Spring 2018 - Project Abstracts

Jaclyn Kachelmeyer - Detecting Agricultural Changes in Finney, Gray, and Haskell Counties, Kansas

Advances in remote sensing have opened the door for new understandings and analyses of historical and ongoing changes in agricultural practices—as well as real-time implications of on-farm management decisions—around the world. This paper uses well-established methodologies and indices, including maximum likelihood supervised classification, RGB change detection, the Normalized Difference Vegetation Index and the Normalized Difference Water Index, to assess agricultural changes in Finney, Gray, and Haskell Counties in Kansas from 1973 to 2014. This analysis focuses predominantly on irrigated agriculture, motivated by ongoing concerns about depletion of the Ogallala Aquifer due to agricultural irrigation. Finney and its neighboring counties are located in southeastern Kansas above the Ogallala Aquifer in a region severely impacted by the wind erosion of the Dust Bowl in the mid-1930s.  Much of Finney County has lost 25-50% of its aquifer water supply from 1950-2013; many of its neighboring counties have seen losses greater than 50%.  More than half of the decline in the water table is likely attributable to irrigation, with declines in annual precipitation credited for the rest.  Thus, this paper seeks to quantify how land area under irrigation has changed from 1973 to 1993 and 1993 to 2014. Secondly, and supplementarily, this paper incorporates in its analysis the effects of the 2013 drought and 2014 drought recovery on the overarching irrigation trends.

Bowen Fang - Studying the Agriculture Change in a Northwest China Artificial Oasis

Zhangye Oasis is an irrigated agricultural zone located the arid Northwest of China. During the past decades it has experienced dramatic change in agricultural techniques and cropping patterns. Understanding such a change would help government and producers make more informed decisions and guarantee food security.

This project consists of two major tasks: the first is to identify, with classification techniques, the regions of double cropping, single cropping and non-agriculture, and the second is to compare the results of three years to detect the change of each category in terms of area and distribution. In the first part, I explored Landsat images and a MODIS Vegetation Index product to conduct the classification. Tasseled Cap Transformation was used to assist the process. To evaluate the reliability, I calculated the summary statistics and obtained scatter plots of each class, and identified certain deficiencies of the classification since classes failed to show either distinct scatter plot patterns or significantly different statistics. Then I used MODIS 16-day composite vegetation index product for another classification and was able to achieve better results especially with EVI bands, using ~ 20 images per year. Based on the above-mentioned partitions I concluded the change detection part as the total agricultural zone has increased since 1990, while the double cropping zone first decreased and then increased. The single cropping area showed an increase from 1990 to 2006 and a slight decrease from 2006 to 2017.

The future steps to enhance and expand this research would be to validate the classification with higher-resolution remote sensing images or ground trothing data, to improve the classification methods for less confusion or broader applications (e.g. to detect cropping patterns in similar landscapes), and to interpret the reasons behind the agricultural change..

Dylan Cicero - A Satellite Remote Sensing Analysis of Cover Cropping Practices in Champaign County, Illinois

This study attempted to analyze cover cropping practices in Champaign County, Illinois.  I used Landsat 5, Landsat 8, and Sentinel 2a images from mid-March 2011, 2014, 2017, and 2018 to compare cover cropping distribution across the county, cover cropping adoption over time, and cover cropping practices for fields previously planted with corn versus fields previously planted with soy.  I then tested a methodology to distinguish cover crop types.  Here, I applied a K-means classification to a November 2017 image of cover cropped fields and paired results to a change detection analysis that looked at differences in NDVI on cover-cropped fields between November and March of that season.  Together, my analyses produced a distribution map showing fields that are most often cover cropped in Champaign County, and determined that fields previously planted with corn generally have less cover cropping than fields previously planted with soy. However, these results have questionable validity for reasons described below.  My analysis of cover cropping adoption over time also lacks validity, also described below.  The tested methodology for cover crop differentiation failed.

Alberto Tordesillas Torres - Crop pattern changes in La Mancha region, Spain

Vineyards have undergone significant changes in the La Mancha region, central Spain, during the last decades. This study analyses the crop pattern changes between 2006 and 2017 and especially focuses on the drop in the vineyard area as a result of the subsidies from the Common Agricultural Policy to remove them.

Crop classification was based on the changes in the NDVI that allowed to distinguish the growing cycle of different crops. For this purpose, Sentinel-2 and Landsat TM multitemporal data was used to classify the crops in 2017 and 2006 respectively. Four different classification methods were used to classify the crops in 2017, both supervised and unsupervised, and both with monthly images from February to September and with only two images – March and August. Supervised classification with only two images proved to have an overall accuracy almost as high as the supervised classification method with the eight monthly images (97.6% versus 99.2% respectively).

After comparing both classified images, the vineyard area was estimated to have been reduced by 17,519 hectares from 2006 to 2017 (56%). The results seem to validate the initial hypothesis and allow an easy and cheap method to monitor the interannual variability of crop patterns in this area.

