OEFS Abstracts - Spring 2012

Julia Osterman - Examining forest cover in northern Madagascar

Ankarafantsika National Park, located in northwest Madagascar, contains one of the island’s largest remaining expanses of dry deciduous forest (Ganzhorn et al. 2001). Assessment of forest cover and other land cover features within the park provides invaluable insight into the land cover composition and status of forest within this the area. My goal for this course project was to use my newfound remote sensing techniques to classify and quantify the various distinct land cover features within the park to give insight into the status of one of Madagascar’s precious natural commodities.

Kayanna Warren - How well can Landsat determine stand age after harvest?

It would be beneficial for forest managers to have a cheap and easy way to assess the amount of biomass in forests. Remote sensing – especially using widely available Landsat imagery – to do this would be a great help. It is unclear whether the 30mx30m resolution of Landsat will be useful for the smaller-scale New England forestry, and this study attempts to explore that. Biomass on a site changes with stand age after a harvest, and this study explored how well Landsat imagery can detect and classify the ages of forest stands that have undergone shelterwood cuts. It found that while Landsat is fairly good at classifying and identifying recent (0-15 years) cuts when using aerial image photography to verify regions of interest, it did not do well at further sub-classifying those sites into ages. The results do suggest, however, that the classification does do well at detecting stand development stage, and the differences between the younger stands in a stand initiation phase were classified differently than the two oldest stands, which – based on ground truth data – were in the stem exclusion phase of forest development. It was also found that a K-means classification of just 4 classes was more useful than either Isodata classifications or a classification based on the NDMI.

Tina Schneider - Forest cover change detection on the Thai-Lao border

The objective of this remote sensing analysis was to analyze the change in forest cover on both sides of the Lao-Thai border. For this purpose, two ASTER images with low cloud cover were obtained for two dates spanning 8 years: March 2, 2003 (127/047, 2011945276), and February 4, 2011 (147, 047, 2093349467). The Mekong river transects the image from west to southeast, with the Pakxan and Borikhan districts in the Lao Bolikhamsay province to the north of the river, and the Buen Kan and Bung Khla districts in the Thai Nong Khai province to the south. These two images show a drastic difference in land cover across the two provinces, with high forest cover in Laos and very low forest cover in Thailand. However, while the remaining forest cover in Thailand appeared stable across the time span upon visual analysis, there seemed to be a pronounced forest cover loss on the Lao side of the border. To investigate the perceived forest cover change in light of proximate causes and driving forces of land use conversion, different methods of land cover classification and change detection were employed in a mixed model approach (Chowdhury 2006) to shed light on deforestation patterns in this region, including unsupervised classification, supervised classification, visual inspection of differences in land cover through additive color overlay, and NDVI comparison through additive color overlay.

Natalie Price - Classifying Greater Cairo to determine high risk areas for avian flu transmission

In the beginning of 2006, avian influenza was introduced to Egypt.  Since then there have been frequent outbreaks infecting not only bird populations, but on occasion infecting humans as well.  Because of lax regulations on animal rearing and lack of infrastructure in urban slums, the majority of people becoming ill from avian flu come from these areas.  In an effort to create a risk map of avian influenza, I first attempt to classify the land cover of Cairo, Egypt and its surrounding areas using remote sensing to distinguish the developed urban areas from the informal settlements or slums where infection rates are high.

Sharon Smith - Land-Use change in the Boa-Esperança corridor, Bahia, Brazil

The state of conservation of the Atlantic Forest within the Conduru-Boa Esperança Mini-Corridor and the Parque Estadual da Serra do Conduru (PESC) was assessed for a twenty-five year period between 1986-2011. Land cover analysis was performed using Landsat TM 5 and Landsat 7 ETM+ satellite imagery to detect changes in land use and understand baselines rates of deforestation and land use change in the region before and after the paving of the highway and the establishment of a national park in the region, both of which occurred in the years 1997-1998. For the period 1986-2001, the annual reforestation rate was 3.0% in the Mini-Corridor, while for the period 2001-2011, the annual reforestation rate slowed to .1%. The increase in forest cover across time, from 49,002 hectares in 1986 to 51,372 hectares in 2011, particularly given the consistent loss of forest cover elsewhere across the state and Atlantic Forest more broadly, indicates that the region has been successful in its conservation efforts, despite a growing population, tourism pressures, and infrastructure development.

