OEFS Abstracts - Spring 2004

Liz Petruska – Examining changing forestry practices in the industrial and non- industrial forested landscapes of Maine

In 2003 the Maine Forest Service published the Fourth Annual Inventory Report on Maine’s Forests. Combined data gathered between 2000 and 2003 indicates that 90% of Maine remains forested and that 97% of forestlands are classified as productive timberlands. The report also notes that significant changes in timberland ownership have occurred during this time period. The Nonindustrial Private Landowner class acquired an additional 1.9 million acres while the Forest Industry class experienced a corresponding decrease of 1.6 million acres. Most forest disturbances within the state are the result of timber harvesting. Landowners in Maine use three types of harvesting systems: selection, shelterwood and clearcutting. Satellite images that cover large areas of Maine offer an opportunity to monitor land use and forest change over time and contribute to inventory and management efforts.

The goal of this project is to use satellite imagery and remote sensing techniques to detect forest change over a 14-year period. More specifically, it will use Landsat TM images to examine the effects of the 1990 Forestry Practices Act on the use of clearcut harvesting practices. It has been suggested that despite the good intentions of the act, clearcuts have actually increased since its passage, creating increased habitat fragmentation and possibly even an increase in the total land area being clearcut. A second objective is to determine whether the management practices of industrial and non-industrial owners have different effects on the composition of the forest landscape.


Erik Hayward – Using Landsat and DEM data to detect landcover change in northwestern Maine

Remote sensing can be a powerful tool for detecting large-scale changes in both landcover type and use of natural resources. It is particularly useful for distinguishing between vegetated and non-vegetated surfaces. For this reason, remotely sensed data is a fitting method for detecting landcover change in the forests of northern Maine. In this project I have used three Landsat TM images as well as a digital elevation model to detect landcover changes in three different ways.

Although timber companies have a large impact on the forests in Maine, legislation known as the Maine Forest Practices Act was enacted in 1991. The Forest Practices Act regulated the harvesting methods of timber companies. It was also the objective of this project to see if remote sensing techniques aimed at landcover change could detect the change this legislation may have had on timber harvesting practices.


Trent Malcolm – Estimating deforestation rates on the island of Sumatra, Indonesia

Tropical forests are some of the most ecologically complex and biodiversity-rich ecosystems in the world. However, ongoing deforestation of tropical forests across the globe poses one of the greatest threats to the Earth’s dwindling biodiversity and is a driving force in global climate change. Indonesia is endowed with ten percent of all of the world’s tropical forests, trailing only Brazil and the Democratic Republic of Congo, but is experiencing unprecedented rates of forest loss. Estimating rates of deforestation in Indonesia has proven difficult, but estimates range from 263,000 to 2,400,000 hectare per year. Deforestation on Sumatra, Indonesia’s second largest island, has resulted in the loss of 6.7 million hectares of forest cover in the past 12 years. If current rates of forest destruction continue unabated, scientists predict that all of Sumatra’s tropical forest will be lost by 2010.

In this study, I used remote sensing techniques to estimate the extent of deforestation near Palembang on the Island of Sumatra. I used standard supervised classification techniques to calculate a coarse estimate of forest loss during a fifteen-year period coinciding with the peak of Indonesia’s transmigration program. I also compared the effectiveness of two different vegetation indices in revealing changes in land cover type - the normalized difference vegetation index (NDVI) and the Tasseled Cap transformation. Vegetation indices are mathematical combinations of different spectral bands known to be “sensitive indicators of the presence and condition of green vegetation”.


Ken Shono – Deforestation and land use change in northern Thailand between 1985 and 1996

Located in the monsoon belt of Southeast Asia, almost the entire land of Thailand is a potentially forested ecosystem. Thailand has undergone rapid development and modernization in the past 50 years, and the process still continues today with the growth of regional cities, industrialization, and improvement of living standards. The country’s natural resources have suffered from degradation as they have been utilized unsustainably to support the expanding population and commerce. Forestry in Thailand has historically focused on utilization of the abundant forest resources, but in the mid 1900’s, the monarchy government began efforts to control exploitation of dwindling forests. Deforestation continued despite these efforts, and forest cover in Thailand had declined from 75% in 1945 to 50% in 1960. In response, various protection and conservation measures were introduced, and the government decreed that forest cover should be maintained at 50%. However, deforestation through encroachment and illegal logging persisted, reaching a peak deforestation rate of more than 2% annually between 1976 and 1978. A series of major landslides and floods claimed hundreds of human lives in 1989. Recognizing that these natural disasters were exacerbated by the loss and degradation of forests, the government introduced a complete ban of logging in 1989 and terminated all existing logging concessions.

