Spring 2017 - Project Abstracts
Brian Lee - Monitoring deforestation in the Moskitia Corridor between Honduras and Nicaragua
This project describes the analysis of land change caused by deforestation in Central America for client WCS. It focuses on the Moskitia region that straddles the border of Honduras and Nicaragua. By using the Hansen dataset from the University of Maryland and Landsat 8 images from the USGS/NASA, we observed and analyzed deforestation as it threatened habitat connectivity for different species. Three pinch points were identified, allowing WCS to refocus their conservation efforts in the region to maintain forest and habitat connectivity.
Holden Leslie-Bole - Deforestation Patterns in the Lowlands of Eastern Bolivia
Deforestation patterns in the Santa Cruz department of Eastern Bolivia demonstrate changes in the agricultural development of the region over the past three decades. Much of the forests in Santa Cruz department remained intact into the 1980s, at which point settlement began in the form of planned communities with cleared fields radiating outward from the nucleus of the town center. In the 1990s and into the 2000s, agricultural expansion changed to drive strip deforestation branching out from existing roads. The mechanized cultivation of soybeans for export by multinational agribusinesses has become the single largest driver of forest conversion in the region, and the lack of a significant change in local absolute deforestation rates suggests that extensive deforestation is occurring in previously unsettled and undeveloped forests. This study additionally found no significant link between soil type and soil desiccation over time.
Benjamin Rifkin - Changes in Vegetation Cover over 14 Years in Southwestern Madagascar
Deforestation is a major environmental concern throughout the island nation of Madagascar, threatening both natural resource sustainability and wildlife habitat. This study focuses on the arid southwestern region of Madagascar where deforestation is thought to be predominantly caused by pressure from a combination of resource extraction, grazing, and climate change. A Landsat 7 and Landsat 8 image from September 2002 and September 2016, respectively, were analyzed using supervised classification, the normalized difference vegetation index (NDVI), and tasseled cap transformation to understand how these pressures are changing the vegetation in a region surrounding a protected area that is managed by the Madagascar National Parks Agency (MNP) known as Bezà Mahafaly Special Reserve. The study found that in this specific region vegetation loss was not as severe as expected, but the composition has shifted, and the climate is drier. Further analysis is needed, using additional satellite imagery and ground truthing to understand the true underlying causes of the changing land cover.
Aaron Lefland - Detecting Connecticut’s 2016 Gypsy MothDefoliation Event Using Landsat Imager
The gypsy moth is an invasive insect that, in outbreak years, can defoliate hundreds of thousands of acres of Connecticut’s forest. The state currently conducts and annual aerial survey to determine the extent of the damaged caused by the gypsy moth, but this survey is expensive and limited in the data it provides. Because the survey is conducted only once per year, there is no data regarding the duration of outbreaks, nor is there any data about continued defoliation or tree recovery. Using the 2016 outbreak as a case study, I sought to use Landsat-8 data to determine if the same, or greater, detail could be obtained using free satellite imagery. Using changes in NDVI values before defoliation and at the peak of defoliation, I was able to observe areas of the state that had been defoliated by the gypsy moth. Using changes in NDVI values at the peak of defoliation, and again later in the summer allowed me to observe areas where continued defoliation had occurred, or where trees has begun to regrow leaves after being defoliated. Supervised classification proved to be an effective tool in mapping the defoliation event. The supervised classification also provided potentially higher resolution and detail about defoliated areas. Unsupervised classification proved to be unsuccessful. The results of this study could be used by both researchers and land managers to learn more about the spread and timing of defoliation associated with gypsy moth outbreaks.