Daniel Hoshizaki - Change detection on the Mattole River watershed

For my remote sensing project I have chosen to examine the Mattole River watershed in order to understand the land cover changes that have taken place over time. Particular emphasis is placed on the changes in distribution of vegetation throughout the watershed. Using a number of different methods including decision tree, guided, and supervised classifications the vegetation distribution is analyzed. Both the supervised and unsupervised classification images are analyzed for change, and the two methods are compared to each other in terms of what they reveal about vegetation cover in the Mattole basin. Qualitative and quantitative analysis of the results provide a general idea of changing boundaries between forests and grassland. Finally, each technique is reviewed in order to identify limitations and opportunities for improvement.

Christopher Dutton - Land cover change in and around the Masai Mara National Reserve, Kenya

Riverine suspended sediment is a major water quality problem and an important transport medium for nutrients and contaminants. The most effective way to control sediment is to mitigate its entry into the river, which necessitates knowing the locations where it originates.

The Mara River Basin of East Africa is well known for its annual wildebeest (Connochaetes taurinus) migration of approximately 1.2 million herbivores. In addition, the Mara River has a growing population of hippopotamus (Hippopotamus amphibus) who reside within the river. Sediment levels in this river are extremely high (turbidities as high as 6,000 NTU) and appear to be increasing over time.

Land cover changes within the Mara River Basin having been occurring since pre-colonial times. More recently, deforestation in the headwaters has been linked to an increase in sediment supply within the Mara River. Minor changes in land cover could potentially lead to a disproportionally high erosional response.

I evaluated two Landsat images, taken approximately 27 years apart, to determine the areas within the Middle Mara and Talek regions that had undergone the highest amount of land cover change. Previous research has indicated that these two regions may be responsible for a high sediment load. I utilized two separate unsupervised classification techniques and grouped the resulting classes into 4 land cover classes (Forest, Bushland, Grass, and Bare) based loosely on the Anderson Classification Scheme Level I.

Change detection between the two images found a trend of grass conversion to bare ground within the Talek region. Inconclusive results were achieved for the Middle Mara region, although there appeared to be up to 22km2 of forest loss from that region.

Frances Liu - Agricultural changes in central Crete, Greece—with a focus on agricultural land use and change detection

Satellite remote sensing has been playing an increasingly important role in monitoring agricultural land use. In particular, it has been widely used to detect land desertification and degradation due to overgrazing and other anthropogenic factors in the Mediterranean region. The Greek island of Crete has been inhabited by human as early as 10,000 BCE. It is home to many great civilizations such as the Minoan, Mycenaean, and Greco-Roman, and its landscape has been constantly under modification by human activities.

Kanchan Shrestha - Describing the changes in Koshi River basin using Landsat

This project examines the land cover change in the rugged terrain of the Sagarmatha National Park (SNP) in Himalayan northeastern part of Nepal. The Park has the world’s highest peaks include Mt. Everest at the altitude of 8848m which is snow covered year around. Understanding land cover changes in terms of vegetation and snow cover change is crucial in the area because of two reasons. The mountain stores freshwater for the millions of people downstream and they are likely to be melting at a faster rate because of climate change. Second, the Koshi water basin in which the park lies has been noted as having higher deforestation rates causing massive sedimentation downstream.

I use unsupervised classification of Landsat images from 1992 and 2009. I use Normalized Difference Vegetation and Snow Index. Unsupervised classification was challenging because of shadows formed by hills and mountains as well as different combination of ice and rocks with ice and sparse vegetation.