This study analyzes land cover change in Chiang Mai Province of northern Thailand near the city of Chiang Mai (18° 47’ 25” N, 98° 58’ 54” E) between 1985 and 1996. Chiang Mai province is one of the most heavily populated in Thailand with a population of over 1.5 million. Chiang Mai city is the regional capital and supports a population 175,000 people. The majority (80%) of Chiang Mai residents earn a living through agriculture, with tourism being the second largest income earner. The results of this study will allow for comparison of deforestation rate around Chiang Mai with the national average. It will also give insight into how modernization and spread of irrigated agriculture has changed land use patterns in the region. The study will also compare two different methods of assigning pixels to classes defined by training regions, namely maximum likelihood and minimum-distance-to-mean, and their effects on the outcome.


Megan Mattox – An analysis of forest fragmentation on the Thames River watershed

Forested landscapes in the Thames River Watershed in Northeast Connecticut are under pressure from alternative land uses including development for housing and shopping malls. As a part of a larger evaluation of the current extent, drivers, and future patterns of change in forested landscapes in this region, this study analyzed the changes in extent of forest cover using the classification of a land cover in a time series of Landsat TM images and the spatial distribution of forest cover based on multiple landscape metrics. The results indicate that while the net change in forest cover on the landscape has changed by less than 3% from 1995 to 2002, the forests of this region are becoming increasingly fragmented across the landscape. These results, in conjunction with concurrent studies help to better understand and therefore predict the threats to the loss of productive forests in the region.


Leigh Baker – Vegetative classification of Grand Teton National Park using satellite imagery

Classifying landscapes into vegetative communities is a tool often used by natural resource professionals. The creation of complete and reliable vegetative maps of a region allows one to make estimates of the overall health of the ecosystem, trends of the distribution of plant species, as well as inferences about the migration routes and habitat of many wildlife species. Remote sensing methods are commonly used to map vegetation for the scientific study and monitoring of large scale ecosystems. While field observations have a higher level of accuracy, they are often time consuming to carry out and rarely cover the entire region of analysis. Images acquired from remotely sensed data can be used to determine an area’s topography, soils, geology, and landscape history. Remote sensing is also useful in the study of biodiversity assessment and trends because it demonstrates the changes occurring in a landscape over time.

The primary goals of this project were to:

  1. Utilize remote sensing methods to analyze the vegetative communities of Grand Teton National Park.
  2. Perform a quantitative assessment of the sensitivity of the supervised classification system to the training regions selected by the classifier.
  3. Perform a qualitative assessment of the accuracy of the classification using land cover base maps and Digital Elevation Models.
  4. Advance my skill level and gain expertise in image interpretation and navigation

Ying Qiu – Application of principle components analysis and entropy to urban sprawl in Beijing, China

The concept of urban sprawl is already well known, although there isn’t a unique definition for urban sprawl. However, most definitions of urban sprawl use common indicators such as increased traffic, auto-dependent development, low-density housing spilling over into farmland, strip malls, and scattered development, etc. It is also accepted that urban sprawl is unsustainable for human communities because of various environmental and economic issues.

Remote sensing and GIS are two technologies developed recently in analyzing urban sprawl. The ability of modeling in multiple temporal and spatial scales makes remote sensing appealing to research in urban sprawl. Many such studies have been conducted successfully. But remote sensing technology is rarely used alone. Instead, GIS data is often tied with remotely sensed data so that they can provide more valuable up-to-date, fine resolution information.

In this project, I applied these two technologies for detecting the change of urban area and measuring the degree of urban sprawl between eight years from 1992 to 2000 in Beijing which is the capital of China. In addition, I proposed to examine the influence of traffic system on urban sprawl by using the entropy method. Beijing gained a rapid development, as many cities in China. In this metropolis city, urban sprawl appears seriously by occupying the open space such as agriculture and forest with the development of economics.


Sharon GulickUrban sprawl in Tucson

“ ‘There’s a two-sentence law of sprawl,’ said David Taylor, a city planner who has charted Tucson’s growth for over a decade. ‘The land is cheaper at the edge. The profit is in the dirt. Why do builders build at the edge? Duh. So where do customers buy their houses? At the edge.’”

Economically, with the current regulatory structure fostering suburban development, there are many incentives to move to the suburbs. One can purchase land and a bigger living space more cheaply. Jobs are also moving to the fringe where large business developments are cheaper to build. In addition, people have more of their own space, neighborhoods are safer and quieter, and the quality of the schools is better.

