland landscape that play a vital ecological role in Canada's boreal forest region. However, they also endanger human life and can disrupt timber resources and other economic assets. Recently, wildfires have ravaged areas of British Columbia, Alberta, California, and other parts of North America, Europe, and Australia. Loss of human life, the economic repercussions in terms of suppression expenses and property damage have been staggering. Many of these fires have occurred near the wildland-urban interface, mostly natural regions increasingly subject to human development. As the population in these areas grows, there is a greater risk of economic impact and human loss. As a result, it is critical to provide an accurate classification of the green spaces as well as which areas pose the greatest risk of fire depending on the ignition sources found in the forested areas. Consequently, developing a fire risk assessment model can be used to effectively locate high risk areas/zones and form a foundational building block for conducting future research for fire prevention strategies or evacuation plans, and policy intervention. This model would also help in locating the low-risk areas/zones to be developed since they could add to the destructive consequences if ignored in the planning and expansion process. To develop an effective assessment model, this thesis has three main tasks. The first task is to provide a thorough review of existing literature, followed by background information that will help to build a contemporary, urban fire risk model. As a case study to form the basis of a fire risk model, this research uses anthropogenic, biologic, topographic, and climatic data from the City of Edmonton (CoE). This is then layered and mapped onto the City’s geographic location using ArcGIS and Python scripting language and then combined with data obtained from large datasets. The datasets in this research are satellite imaginary, aerial LiDAR dataset, urban Primary Land and Vegetation Inventory (uPLVI), and Road Weather Information Systems (RWIS) data, used to extract 12 variables that represent the fire risk assessment model. Fire risk assessment maps are subsequently generated by processing and analyzing all the datasets using the analytical hierarchy process (AHP) technique. The output of this research effort is a fire risk model that identifies the locations with the highest risk of wildfire within the CoE. Secondly, this study forecasts the wildland fire risk for 2050 and 2080, given the climatic projections from IPCC RCP4.5 and RCP8.5 (Representative Concentration Pathway). This analysis offers a better understanding of the forecasted climate change by highlighting transportation development and evacuation planning and integrating a multitude of data sources, including temperatures, precipitation, wind speed, and humidity levels. Results indicate that from 2021 to 2050, the fire risk may increase by almost 20%. Furthermore, the risk will increase by another 11% from 2050 to 2080 for the City of Edmonton. Finally, a comprehensive discussion that illustrates all the findings of the fire risk maps, current and forecasted, is presented. The fire risk map and the road map of the CoE are overlayed to facilitate insight into transportation development and evacuation planning. To help create a climate resilient municipality, an ecological vulnerability classification map is constructed to identify developable areas and areas that should remain under preservation. Since creating awareness for climate adaptation and zone resiliency is a shared goal among stakeholders, a brief discussion on the role of each stakeholder is provided. The discussion covers strategies for fire prevention and mitigation in high-risk areas/zones, as well as establishing several cornerstones for strategic planning and action to strengthen climate resilience and the transportation development foundation of urban communities.

Advancements in remote sensing technology have resulted in petabytes of remote sensing data being made publicly available. The widespread use of smart devices and GPS technology has also led to the availability of highly accurate geographical features. This increase in the amount of spatial data has allowed for greater research opportunities in many scientific domains including hydrology, political science, environmental science, and agriculture. In these applications, scientists rarely base all their analysis on a single dataset but they usually need to combine multiple datasets in their analysis. Machine learning is a popular tool used by these scientists and often requires combining different datasets into a form usable by the machine learning algorithms. Spatial data is generally available in two representations, raster, and vector. The best data science and machine learning applications need to combine multiple datasets of both representations which is a data and compute-intensive problem.My dissertation proposes a new system called Raptor that bridges the gap between raster and vector data. It is an end-to-end system for efficiently processing raster and vector geospatial data concurrently. First, it discusses an initial approach to parallelize the zonal statistics operation called DARaptor. Second, it proposes Raptor Zonal Statistics, a system implemented in Hadoop that can be used to perform the zonal statistics operation for big raster and vector datasets. Third, it proposes Raptor Join which is modeled as a relational join operator in Spark that can be easily combined with other operators, while also offering the advantage of in-situ processing. Raptor Join is flexible to support ad-hoc applications and has been used for various real-world applications such as wildfire modeling, area interpolation, and crop yield mapping. Finally, this work proposes RDPro to add distributed raster pre-processing capabilities to Raptor that can scale to big data. The experimental evaluation on large-scale satellite data with up to a trillion pixels, and big vector data with up to hundreds of millions of segments and billions of points has shown that the proposed system is promising and can scale to big data.

A practical method for calibrating LANDSAT multispectral scanner imagery from estuarine and coastal water areas in terms of the standard output quantity of near-surface irradiance reflectance ratio has been developed and tested. The algorithm consists of radiometric correction, atmospheric and surface corrections, and the calculation of reflectance ratio - found to be the most useful output parameter. The algorithm has been extended to produce a multi-date calibration in terms of suspended sediment concentration for imagery of the Solent area. This was obtained through statistical analysis of satellite-data-derived reflectance ratios and synoptic measured sediment concentrations. It achieved a correlation of 0.98 and the ability to predict sediment concentration to a root mean square error of 3.7 mg/l in a concentration range from 5.9 to 53.2 mg/l. The development of the algorithm includes the review of previous work, the study of the satellite system and data content and format, and an analysis of the theoretical arguments for atmospheric and surface corrections and subsurface optical processes including depth and sea bed reflection effects. Conclusions are reached about the expected size and source of errors in the calculation of reflectance ratio and the use of relevant standard image processing techniques - some of which were further developed in this work. The algorithm has been implemented at the NRSC, RAE Farnborough and applied to imagery from the Bristol Channel, the Mersey/Dee area, the Christchurch and Poole Bays area, and the Solent area - mainly for testing of the model rather than for oceanographic interpretation. Boatwork was carried out to try to test the algorithm by comparison of processed imagery with measured suspended sediment concentration and reflectance ratio.

