Spring 2016 - Project Abstracts

Chendan Yan - Observing Change in Urban Heat Island Effects and Vegetation in Shanghai

The term urban heat island refers to dense urban areas that are hotter than surrounding suburban or rural areas. 1–3°C difference in temperature can constitute what is traditionally defined as urban heat island effect (Environmental Protection Agency). Urban heat island effect has caught great attention with the rapid urbanization processes in recent years. Its effects are more significant in summer months due to intensely warmer temperature. Effects of UHI are quite concerning, ranging from “increasing summertime peak energy demand, air conditioning costs, air pollution” to “greenhouse gas emissions, heat-related illness and mortality, and water quality” (Environmental Protection Agency). It is important for city governments and planning agencies around the world to understand the likely consequences caused by heat island effect and come up with affordable and novel solutions.

Stephanie Ng - Examining the Urban Heat Island effects in Singapore

Many studies have shown the relationship between land use and ambient air temperature, in particular various studies have looked specifically at the role of vegetation and green spaces, within urban environment, in lowering ambient air temperature. Singapore adopted a comprehensive land use plan, Concept Plan 1991, followed by a sustainability plan in 1992, The Singapore Green Plan which introduce green spaces within the urban environment. Through the use of satellite images from Landsat, this study investigates the influence of these policies changes on the land use between 1991 and 2013, the change in green cover over the same period and the relationship between land use and ambient air temperature. Results showed an increase in land area of ‘Parks and Open Space’ and ‘Urban residential’ classes as well as an increase in green cover from 29% to 39% between 1991 and 2013. The study also found that ‘Parks and Open Space’ class is about 1^oC lower in ambient temperature compared to the urban land use classes.

Kate Farley - Recovery of Mountain Top Removal sites in eastern Kentucky over time

The Appalachian Mountains have a long history of coal mining. In recent years, however, traditional underground mines have been replaced by surface mines. The practice of surface mining in the Appalachians is commonly known as “mountaintop removal mining” (MTR) because mountain tops are leveled and valleys are filled in order to reach coal seams within the mountains. MTR has been a valuable economic activity for many in the region—Appalachia is one of the poorest regions of the country, and coal mining has provided a livelihood for many families in the area for generations. However, many environmentalists and local residents are concerned with the ecological impacts of MTR. MTR drastically changes the contours of the landscape and disrupts local ecosystems, and is linked to erosion and leaching of toxic materials into local watersheds. MTR operators often include plans for site remediation or restoration to be implemented once mines are exhausted, but skepticism remains about the effectiveness of such plans. In this paper, I use NDVI as a tool for visualizing the impact of MTR on the local landscape. Is active MTR and/or recovery from MTR operations detectable using remote sensing?

Rachel Gulbraa - Land use change on Easter Island

On an island once denuded of all its trees, erosion is an undeniable problem on Easter Island and a threat to the archaeological tourism that its economy now depends on (Legrand 2013). However, erosion on the island has only recently begun to be addressed in the academic community (Mieth and Bork 2005). Likewise, urban expansion— in large part due to efforts to accommodate such tourism— is another potential threat on an island that only has 170 square kilometers of land total. This study aimed to use remote sensing via Landsat imagery, classification, and change detection, to deduce how erosion and urban expansion have increased over the last decade and a half, and whether efforts to plant trees to counteract erosion had expanded. Two Landsat images from fifteen years apart were selected, spatially subset, and masked of clouds. A supervised minimum distance classification was then run on both images to compare changes in land class. Change detection statistics were completed, revealing an increase in erosion, urbanization, and tree range. However, results were confounded by classification errors, especially in the urban class. While the results may accurately indicate a general trend, due to the classification errors they should not be taken as reliable accounts of the specific change in area of these land class categories. Excessive mixed pixels in urban and vegetated areas, as well as cloud shadows aided in this confusion. In the future, a higher resolution image could be used to produce a more accurate classified image and subsequent change statistics, but the resolution of the Landsat images makes them insufficient as tool for this task.