But there is also an extensive list of often hidden or intangible negative consequences, which include:

  • Increased air pollution due to longer commutes
  • Fragmentation of native habitats
  • Permanent alterations to natural resources and an unbalanced loss of resources in relation to the growth of the population
  • Increased pressure on scarce water resources—Depletion of the aquifers causes Tucson to subside and average of ~1 inch per year.
  • Decreased water quality
  • Economic impacts including erosion of the tourism industry
  • Increased pressure on infrastructure
  • Inner city decay and a declining tax base

Balancing residents’ desire to move to the suburbs with the negative consequences of sprawl pose significant challenges to city and local management efforts. Initiatives used to address sprawl in Tucson by local planners include development restrictions in specific natural areas, tax incentives, and zoning regulations. Temporal analysis of satellite imagery could be invaluable in helping planners and managers to understand the dynamics of sprawl in their city and aid them in making informed decisions regarding urban growth.


Michael RitgerLandcover change analysis of Albuquerque, NM and the New York Philadelphia metro areas

This project used supervised classification as a tool to identify land cover change during approximately the 1990s in Albuquerque, New Mexico and the New Jersey area (“NJ”).  Landsat TM data from the late 1980s and Landsat ETM+ data from 1999 and 2002 were used in the analysis.  Of particular interest was the phenomenon of suburban sprawl, which was prevalent in these locations during the study period.  The analysis clearly identified areas of development in arid Albuquerque, but was unsuccessful in NJ; the difference may have been the consequence of meteorological factors, the differences in data types, or changes in the properties suburban land cover.


Jessica BarnesInterannual variability in the agricultural land cover of the Khabour River Basin, 2000-2003

In the summer of 1999, the Khabour River in Northeast Syria – one of two major tributaries to the Euphrates – ran dry for the first time in its recorded history (Zaitchik et al, 2003). The marked environmental changes which have taken place in this river basin over the past fifty years, as water has been diverted into state-run irrigation canals and wells have been sunk to tap the groundwater have been documented as part of Yale’s South West Asia Project (SWAP).

The fact that the river ran dry in 1999 was seen as a critical indicator of the severe consequences of this transition in irrigation regimes. But what has happened since that date? Has agriculture been permanently impacted by the changes in the river’s regime? How much agricultural production has there been in recent years? Is the interannual variability in cropped area anthropogenic or climatic in origin? These are some of the questions which this paper will seek to answer.


Charlie LiuUsing NDVI to compare changes in deciduous leaf out in the US between 1989 and 1995

Hypothetically, leaf out patterns and their corresponding NDVI signature, as well as deciduous forest range, would be expected to shift over the years as a result of global warming, and this has been qualitatively observed in Northern latitudes areas such as the Canadian Boreal forest.

This paper examines the use of 1 km resolution, two-week composite NDVI data from AVHRR images of the contuminerous United States to detect two phenomena that would indicate vegetation change in the Eastern United States due to global warming: 1) earlier leaf out date or “onset of vegetation green-up” (OVG) in deciduous trees in more recent years and 2) a northward shift of deciduous trees in the Eastern U.S. between 1989 and 1995. No trend in OVG was observable due to “noise” from anomalous lower NDVI signatures in 1992, which may have been caused by global cooling due to the eruption of Mt. Pinatubo. The data obtained was likewise insufficient to convincingly show a northward shift of deciduous trees, especially as the results were very dependent on the method of data processing. While no meaningful results were obtained this time around, a number of techniques were developed in the process that may be of some use to future remote sensing analyses of change over time, so this paper will proceed to discuss the project performed.


Heidi BrownClassification of urban areas to facilitate prediction of mosquito density

As West Nile virus continues to spread across the nation and southward through the Americas, mosquito surveillance is becoming more important. The goal of this classification project is to use remote sensing to predict optimal light trap placement to facilitate the detection of favorable urban mosquito environments. The high variability of urban areas makes classification difficult, but we were able to classify the New Haven area into ten distinct regions. We found that limiting supervised classification to bands 3, 4, and 5, after a preliminary unsupervised classification, produced the most distinct, relevant classes. Future work is to test the classes using light traps


Laura KruegerUrban and suburban growth into tick and mosquito areas leading to risk predictions