Mitchell Weldon - Investigation of Sentinel-2 Data for Distinguishing Conifer Species in the Catlins
Satellite remote sensing for the identification of tree species faces significant barriers due to the relative similarity of tree spectral signatures, the low spectral resolution of most satellite equipment, and the myriad of other factors which affect spectral reflectance values from forested areas. However, the enhanced spatial and spectral resolution of Sentinel-2a (S-2a) offers the possibility of greater distinction of forest types using satellite remote sensing. A study of the Catlins, a heavily forested region on the southern tip of New Zealand’s South Island, was conducted in order to investigate the potential of S-2a data for differentiation of the conifer species, native and introduced, which form the bulk of New Zealand’s tree biomass. A scene from February 16, 2017 was selected after screening for clouds, and was subjected to unsupervised IsoData classifications in order to first create a mask of forest cover and then to classify forested areas in the study area. This methodology was performed on an uncorrected Level-1C Top of Atmosphere scene, a scene corrected using the QUick Atmospheric Correction module, and a spectral subset of the Level-1C data. Aspect, slope, and shaded relief rasters were produced from a digital elevation model (DEM) in order to investigate the influence of topography on the classification process. The most sensible results were obtained from the spectral subset containing bands 2, 4, 5, 11, and 12, which clearly delineated the exotic timber plantations of the region and provided some additional distinction between native vegetation that is consistent with information pertaining to the forest ecology of the Catlins. However, shading greatly impacted the ability to differentiate forest types. Definitive interpretation of the classes was not feasible with the given information, but the methodology demonstrated Sentinel-2a’s potential for finer distinctions between tree types in evergreen forests.
Kathryn McConnell - New Zealand’s Emissions Trading Scheme and Land Use Patterns: Exploring the Forest to Dairy Conversion Hypothesis
New Zealand’s introduction of a country-wide Emissions Trading Scheme linked the forestry sector to broader price trends of the New Zealand Unit (NZU) – the newly created emissions currency. When NZU value declined dramatically around 2012, researchers hypothesized that plantation forests were being converted to dairy production, an unregulated land use under the Trading Scheme. If this trend were to be confirmed at a large scale, it would raise significant questions regarding the ability of the Trading Scheme to effectively decrease greenhouse gas emissions. Most research on this topic focuses on economic data or qualitative interviews; physical land changes have not been systematically analyzed to investigate the forest to dairy conversion hypothesis.
This project attempts to determine whether such a land use change can be detected from satellite imagery, focusing analysis on the Taupo District of the North Island as a proof of concept. Comparing images from 2009 (Landsat 7 ETM+) and 2016 (Landsat 8 OLI) found some evidence for a decrease in plantation forest cover over time and specific sites of apparent forest to dairy conversion. However, challenges with classification confusion (both K-Means and Supervised) of land categories made precise quantification of this decrease impossible. Mean NDVI data was also taken from Proba’s N-Daily Compositor for the month of November from 2013 to 2016. Layerstacking these images allowed changes in vegetation to be analyzed temporally. Evidence was found of forest to dairy conversion, but could not be accurately quantified across the entire Taupo District.
Rose Sulentic - Urbanization in Cluj-Napoca: Examining Expansion After Joining the EU
Romania joined the European Union (EU) in 2007. Doing so ensured Romania would enjoy the free movement of people, goods, services, and capital within the EU’s internal market. The foremost – movement of people – has particularly significant implications for Romanian city development. This project is focused on examining the urbanization of Cluj-Napoca, Romania’s second largest city, since its inception into the EU. Landsat satellite images from 2007 and 2017 are used to confirm the change in urban scenes. These images were spatially subset, and then classified using maximum likelihood supervised classification. Change detection statistics revealed urbanization and an increase in farm and grassland. There was an overall decline in tree coverage and water distribution. However, these results were highly confounded by classification errors. While the general trend of urbanization can be cautiously accepted, higher resolution images should be analyzed for a more accurate understanding of Cluj-Napoca’s change since joining the EU.