Martin Becker - Land use change in the Araucarias Biosphere Reserve in Chile

The objective of this project is to generate an ecosystems classification that can be used by land managers as a reference about the main vegetation types and land uses present in the Araucarias Biosphere Reserve. A secondary objective is to identify areas where land uses changes are evident between the years 2005 and 2016, especially the impact of the forest industry in terms of substitution of agricultural lands and native forest by exotic conifer plantations.
The project was conducted using two Landsat images (Landsat 4,5 from Match 2005 and Landsat 8 from March 2016) at a spatial resolution of 30 meters. Manual classification techniques (maximum likelihood) showed much better results than unsupervised classifications (K-means and Isodata). A digital elevation model was also used to refine the image classification based on elevation ranges.
The main results show that the Andean region is dominated by different forest types depending on elevation ranges. On the drier areas at the eastern side of the Andes mountain ranges, steppe ecosystems clearly dominate the landscape. In contrast agricultural croplands and grasslands are predominant in the central valley at the western portion of the study area. In terms of land use change, three major areas were identified where substitution of native forests or agricultural lands by exotic conifer plantations has been significant. Policy makers and land managers should consider evaluating possible environmental impacts arising from these land use changes in the identified areas.

Chelsea Judy - Land use change and forest degradation in Ethiopia

Deforestation and forest degradation continue to be enduring biophysical and socio-economic challenges across the global tropics. Although relative consensus has been reached regarding what constitutes deforestation mechanically, and how to measure it, contention still exists between scholars across disciplines regarding the definition of forest degradation, and perhaps more importantly, how to measure it. This project aims to use a myriad of remote sensing methodologies to distinguish between deforestation and forest degradation processes in the Oromia regional state of Ethiopia. Using Landsat images from 2001 to 2015, change detection results, while confirming that there has been significant land cover change in the study region, remain mixed across methods. Unsupervised and supervised classifications proved the most robust tools for detecting deforestation between 2001 and 2015 in the region, but remain too blunt for detecting forest degradation. Two methods of detecting forest degradation were attempted: a temporal Tasseled-Cap transformation analysis and Linear Spectral Unmixing with Decision Tree classification. Both methods, while providing useful and more nuanced information about the extent and kind of change detection than the supervised classification method offered, still remain too erroneous to confidently measure the extent of forest degradation in the study region. This project concludes that while deforestation is relatively straightforward to quantify with some degree of accuracy, using remote sensing methodologies for detecting forest degradation remains difficult and highly dependent upon one’s definition of forest degradation. However, the use of Linear Spectral Unmixing, while in this project became unfeasible, promises future pathways forward in terms of both developing a conceptual methodological model for detecting forest degradation over time and strengthening the degree of accuracy in this detection.

Serena Lau - Land cover change in California’s Central Valley

I examined the change in land cover over a study region in California’s Central Valley at two distinct time-points: July 1985 and July 1991. With a focus on how agricultural production as measured by vegetation intensity and distribution could potentially have been affected by one of the longest drought events (at the time) in California’s history, I utilized NDVI and explored how to better classify cropland from bare soil, and from cities’ urban footprints, in the hopes of capturing the nuances of this dynamic region. Overall, a slight decrease in the amount of vegetated area, as well as an overall decrease in the intensity of the vegetation as measured by NDVI, was seen.

Rachel Lowenthal - Study the Farmington River watershed

Detecting change in land cover composition over time has tremendous value in the field of water resources management. This study closely examines land cover in the Farmington River watershed, located in north-central Connecticut and south-central Massachusetts, in order to better inform water quality research in the region. Two Landsat images one from 1996 and another from 2014, were classified for land cover using minimum distance and maximum likelihood classification procedures, which were compared for accuracy. Also, change detection analyses were performed to determine whether parts of the watershed had experienced growth in urban and suburban development. Finally, the accuracy of the 2014 classified image was also investigated by groundtruth, revealing that some of the suburban regions were not accurately classified due to the land cover heterogeneity in developed areas. Ultimately, while the Farmington watershed has experienced suburban growth in the past 18 years, the details of that land cover change remain unknown. Specifically, the extent to which forested, agricultural, or wetland areas were changed to make way for suburban and urban land cover was not clearly determined. Perhaps these details would have been more noticeable by analyzing images with more seasonal contrast.