Meredith Martin - Land-use change around Jenaro Herrera in Loreto, Peru from 1985-2011

Satellite data used in remote sensing to estimate changes in a landscape, either anthropogenic or natural, can be a key tool to understand the dynamics of an area.  Remote sensing has been useful in understanding climate patterns, deforestation dynamics, changes in hydrology and glacial ice, urbanization, epidemiology, and more.  In the tropics, using remote sensing to study land-cover changes can be particularly useful in regions that may be difficult to access or where other record may not exist. This project uses remote sensing to examine land-use change and hydrological shifts in the area around the town of Jenaro Herrera, Loreto, Peru.  The project supplements field work conducted in the region this past summer studying the economic botany of a wild-harvested fruit, camu camu.  Although the spatial resolution of the Landsat images used is too coarse to capture shifts in the camu camu popuations themselves, changes detected in land-use in the area can be an indirect measure of anthropogenic pressure on the landscape.  Additionally, camu camu is a riparian species that colonizes the edges of oxbow lakes, so further understanding of the dynamics of these riparian and lake systems will give insight into changes in the camu camu populations. 

Jennifer Kasbohm - A geological classification of the Sinclair Group, Namibia

The Mesoproterozoic sediments and extrusive igneous bodies of the Sinclair Group in southern Namibia may yield greater insight into Precambrian tectonic processes, such as the formation of the supercontinent Rodinia. Greater paleomagnetic sampling of Sinclair rocks with magnetic minerals is required in order to discern the paleogeography of the Kalahari Craton. With this purpose in mind, I pursued a geological classification of the Sinclair Group. Guided by GPS points and spectral data from rocks collected in the field, two different supervised classifications were performed using an ASTER image. The first utilized the 15 m resolution visible and near-infrared bands to identify landscape features to construct regions of interest, overlain on the full spectral stack of ASTER bands, and input into a Maximum Likelihood classification. The second utilized ASTER’s thermal bands to create ROIs from a false-color image that differentiated silicate, carbonate, and basic rocks.

When compared with a geologic map created from field observations, it was found that carbonates and mafic rocks were well-classified, while silicates and felsic rocks were confused. The full stack classification was more successful at determining the formation to which the rock belonged, while the thermal classification could only discern rock type. Dykes, which are effective targets for paleomagnetic sampling, could not be grouped into their own class without introducing significant error to the classified image. However, the high spatial resolution of the ASTER imagery allows for remote reconnaissance of new sampling sites.

Rafael Lefkowitz - Remote sensing of the Dead Sea region

This is a descriptive study of the Dead Sea basin using Landsat TM 5 imaging.   The first goal of the project is to determine the change in Dead Sea area between over a 26 year period, using multiple classification methods and trials.  The predicted amount of lost Dead Sea area would be 104 kilometers squared.  Discrepancies between results produced by image processing will be explained based on principles of remote sensing.  The second goal of the project is to describe the unique mineral distribution along the shore of the Dead Sea.

Rachel Kramer - Mapping the forest-agricultural boundary along a national park in Madagascar

Enhanced understanding of anthropogenic land use along protected area peripheries is essential to informing development planning that integrates local livelihood concerns with biodiversity conservation priorities. Marojejy National Park in northeastern Madagascar provides refuge to one of the world’s 25 most endangered primates, yet also borders one of the most densely populated regions of this Indian Ocean island. In 2008, a revised Park boundary was delineated by an independent third-party field auditor that conceded contiguous areas converted for agriculture inside the original Park periphery to forest-bordering settlements.  A 2011 socio-economic survey in the region documented continued demand to access agricultural plots and fallows remaining within the new legal boundary, where soils are more productive than the surrounding degraded landscape.  This study uses remote sensing as a tool to perform a land cover classification of the Marojejy Massif and map the distribution of recent production and fallow areas within the former and current extents of Marojejy National Park. This mapping enables the identification of priority regions for investment in agricultural extension to increase crop yields on cleared surfaces outside the legal periphery, where pressure to continue accessing Park soils is greatest. Smallholder intensification on existing cleared lands is considered an essential strategy for ensuring food security in the region while preserving the integrity of the current Park boundary and protecting remaining habitat for a critically endangered species.