Remote sensing data has been used to investigate the relationship between landscape characteristics and vector borne disease risk.  Urban and suburban growth into areas endemic for disease vectors has occurred in Connecticut over the course of the last 15 years. This study attempts to identify and quantify land cover change around the Hartford metropolitan area from 1985-2000 using remote sensing.  Human risk for West Nile virus and Lyme disease was evaluated using the NDVI, Tasseled Cap greenness, and Tasseled Cap wetness vegetation indices


Lee CohnstaedtAbiotic factors influencing mosquito distribution and disease risk in Sardinia

Malaria was eradicated from the island of Sardinia, Italy over 50 years ago but the principle disease, Anopheles labranchaie is still present which creates the ever present question, can a malaria epidemic return to the island?  By using a single day’s remotely sensed reflectance and temperature values combined with elevation information I was able to gather enough information to predict on a course scale the amount of area for each land cover type on the island. I used the basic premise that organisms and their environments are intricately linked to each other, either can be used to define the other. I identified land cover regions based on the organisms found within them (supervised classification) and compare those regions to land cover regions created based on spectral characteristics (unsupervised classification).    All classification methods used to identify populated areas bordering vector prevalent areas created disease risk maps of Sardinia, Italy were created from the classification methods and based on human distribution and malaria vector populations.


Annie Gatewood – Characterization of deciduous forest phenology in the northeastern US using MODIS

The black-legged tick, Ixodes scapularis, is the vector for at least three zoonotic disease agents in the central and eastern United States, including Borrelia burgdorferi, the causative agent of Lyme disease. Risk for Lyme disease in this area depends on the presence of I. scapularis, which in turn is dependent on the availability of a suitable habitat. I. scapularis requires a temperate climate, and a humid, shady habitat typically found in leaf litter in deciduous and mixed forests. I. scapularis is sometimes called the deer tick, since the adult stage of the tick feeds on deer as its primary host, and it has been shown that deer are essential for the maintenance of I. scapularis populations. Although it is generally understood that climate and land cover are important environmental determinants of I. scapularis presence or absence (and in turn, of Lyme disease risk), the exact factors that influence the seasonal abundance of host-seeking ticks in an area where I. scapularis is established are complex and difficult to quantitatively define.

A remotely-sensed correlate of measured tick seasonal abundance data from this study would be useful for both planning the timing of future field sampling and for predicting the seasonal risk of Lyme disease. Since both deciduous forest phenology and seasonal tick abundance seem to be related to annual cycles of temperature and photoperiod, an attempt to correlate deciduous forest phenology with seasonal tick abundance cycles is a logical first step toward building a model of seasonal Lyme disease risk. The 16-day composite MODIS normalized difference vegetation index (NDVI) product provides a simple way to characterize deciduous forest phenology in the proposed tick sampling sites for this study by providing a measure of vegetation density for 16-day intervals over the course of a growing season.


Kristin Anderson – Marsh subsidence on the Mississippi Delta

The Mississippi delta problem is one that is ideal for remote sensing.  Wetlands are delicate areas, and extensive ground studies of the region are likely to cause further damage.  By using satellites, we are able to measure change in the region and assess the effectiveness of potential solutions to the subsidence problem, without contributing to the ongoing damage.

For this project, satellite images were applied to analysis of land cover change, possible subsidence, and overall marsh vegetation health.


Robin Barr – Classifying the forests of Mt Kenya for use in change analysis 1986-2000

In this study, a Landsat ETM image taken on February 5, 2000 of the Mt. Kenya region in the nation of Kenya, East Africa (see Figure 1) was used to compare different classification techniques for identifying forest and agricultural belts surrounding Mt. Kenya. The most interesting aspect of this tropical montane system is the huge diversity of ecotones and plant communities which occurs over small areas of space due to dramatic rises in elevation which coincide with dramatic changes in climate. Altitudinal gradients in temperature and aspect-related gradients in rainfall create ‘belts’ and ‘zones’ of vegetation surrounding the mountain. Such ‘belt-like’ diversity is commonly found in tropical montane regions. For this reason, knowledge regarding effective techniques that can be employed to classify such regions is widely applicable.