Ana Lambert - Change detection in Lake Chapala

Lake Chapala, Mexico’s largest lake, faces a growing tension between the promotion of economic development (mainly through agribusiness and livestock farming), and the need to strengthen conservation efforts to preserve this important ecosystem. High water extraction rates, siltation coming from extraction of sediments in highlands and nutrient loading from agricultural fertilizers are just a few of the many threats Lake Chapala is facing.  This paper documents a change analysis of Lerma-Chapala sub-basin watershed landscape with the support of remote sensing.  The objective of this research is to define and prioritize water management activities in the basin.  Landsat 5 images from two dates (2011/1/12 and 2000/1/14) were compared to detect shifts in the lake in regards to: (a) vegetation-agricultural-urban-water land coverage; (b) sediments extraction activities (c) lake surface area; and (d) Normalized Difference Vegetation Index (NDVI) in the lake. As a conclusion of this study, a trend of increasing urbanization, evidence of sediments extraction and a relationship between algae hotspots and agricultural runoff from water streams were detected.  The study pinpoints that the integration of all of these threats should be considered for government policy makers.

Aisha Pasha - Changes in the Chernobyl exclusion zone

The Chernobyl Exclusion Zone is a 2,600 sq km area that was declared after the 1986 nuclear disaster that filled the region with devastating levels of radiation. In the years that followed the Soviet and later Ukrainian governments took steps to relocate approximately 120,000 people who inhabited the area. By the mid–]1990s the region was almost completely uninhabited and has remained so ever since. The purpose of this project was to investigate land cover change in the exclusion zone after nearly years of abandonment. In particular this research was interested in assessing the relationship between vegetation re–]growth and built–]environment decline, which have occurred simultaneously in the region.

Shu Tao - Pripyat - Changes 30 years after the Chernobyl Nuclear Disaster

Using three images from 1985, 1986 and 2009 acquired by Landsat 5, this project detects the changes 23 years after the Chernobyl nuclear disaster in the Chernobyl Exclusion Zone. 900 km2 subsets of these original images were used for the analysis. Clouds were removed from the images through band math using brightness temperature and reflectance. With brightness temperature, the project explores the thermal difference of the Chernobyl nuclear power plant pre- and post-disaster. The results indicate that Landsat 5 helped to confirm the nonfunctioning of the Chernobyl Nuclear Power Plant. NDVI images were used to determine the change in vegetation. Mean NDVI increased from 0.19 to 0.21. A supervised maximum likelihood classification of these images helped to determine the area in square kilometers of different classes and allowed the calculation of change statistics. The results of this analysis suggest 1) the conversion from an urban environment to coniferous forest and from agricultural land to natural vegetation, 2) construction extension due to New Safe Confinement construction and 3) deforestation in a small region due to commercial logging.

Kara Fikrig - Rat habitat for studying Leptospirosis in Salvador, Brazil

Leptospirosis, a disease caused by a bacterial infection, is the source of a large global burden of disease. The bacterium is transmitted to humans via contact with infected animals or water contaminated with the urine of infected animals, such as rats. The contribution of land cover to disease risk is not well understood; little is known about the correlation between land cover type and either rat activity or leptospirosis incidence. I analyzed a WorldView-2 image of a study site in the urban slums of Salvador, Brazil, where the Yale School of Public Health has an ongoing leptospirosis study. I used three methods to create land cover classifications that included at least three of the four features of interest: vegetation, soil, manmade structures, and wet ground. The three types of classification methods implemented were 1) a decision tree of the index layer stack, 2) a maximum likelihood supervised classification of the index layer stack, and 3) a maximum likelihood supervised classification of the calibrated image. I then compared the three classification methods in order to determine which method was most appropriate for the slum context. The two main challenges of classifying the slum landscape were the high degree of small-scale heterogeneity and the similar reflectance properties of the red tile roofs and soil. The decision tree method was the best able to capture the heterogeneity of the landscape, whereas the supervised classification of the calibrated image was able to differentiate between the red tile roofs and soil. The next step will be to use the decision tree land cover classification to correlate land cover with rat activity and leptospirosis incidence at the study site.