The soil water repellency spatial and temporal dynamics remain ambiguous. Water repellency is an inherent soil property that refers to the impedance in dry soil wetting. This phenomenon was ascribable to the hydrophobic compounds coating the soil particles and has emerged as a recalcitrant issue impacting multiple processes upon agroecosystems. The apprehensions around soil water repellency include its impact on surface runoff, plant growth, and nutrients losses (e.g. phosphorus). The soil hydrophobic compounds, which are intrinsic constituents of the soil carbon pool, have different sources including plant leaves and roots, soil microbial communities and fungi. Previous methods for water repellency measurements are laborious, time-consuming and costly. The raison d'être of this thesis was to i) explore and test novel approaches for estimation of soil water repellency in pastoral ecosystems, and ii) study the factors controlling soil water repellency and assess its impact on surface runoff volumes and phosphorus losses in surface runoff. In the present work, multiple remote sensing approaches were tested to assess and map soil water repellency at multiple scales. The liaison between water repellency and soil surface reflectance was exploited to access the water repellency using the satellite multispectral reflectance and hyperspectral satellite data. A novel approach implicating the use of time series of surface reflectance and water deficit data was used to study the impact of both surface biomass and soil moisture temporal dynamics on the occurrence of water repellency and carbon content in pastoral systems. Multispectral broadband data from both Landsat-7 and Sentinel-2 satellites showed big potential for assessing soil water repellency and carbon content in permanent pastures. Partial least square regression models were calibrated and cross-validated using topsoil measurement of water repellency and soil carbon from 41 and 35 pastoral sites that were matched with reflectance spectra from Landsat-7 and Sentinel-2, respectively. Soil carbon showed higher predictability compared to water repellency with R2v=0.50, RMSEv=2.58 when using Landsat-7 spectra. The higher predictability performance for water repellency persistence was reached using Sentinel-2 spectral (R2v=0.45; RMSEv=0.98). However, using hyperspectral narrowband data from the Hyperion satellite showed a higher prediction accuracy (R2v=0.78; RMSEv=0.58). Prediction performance was generally higher when using the calibration sets, indicating the possibility of improving these prediction models when using larger datasets. A novel approach was tested using multiple predictors for soil water repellency occurrence. The predictors included time series of surface biomass assessed through normalised difference vegetation index (NDVI) and soil moisture data estimated through water deficit and synthetic aperture radar satellite data. The results showed an attractive opportunity for water repellency and soil carbon mapping. Three machine learning algorithms including artificial neural networks, random forest, and support vector machine were trained and cross-validated using multiple configurations of satellite time-series data and topsoil measurement from 58 pastoral sites. Random forest and support vector machine (RMSEv=0.82 and 0.87, respectively) outperformed artificial neural networks (RMSEv=1.23). With increasingly available remote sensing data, the use of satellite time-series data will open unprecedented opportunities for soil carbon, water repellency mapping, and potentially other functional chemical and physical soil attributes. To understand water repellency dynamics and evaluate their impact on surface runoff and phosphorus losses in pastoral soils, two experiments were conducted. The first experiment aimed to understand the relationship between the actual water repellency persistence and water content in drying hydrophobic soils. The second experiment had the objective to evaluate the impact of soil water repellency on the surface runoff and phosphorus losses in runoff. Results from the first experiment showed that the actual water repellency increased dramatically when water content decreased, especially when moisture dropped below a critical value. Using lab measurements, the actual water repellency was modelled using a simple sigmoidal model, as a function of water content, the potential water repellency, and two characteristic parameters related to the response curve shape. Results from the runoff trial showed that the surface runoff was influenced by soil water repellency to some extent (R2=0.46). Although more than 90 % of phosphorus losses happened in incidental losses following fertiliser application, the data point to non-incidental phosphorus loads being related to soil water repellency (R2=0.56). These results bespoke the effect of soil water repellency on background phosphorus losses through surface runoff during post-summer runoff events in pastoral ecosystems.

The remote sensing community has extensively used Pseudo-Invariant Calibration Sites (PICS) to monitor the long-term in-flight radiometric calibration of Earth-observing satellites. The use of the PICS has an underlying assumption that these sites are invariant over time. However, the site’s temporal stability has not been assured in the past. This work evaluates the temporal stability of PICS by not only detecting the trend but also locating significant shifts (change points) lying behind the time series. A single time series was formed using the virtual constellation approach in which multiple sensors data were combined for each site to achieve denser temporal coverage and overcome the limitation of dependence related to a specific sensor. The sensors used for this work were selected based on radiometric calibration uncertainty and availability of the data: operational land imager (Landsat-8), enhanced thematic mapper (Landsat-7), moderate resolution imaging spectroradiometer (Terra and Aqua), and multispectral instrument (Sentinel-2A). An inverse variance weighting method was applied to the Top-of- Atmosphere (TOA) reflectance time series to reveal the underlying trend. The sequential Mann–Kendall test was employed upon the weighted TOA reflectance time-series recorded over 20 years to detect abrupt changes for six reflective bands. Statistically significant trends and abrupt changes have been detected for all sites, but the magnitude of the trends (maximum of 0.215% change in TOA reflectance per year) suggest that these sites are not changing substantially over time. Hence, it can be stated that despite minor changes in all evaluated PICS, they can be used for radiometric calibration of optical remote sensing sensors. The new approach provides useful results by revealing underlying trends and providing a better understanding of PICS' stability.

Abstract: Detecting and interpreting image abnormalities and restoring images are essential to many processing pipelines in diverse fields. Challenges involved include randomness and unstructured nature of image artefacts (from signal processing perspective) and performance constraints imposed by resource-constrained systems (computational perspective). This thesis studies three such scenarios involving different sensor modalities, and proposes novel approaches to address such challenges using computer vision and machine learning techniques. The three scenarios are: GPU-friendly debanding for mobile HDR, MRI abnormality detection, and InSAR signal recovery for WAM. Below, we summarize each scenario, and outline the unique challenges involved in solving them. In the first scenario, in High-Dynamic-Range (HDR) imaging domain, we propose how to perceptually eliminate quantization artefacts resulting from dynamic range conversion, without distorting image contents, by adding noise patterns in the mobile Graphics Processing Unit (GPU) computing environment with limited computational resources. Traditional filtering methods are computationally non-ideal due to limited access to neighborhood information. We introduce a pixel based solution which does not rely on any neighborhood information. We impose a real-world transmission scenario where the receiving end cannot access the un-quantized (original) image. Most existing methods assume access to the original image, and this makes the image restoration problem easier to address. Our challenge is designing noise patterns that are perceptually pleasant, i.e. blended into the image content naturally based on the intensity profile and dynamic range conversion characteristics of the image. In the second scenario, we apply our methods to medical imaging: Magnetic Resonance (MR) images of preterm infant brains. We propose abnormality detection without using tissue priors (atlas) from a single acquisition sequence (T1-weighted). The rate of preterm births is increasing worldwide at an alarming rate. Preterm infant brains are at extremely high risk of developing abnormalities, which deter neuro-development. Traditional segmentation-based lesion detection methods rely on brain atlases to guide segmentation. For rapidly evolving preterm infant brains, such reliable atlases are unavailable, and low signal-to-noise ratio of small preterm infant brains complicates the restoration of tissue intensities. This motivates our solution design using more generic structural assumptions and heuristics for atlas-free WMI detection. In the third scenario, for Wide Area Monitoring (WAM) via ground movement prediction from noisy interferometric images, e.g. nation-wide monitoring for earthquake prediction or landslide prediction at mining sites, we propose image detail recovery through unsupervised learning-based filtering and filter output confidence prediction due to unavailability of clean training data. Filtering and confidence estimation are crucial steps in interferometric image processing pipelines. Challenges arise due to atmospheric factors, e.g. moisture, corrupting the images during acquisition. Also, corresponding pixels in time series images get decorrelated due to other undesired movements on the ground, e.g. moving vehicles on roads, vegetation and flowing water. Intermediate results of the filtering and confidence estimation steps can also influence the quality of the pipeline’s final result. Traditional algorithms in this area were not designed by prioritizing scalability via parallelism, without sacrificing accuracy, which are crucial for WAM, and addressed in this thesis. Thus, our main contribution is proposing novel solutions in scenarios where traditional image abnormality detection / restoration approaches are inadequate to address the signal processing and computational challenges involved.