Alyssa Parpia - Kawasaki Disease and Dust Storms in Latin America and the Caribbean: An Exploratory Analysis using Meteosat and MODIS
Cases of Kawasaki disease, the leading cause of acquired cardiac disease among children, in Latin America and the Caribbean have been linked to reports of increasing incidence of dust storms originating from the Saharan Air Layer, transporting dust across the Atlantic. As climate change continues to take its course, it is projected that an increase in dust storms will occur, with a potential accompanied increase in density of the transported pathogen thought to be the causative agent of Kawasaki disease. Using SEVIRI-Meteosat 9 and 10 data as well as MODIS-AQUA data, dust is visualized using a combination of thermal band subtraction (IR 12.0-10.8 or 12.02-11.03μm, 10.8-3.9 or 11.03-3.959μm, and 3.9 or 3.959μm) layer stacking to generate a composite image, and isolated using unsupervised k-means classification. Dust classes were observed to have emissivity values for the differences between IR 12.0 and 10.8μm bands of -0.57 (range: -1.01 to -0.32), and emissivity values for the differences between IR 11.03 and 3.959 of 7.59 (7.24 to 7.88). Next steps will involve identifying an association or lack thereof between incident case reports of Kawasaki disease and dust storms in this area of interest, in addition to using the composite images to quantify the amount and concentration of dust.
Adam Eichenwald - Seasonal Distributions of Predators and Prey: Mapping Presence of Falcons and Grouse with Maximum Entropy Modeling
Predators are known to affect population abundance of their food sources through capture and consumption. However, the mere presence of carnivores can also result in behavioral and physiological changes in prey. Such changes are referred to as non-consumptive effects and can also affect the food source of said prey species, indirectly linking carnivores to plants. This concept, called a trophic cascade, is not well studied in avian systems. I hypothesized that a trophic cascade was present in an arctic food web that linked gyrfalcons to carbon storage through predation of species of ptarmigan. However, I had little to no information on whether this was an avenue worth pursuing. I hypothesized that if a trophic cascade existed, I would see spatial effects on ptarmigan as they browsed in locations most likely to keep them safe from predation. I also predicted that such effects would be most visible in the winter, a time where cover is scarce and effects of ptarmigan on carbon storage are at their highest.
I used presence data of gyrfalcons, rock ptarmigan, and willow ptarmigan gathered from online databases as well as environmental raster datasets taken from various satellites. These were used for maximum entropy modeling of the three species in both summer and winter, although rock ptarmigan and gyrfalcons were not modeled in the winter due to a lack of data points. Ptarmigan appeared to alter their foraging habitat in the winter to move away from areas with high levels of gyrfalcon predation, suggesting the presence of predation effects that could result in a trophic cascade.
Luke Menard - Identifying Grazing Halos from a Series of Depth-Corrected Satellite Images
This study utilizes field-collected depth sounding data to perform Lyzenga Depth Correction on a time series of high-resolution coastal satellite images covering a marine preserve off the coast of Belize. The methodology attempts to correct for the exponential attenuation of light with depth in water and generate novel layer stacks of depth-invariant bands that accurately estimate the spectral signatures of distinct benthic substrates. Utilizing unsupervised classification techniques and feature extraction, the study then attempts to generate accurate classifications of benthic cover, with a particular focus on the grazing halos surrounding patch reefs. Finally, changes in grazing halo area are estimated between 2005 and 2015. While an overall gain in grazing halo cover is estimated over time, limitations to the current depth correction methodology reduce the applicability of these findings. New approaches to these techniques, including the need for more nuanced spatial subsetting when classifying depth-corrected images, are discussed.
Juliana Hanle - Remote sensing to track Thick-billed Murre habitat change on Prince Leopold Island, Nunavut
Seabird colonies significantly impact their environment, are sensitive to environmental change, and can, in arctic latitudes, be highly difficult or impossible to monitor, all of which uniquely suit colonies to assessment by satellite remote sensing. The birds affect their habitat through the transport of nutrients: the guano of hundreds of thousands of birds breeding in dense communities can result in vegetation growth across large areas in difficult-to-access arctic regions. Surveys of remote seabird colonies can provide insight into not only species population dynamics, but also larger ecosystem health. This project to investigated using change in the vegetation index NDVI to follow change in a large cliff-nesting seabird colony, asking, is it possible to create an effective metric for evaluating changes in cliff-nesting seabird colonies based on NDVI? I used two landsat images, from 2005 and 2016, and ASTER digital elevation information, processing them in ENVI, ArcGIS, and R, to explore the relationships between landscape and thick-billed murre (Uria lomvia) colony change on Prince Leopold Island, Nunavut, Canada—where the population of roughly 200,000 has steadily increased since the 1970s. The correlations between inordinately high increases in NDVI, proximity to cliffs and ocean, and increases in albedo due, I reason, to guano, suggest that colony-specific metrics could be created by experienced researchers to follow population changes from satellite imagery. Pursuing an informed combination of spatially subset and de-trended metric of changes in NDVI and albedo could help scientists to monitor arctic seabird change.