Rachel Arnesen - Predicting schistosomiasis risk in Zimbabwe using remote sensing data

This project is an investigation of how remote sensing methodology can be used to predict regions of high schistosomiasis risk in Mozambique. Schistosomiasis is a disease caused by parasitic worms affecting over 240 million people worldwide. It is very easily and cheaply cured and thus is a focus of public health efforts. As good ground truth data on prevalence is frequently unavailable, remote sensing methods could be useful to direct targeted drug distribution efforts. This project investigated elevation, NDVI and LST to see if there was a detectable difference between high prevalence and low prevalence regions. MODIS daily scenes and MODIS time series data from June 2005 to August 2007 was used to investigate. The districts of Morrumbene and Muecate were chosen as the former had a <10% prevalence rate while the latter had a >90% prevalence rate.

The images were downloaded from MODIS and were subsetted and masked to the districts of interest. Both the daily scenes and the time series data had to be pre-processed in order to be imported and manipulated in ENVI. Band 5 of the daily scenes had an error so it was removed. There were also issues with night LST for Muecate and thus various pixels had to be replaced with a constant temperature of 18 to avoid falsely low readings. Overall, no significant difference was seen between the two regions for either NDVI or minimum LST. Both regions were very flat, but Muecate (high prevalence) was at a higher elevation than Morrumbene (low prevalence), the opposite of what was predicted in the literature. Overall, this project was not successful in using remote sensing technology to detect high prevalence schistosomiasis areas but the field remains open for future exploration.

Olivia Walker - Detecting Land Use Patterns in Archeological Sites in the Tao River Valley

The Tao River Valley in the Gansu Province is home to many Neolithic archeological sites, known and unknown. I was provided with a very high-resolution WorldView2 image of a large portion of the Tao River Valley with 35 known archeological sites included in the image. Sampling 10 of those sites, I used the unique spectral bands featured in the WorldView2 image to search for a pattern in the archeological sites. Eventually finding that any search for a pattern would be foiled by the fact that the sites are overlain with flooded rice patty terrace farms, with either very high reflectance or very high absorbance, I looked to elevation data to seek a pattern. While I did find a pattern with that data, the result was a bit obvious (the sites faced the river and were on steep, but not too steep, slopes of the river valley). Overall, I was unsuccessful in my attempt to find a pattern, but was very successful in learning the process of trial and error with the ENVI software.

Melissa Castera - Algal Blooms In the Gulf of Mexico

This study analyzes algal blooms in the bay area delimited by the discharge of the Mississippi River and the discharge of the Apalachicola River, in the northern part of the Gulf of Mexico. Algal blooms are triggered by the input of nutrient-rich waters of the Mississippi River and are associated with severe impacts on aquatic biota, human health and local economy. The oxygen-depleted zone (dead zone) developed in 2015 on the west of the Mississippi River’s discharge, was larger than the average from 2010 to 2015. This study analyses algae biomass, measured as Chlorophyll-a (Chl) concentration, in the study area for the year 2015. The input data comprised Chlorophyll-a ocx products of MODIS Aqua. The data used were of the type L3 data 8-day composite with 4km of resolution. Images with high amount of clouds in the study area were not considered. A total of 19 images were used for the analysis, representing around 40% of the year 2015. The selected images were classified in 5 classes, according to Chl concentration (low class number corresponds to a low Chl). The class 5 represented algal blooms. The accumulated intensity of chlorophyll for a given pixel was obtained from the sum of the classes of all the images. The bay area studied showed areas of high values of algae biomass, particularly in areas close to the shore. The western half of the bay showed highest values of accumulated intensity of chlorophyll, inferring a clear connection between algal blooms and the discharges of the Alabama River and the Mississippi River. Even more interestingly, it was the identification of an area that presented a high accumulated intensity of chlorophyll. Although it was not connected directly to any river affluent to the bay, this area has a trapping effect of the nutrients that flow into it. The study also included the identification of sectors considering their variation of classification throughout all the pictures. Higher changes are more likely to occur along the shore, particularly in the eastern half of the bay that could be associated with sea level variation. Finally, the highest peak in Chl concentration happened during September. Considering that the available data in this study did not cover the entire year, it is not possible to conclude if this peak corresponds to the most severe algal bloom or is caused by the same reasons as the dead zone. Overall, this project has demonstrated that remote sensing is a valuable tool for the assessment of algal bloom in the northern part of the Gulf of Mexico. However, in future studies more dense information should be used to improve the quality of the results.