A major issue with data-hungry deep learning algorithms is the lack of annotated ground truth for specific applications. In this thesis, we explore the challenges of applying artificial intelligence (AI) techniques on remote sensing data due to the unavailability of large annotated datasets for training in comparison to regular imagery data. We first tackle the problem of improving object tracking in Wide Area Motion Imagery data by using a semantic segmentation model to predict a building mask to filter false tracker points on buildings. The combination of AI image understanding techniques with tracker output significantly improved tracking quality. However, manual annotation of training data limited the performance of the semantic segmentation model.We propose a solution for this problem by developing a framework for automated annotation of remote sensing data. We pick satellite imagery as our data source, due to the high volume of Earth Observation Satellites available today coupled with crowd-sourced map data that can enable a new means for automated annotation of remote sensing data. In the second part of this thesis, we present a modular pipeline for collecting and labeling satellite imagery using geo-databases and OpenStreetMap data to facilitate training custom deep learning models. We demonstrate this approach by automatically collecting labeled imagery of solar power plants and building a classifier to detect the presence of such structures. The framework developed in this thesis can collect labeled satellite imagery of any object mapped by spatial databases creating use cases for geospatial and defense applications previously constrained by the lack of annotated datasets.

This dissertation consists of three essays that address the role of infrastructure in different sectors- education, manufacturing, and agriculture- in Sub-Saharan Africa. In the first essay, I estimate the impacts of access to electricity on student performance in the 9th and 12th grades examinations in Ghana. In Particular, I use nighttime satellite imagery to estimate the impact of luminosity on student performance. The results suggest luminosity matters the most when there is a considerable study time constraint. In the second essay, using the World Bank Enterprise Surveys, I investigate the link between access to infrastructure-electricity, transportations, and telecommunications- and the productivity of manufacturing firms and its role in firms’ location decision in 29 Sub-Saharan countries. These manufacturing firms are from seven different industries- Food, Textiles, Garment, Chemicals, Non-Metallic and Basic Metals, Fabricated Metals and Machinery, and Wood and Furniture. These relationships are less known in the context of Sub-Saharan African economies given the challenges for many countries in the region to build adequate infrastructure. The results shed lights on the importance of specific infrastructure in a particular industry. The last essay focused on the quality of trunk (major) roads and agricultural yield in Tanzania. Given the poor quality of roads in Sub-Saharan Africa, this essay tries to evaluate the impacts of recently rehabilitated trunk roads (2,500 Km) on agriculture yield in Tanzania- a country whose economy is largely based on agriculture led by small agricultural households.

Climate change will increase the frequency and intensity of many natural hazards such as storms or fires. Governments and public officials have the duty to protect their citizens and to prevent these hazards from turning into disasters. To do so effectively, they first need a scientific assessment of the disaster risk, which they communicate to affected people in the second step. This thesis examined both steps, by assessing past events of riverbank erosion in Bangladesh using radar imagery, and by investigating how the resulting disaster risk information could aid in risk communication. I could show that analyzing radar imagery on the Google Earth Engine (GEE) can assess riverbank erosion shortly after the end of the monsoon, and hence earlier than it would be possible with optical imagery. I developed an interactive online tool allowing the user to explore where riverbank erosion has occurred along Jamuna River in the last five monsoon seasons (2015-2019). Further, the source code of this tool is made publicly available, providing an option to apply the algorithm in other geographical settings. This can be attractive for authorities in low resource settings, given that the GEE can be used free of charge. In the second part of this study, I conducted an online experiment in seven coastal US states to investigate how aerial photographs containing information on past events of coastal erosion can help to decrease framing effects inherent in disaster risk communication. I found no framing effects in a risky choice situation and a goal framing setup, casting doubt on the strength of framing effects in real world scenarios. Adding the aerial photographs to the textual description of the scenario made respondents more risk seeking and increased their stated behavioral intention to take preventive measures against coastal erosion. The overall awareness among respondents about the issue of coastal erosion was high, resulting in few risky choices and high levels of stated behavioral intentions. Further research is required on the nature of framing effects in more realistic scenarios and on how people read and interpret aerial photographs containing information on past disaster events. In addition, I will test the findings from the online survey in a large scale household survey in Bangladesh.

Greater projected demand for food, fuel, and fiber will require substantial increases in global agricultural production over the next three decades. Climate change is also forecasted to make weather events more extreme and variable. Efficiency will become more important as demand for food products increases and the availability of fertilizer and land decreases. Technology may be of paramount importance for pushing the boundaries of production while remaining sustainable for generations to come. The first chapter of this dissertation investigated the importance of rate and timing of the plant growth regulator trinexapac-ethyl to malting barley in Virginia. Plant growth regulators can help plants remain upright during strong winds, thereby preserving grain quality and yield. However, this study demonstrated that risks of plant injury also exist. Application should be restricted to fields with greater risk of lodging and made only after the barley crop has broken dormancy and a substantial increase in air temperature is not forecasted in the week following application. Chapter two compared the efficacy of eight vegetation indices calculated from three satellites (Landsat 8, Sentinel 2, and Planet) for estimating cover crop biomass. Cover crops can have beneficial effects on agricultural land as well as groundwater and surface water, but only when adequate biomass is established to reduce erosion and nutrient leaching. Satellite imagery was able to estimate multi-species cover crop biomass more accurately than field-based sensors, although the most accurate vegetation index was dependent upon which satellite was being tested. Chapter three investigated the potential of Arabidopsis thaliana ipk1-, a loss-of-function mutant which exhibits decreased growth at elevated phosphorus concentration, for serving as in indicator of plant available phosphorus. An indicator crop could provide greater spatial resolution compared to soil testing, as well as represent plant available nutrients as opposed to chemically extracted nutrient estimations. Plant response exhibited a quadratic relationship with media P concentration in the range of fertilizer decision making for maize, providing valuable insight for potential yield response in agricultural fields below 'very high' phosphorus concentration.

Global policy challenges, such as addressing climate change and improving global health, require coordinated international efforts. Enacting such policies is often hampered by difficulties in quantifying the public benefit brought about by action. Without this information, policy decisions may focus primarily on the costs of action and thus disproportionately harm the groups for which inaction enacts the most damage. While classical econometric approaches are designed to identify policy impacts, they fall short when globally representative data is missing or when present or future contexts have no clear historical precedent. In these situations, new approaches are needed to incorporate evolving global data sources, such as satellite imagery, and process-based understanding from relevant fields, such as climate science or epidemiology. This dissertation leverages recent computational advances to develop and apply scalable tools for incorporating these data and models into policy impact estimation. Specifically, the first chapter describes a novel method for quickly extracting socio-environmental measurements using satellite imagery and machine learning. The second and third chapters describe applications of interdisciplinary methods to (a) project climate-driven changes to hurricane damages and (b) estimate the effect of COVID-19 policies on the pandemic's spread. In general, the findings suggest that the previously unmeasured benefits of global action in these contexts are substantial, and they demonstrate how such interdisciplinary approaches can better inform policy-making for complex global challenges.