There is also a great diversity in the agricultural regions surrounding Mt. Kenya. Most of the agriculture surrounding the mountain is smallholder agroforestry, with a portion of the land usually dedicated toward a cash-crop appropriate to the climate. Higher-altitude belts of smallholder tea, followed by mid-altitude belts of smallholder coffee and horticulture, and lower altitude belts of smallholder cotton and tobacco can be found outside the protected areas of Mt. Kenya. In addition to altitude, the aspect of the mountain has resulted in the dry, northern face of the mountain which is now used for large scale agriculture and grazing.  Since smallholder agriculture is the dominant form of agriculture throughout the developing world, questions of whether or not remote sensing can be a useful tool is assessing farmed landscapes are also very important


Christopher Riely – Land use and cover change in and around the Yale Meyer’s forest between 1985 and 2001

The goal of this remote sensing project was to perform a qualitative (observational) and quantitative analysis of land cover and land use change between 1985 and 2001 on a 165-square mile tract of land in the vicinity of the Yale-Myers Forest.  The forest is located in rural northeastern Connecticut, but lies within an hour’s drive of three medium-sized cities.  ER Mapper software was used to process two Landsat TM images taken in late April 1985 and 2001. 

Results were based on supervised classification of the two images and a comparative NDVI analysis.  Warm spring weather appears to have arrived earlier in 2001 than 1985 – the 2001 image shows more growing fields and the 1985 image shows more bare ones, but the total acreage classified as fields was nearly the same in both images.  An “urban” cover type composed mostly of structures and roads increased by 38%.  Most development occurred along roads in rural areas and around the shores of lakes and ponds.  Compared to some similar locations in southern New England (historically rural areas within 50 miles of cities), land cover and land use in the study area remained relatively stable.  Questions and problems encountered demonstrated that land cover and land use change analyses are only as accurate as the methods used to classify the different land cover types


Alexandra Ponette – Landuse/Landcover classification of Central Veracruz, Mexico

Growing interest in local and landscape level interactions between forest patches and adjoining ecosystems has fueled a fast-growing literature on matrix environment.  These studies are particularly useful in areas dominated by farmland where forest remnants are embedded in a variety of land use/land cover types differing in vegetational and structural complexity, and land use intensity. The aim of this project is to conduct a land use/land cover classification of a fragmented forest region in Central Veracruz, Mexico.  This is the first phase of a long-term study that will examine the effects of matrix heterogeneity on edge-related gradients in cloud forest fragment microenvironments


Julie Velásquez Runk – Landcover change in Wounaan Village of Eastern Panama between 1985 and 2003

In this study I assess land cover change in eastern Panama from 1985 to 2002 using Landsat TM imagery and then look at those changes as a finer scale by examining the fit between these remotely sensed data, georeferenced land use data, and ethnographic data, and historical data in a Wounaan indigenous community in the region.


Nadia Khan – Change analyses east of Aleppo and near Lake Assad, Syria (1984 -1999)

The land east of Aleppo and next to Lake Assad, Syria is a region in which irrigated land use has change dramatically from 1984 to 1999. Remote sensing analyses provides the means to determine and quantify the changes in vegetation in this region. Using two  Landsat TM images from October 1984 and October 1999, I will georeference the two raw images using 12 ground control points. I will then use a Normalized Difference Vegetation Index to quantify changes in vegetation over this 15 year period of time.


Rosemarie Mannik – Studying changing agricultural practices in the Middle East

In 1994, Syria passed a ban on rain fed agriculture within the arid rangelands, or Badia, covering approximately 55% of the country, located mainly in the southern, central and north-central regions of the country.  This region receives 200mm or less of rainfall per year.  This ban was passed based on perceived degradation occurring in the region as agricultural crops were grown on marginal lands.  Barley is the main crop grown in this area.

This analysis is an attempt to find a good method of analysis to identify agriculture remaining in the area using satellite images.  Four methods were analyzed to determine how well they classified current agriculture within a portion of the region affected by the 1994 ban.  Of particular importance was to differentiate current agriculture from naturally occurring vegetation.


John Williams – Radiative characteristics of a super typhoon

Several properties of clouds can be examined through satellite remote sensing.  The goal of this project was to look at a few images of a super typhoon captured on the MODIS/TERRA instrument and specifically examine selected properties that can be determined from the 16 brightness temperature bands.  Since multiple images were obtained over the course of the typhoon’s weakening, an effort was also made to relate the observed properties to the strength of the storm.  The first property looked at was cloud altitude.  A calculation of cloud altitude requires both an accurate method of determining cloud top temperature along with a means to relate cloud temperature to height.  Secondly, the independently varying emissivity of ice and water makes possible a method of determining the thermodynamic phase of clouds.  Both of these properties were compared over the different images and also were compared to each other, since it is reasonable to assume that cloud temperature is related to phase.  Lastly, some information regarding the relative humidity around the storm and within the eye was obtained through use of a band whose intensity is related to both temperature and water vapor content.  The question to be answered is how these properties correlate with the known strength of the typhoon as it weakens.