Bart DiFiore - Grazing halos around coral reefs

Grazing by herbivorous fish and echinoderms is an essential process in the maintenance of hard corals in coral reef ecosystems.  Due to the vulnerable nature of the world’s coral reefs, quantifying grazing and shifts in benthic cover through time and space from remote sensors is essential to successful management of coral reefs.  Using three high-resolution satellite images (World View 2, GeoEye, and Quickbird), this study explored the methodologies associated with preprocessing coastal imagery in order to quantify shifts in benthic cover.  Specifically, this study tested the efficacy of Stumpf’s bathymetric derivation and Lyzegna’s water column correction towards improving the classification of benthic cover in shallow (>30 m).  A full 2-m resolution bathymetric model was developed for the northern portion of the South Water Cay Marine Reserve, Belize.  Without in situ field data to validate the classifications, it proved difficult to immediately judge the different depth correction algorithms.  However, analysis of the various depth corrected classifications based on visually interpreted ground truth points suggested that depth correction techniques may not be necessary within the studied images.  The depth correction techniques did improve accuracy, however, when the classifications were conducted as unsupervised K-Means classifications.  Initial change detection for the years 2005, 2009, and 2015 was undertaken on a spatial subset of the overall scene and suggested significant changes in benthic cover over the period.  Higher resolution depth data and in situ training regions must be pinpointed before such changes can be validated.

John McNamara - Air Pollution Effects on the Radiative Properties of Clouds in West Africa

Lagos, Nigeria is one of the largest and fastest-growing cities in the world, with   As the world’s population becomes more concentrated in urban centers anthropogenic climate change threatens to increase the frequency of severe weather events such as droughts and heat waves, it is important  to understand how these changes will affect the world’s largest cities.  I used Landsat 5 TM and Landsat 8 OLI data to examine changes in land-cover and urban heat island effects in Lagos between 1984 and 2015.  In addition to calculating and comparing brightness temperature and albedo, I performed supervised maximum-likelihood classifications and tassel cap transformations on both images.  The data shows a significant amount of urban sprawl developing to the north of Lagos between 1984 and 2015, as well as more severe urban heat island effects, and a higher overall albedo in the city.

Christine Tsai - Cloud Phase Discrimination using MODIS Imagery

Cloud microphysical and optical properties modulate cloud radiative properties and thus play an important role in the climate system. Determining cloud thermodynamic phase is the first step in calculating these parameters when derived from satellite imagery and when representing them in global climate models. In this project, I discuss and test a variety techniques used to distinguish liquid water and ice clouds using bands from the thermal infrared, shortwave infrared, near infrared, and visible spectrum. I compare these results to the MODIS Cloud Product phase determination and temperature thresholds. This project looks at clouds over the South Island of New Zealand as a case study.

Annie Bui - NDVI study of drought conditions in California

California has been experiencing cycles of drought, ranging from abnormally dry to exceptional drought between 2000 and 2015. Insufficient precipitation due to drought can lead to poor ecosystem health and generate unpredictable economic costs. Normalized Difference Vegetation Index (NDVI) has been a useful tool for monitoring drought conditions, which can be used for communities to develop response plans to drought. This study uses MODIS time series data from 2000 to 2015 to determine if there is a correlation between NDVI and drought in Central California by examining inter-annual variability of NDVI values, and to determine if there is a correlation between NDVI, drought, and land cover classes. The results indicate there is a negative correlation between NDVI and drought intensity. There was no conclusive evidence for a correlation between land cover classes and NDVI.