Nikki Grigg - A seasonal comparison of vegetation and elevation in Chaco Canyon

This paper uses remote sensing classification tools and digital elevation data to outline the environment of the Chaco Canyon area, home to a well-known archaeological site, and compare land cover change in the wet and dry season. The results demonstrated the complexity of the landscape in the region and the need for detailed ground truth data when performing supervised classifications.

Denise Soesilo - Monitor vegetative growth against population and density in Berlin, Germany

This paper looks at and compares urban vegetation change in Berlin from the year 1989 to 2011, where the rate of change between the former East and West are compared. The normalized difference vegetation index (NDVI) is applied to two sets of Landsat Thematic Mapper (TM) images to serve as a proxy for total vegetation.

NDVI is a commonly used vegetation index not only for land-cover classification but also to remotely sense vegetation health among others. In an urban context it is often used to estimate temperature related urban phenomena such as the urban heat island effect. However, some studies have made use of the NDVI in order to assess overall “greenness” of city space. Mansfield et al. were probably the first to use NDVI transformed Landsat TM images to estimate overall urban greenness in the Triangle region of North Carolina to provide a quantitative measure of the density of trees and other green micro-areas within a city. Before Mansfield et al. much more time consuming methods were employed to measure a city’s greenness. In some instances, high resolution aerial photographs were examined and trees contained in each plot counted8. For this project, Landsat TM images are used and methods applied to provide a measure of overall greenness of Berlin’s urban environment.

Paulo Quadri Barba - Land use and temperature variations across eastern Mexico City Valley

Urban thermal analysis with remote sensing techniques is a challenging task because of the complexity of urban surfaces and the lack of more quantitative descriptors for such surfaces reflectivity and emissivity. However, the ability to approach and map differences in surface temperature and its relationship to various land uses / covers, is extremely important in urban centers where the local and surrounding ecosystems have been very degraded or impacted by human activities related to urban development.

In this study we explore the potential of Landsat5-TM imagery to contribute to the study of local and regional climate change processes related to land use / land cover. Broadly the main goals of this project are:

1. Explore the potential of Remote Sensing techniques to understand how land use affects local temperature.
2. Identify what classes of land uses are predominant in the eastern portion of Mexico City Valley.
3. Compare land surface temperature results obtained with remote sensing methods to local climate data obtained through the National Meteorology Service.
4. Describe potential environmental and social impacts created by local climate change induced by land uses.

Catherine Chen - The effect of drought on Mongolia

This study will examine drought in the arid and semi-arid lands of central Mongolia. Using a study area that covers most of the Arkhangai aimag and part of the Bayanhangor aimag, it will examine how the region changed from 2007—a normal year—and 2009—a drought year. The project will analyze these images using the Normalized Difference Vegetation Index (NDVI), the Normalized Difference Moisture Index (NDMI), the Normalized Difference Snow Index (NDSI), the Normalized Multi-band Drought Index (NMDI), and surface temperature, comparing and contrasting the utility of these indices. It will also conduct a supervised classification over the two images and a change detection on the classified images. Overall, the results of the study suggest that the NDVI and NDMI were most effective in sensing changes from drought. It also suggested that the maximum likelihood classification offered the most effective classification change analysis, although difficulties in accurately classifying the image hindered the effectiveness of classification change analyses.

Rebecca Schultz - Interpreting atmospheric aerosols with remote satellite imagery

Atmospheric aerosols represent an area of large uncertainty with respect to the Earth’s heat budget and climate and meteorological trends. Some particles, like sulfates, reflect incoming solar radiation, and some like black carbon absorb it. The presence of aerosols in the atmosphere further effects cloud formation, which in turn has varying impacts on local precipitation and convection patterns around the world.  Scientific understanding of these phenomena is being improved through the use of satellite remote sensing, which can detect the presence and character of particles through signatures of electromagnetic extinction in the atmosphere. Taking the municipality of Beijing as a case study, this project will explore rudimentary methods for aerosol detection during an extreme air pollution event, spanning from February 21 to February 24, 2011.  It will investigate a MODIS Level 1 image of the pollution event, comparing band reflectances with those from a low pollution day image four weeks earlier, on January 29, 2011.  This project will use the following experimental methods to explore techniques for identifying aerosol pollution over Beijing:

1. Dark object subtraction;
2. Qualitative investigation of spectral signatures;
3. Image subtraction; and
4. Use of multi-band indices.