Understanding migrants' lived experiences are integral to explorations of lifestyle migration. Examining mobility experiences over the entire life course enables insight into decision-making processes, how lifestyle migration may link to childhood experiences, as well as to life stage and tourism-related mobilities. Capturing the rich detail of migration(s) as lived experience, thus requires specialised research design that not only charts migration trajectories, but also synthesises the various components of migrants' mobile lives over their life course. The papers presented as part of this thesis draw from an innovative research design. Participants recalled their lived experiences over their life course using narratives, visual imagery and virtual visitation. One part of the method drew from in-depth biographical interview data with international and domestic migrants to Maroochy in Queensland, Australia. During the interviews, each participant narrated their residential moves over their life course. Our experiential journeys navigated multiple homes and places, lived and left as a migration biography. In positioning the places lived and left as important to the most recent migration decisions to move to Maroochy, I also reflected on my own migration biography - from Maroochy - and the influence of my situated knowledge. The visual and virtual components of the research method provided a platform to advance the use of geospatial technologies in presenting migrants' experiences. Using Google Earth satellite imagery, participants' migration narratives proceeded alongside their visitation of past residences, and their accompanying place-based stories. I recorded this synchronous combination of oral narrative with visual journey using audio and screen-capture software, and created mapped migration biographies combining the transcribed narratives and satellite imagery. I discuss the difficulties and opportunities for using geospatial technologies to represent migrants' experiences. Digital mapping added visual, multiscalar, tactile and mobile aspects to the migration narratives, providing opportunity for virtual visitation of past migrations. The participants interacted with the interface as a 'digital body', (re)performing their lived experiences of migration. A key thesis finding relates to how the method extended the capacities of oral and textual analysis to incorporate the visual features and spatialities of my participants' mobile lives.

Indonesian national park management utilises a zoning classification process, which is crucial to maintaining forest integrity, and ultimately providing the widest possible benefits to diverse human and non-human communities. Appropriate zoning management is essential in the regulation of spatial land use purposes, definition of natural resource use, and the imposition of land use limitations. As part of zoning development, current land cover information is fundamental for reliable environmental databases. These data represent the actual conditions of forest structure. Satellite imagery is a powerful resource for identifying land cover though it is not easy to provide updated information with this imagery. To fill this gap, a new approach with aerial surveys by Unmanned Aerial Vehicle (UAVs) has been adopted in this study to monitor national park forest coupled with GIS analysis techniques (Arc Map 10.6) to provide valuable and up to date information to evaluate zoning management in Aketajawe Lolobata National Park, Indonesia. A remotely piloted UAV captured high-resolution images from 687 hectares in a relatively more efficient time allocation (4 hours 19 minutes and 10 seconds) than conventional GPS surveys. This aerial survey clearly identified indigenous people's house distributions, their home range activities in Tayawi and some coconut plantations in Tabanalau and Akejawi. Following land cover analysis using Landsat 8, it was shown that Aketajawe primarily consists of primary forest at 65,590.3 ha (84.3%), followed by secondary forest at 9748.9 ha (12.5%), open area 2045.5 ha (2.6%), and water bodies 396.1 ha (0.5%). These data combined with several spatial data in order to generate spatial planning models for Aketajawe National Park. Two spatial models (zonal and eco-social) were developed to identify areas of priority in terms of ecological and social value in zoning classification for Aketajawe. In both models, the distribution and location of core, wilderness, and rehabilitation zone were relatively similar, even though the total area is slightly different, possibly due to the digitising process. However, in the spatial planning related to indigenous protection and livelihood, there is a different distribution of the traditional and utilisation zone in eastern and southern parts of the Aketajawe region. This study provides evidence that the combination of a zonal model and an eco-social model can be used for improved national park spatial planning, as more accurate and current information better represents actual field conditions. This study also demonstrated that the application of UAV technology for periodic forest surveys to update land-use patterns has the potential to contribute valuable information for national park programmes or policy evaluation.