Lily Hahn - Impact of 2012 Colorado Drought on Fire Risk Factors

In 2012, more than half of the United States experienced record-breaking warmth and exceptionally dry conditions. With moderate to extreme drought dominating more than half the nation, the average size of wildfires in 2012 was the largest observed since 2000. As climate change is increasing the frequency and severity of drought conditions in many regions of the world, understanding the effects of drought on fire risk is becoming increasingly important for informing fire mitigation strategies. The objectives of this study were (i) to examine the correlation of drought and fire risk change in Colorado for different classes of vegetation; (ii) to determine which fire risk factors were dominant; and (iii) to assess the fire risk index proposed by Chowdhury and Hassan [1]. MODIS/Terra 8-day composite surface reflectance and surface temperature images were obtained and subset to Colorado for a period prior to the drought (June 10 – June 17, 2010), during the drought but prior to the 2012 Colorado summer wildfires (June 9 – June 16, 2012), and after fire containment (July 11 - July 18, 2012).* Images were processed to exclude poor quality pixels, and a supervised classification was performed for the June 2010 and June 2012 surface reflectance images. Using a modified version of the Chowdhury and Hassan fire risk index, pixel-level fire risk for the June images was determined using three risk factors: Normalized Difference Vegetation Index (NDVI, derived from surface reflectance images), Normalized Multi-band Drought Index (NMDI, derived from surface reflectance images), and surface temperature (ST, given in surface temperature images). Lastly, the July 2012 post-fire image was compared to the June 2012 pre-fire image to depict fire locations and to validate six USGS fire perimeter shapefiles, which were then used to subset the June 2012 fire risk data and to assess the fire risk index. Analysis revealed that during the drought (i) fire risk increased for all vegetation, with shrubland and grassland experiencing the greatest change; (ii) fire risk was dominated by NMDI and ST risk factors for agriculture, NDVI and ST risk factors for shrubland and grassland, and the NMDI risk factor for forest; and (iii) 66.5% of the fire pixels fell within the “very high” to “moderate” fire risk categories. While the original fire risk index was not highly predictive of fire regions, a revised forest fire risk index indicated that 98.6% of the forest fire pixels fell within the “very high” to “moderate” risk categories. More work is necessary to refine and validate the fire risk index, but these results illustrate the severe impact of drought on fire danger and the relevance of satellite data for understanding and forecasting fire risk.

Nicholas Brown - Elevation and vegetation Relationship in the Peruvian Central Highlands

This report attempts to classify ecological zones across the vertical landscape of the Peruvian Central Andes on the basis of vegetation (NDVI) as perceived through remote sensing Landsat 7 ETM+ satellite imagery. An unsupervised IsoData classification was carried out of an Landsat NDVI time series (September 2000, December 2000, June 2001) for the 4,865 km² study area, located in the Chaupihuaranga valley of highland Pasco and Huánuco, in order to identify ecological zones that react similarly to changes in intra-annual water availability (precipitation). This vegetation-based classification is compared to the widely accepted altitudinal zonation for Central Andean geography, as elaborated by Pulgar Vidal (1938) and Parsons et al. (2000). Comparison between altitudinal thresholding and NDVI time series classification of the same landscape revealed significant differences that highlight factors other than elevation that influence ecological productivity, including irrigation potential and slope steepness.

Maria Pozimski - Effects on Treeline Classification on Landsat Imagery

The correction delineation of treelines is – both conceptually and in reality – fraught with difficulty. The goal of this project is to test whether the incorporation of different environmental gradients such as altitude, temperature, physiological anomalies at high altitudes, aspect and slope can help to more accurately classify pixels as forested areas at altitudinal treelines.  This analysis will concentrate on the White Mountains in New Hampshire and investigate if treeline position can be traced using Landsat satellite imagery analysis. Furthermore it will be investigated whether different abiotic factors such as temperature, wind and solar insolation have a measurable influence on the location of the treeline.

Grace Stonecipher - Using change detection to identify wildfire areas in Alaska

Climate change is occurring at an amplified rate in northern latitudes, and is causing an increase in the frequency and severity of fires.  Different methodologies for identifying and defining the burn scars of two fires in Alaska, one in 1991 and the other in 2005, were investigated.  Landsat 5 TM images from 1995 and 2006 were analyzed using NDVI change detection and scatterplots, unsupervised classification, supervised classification, and the normalized burn ratio.  The resulting areas were compared to Alaska Forest Service maps to assess accuracy.  The effectiveness of all methods was decreased due to shadows of nearby mountains being misclassified as burned area.  Supervised classification was most successful for defining the burned area of the 1991 fire in the 1995 image and the 2005 fire in the 2006 image.  None of the methods were very successful at defining the 1991 fire in the 2006 image, showing that different methodology may be necessary for detecting older burn scars.