Rick Russotto - Saharan dust and the DOMEX project

This project attempted to use Aqua MODIS data to determine whether Saharan dust was present in the vicinity of Dominica during the April-May 2011 DOMEX field campaign. I employed two image processing algorithms: a false-color enhancement developed by Miller [2003] that uses the visible and NIR bands, and an index developed by Hao and Qu [2007] that uses the thermal IR bands. After comparing these images to the examples in the published papers to ensure that I had implemented the algorithms correctly, and applying each algorithm to the example image for the other to compare the two methods for test cases, I applied both methods to an image of the Caribbean taken on April 24, 2011, the day of DOMEX flight RF10. I concluded that there was no “dust outbreak”, defined as a dust layer having an optical thickness (AOT) of greater than 1.5, in the vicinity of Dominica on this day, but could not draw any conclusions regarding the presence of smaller quantities of dust. Future work will involve applying the algorithms to other days during DOMEX; comparing results to the NASA MODIS AOT product; and, if possible, applying these methods to a known dust outbreak over the Caribbean with ground truth data for comparison.

Erin Raboin - Assessing biomass across a mixed-use landscape in Panama

Panama is a pilot country in the UN REDD program, and is attempting to inventory biomass reserves across the entire country.  To achieve this, accurate land cover maps must be developed to extrapolate out to landscape scale carbon stores.  I compared a variety of classification techniques in order to quantify land-use coverage across the Panama Canal Watershed.  Landsat TM 4-5 images were analyzed from February and March of 1998.  Indexes of NDVI and NDMI were developed, and pre-processing masking techniques were employed.  Post processing included majority analysis to smooth data, and class statistics to calculate land cover.  I used band math to normalize NDVI images, and used layer stacking to amplify seasonal variation.  I review my results from unsupervised K Means classification, Unsupervised ISO classification, and Supervised Minimum Distance Classification.  I compare and contrast the results, and the relative strengths and weaknesses of the techniques.

Ankur Garg - Identify potential for energy generation from non-wood based biomass

The objective of the study was to identify the potential for energy generation from agricultural waste in Kisumu, Kenya using remote sensing. Three different remote sensing methods were used to derive the estimates. In the first two methods, a 30 meter resolution Landsat – TM 5 image was analysed before and after applying Tassle cap transformation and then K-means unsupervised classification was performed to derive 5 classes. The third method used 24 image time series stack of 250 meter resolution MODIS imagery. Z-profiles were observed to see the seasonal variation in Enhanced Vegetation Indices (EVI) to differentiate between forests, urban agriculture and peri-urban agriculture. Finally the estimates derived from the three methods were compared. Estimates provided by methods 1 and 2 were in close agreement with each other whereas those derived using method 3 were significantly different. The reason for this difference could be the low spatial resolution of MODIS images and high sensitivity towards EVI thresholds selected for differentiating the classes.

Based on the estimates of peri-urban agriculture provided by the remote sensing methods, yield data and energy factors from literature, 5-10 Gwh of energy could be generated annually from corn stover and bagasse in Kisumu.