Yanping Li – Studying clouds in the Great Plains using MODIS

The mountain-plain circulation over Rocky Mountains shows such phenomena during warm season: in the morning, cumulus clouds form over the ridges and propagate slightly downwind. Steering winds carry these clouds away from their source, and then these clouds dissipate. Later, larger clouds replace them when mountain circulations get stronger and when the atmosphere above the ridgetops get moister by evaporating cloud droplets. The newer clouds move downstream and dissipate, and are replaced by larger cumulus clouds near the ridgetops. From these cloudy regions, a few deeper clouds emerge and develop self-sustaining circulation which allow them to move away from their source without dissipating. A few of these deeper clouds become thunderstorms. These mountain thunderstorms can travel great distances. Some dissipate in the mountains. Some move to the edge of the mountains before dissipating, their outflows or gust fronts spill over into the adjacent plains, which become convergence lines and participate in the generation of afternoon convection over the plains. Some mountain thunderstorms propagate directly onto the plains, affecting areas downwind of the mountains. A few such storms become part of longer-lived systems (eg: mesoscale convective complexes) which last from one to several days and propagate far away from the Rocky Mountains.

Knowledge of these clouds properties and their variation in space and time is crucial for us to understand their formation mechanism and how they can sustain such a long time after they are away from the generation regions. To understand the cloud properties is also helpful for modeling studies.

Cloud properties can be achieved through remote sensing. There are many approaches for detecting clouds by using MODIS observations. In this paper, some current cloud detection methods are introduced and the wavelengths used in the MODIS clouds distinction are discussed.


Katherine Lin – Asian dust storms

Asian dust (AD) events have recently garnered much attention in the realm of scientific research because of the potential effects that the particles and aerosols, which comprise these episodes, could have on global climate and geochemical mass cycles.  Of particular interest to this project will be whether or not Asian dust particles and aerosols can be identified in satellite images and differentiated from cloud droplets.  The satellite images that were used in this project were from the MODIS-Terra sensor.


Brett Galimidi – Using MODIS to differentiate between clouds and particulate matter in the Himalaya

Understanding atmospheric pollution is of critical importance today, particularly with relation to climate change. As the existence of the climate change problem is generally agreed upon, the causes and effects are still under considerable study. Remote sensing, and the MODIS sensor in particular, has created an invaluable path for understanding atmospheric issues due to the presence of aerosols. Anthropogenic aerosols are the result of burning biomass and fossil fuels, as well as industrial processes. MODIS research teams are working to formulate algorithms to analyze aerosol composition in air pollution. The work, however, is still new and complex. In this project, I set out to find a process by which haze could be isolated using currently available tools. I attempt to answer the simple question: to what extent is it possible to isolate haze from a MODIS image, using commercially available tools and known remote sensing techniques, in lieu of NASA’s extensive resources? The significance of this lies in the ability isolate and eventually measure aerosols in an image if they are either the desired study target, or an impediment to the study of other features. My focus here will be primarily a process determination. While some quantification is presented, my goal is to establish an initial methodology for isolating haze and aerosols in a MODIS image. I present two approaches for isolating aerosols. The final results point to a process that can be valuable as a proxy measure for aerosol optical thickness in MODIS images


Peter Isaacson – Mineral identification and hydrothermal alteration mapping at Cuprite-Goldfield, Nevada, using Landsat ETM+, ASTER, and AVIRIS data

This project attempts to utilize three different types of data (Landsat ETM+, ASTER, AVIRIS) to map mineral abundances and to identify zones of hydrothermal alteration at Cuprite-Goldfield, Nevada. Various spectral mapping algorithms were used, including ratio images for Landsat, Spectral Angle Mapper and Spectral Unmixing. These routines produced varying degrees of success; the Spectral Unmixing routine proved to be the most concordant with the Landsat results, which were taken as the reference to compare the other results against. Hydrothermal alteration (primarily of the argillic alteration zone) was mapped reasonably well with all sensors, though more specifics could be resolved with the ASTER and especially AVIRIS data, where subzones of the argillic zone could be identified, which was not possible with the Landsat data. The AVIRIS hyperspectral data allowed more specific identification of minerals, and allowed mapping of more endmembers, but mineral identification did prove successful in the ASTER data as well.


Nell Larson – Examining the 1998 ice storm damage to the northern Adirondacks

I used remote sensing image interpretation to assess and quantify the damage done to the northern Adirondacks during the Ice Storm of 1998 at different elevations.  I used several methods, focusing on NDVI, all of which yielded similar results.