Qiying Kuang - Explore Changes of Panda Habitat Over Time
Giant panda has recently been removed from IUCN’s list of ‘endangered species’, but the survival of this species remains challenging. The major survival challenge comes from the limited and fragmented nature habitat. To understand how panda habitat suitability changed in Wolong Nature Serve from 2005 - 2016, this study identified four major factors determining giant panda habitat suitability- elevation, slope, bamboo density and vegetation, and developed suitability evaluation model based on these factors with Analytic Hierarchy Process (AHP). DEM dataset was used in elevation and slope analysis. Regression model was applied in bamboo density estimation. Supervised classification of Landsat satellite images was performed for vegetation classification. Results indicate that from 2005 – 2016, area of suitable giant panda habitat remained stable (41.83% of the Nature Serve), but region with higher suitability level tended to decrease from 2009 -2016. 3D model of suitability indicates that suitable giant panda habitats were connected vertically but were significantly segmented horizontally.
Rachel Renne - Relating Climate Zones and Vegetation in the Sagebrush Region of the United States
Big sagebrush (Artemisia tridentata) dominates one of the most geographically extensive biomes in western North American. Sagebrush is well adapted to seasonal drought and this shrub relies on soil-water stored at depth during the cold-season for survival during the warm season. Seasonal reliance on deep soil-water allows sagebrush to co-dominate with perennial grasses through soil-water resource partitioning. Although the deep soil-water storage required for sagebrush is achieved across a large geographic area in the western U.S., precipitation patterns vary dramatically within this region. Differences in precipitation patterns result in variation in water availability in time and space in shallow soil layers, and these differences should predictably influence the quantity and importance of plant types across the sagebrush region. Weekly 30-year average precipitation and temperature data were used to define 14 climate zones within the sagebrush region using the unsupervised classification technique, ISODATA. A large vegetation dataset from the Bureau of Land Management was used to compare plant community composition between the eight largest climate zones. Univariate and multivariate analysis of variance demonstrate that there are significant differences (p < 0.05) in plant community components and composition between climate zones. To determine if these differences are detectable using remote sensing products,16-year average, 16-day composites of Normalized Difference Vegetation Index (NDVI) and Normalized Difference Snow Index (NDSI) values were calculated for each climate zone. Regions with a large perennial grass component exhibited strong seasonal NDVI and NDSI fluctuations. The seasonal fluctuations in vegetation and snow for other regions were weaker due to high reflectance of bare ground or interference by conifers. Further investigation into remote sensing techniques may provide a more reliable means of discerning regional differences in sagebrush plant communities. Classifying the sagebrush region into distinct climate zones demonstrates that plant communities vary significantly with differences in precipitation seasonality and temperature. Understanding how different components of sagebrush plant communities respond to differences in climate provides a valuable method of predicting potential vegetation for management under both current and future climate conditions.
William Koh - NDVI Growing Season Analysis: Currant Creek Ranch
This summer, the Ucross High Plains Stewardship Initiative, Trout Unlimited, and Wyoming Game and Fish are coordinating to create a Rangeland Management Plan for Currant Creek Ranch. In a baseline contribution to that effort, this study analyzed eight Sentinel-2A images of the property in order to measure changes in biomass over the course of the 2016 growing season (March to November). Normalized Difference Vegetation Index (NDVI) was used on layer stacked images to identify when and where the ranch experienced both peak and limited vegetative growth. This study found that the riparian zone represents the heart of plant activity on the ranch and last year the growing season climaxed sometime close to June 19th. While pasture areas close to the ranch house exhibited healthy signals, areas on the western and eastern edges of the property’s riparian zones consistently demonstrated limited NDVI. When considering grazing current grazing practices, this study recommends the rancher retain cattle populations close to the ranch house and avoid sparsely vegetated grazing areas in the riparian zone’s extremities.