Matthew Long - An examination of turbidity in the upper Chesapeake Bay

This study explores the spatial and temporal variation of suspended sediment concentration within the Chesapeake Bay using regression analysis to derive total suspended matter (TSM) concentration from remote sensing reflectance for Band 1 of the MODIS/Aqua Level-2 Ocean Color product. There are several advantages to using the Level-2 Ocean Color product as opposed to other satellite images, such as higher resolution Landsat TM or ETM+ data. First, the return time for MODIS is daily, which allows for a much greater temporal resolution of changes in total suspended mater than could be achieved with the higher resolution Landsat data, which has a typical return period of 16 days. Second, the Level-2 Ocean Color product is atmospherically corrected to account for the effects of atmospheric particles, clouds, aerosols, sea-surface reflectance, and sun glint. The models and calibration of those models are complicated and vary both spatially and temporally. Therefore, the atmospheric correction included in the Level-2 product saves the analyst a great deal of time and effort. Third, the MODIS/Aqua Level-2 Ocean Color product includes two high-resolution bands (bands 1 and 2, 250m resolution) that can be used to estimate TSM from remote sensing reflectance or normalized water-leaving radiance (nLw). Forth, unwanted and invalid pixels are masked out. These masks, or flags, include land, sun glint, shallow water, clouds, and over-saturated pixels. Finally, probably due to the reasons listed above, several models have already been developed that utilize the MODIS/Aqua Level-2 Ocean Color product to estimate TSM.

Sebastian Ramirez - Vegetation response to summer monsoon, elevation and aspect in Mexico

An analysis of the relationship between changes in Normalized Difference Vegetation Index and changes in albedo as a result of the North American monsoon. This relationship is analyzed using pre monsoon and post monsoon LANDSAT imagery. The presence of thick deciduous dry tropical forest is proposed as an alternative hypothesis for the positive correlation between NDVI and Albedo in mountain slopes of the sierra Madre. Finally the paper explores the limitations of using short wave albedo or visible albedo as a proxy for broad band real albedo.

Judith Ament - Effects of the 1997/1998 El Nino events on the water levels of Lake Kyoga, Uganda

In the year between March 1997 and February 1998 one of the strongest El Nino events of the previous century occurred. Arround the world, weather patterns were disrupted. This study investigates the effects of the 1997/1998 El Nino event on the Lake Kyoga basin in Uganda, Eastern Africa. Floating islands of papyrus became adrift due to higher than normal water levels and became caught in the lake outlet, blocking water from leaving the lake and causing floods all around the lake shore. This study attempts to quantify the amount of land loss to the floods after the 1997/1998 El Nino event, and characterize it by land use type.

Andrew Womack - Detecting degradation in archaeological sites over time in the Chengdu Plain

Over the last twenty years archaeologists have begun to utilize satellite remote sensing capabilities for identifying and mapping archaeological sites from space; few however have explored the possibilities of using this technology to monitor change in archaeological sites over time. Here very high resolution (CORONA; Google Earth) and medium resolution (ASTER) satellite imagery is used to attempt to detect change over time at eight Neolithic sites on the Chengdu Plain in Sichuan, China. In the first case CORONA images from 1971 are compared with modern Google Earth images to determine if above ground site features have degraded over time. In the second section ASTER imagery from 2001 and 2011 is classified and compared to determine changes in land use patterns around sites. Conclusions are then drawn about both the usefulness of these methods for archaeology and about the state of preservation of archaeological sites on the Chengdu Plain.

Danielle Rappaport- Land cover classification with a focus on cloud and cloud shadow removal

Brazil’s Atlantic Forest harbors both high levels of species diversity and endemism.  Yet its extensive history of anthropogenic disturbance has resulted in equally striking levels of forest fragmentation.  Forest fragmentation – the process of splitting a once-contiguous habitat into smaller, more isolated patches – is one of the central causes of native species loss and habitat degradation, resulting in both a reduction in the original habitat area, and a modification in the spatial configuration of remaining fragments.  To safeguard these imperiled resources, there should be a concerted conservation effort to restore forest connections within this region.  Yet, in order to optimize restoration actions aimed at increasing connectivity, practitioners need landscape-scale spatial information that can assist them in analyzing the landscape structure (i.e. composition and configuration of forest fragments). 

This paper intends to support such goals within the 69,170 ha landscape mosaic in southern Bahia that extends south from the Serra do Conduru State Park to the Boa Esperança Municipal Park. These 69,170 hectares are known as the Conduru-Esperança Mini Corridor.   In this paper, I argue that such analyses can be produced through generating a supervised classification using Landsat satellite imagery.  Owing to its 30 m multispectral spatial resolution, and broad accessibility (images since 1972 available at no cost), Landsat is frequently employed to generate land cover classification analyses.