INTRODUCTIONGiven that concentrated poverty is deepening around the world, the international development community now has even more reason to address this issue. Development aid has ostensibly served as an important policy instrument for promoting the welfare of marginalized communities in the Global South. The effectiveness of such efforts can be evaluated from varying angles but the first test to pass is its relevance to the lives of the most marginalized. This dissertation evaluates the extent to which aid activity is suited to the needs and priorities of recipients using three lenses: 1) needs assessment at the global level, 2) the design of interventions at the country level, and 3) evaluations at the sub-national level. The first chapter identifies the salient dimensions of poverty from the monetary and capability perspectives, using a cross-country analysis for 188 developing countries. The second chapter introduces a framework for analyzing two community development models in Myanmar as a country case study. The third chapter explores whether community development projects reach the poorest villages. It combines satellite imagery with spatial analysis to evaluate sub-national aid distribution. This study suggests strategies to deploy aid resources in a way to maximize their impact on people living in absolute poverty and data-sparse contexts. CHAPTER 1. THE DISCREPANCY BETWEEN TWO APPROACHES TO GLOBAL POVERTY: WHAT DOES IT REVEAL?For decades, development communities have attempted to develop poverty measures that can be used to inform need assessment and aid allocation. Building on these efforts, this paper examines the discrepancies between global poverty measures and brings that analysis to bear on identifying the salient dimensions of poverty in developing countries. It first compares the monetary and capability approaches to poverty and identifies comparable indices from each approach: the poverty headcount ratio (P0) and the multidimensional poverty headcount ratio (H). This paper then describes the degree of overlap and discrepancy between P0 and H for 118 developing countries from 2000 to 2014, synthesizing the Multidimensional Poverty Index (MPI), World Development Indicators, and OECD aid activity data. On average, there is a high correlation between the two poverty measures, but considerable discrepancies surface for some countries. I analyze the position of these countries with respect to the fitted line of the two measures, classifying them into income-poor and capability-poor countries. Countries such as Pakistan and Ethiopia, for example, are experiencing “capability poverty” while Uzbekistan and Zimbabwe are experiencing “income poverty.” I examine whether aid composition corresponds to the country’s relative income and poverty status, finding that capability-poor countries receive marginally higher social sectoral aid compared with economic sector aid. This study suggests that discrepancies between measures of international poverty can be used to target, monitor, and evaluate global aid distribution. CHAPTER 2 ALTERNATIVE MODELS OF COMMUNITY-LED DEVELOPMENT: IMPLICATIONS FOR POLICY AND PRACTICEReconciling the dual imperatives of legitimate state building and efficient service delivery, Community-led Development (CLD) has been praised as “a new form of engagement” in providing aid to fragile states. However, whether or how the CLD represents a new model of practice remains poorly understood. The absence of an analytical framework for distinguishing alternative CLD approaches to development aid hinders both the design of context-specific interventions and the evaluation of their impacts. This dissertation aims to compare two alternative models of CLDs against a backdrop provided by the framework of community-led development. Using document reviews and stakeholder interviews, this paper analyzes two aid projects in Myanmar: the Korean government-supported Saemaul Undong (SMU, New Village Movement), which reflects the perspective of the developmental state, and the World Bank-supported National Community-Driven Development Project (NCDDP), which reflects the perspective of the revised neoliberalism. Next, this study proposes the Agency-Power-Dimension (APD) framework for use in describing donors’ general CLD aid policies in conjunction with specific CLD projects in Myanmar. The Agency-Power-Dimension (APD) framework is proposed to describe donors’ general CLD aid policies in conjunction with specific CLD projects in Myanmar. This study finds that the intervention strategies of SMU and NCDDP differ regarding the main agency of change, the handling of power, and the objectives of projects. SMU engages with government extension workers as the main change agent, and its accountability comes from the performance of projects that focuses on agricultural production. In contrast, NCDDP works with private facilitators, emphasizing the processes of inclusion in the context of public infrastructure development. Previously, impact evaluations of CLDs set hypotheses based on the logical progression of the projects whose indicators are diffused over broad socio-economic domains. The APD framework identifies the main facets of treatment arms in future experimental studies. Policymakers seeking development opportunities in other fragile states can compare East Asian/Southern and Western/Northern approaches and apply it to varying local conditions. CHAPTER 3: MAPPING COMMUNITY DEVELOPMENT AID: SPATIAL ANALYSIS IN MYANMARAid policy has the potential to alleviate global poverty by targeting areas of concentrated need. However, few aid-determinant studies have analyzed the characteristics of poverty at the sub-national level, and even those studies were conducted with their units of analysis at a high administrative level such as the state. This study intends to fill this knowledge gap by portraying poverty at the granular level, and promoting the evaluation of aid towards the most marginalized communities. The goal of this study is to explore the extent to which community-led-development (CLD) projects take place in poor villages, using the case of Myanmar. It also analyzes how two CLD models, National Community-Driven Development Project (NCDDP) and Saemaul Undong (SMU) target needs differently. To collect outcome variables, I develop web scraping algorithms to create comprehensive and up-to-date locations of CLD participating villages (n=12,282). As for exploratory variables, radiance values from nighttime satellite imagery are extracted to estimate wealth at the community level. In addition, I spatially interpolate the DHS wealth index to make inferences on poverty in aid sites. By geospatially matching aid and wealth related data, I test factors that explain variation in the distribution of CLD and different approaches to community development. The results show mixed evidence of poverty-oriented targeting. First, as each increment of the share of a vulnerable population rises, the likelihood of aid presence in that community declines by 4%. Next, the density of community development projects is higher in areas shining brighter. A one unit increase in the nightlight intensity increases the number of projects by 86 within a two-degree radius of a DHS village cluster. Among villages of similar levels of nightlights and population, however, aid goes to areas with lower assets. Last, NCDDP, which emphasizes inclusion and collaboration, supports poorer villages farther away from conflict events. In contrast, SMU, which considers competition conductive to performance, supports more established areas including villages near conflict zones. Unlike previous studies finding that state-level aid allocations favor the richest, this more fine-grained analysis suggests that a need-based allocation is also in place. The nuances captured in nightlight luminosity are also shown to improve predictions of aid distribution. Synthesizing new sources of data can be used to assess area-based interventions in the context of poverty and conflict where traditional survey is too costly. CONCLUSION This study draws attention to alternative forms of evidence-based targeting, design, and evaluation of aid from poverty-oriented perspectives. The first chapter reveals that there are 1.5 times more capability poor countries than income poor ones, and the capability poor countries receive marginally higher social sectoral aid relative to economic sector aid. The second chapter finds that the intervention strategies of the revised neo-liberal (NCDDP) and the developmental state (SMU) model differ in terms of the main agency of change, the handling of power, and the primary dimension of projects. The third chapter highlights that community development aid in Myanmar flows to villages with low assets but also with higher nightlight luminosity and a lower proportion of vulnerable populations. These three chapters also speak to the evolution of an aid landscape with a distinctive way of delivering aid and generating empirical evidence. This study concludes with a call for both research and practice to return to the basics, and to begin by considering client and user needs. Grounding development policy in more contextualized knowledge, the development community can better serve the “bottom billion.”

As urbanization increases in cities prone to earthquakes, increasing disaster resilience, or the ability to absorb shock of the disaster, is increasingly important to preserve the integrity of critical infrastructure and save human lives. This study explores the response and resilience of Haiti following the 2010 M7.0 earthquake. Traditional methods of measuring resilience following a major earthquake require census data. Census data is seldom available at a great level of detail. As an alternative to census data, satellite imagery provides an objective measurement of the history of the earth, consistent both in temporal and spatial resolution. The currently available Very High Resolution (VHR) remote sensing sensors observe objects on the ground as small as 0.3 meters. The additional dimensions of volume and shape of the buildings provide the ability to distinguish building functions when compared to the traditional two-dimensional data. From the land cover and land use classification results for each year, a time series analysis analyzes the changes through the years of the individual buildings and building types. Using the building type classification results, the changes in resilience indicators are analyzed by year. Elasticity, amplitude, and malleability are the three indicators used to measure resilience. Elasticity refers to the recovery duration of the city to a stable state after the earthquake; Amplitude refers to the changes in the built-up area caused by the earthquake, essentially how much the city is impacted by the earthquake; finally, malleability refers to the city’s new development after the earthquake, compared to its original state. The results are compared to census data to illustrate the correlations between the observed dynamics and the given data, as well as to draw conclusion about the recovery processes.Using satellite images to characterize the resilience of a built-up area is feasible, and change detection analysis can be used to do a change series analysis of imagery spanning an earthquake. This analysis found that the recovery of the building system in Haiti could be characterized as elastic, as it increased to pre-shock levels following the earthquake, and malleable because there was not a net negative difference in the building area; Looking at amplitude, the city did not reach the threshold level of strain that prevented return to the original state. While the definitions of the three resilience indicators would lead a reader to believe that Haiti was resilient to the impact of the 2010 earthquake, future work needs to be done to disaggregate the effects of increasing urbanization and development that were occurring regardless of the earthquake from the impact of the earthquake itself.Overall, this project provides a case study that there is capacity to carry out traditional resilience analysis in areas without substantial census data, but that more datasets in the time series analysis are necessary in the years prior to the earthquake. The methods can be improved to be applied in the case of future global earthquakes, addressing questions of resilience in a more comprehensive and timely manner.

Sierra Nevada forests have provided many economic benefits and ecological services to people in California, and the rest of the world. Dramatic changes are occurring in the forests due to climate warming and long-term fire suppression. Accurate mapping and monitoring are increasingly important to understand and manage the forests. Light Detection and Range (LiDAR), an active remote sensing technique, can penetrate the canopy and provide three-dimensional estimates of forest structures. LiDAR-based forest structural estimation has been demonstrated to be more efficient than field measurements and more accurate than those from passive remote sensing, like satellite imagery. Research in this dissertation aims at mapping and monitoring structural changes in Sierra Nevada forests by taking the advantages of LiDAR. We first evaluated LiDAR and fine resolution imagery-derived canopy cover estimates using different algorithms and data acquisition parameters. We suggested that LiDAR data obtained at 1 point/m2 with a scan angle smaller than 12°were sufficient for accurate canopy cover estimation in the Sierra Nevada mix-conifer forests. Fine resolution imagery is suitable for canopy cover estimation in forests with median density but may over or underestimate canopy cover in extremely coarse or dense forests. Then, a new LiDAR-based strategy was proposed to quantify tree growth and competition at individual tree and forest stand levels. Using this strategy, we illustrated how tree growth in two Sierra Nevada forests responded to tree competition, original tree sizes, forest density, and topography conditions; and identified that the tree volume growth was determined by the original tree sizes and competitions, but tree height and crown area growth were mostly influenced by water and space availability. Then, we calculated the forest biomass disturbance in a Sierra Nevada forest induced by fuel treatments using bi-temporal LiDAR data and field measurements. Using these results as references, we found that Landsat imagery-derived vegetation indices were suitable for quantifying canopy cover changes and biomass disturbances in forests with median density. Large uncertainties existed in applying the vegetation indices to quantify disturbance in extremely dense forests or forests only disturbed in the understory. Last, we assessed vegetation losses caused by the American Fire in 2013 using a new LiDAR point based method. This method was able to quantify fire-induced forest structure changes in basal area and leaf area index with lower uncertainties, compared with traditional LiDAR metrics and satellite imagery-derived vegetation indices. The studies presented in this dissertation can provide guidance for forest management in the Sierra Nevada, and potentially serve as useful tools for forest structural change monitoring in the rest of the world.