TC Chakraborty - Investigating associations between air temperature and surface temperature over North America
Surface temperature and near-surface air temperature are important variables that dictate land-atmosphere interactions. Previous studies have looked at the coupling between the two at different spatial resolutions. In the present study, I use the newly released Daymet version 3 dataset and MODIS satellite products to investigate this coupling at a 1 km x 1 km resolution over the continent of North America. I compare this coupling across the five main Koppen-Geiger climate classes for different vegetation densities. The strongest coupling was seen for the Temperate climate zone, with a 1-to-1 coupling between both maximum and minimum surface and near-surface air temperature for high vegetation cover. However, there are significant differences in this coupling for the different climate zones, especially for daytime.
Dimitri Diagne - Lac de Guiers, 1987 and 2016: Interannual and Seasonal Surface Area Change in a Sahelian Lake
Lac de Guiers is a perennial lake located in the Sahel of Senegal, West Africa, and fed by the Senegal River. Desertification and shifts in precipitation regime caused by global warming could have a significant negative effect on water availability in the Sahel, but current studies reach various conclusions on the nature and extent of this potential change. Large-scale climatic changes occur in conjunction with rapid population growth and agricultural expansion in this region, leaving its water resources doubly vulnerable.
In this study, I consider Lac de Guiers as a potential indicator of decreasing water availability in the Sahel. By classifying and comparing a series of Landsat images, two from 1987 and two from 2016, I attempt to measure seasonal and interannual changes in the surface area of Lac de Guiers. I discuss the potential causes of these changes, and their implications on water availability in the Sahel.
Khalid Cannon - The Impact of Climate Change on the Coastline of Newtok, Alaska
Newtok, a small village on the western coast of Alaska, is in danger of being washed away due to the effects of global warming. Rising temperatures in the Arctic has led to the melting of permafrost, rising sea levels, and flooding. The goal of this project is to calculate and report the area of land that the Newtok natives have lost over the past quarter century. This goal was accomplished by analyzing Landsat imagery with unsupervised classifications. Direct results from these classifications were not revealed due to poor resolution. Instead, a handcrafted ROI polygon was used to calculate the lost area. The results concluded that the area of the village had been reduced by over 100% since 1975.
Hannah Walchak - Drought and Recovery at Lake Oroville, California Exploration of Planet Labs Image Capabilities, Processing & Analysis Methods
The state of California has just emerged from a historic drought that lasted five years, and parched riverbeds, mountainsides, cities, and agricultural areas alike. The drought’s end came with a bang, with reservoirs like Lake Oroville swelling to dangerous levels under torrential rains. This project aimed to explore the cycle of drought and recovery in California through twin goals: 1) observe and quantify the effects of changing precipitation at Lake Oroville and 2) explore the uses of Planet Labs’ Open California tool through its RapidEye satellite, focusing particularly on applications of the Red Edge band. Findings included a doubling in water area on and around the reservoir from 2015 to 2017, as well as indications of early stress in vegetation surrounding the reservoir that ultimately led to damaged and dead vegetation.
Ajit Rajiva - Observing Vegetation-Temperature Relationships in Western India
In the recent past, spells of extreme heat associated with appreciable mortality have been documented throughout the globe. Of these, but for the recent past, there are fewer research reports available from developing countries or specific cities in South Asia. Since 2013, India has made progress in addressing these concerns with a targeted efforts to produce city specific action plans for heat waves as well as increase. The first of these plans, The Ahmedabad Heat Action Plan was rolled out in response to a massive heat wave event in May of 2010 that claimed over 1300 lives (GS Azhar, 2014). The study showed that relationship between the temperature and mortality during the heat wave and began the first steps to quantifying the response to heat stresses in tropical climates.