Ali Macalady – Europe’s 2003 heat wave from space – A look at changes in vegetation

Climate modelers predict that European summers will become hotter and more variable as the concentration of green house gases increase in the atmosphere. In the summer of 2003, summer temperatures broke records in Switzerland and across central Europe, and some scientists contend the conditions are a window into summer conditions in the future. In order to begin to understand how elevated summer temperatures might affect other environmental variables, this study looks at a remotely-sensed vegetation index, the Normalized Difference Vegetation Index (NDVI), over four Augusts in central Europe. For a 16-day period in August 2003, NDVI declined
by 14% over the study area, as compared to August NDVI across the more ‘normal’ summers of 2000, 2001 and 2002. Crops and grasslands experienced the greatest declines in August NDVI, but NDVI for forests and woodlands also declined. These preliminary findings may be contrary to the results of previous studies, in which yearly average NDVI increased with higher temperatures across midhigh latitude Eurasia. However, it is difficult to know whether these results are an accurate portrayal of the true interannual or seasonal variability in central European NDVI. In a
longer time series, the ‘average’ NDVI would include a broader range of yearly variability. Compared to that longer average, 2003 might not look so dramatic


Geri Kantor – Using Landsat imagery to evaluate vegetation cover change in Vietnam following defoliation with Agent Orange

Having obtained 1973 and 1992 Landsat images of Vietnam, just north of the Rung Sat Special zone, I was interested in determining if I could see mangrove regeneration in the two decade interval, along the east coast of Binh Thuan province and along the waterways leading inland from its capital city, Phan Thiet.  Phan Thiet is in the most southern portion of Vietnam’s central highlands.  It is a coastal city with a population of 100,000 just north of Saigon, and fishing is its major industry.  Phan Thiet lies in Corps Tactical Zone II, just north of the boundary with Zone III (see Figure I).  Mangrove forests occur along the southern coast of Vietnam and I suspected that the spectrally distinct feature that appears along the coastline in 1992 but not in 1973 (two years after spray missions had ended) would turn out to be mangroves. 

In addition, researchers at Columbia University are building a GIS database of herbicide spray missions during the Vietnam War, which to date includes more than 9,000 flights.  Although I would need satellite data from just before and just after herbicide application to firmly attribute mangrove destruction to Agent Orange, overlaying a vector file of spray missions onto the remotely sensed images and noting vegetation change between the two time periods might provide a very telling correlation.


Martha Bell – Land classification and elevation change in the Central Andes

For this project I explored the issue of land classification of mountainous terrain, focusing specifically on the Huarochiri region of the Peruvian Andes.  This is a ~6000 sq. km region located on the westernmost slopes of the Andes, directly east of the city of Lima.  The region  spans a range of elevations, from low river valleys (~1000m) to peaks as high as ~5000m (Atlas Departamental del Peru).

The goal of this project was to use Landsat images to identify and quantify (in terms of area) different types of ground cover, and to determine the relationship between these cover types and elevation.  Although classification proved difficult, several important findings (both environmental and analytical) were reached.   This remote sensing study is part of a larger archaeological study of current and prehispanic land use patterns in this region, image analysis can also be used to assess these patterns.


Jeremy Goetz – Exploring vegetation changes over topographic gradients in the Congo Basin

The goal of this research was to compare broadband hyperspatial IKONOS data and advanced multispectral Landsat-7 Enhanced Thematic Mapper (ETM+) data through classifying complex rainforest vegetation over topographic features.  For this purpose, IKONOS, ETM+, and Digital Elevation Model (DEM) data were acquired for southern Cameroon, a region generally considered to be representative of tropical moist evergreen and semi-deciduous forests.  Image data, in combination with elevation data, were used in an attempt to characterize forest land cover classes (upland vs. lowland).  

The study established that the broadband sensors (both IKONOS and ETM+) had serious limitations in classifying forest landcover classes.  This study was undertaken without ground-truth data so the overall accuracies for unsupervised classification of 5 complex rainforest landcover classes could not be determined.  However, it was evident no clear pattern of landcover change could be detected as elevation increased.  Unsupervised classification varied depending on the specific methodology used to generate the classification (including DEM or excluding).  This suggests that caution is needed when making comparisons between classification accuracy reported by different studies, unless their methodologies are clearly identified.  Further, the widely divergent classification results when including or excluding DEM data suggests care must be taken in deciding how best to include elevation data in an analysis of landcover change.  