Chenyu Ma - Surface Melt of Central and Western Chugach Mountains

The melting of the global cryosphere has been widely observed through remote sensing. In the past decade, with the rising global temperatures, the primary cause of ice sheet and glaciers loss has changed from ice calving to surface melting. Mountain glaciers, although representing <1% of the global glacier ice volume play a vital role in regional climate. This study looks at surface melting on the western and the central Chugach Mountains in Alaska from 1989 to 2015 using satellite remoting sensing techniques. A calculation of normalized difference ice index shows that compared to the 1989 values, the mean ice index value of 2015 has decreased by 20 percent, indicating a drastic surface melt event independent of local weather. However, the study also questions the difference between the Landsat 5 and Landsat 8 sensors, as thermal data, on the other hand, show no melting event in 2015.

Weijing Soh - El Niño Impacts on the Stress Response of Forests and Palm Plantations

Climate change is one of the most urgent problems we face today. The expansion of palm oil plantations in Southeast Asia in response to global demand for palm oil products has resulted in widespread deforestation and the loss of critical carbon sinks that help buffer greenhouse gas rises from industry. Rising global temperatures and increasing frequency of drought are some of the effects brought on by El niño events as a result of climate change in Southeast Asia. The El niño event of March 1997/1998 was one of the worst recorded in recent history and the objective of the study was to compare the resilience of palm oil plantations and forests in response to climatic anomalies brought on by El niño. Two Landsat images from the eastern peninsular Malaysia (path 126, row 67) on March 1998 and March 1999 were used in this study. The image was subset to include both forest and oil palm plantation and had an area of 10,612 km2. Due to its geographical position near the equator, cloud masking was required; using a combination of temperature thresholds and band math temperature and reflectance thresholds to cater to removal of clouds in each image. Supervised classification was performed using 8 classes with the focus of the analysis centered on forest, young palm plantation and mature palm plantation. Mean surface temperatures were distinctly different with forests being cooler than oil palm plantations. Mean NDVI was lower across all vegetation classes in 1998 showing that El Nino had a physiological effect on all types of plants ( NDVI Forest > Mature Palm > Young Palm). However, forests were the most resilient in terms of green-ness measured. Mean NDWI of Forest > Mature Palm > Young Palm for both years, indicating that forests are the least moisture stressed of the group. However, there was negligible difference across years within classes suggesting little stress response from drought. This could be due to forests being able to access further/ deeper sources of water and oil palms benefiting from irrigation practices. NDDI, a relatively new drought monitoring tool, was tested and determined to not be a good indicator for drought monitoring for tropical forests and oil palm plantations. As the spectral signature of forest and mature palm were similar, there was misclassification between these two classes that was a limitation of the study. Further work could focus on de-tangling these two vegetation types to produce better accuracy.

Rain Tsong - Vegetation response to El Niño in Southern California

Hot urban temperatures, the patterns of which are termed urban heat islands, pose a lingering threat to many urban centers. The study of these temperatures is important to reducing heat-related hazards to human and environmental health. Vegetation is known to reduce temperatures in urban centers through evapotranspirative cooling. This study focuses on vegetation and temperature differences across 2015-2016 in order to observe temperature responses to both seasonal and climatic change, focusing on the time period of the end of the California drought and the onset of the 2015-2016 El Niño. The particular place of study is Los Angeles, located in Mediterranean climate and therefore susceptible to extremely high temperatures during the dry summers.

City distribution of vegetation is also connected to socioeconomic factors. Inequality in access to parks and inequal distribution of green space across the city has been documented. This study proposes to follow the temperature responses discussed above through the lens of socioeconomic data. This study finds correlations between neighborhood median income and Greenness and temperature, as well as temporal variations of these correlations. As Los Angeles pushes to mitigate urban heat island effects, this study finds that more emphasis should be placed on low-income, low-green space areas, areas that also have the highest temperatures year round.

Wenjun Wang - Large scale monitoring of snow cover change in Tibetan-Qinghai Plateau

Qilian Mountains, one of the major mountain ranges in Northwestern China, has experienced rapid snow cover reduction in the past several decades, leading to serious risks for both local ecosystem and irrigated agriculture. This project uses MODIS datasets from 2000, 2008 and 2016 to quantify the change in snow cover. Both K-Means and NDSI with Thresholding show a dramatic decrease in snow cover percentage over the period. Results of NDSI with Thresholding suggest that about 69% of snow cover has disappeared. In addition, this project also investigates the spacial distribution of the change using RGB Change Detection and 3D Surface View. Results show that snow cover reduction mainly happens in the southern and middle part of the  range where the altitude is relatively lower.