In this study, I tested the potential for remote sensing data with a high spatial resolution to model breeding forest bird species and their habitat at a fine spatial scale. The research took place on ridgetops in a large, relatively contiguous Appalachian mature deciduous forest in northwestern WV, USA. The remote sensing data sources were a leaf-on QuickBird satellite image (0.6-m panchromatic and 2.4-m multispectral) and a 3-m digital elevation model (DEM). For the first part of the study, I extracted spectral and textural measures from the satellite image and terrain information from the DEM. I then used these data to analyze avian community survey and habitat data collected at circular plots (n = 68) distributed across the ridgetops. The primary results of this analysis indicated that the satellite image provided information about trends in forest composition and structure across the study site, and further that a relatively simple plot-level measure of image texture (the panchromatic pixel standard deviation calculated at plot radii of 50 and 100 m) was a useful proxy of environmental heterogeneity for predicting the distributions of certain forest canopy gap-dependent bird species. For the second part of the study, I analyzed the habitat and remote sensing data at a finer spatial scale to develop remote sensing-based indices of forest structure and composition. These indices provided further insight into local variation in forest characteristics (e.g., in relation to topographic aspect) on the ridgetops. I also tested these indices, the DEM, and anthropogenic forest edge for modeling the breeding territory distributions of three focal species (Cerulean Warbler, Setophaga cerulea; Hooded Warbler, S. citrina; and Ovenbird, Seiurus aurocapilla) mapped over ~11 km of ridgetop transects. These models indicated the importance of local influences of terrain (e.g., east-facing aspects for Cerulean and Hooded Warbler, west-facing aspects for Ovenbird, and knolls for Cerulean Warbler), and forest edges (positive for Cerulean Warbler and negative for Ovenbird) on their distributions. Among the remotely-sensed indices, the index of forest structural complexity was primarily useful as a strong predictor of the distribution of the canopy gap-dependent Hooded Warbler. For the third and final part of the study, I used the locations of singing males of the three focal species collected across a greater extent of the site (~28 km of ridgetop transects) in point pattern analyses that incorporated the remote sensing data and the potential for intraspecific interactions (attraction and repulsion) between neighboring individuals. The results of these analyses supported that intraspecific interactions in addition to environmental influences as indicated by the remote sensing data explained the species’ fine-scale distribution patterns. While the individuals of all three species exhibited regular spacing over short distances that was consistent with competition for territorial space, Cerulean Warbler individuals exhibited more clustering than could be statistically accounted for by the remote sensing data, suggesting the importance of conspecific attraction in its distribution. In summary, my findings supported the potential application of fine-scale remote sensing data for purposes such as complementing coarse-scale environmental data (e.g., land cover maps) in predicting forest breeding bird species distributions, and for comparative analyses of the local spatial distributions of these species. The capacity for remote sensing data to provide useful environmental information at a fine spatial scale is likely to improve as the technology continues to develop.

Land Cover Land Use (LCLU) change affects land surface processes recognized to influence climate change at local, national and global levels. Soil organic carbon is a key component for the functioning of agro-ecosystems and has a direct effect on the physical, chemical and biological characteristics of the soil. The capacity to model and project LCLU change is of considerable interest for mitigation and adaptation measures in response to climate change. A combination of remote sensing analyses, qualitative social survey techniques, and biogeochemical modeling was used to study the relationships between climate change, LCLU change and soil organic carbon in the semiarid rural zone of Senegal between 1960 and 2050. For this purpose, four research hypotheses were addressed. This research aims to contribute to an understanding of future land cover land use change in the semi-arid West African Sahel with respect to climate variability and human activities. Its findings may provide insights to enable policy makers at local to national levels to formulate environmentally and economically adapted policy decisions. This dissertation research has to date resulted in two published and one submitted paper.

We develop two different descriptors which can be utilized to describe satellite imagery. The first, the differential-magnitude and radius descriptor, describes a scene by computing the directional gradient of the scene with respect to a vector field whose solutions are circles around a pixel to be described, and then counts pixels in a descriptor matrix according to the magnitude of this gradient and the distance at which this magnitude occurs. The second, the radial Fourier descriptor, extracts from the scene a sequence of annuloid sectors, and uses this to approximate the behavior of the image on a circle around the point to be described. The fast Fourier transform is then used to obtain a description of this function in the frequency domain; the absolute values of these complex-valued frequencies form the descriptor. A set of data to test and perform parameter selection for these procedures using 79 Landsat 8 imagery scenes was constructed. A cellular evolutionary algorithm was then utilized for parameter selection by training and testing support vector machine classifiers using LIBSVM from this dataset utilizing classification accuracy as a fitness function. We then analyze the classification success associated with the two methods equipped with their optimized parameters.

From space, the Earth can be observed over time. Satellite imagery has documented human influences on the landscape. I examined the effect of economic fluctuations on the landscape across the urban to rural gradient in Southeastern Michigan. Using Landsat satellite imagery, I described and compared land-cover changes observed from Landsat during 2001 – 2005, a five-year period before the Great Recession, to those observed during the period 2007 – 2011, a five-year period during and after the recession. I used dense time-series satellite observations to observe changes in impervious surfaces, compared these over time, and related them to socioeconomic variables collected by the US Census Bureau over census tracts. The results suggest that a suite of socioeconomic factors and landscape characteristics influence the amount of impervious surface change occuring within each census tract. On average impervious surface areas did not increase at a faster rate during the period before the recession, decrease at a slower rate, or change from increasing to decreasing, when compared to the post-recession period. In addition, the socioeconomic composition of individual communities was strongly associated with how the landscapes changed through time and space. Overall, I demonstrate that the socioeconomic characteristics of communities have land use, ecological, and carbon sequestration implications.