When compared to each other, neither IKONOS nor ETM+ data could detect a change in the spectral signature of vegetation as elevation increased.  This result held true throughout all bands and for the resulting NDVI calculations.    

It is further argued that a low accuracy landcover map (as assumed to be here) still makes a valuable contribution to our knowledge of this hitherto poorly studied environment, provided that its limitations are understood and respected.


Elena Traister – An investigation into whether alkalinity export varies with land use over time

With atmospheric carbon dioxide concentration increasing due to anthropogenic emissions and with concerns over the ensuing greenhouse effect and further global climate change, scientists are examining the carbon cycle in detail to make predictions about the environmental response to continued CO2 emissions. An intriguing focus of carbon cycle research is concerned with the so-called “missing carbon-sink” that arises in estimating the global carbon budget. According to Sedjo, “Two very recent attempts to account for the fluxes by identifying carbon sources and sinks have provided evidence, of a missing carbon sink and that the missing sink is located somewhere in the temperate region of the Northern Hemisphere (1992).” Nemani et al. suggest that, “The strong coupling between carbon and hydrologic cycles implies that global carbon budget studies, currently dominated by temperature analyses, should consider changes in the hydrologic cycle (2002).” Raymond and Cole identify two terrestrial carbon sinks: the conversion of carbon dioxide to organic carbon by plants, and the sequestration of CO2 by chemical weathering, which results in a contribution of carbonate and bicarbonate ions (alkalinity) to surface water. They have documented an increase in alkalinity in the Mississippi River, as well as a relationship between alkalinity export and land use. This project introduces a method of using remote sensing coupled with GIS analysis and stream alkalinity data to examine the relationship between land use and fluvial inorganic carbon export in the Ohio River Subbasin of the Mississippi River Watershed.


Jonathan Cox – Desert landscape changes in the American Southwest

A Landsat TM image from May 19, 1989 and a Landsat ETM+ image from May, 31 2002 were acquired and used for this change detection and desert landscape classification project.  A NDVI composite image of these scenes as well as climate data from the NCDC suggests that more natural vegetation grew in the region southeast of Albuquerque prior to the acquisition of the 1989 image compared to the 2002 image.  Further change detection analysis was not conducted due to the inability to derive acceptable landscape classification images.  Several classification schemes were developed and are discussed.  A discussion on the future steps required to complete this study conclude the project paper.


Deborah Fillis – Examining landcover adjacent to salt marsh ecosystems

Salt marshes in Connecticut are especially critical habitat. Over half of the coastal wetlands have been directly destroyed by human development, and the remainders are threatened by non-point sources.  Therefore, remotely sensed land use/land cover data is essential to investigating the health of these marshes. Utilizing this information in an intelligent manner is the challenge of all environmental professionals. In order to do so we must understand the shortcomings of this data. Two classification techniques are analyzed to determine how these schemes diverge and to reveal the causes of variation.

The classification methods analyzed include minimum distance, which is not sensitive to variance and maximum likelihood which incorporates variance and covariance into the statistics. In the minimum distance classifier a pixel may be assigned to a class that is close to the means of a class. It is “not used where spectral classes are close to one another in the measurement space and have high variance.”  The minimum distance classifier can differentiate between classes that have similar mean, but differ in the way the means change between bands (covariance.) This classification scheme also takes variance into account often resulting in more robust results.


Sparsh Khandeshi – Vegetation variations between alpine meadows

The Inyo National Forest (INF) extending 165 miles along the California and Nevada border, is home to many natural wonders, including Mt. Whitney, Mono Lake, Mammoth Lakes Basin, and the Ancient Bristlecone pine Forest as well as seven Congressionally-designated Wildernesses.  The Ansel Adams and John Muir Wilderness Areas constitute over 800,000 acres of this region. These regions are mandated, by the Wilderness Act of 1964, to preserve ‘natural conditions’ and minimize human development, while simultaneously promoting recreational activities. In this expansive and rugged mountain terrain, meadows are conspicuous hotspots not only for productivity and biodiversity, but also recreational use.  The use of recreational commercial packstock (e.g., horses and mules) that are allowed to graze in these mountain meadows is a focal point of ecological concern and management.  Efficient management of these ecologically sensitive areas requires a thorough understanding of use patterns and ecosystem responses to use.

Because of considerable variation of characteristics both within and among meadows, it is difficult to generalize, over a broad geographic area, vulnerability to recreational impacts. The purpose of this study is to explore the potential of using multi-spectral satellite imagery to measure key meadow attributes that the literature has cited as being especially vulnerable to packstock impacts.