Among the plans long term goals was the promotion of urban greening within the city in order to reduce temperatures during hot spells. While largely a common sense initiative, it is useful to be able to quantify the rate of change in temperature in order to properly estimate the impact such an initiative would have on reducing heat related mortality and morbidity. This project represents the initial steps in this process and outlines the usefulness remote sensing will have in evaluating this task.
Nina Lagpacan - Detecting Grassland to Cropland Conversion in Southeastern Montana
Over 53 million acres of grassland has been lost across the Great Plains since 2009. This is an area larger than the size of Kansas.1 The objective of this study was to detect changes from grassland to cropland in Yellowstone County, Montana using remote sensing techniques. One method that has been proven successful in classifying agricultural fields involves using a multi-temporal analysis and the normalized difference tillage index (NDTI) to differentiate between varying levels of crop residue cover (CRC).2 Seven images (Path 38, row 27) from Landsat 4-5 TM from 2006 and 2011 were analyzed using the multi-temporal percentage change NDTI method. All images were corrected for clouds and cloud shadows using the Fmask Algorithm and IsoData unsupervised classification tools. Percent change in NDTI over the growing season (March – August) was used to classify images from 2006 and 2011 into three classes: non-conservation tillage, CRC 30 – 70% and conservation tillage. Changes between the 2006 and 2011 classified images were evaluated using change statistics and change visualization techniques. Results from this study show that there is a trend towards using more sustainable agricultural practices and leaving a greater percentage of CRC after harvest. Opportunities for further study include applying the multi-temporal NDTI method to other study areas, utilizing other indices that detect CRC and incorporating policy and economic drivers in analysis.
Sophie Ruehr - The Spatial Effects of the Moroccan Green Plan: Using ENVI to detect change in agricultural technology near Agadir
The Moroccan Green Plan has resulted in massive economic and agricultural shifts in Morocco in the last decade. Having studied abroad in Morocco and talking with farmers in the agricultural region outside of Agadir, I was curious to see whether the effects of the Moroccan Green Plan can be observed and quantified on the landscape. I compared images from 1987, 2010 and 2016 to track changes in the amount of covered agriculture and cereal production in the area. After preforming several processing algorithms on each image, including tassled cap transform, band math and masking, to isolate covered agriculture from surrounding land uses, I calculated the total land conversion to covered agriculture to find the rate of increase in the thirty-year time
period. I also used supervised classification to find that there has been less cereal farming in the past six years in the region. This analysis exposes the extent of the change induced by the Moroccan Green Plan with implications on food sovereignty and farmers’ autonomy over their cultivation methods and land.
Samantha Garvin - Development and Degazzetement: land cover change over time in northern Botswana
The greatest challenge facing conservation today is making space for wildlife while balancing needs for economic development. In Kasane-Kazungula, in Northern Botswana, there is still space for wildlife – for now. Infrastructure development over the past thirty years, from tarmac roads to an airport to a bridge, make tourism and other economic opportunities more viable. With this development come more people seeking economic opportunities. As a result, residential areas are encroaching on a protected area, the Kasane Forest Reserve. This study uses three Landsat images across a thirty year span to analyze the change in land cover over time, capturing the before and after of these major infrastructure developments. Supervised classifications with maximum likelihood were run on all images. Change in class composition over time indicated a loss in forest over time and an increase in cleared areas. This study documents a loss of 2.25 km2 of Kasane Forest Reserve to residential development. With no enforcement of the Forest Reserve’s protection status, this default degazettement sets dangerous precedent for the future of this and other protected areas.
Joanne Choly - Change in Vegetation and Land Cover/Land Use in the Mount Monadnock Area of Cheshire County, NH
Is the forest re-ascending to the Mount Monadnock peak (Cheshire County, southwest NH)? Is wild area decreasing in the Mount Monadnock area? Is wildlife connectivity changing in the area? What can 30-meter spatial resolution Landsat satellite images add to answering these questions?