The last decade saw major breakthroughs in geospatial image technology in both the spaceborne and in airborne domains. The commercial use of high-resolution satellites was enabled by the Presidential Decision Directive 23 issued by the Clinton Administration in 1994. High-performance large-format digital cameras have been replacing traditional large-format aerial film cameras. Both technologies are digital providing high image quality as well as efficient data acquisition capability. However, there are still significant technical challenges in exploiting these images in achieving high-accuracy 3D mapping performance. An ongoing challenge in high-resolution satellite image (HRSI) processing has been the establishment of accurate epipolar geometry which is a pivotal in any stereo-image processing for 3D topographic mapping. However, the pushbroom camera of most high-resolution satellites does not produce straight epipolar lines. Furthermore, the epipolar curve pair does not exist for the entire stereo image area. These properties make it difficult to establish the epipolar geometry of a pushbroom camera limiting accuracy of the epipolar image resampling. In this study, a new method of epipolar curve pair determination and epipolar image resampling of spaceborne pushbroom imagery is proposed. First, the new method determines the epipolar curve pairs. Next, accurate epipolar image resampling is achieved by arranging the determined epipolar curve pairs to satisfy the epipolar resampled image condition. In this research, the characteristics of epipolar curves in the image were investigated and the proposed approach was validated. The experimental results showed that the new method could not only successfully establish the epipolar curves over the entire stereo HRSI enabling the accurate epipolar image resampling, but showed superior results with better accuracy than the conventional method. In airborne imaging systems, alternative automated georeferencing is required to serve as a backup in case when GPS/INS-based (Global Positioning System/ Inertial Navigation System) georeferencing is not feasible. This is particularly important when the GPS signal availability is not assured jeopardizing the mapping accuracy; note that GPS is subject to interference and jamming. In this study, the potential of stereo HRSI as ground control information is investigated for georeferencing of aerial imagery. The use of a stereo HRSI can provide the 3D ground control information by restitution avoiding the apparent leaning of the on-ground objects in HRSI introducing positional errors. A robust and efficient image matching scheme based on SIFT (Scale-Invariant Feature Transform) and RANSAC (RANdom SAmple Consensus) was developed to handle image differences between aerial images and stereo HRSI. In the georeferencing step, the bundle adjustment with rigorous outlier removal is utilized. The proposed method was tested for a strip of aerial images and a stereo IKONOS images. The matching scheme was able to successfully extract 3D ground controls from the stereo IKONOS by mitigating the outliers. Georeferencing of aerial images that lack distinct features such as forest was possible by utilizing tie points to adjacent aerial images. The fully-automated approach has potential not only for aerial image georeferencing practically all over the world, including inaccessible areas, but also provides a good basis for image-based aerial navigation under GPS denied-conditions.

Excavating cairns in southern Arabia is a way for anthropologists to understand which factors led ancient settlers to transition from a pastoral lifestyle and tribal narrative to the formation of states that exist today. Locating these monuments has traditionally been done in the field, relying on eyewitness reports and costly searches through the arid landscape.In this thesis, an algorithm for automatically detecting cairns in satellite imagery is presented. The algorithm uses a set of filters in a window based approach to eliminate background pixels and other objects that do not look like cairns. The resulting set of detected objects constitutes fewer than 0.001% of the pixels in the satellite image, and contains the objects that look the most like cairns in imagery. When a training set of cairns is available, a further reduction of this set of objects can take place, along with a likelihood-based ranking system.To aid in cairn detection, the satellite image is also clustered to determine landform classes that tend to be consistent with the presence of cairns. Due to the large number of pixels in the image, a subsample spectral clustering algorithm called "Multiple Sample Data Spectroscopic clustering" is used. This multiple sample clustering procedure is motivated by perturbation studies on single sample spectral algorithms. The studies, presented in this thesis, show that sampling variability in the single sample approach can cause an unsatisfactory level of instability in clustering results. The multiple sample data spectroscopic clustering algorithm is intended to stabilize this perturbation by combining information from different samples. While sampling variability is still present, the use of multiple samples mitigates its effect on cluster results.Finally, a step-through of the cairn detection algorithm and satellite image clustering are given for an image in the Hadramawt region of Yemen. The top ranked detected objects are presented, and a discussion of parameter selection and future work follows.

Shoreline mapping is of critical importance for coastal communities. It supports nautical charting, coastal zone management, legal boundary determination, analyzing shoreline erosion and other threats of climate change, and a host of other applications. A tide-coordinated shoreline is representative of the shoreline at certain tidal stage, e.g., Mean Lower-Low Water (MLLW) or Mean High Water (MHW). Since the establishment of the Survey of the Coast in 1807, mapping of tide-coordinated shorelines has been a challenging operation. In the past, analog devices, such as plane tables and telemeter rods, had been used in mapping the tide-coordinated shoreline for nautical chart production. Although shoreline mapping from such techniques provided high-quality shoreline, the procedure requires massive amount of time, manpower and budget. The standard method to map tide-coordinated shorelines was later shifted to aerial photogrammetry in 1927. Advantages of conducting coastal surveys using aerial photogrammetry over the conventional ground survey include enabling surveying of large areas in a short time, while maintaining good accuracy. Today, approaches to tide-coordinated shoreline involve recent advances in technologies and methods, including global positioning system (GPS), light detection and ranging (LIDAR), satellite imagery, and auto-feature extraction from imagery. Additionally, LIDAR has currently begun to be utilized in NOAA’s national shoreline production for MHW shoreline, along with MLLW shoreline from photogrammetric compilation deriving tidally-referenced aerial imagery. This thesis discusses coastal mapping and surveying work focused on approaches to tide-coordinated shoreline. A brief history of shoreline mapping in the United States, dating back to the early era of U.S. coastal survey, is presented as well as current standard procedures of tide-coordinated shoreline mapping, including quality of work and efficiency of the work process. Definitions of a shoreline and some of its variations, usually misinterpreted and misused, are addressed to give a basic understanding. Implementations of modern technologies and methods in derivation of tide-coordinated shoreline will then be reviewed and discussed regarding the accuracy and efficiency of the shoreline products. Finally, trends and developments of tide-coordinated shoreline mapping approaches regarding advancements of technologies in the future are analyzed and discussed in the thesis.

In a wide range of applications, images convey useful information about scenes. The “utility” of an image is defined with reference to the specific task that an observer seeks to accomplish, and differs from the “fidelity” of the image, which seeks to capture the ability of the image to represent the true nature of the scene. In remote sensing of the earth, various means of characterizing the utility of satellite and airborne imagery have evolved over the years. Recent advances in the imaging modality of spectral imaging have enabled synoptic views of the earth at many finely sampled wavelengths over a broad spectral band. These advances challenge the ability of traditional earth observation image utility metrics to describe the rich information content of spectral images. Traditional approaches to image utility that are based on overhead panchromatic image interpretability by a human observer are not applicable to spectral imagery, which requires automated processing. This research establishes the context for spectral image utility by reviewing traditional approaches and current methods for describing spectral image utility. It proposes a new approach to assessing and predicting spectral image utility for the specific application of target detection. We develop a novel approach to assessing the utility of any spectral image using the target-implant method. This method is not limited by the requirements of traditional target detection performance assessment, which need ground truth and an adequate number of target pixels in the scene. The flexibility of this approach is demonstrated by assessing the utility of a wide range of real and simulated spectral imagery over a variety ii of target detection scenarios. The assessed image utility may be summarized to any desired level of specificity based on the image analysis requirements. We also present an approach to predicting spectral image utility that derives statistical parameters directly from an image and uses them to model target detection algorithm output. The image-derived predicted utility is directly comparable to the assessed utility and the accuracy of prediction is shown to improve with statistical models that capture the non-Gaussian behavior of real spectral image target detection algorithm outputs. The sensitivity of the proposed spectral image utility metric to various image chain parameters is examined in detail, revealing characteristics, requirements, and limitations that provide insight into the relative importance of parameters in the image utility. The results of these investigations lead to a better understanding of spectral image information vis-à-vis target detection performance that will hopefully prove useful to the spectral imagery analysis community and represent a step towards quantifying the ability of a spectral image to satisfy information exploitation requirements.

As world population increases the demand for accessible water resources also increase, particularly in developing nations. Groundwater deposits are the largest source of available fresh water on the planet; however, exploration for such deposits is expensive and time consuming. Many recent studies have demonstrated the vast potential of the employment of satellite imagery in the search for reliable groundwater sources. This thesis made use of two freely available satellite datasets, LANDSAT TM+7 and SRTM, to explore the Central Plateau Region of Haiti for shallow subsurface water. The satellite imagery was used to create seven GIS terrain analysis layers: slope, aspect, curvature, planform and profile curvature, catchment area, and compound terrain index; and two remote sensing images: linear features overlay and thermal signature. These layers each represent independent variables used in this research. Seventy-six wells within the Central Plateau Region were overlaid onto each layer to determine a relationship drilling well success rate and the respective variable. Descriptive statistics performed served to outline the variability and data range for each variable. Categorizing the data was also used to compare the distribution of data overall the range of the variable to determine optimal conditions. The GIS terrain analysis layers were categorized to related findings from generated histograms to the study area. The remote sensing images were used to determine the drilling success rate of the two variables.The results of the analysis showed three relationships between drilling success rate and proximity to a linear feature, thermal signature digital number, and planform curvature. As the distance from the nearest linear feature increases the probability of drilling a successful well decreases. The probability of drilling a successful well increases as the thermal signature digital number value decreases. The more concave the value of planform curvature the higher the potential for drilling a productive well. These three relationships indicate that groundwater pumped out of hand-pump wells is accumulated precipitation water caught in an unconfined aquifer. The other spatial variables did not show statistically significant relationships which may be a result of many unforeseen factors including the depth, productivity reading, accuracy of well location, and the spatial resolution of the satellite imagery used was not appropriate for the relative size of the groundwater deposits. Continued research into the interactions of the hydrologic cycle, geology, and geomorphic features need to be undertaken to more accurately utilize satellite imagery in areas like the Central Plateau of Haiti.

This dissertation analyzes alterations in a closed-basin wetland system resulting from the construction of a large-scale irrigation project in its catchment. The study was conducted at the Sultan Marshes ecosystem (Develi Basin, Turkey), which has been severely degraded within the last 20 years due to diversion of its major water sources for agricultural irrigation. Spatial changes in the Sultan Marshes from 1980 to 2003 were analyzed using satellite remote sensing. Changes in the areal coverages of lakes, marshes, agricultural, and steppe areas determined by unsupervised classification of four Landsat images showed that both lakes and marshes became smaller after construction of the irrigation project. Steppe areas expanded onto wetlands. Significant portion of northern (Kepir) marshes were converted to agriculture. Hydrologic changes in the Sultan Marshes were analyzed statistically and used to develop a dynamic hydrologic model of the system. Water levels dropped more than one meter in the lakes and marshes from 1993 to 2003, and decreases were observed in ground-water levels and spring flows, although precipitation and evaporation rates remained mostly stable. Simulations with the hydrologic model showed that even if surface water continues to be used for irrigation, reductions in appropriations from ground water and springs would restore and protect water levels in the marshes. Agricultural and environmental changes in the Develi Basin were analyzed after the irrigation management was transferred from state to "irrigation associations" in 1994. The analyses showed that irrigated areas and water use in the Develi Basin showed significant fluctuations. The area allocated to production of high water-consuming plants increased. Water fee collection rates were lower than 100%. Although soil and water quality in the Develi Basin did not change significantly, ground-water levels, flow rates from springs and water levels in the Sultan Marshes all dropped. Four recommendations were developed that would help to resolve the conflict between agricultural and wetland water requirements: (1) a basin-wide approach water planning, (2) more realistic water pricing, (3) demand-based irrigation scheduling, and (4) rehabilitation of the irrigation system. Economic costs and benefits associated with water diversions from agriculture to the wetlands were estimated, and the optimum or economically-efficient amount of water diversion was determined. When only direct-use values of the wetland (reed cutting, animal grazing and ecotourism) were included, the annual optimum amount of water diversion to the wetlands was found to be 5.2 million m 3 yr -1 (165 L s -1 ) compared to about 62 million m 3 yr -1 (1,957 L s -1 ) used in irrigation. Diversion of 5.2 million m 3 yr -1 water would be sufficient to restore the conditions in the marshes. The analysis showed that economically-efficient restoration of water levels in the Sultan Marshes is feasible with moderate amounts of water diversion.

The sea breeze circulation forms as a result of differential heating across the land-sea surface interface. The resultant pressure gradient force (PGF) induces an onshore flow at the surface and a return flow higher in the atmosphere. The sea breeze front is a reflection of this circulation, a boundary between the cool, moist, maritime air mass advancing landward and the warm, dry ambient air mass in place over inland areas. Studies have shown that the circulation forms more often in the spring and summer months when the temperature difference between land and sea surfaces is greatest. The following study is based on the analysis of satellite imagery and standard hourly measurements of air temperature, dewpoint, wind speed, and wind direction recorded at four coastal sites: Savannah, GA, Charleston and Myrtle Beach, SC, and Wilmington, NC. One of the objects of this study is to establish specific values associated with the changes induced by passage of the sea breeze front, and to examine differences in the station-to-station incarnation of the sea breeze circulation. Variability from station to station in the nature and timing of sea breeze frontal passage was found to be a function of relative proximity to the coast. For example, sea breeze frontal passage was found to occur earliest at Myrtle Beach (the closest station to the coast), around 1300 LST on average. Savannah, the farthest of the four coastal sites from the water, was affected by the sea breeze last, with an average passage time of between 1630 and 1700 LST. Previous studies indicate that the extent to which the circulation and associated front penetrate inland is usually on the order of about 20-60 km. GOES satellite imagery was accumulated from the North Carolina State Climate Office and analyzed with GIS (Geographic Information System) software for the purpose of determining the inland horizontal extent of the sea breeze circulation. Penetration distances of 20-40 km were common, but occasionally the sea breeze penetrated as far as 80 to 120 km. The second part of this study attempts to develop a scheme for the prediction of the development and evolution of the sea breeze front. The factors most significant to this prediction include the synoptic wind flow regime and values for the temperature difference between land and sea surfaces. North American Regional Reanalysis data were downloaded and analyzed to gather daily geostrophic wind vectors. Water temperature data were collected and compared with air temperature measurements to determine the temperature difference, the driving mechanism for sea breeze development. Data for the Myrtle Beach site were analyzed first to provide a framework for the other stations because of relative completeness and reliability of the data sources. At Myrtle Beach, ninety percent of sea breeze development occurred with land-sea temperature differences of 2.0 degrees Celsius or higher.