Preparatory work included downloading Landsat images – one each from June 1992 and 2016, sub-setting the images, top-of-atmosphere (ToA) and Harris ENVI quick atmospheric correction (QUAC). Visual inspection and comparison of years was made with RGB, SWIR-NIR-R and NIR-R-G views; a Google Maps image was georeferenced and draped over a DEM of the area. Supervised Classification was run, rejected due to difficulty classifying the predominately mixed forest. K-means and ISODATA unsupervised classifications were run, with K-means being the technique chosen to continue work. Analytical techniques, largely for change detection, were applied to the classified images to examine vegetation at the top of Mount Monadnock, the forest and the entire greater area: ENVI change detection (descriptive statistics), RGB change detection, NDVI and rudimentary Tasseled Cap Work.
Findings, partial paths pursued and struggles with the images suggest additional work, including: try additional classifications; do additional tasseled-cap work; look further into forest cover trends; consider the impact of change and flux on biodiversity; noting that change and flux can sometimes increase biodiversity; consider connectivity issues.
Genora Givens - Land Cover Change in Apia, Samoa
Apia, the capital of the island nation of Samoa, has experienced significant population growth in the last decade. With this increase in population it is likely that will be urban expansion from the city center into surrounding forest areas. Interestingly, there is a simultaneous effort by the Ministry of Natural Resources, Environment, and Meteorology to increase forest cover on the island. This study uses image classification and change detection to
better understand trends in land cover change around Apia. Two Landsat images were selected, one from 2003, the other from 2017. The images were spatially subset, clouds were removed, and a supervised maximum likelihood classification was run on both images to identify change in land classes over time. Class statistics indicate there has been a significant decrease in forest cover around Apia and an increase in urban area. While these trends are legitimate, the details of changes in land cover area are difficult to discern due to classification challenges such as mixed pixels. Future studies should use higher resolution, cloud free images.
Andrew Wilcox - Integrating High Resolution Imagery with Satellite Data
The technologies that produce high resolution imagery, either by satellite or unmanned aerial vehicle (UAV), are advancing rapidly in terms of precision and resolution while also dropping in cost. Yet there is little consensus about how exactly to integrate this type of data at scale with other remote sensing methods in a systematic and automated way. Frequently, high resolution is used for visual inspection of images, a form of remote ground-truthing. It is also commonly used to investigate discrete features, such as an expanding mine, construction site, or a linear feature that requires precise delineation. There are few attempts to integrate high resolution imagery with landscape scale analysis, partially due to the effort and cost of data acquisition, the data storage needs, and also the heterogeneity of the images themselves. This heterogeneity being that they are often off-nadir when acquired by satellite and also acquired at different times of day, resulting in at times drastically different images due to the solar azimuth. This study represents an initial attempt to explore the potential for techniques that analyze neighborhoods of pixels, rather than individual pixels themselves, for distinguishing between forest cover types and stands.
Qingyang Chen - Texture-based Land Cover Classification
Whereas traditional classification methods use the band values of pixels to classify individual pixels, this project explores the classification of land cover in aerial imagery using the textures of the land cover. The project explores proof-of-concept implementations of supervised and unsupervised classification algorithms that use texture signatures as predictors. The texture signatures are extracted using Segmentation-based Texture Analysis. The classification method produced comparable results to spectrometric band-based methods using only grayscale images, but was slower and had lower classification resolution. The implementation is only a proof-of-concept and not production-ready, but opens the field to more extensive studies into a generalized topological classification method based on textures.
Cameron Yick - Day & Night: Combining High-Resolution Night Photography with LANDSAT to Understand Land Use in Dubai
Assessing land use within cities is traditionally performed using imagery from day-time remote sensing products, due to limitations with spectral and radiometric resolution in current night-time imaging products (VIIRS, DMSP). This papers explores how photography captured by high-resolution cameras on the International Space Station could be processed such that it could be used together with Landsat to analyze urban features using spectral characteristics from day and night-time. This paper focuses on methods, results, and challenges with respect to using this combination of data to perform unsupervised classifications on imagery of Dubai. Initial applications of this method include detection of island occupancy, detection of shipyards, and assessment of the busyness of various road features.