Rachel McMonagle - Assessing the vegetation effects of the 2006 drought and subsequent conflict in Syria

Syrian agricultural production and soil fertility have been severely affected by extreme drought events coupled with mass migrations of land stewards due to recent conflict (Kelley, 2015; Naranjo, 2018). Climate and conflict have affected the country’s landscape and agricultural yields at large (FAO, 2017). As a result, Syria’s food and nutrition security have been called into question (Dobiasova, 2016). This proposal intends to analyze a timeline of state of vegetation states in the region as a means of evaluating agriculture and food security. To better understand the effects of drought, conflict, and mass exodus in Syria, the proposal studies satellite images depicting surface reflectance from three years: 2005 (pre-conflict, robust agricultural sector), 2011 (early conflict, drought officially declared over), and 2018 (ongoing conflict with residual effects of drought, conflict, and agricultural abandonment due to fleeing). For the purpose of cross-checking outcomes, this study’s remote sensing imagery analysis methodology includes duplicative approaches: 1) a normalized difference vegetation index to assess presence of vegetation using red and near infrared satellite bands and 2) an unsupervised classification to categorize land cover using six out of seven of the satellite bands’ data. The maps created will highlight regions where vegetation has decreased from historical contexts as well as regions that have undergone vegetation expansion during that past decade. Defining the boundaries of these evolving areas has the potential to inform international aid’s prioritization of the most vulnerable populations in the region from an agricultural perspective (Alkhaled, 2015; Ghaleb, 2015).

Alexandra Savino - 2016 Drought Impact on Reservoirs and Large Lakes in New Jersey

The goal of this project is to analyze vegetation health and the surface area change of reservoirs and large lakes in New Jersey during and after the 2016 drought. This project compares two Landsat 8 images centered on Morristown, New Jersey from October 15, 2016 and October 2, 2017. Subsets of the 2016 and 2017 images were created for use in this analysis in order to focus on six distinct waterbodies- Wanaque Reservoir, Boonton Reservoir, Lake Parsippany, Lake Hopatcong, Spruce Run Reservoir and Round Valley Reservoir. The normalized difference vegetation index (NDVI) increased to 0.74 from 0.70 following the drought. A supervised minimum distance classification revealed a modest pixel count increase in tree cover post-drought; however, the total pixel count for water decreased and the total pixel count for bare soil increased despite recovery from the drought. Due to restricted access by local water agencies, ground truthing at the Wanaque and Boonton reservoirs was challenging and not as fruitful as it was at the other sites. It is suggested that topographic analysis, capacity of the waterbodies and water depth be explored in future studies.

Sharada Vadlamani - Before and After the Drought - Analysis of Changes in Nebraska’s Agricultural Productivity from 2010-2016

This study was developed to analyze the 2012-13 drought in Nebraska, which was the worst drought conditions the state had seen since the Dust Bowl Era of the 1930s. While the impacts of this drought were seen across the lower Mississippi Valley, and other Midwest states, Nebraska had among the most severely affected regions, with 70% of the state reporting severe exceptional drought by August 2013.

Through the use of MODIS data from 2010 to 2016, the objective of this project is to identify indicators of drought development in the state across two time spans – one from 2010 leading up to 2013, and the subsequent recovery from 2013 to 2016. This study explores change detection utilizing the band math as a means to identify spectral indicates of drought onset and recovery. The methods considered included NDVI, identification of net change using NDVI difference, and the use of Tasseled Cap Indices, particularly Brightness Index which is used to indicate soil moisture levels.

Sushant Banjara - Detection and Interpretation of Land Cover Changes Caused by the Koshi Flood, 2008

The primary purpose of the study is to apply remote sensing to study the main reasons behind land cover changes in western part of Sunsari district of Nepal and bordering region of India in the wake of 2008, Koshi flood. Three anniversary date Landsat TM images each from before, during, and after the flood, was analysed to observe the changes in broadly classified land covers—water, vegetation, and barren land. Normalised Density Vegetation Index (NDVI) and supervised classification were applied to analyse the images. From the preliminary analysis of the temporal data, I was able to hypothesise that the primary reason for the increase in NDVI after the flood was not agriculture but wild vegetation generated as a result of nutrient deposition by the flood.

To look for the evidence supporting the hypothesis, NDVI raster and original corrected Landsat images were classified using unsupervised and supervised classification respectively. While ISOData unsupervised classification was found to be unsuitable for the purpose, K-means unsupervised classification and supervised classification were able to produce results that indicated the increase in wild vegetation. However, it should be noted that sediment deposition during flood events is mostly a hydraulic phenomenon and remote sensing alone—at least with the rigour applied in this study—cannot provide exhaustive evidence to support the reasoning.

Cori Grainger - Determining flooding impact in Almora, India from 2008 – 2018 using MODIS NDVI data

Almora, India is frequently impacted by flooding events, and the steep terrain of the
region and rapid urbanization in the area makes people living in the region especially vulnerable to the effects of these flooding events. NDVI depicts how vegetated an area is and examining NDVI before and after events, as well as over a longer time series can help determine how flooding affects a region. A decrease in NDVI after a flooding event may depict the presence of water in that pixel or loss of vegetation. The more floods a region is exposed to may also decrease the average NDVI in that region over time.

Tina Huang - Damage and Recovery Assessment in Puerto Rico Following Hurricane Maria

Hurricane Maria made its landfall on September 20, 2017 and the aftershock was regarded as the worst natural disaster on record in Puerto Rican history (Brindley 2018). The immediate damages comprise the uprooting of trees, destructions of homes and roads, and widespread blackouts (Arduengo 2018).  Moreover, the tardiness, insufficiency, and unfairness of the rebuild efforts in Puerto Rico, along with an enduring economic recession have led to an escalation of recent protest on International Workers’ Day (Romo and Florido 2018). This project aims to utilize remotely sensed data to assess the damage and recovery levels in Puerto Rico, after Hurricane Maria. In terms of damage assessment, I ask: what are the top 10 most damaged counties in Puerto Rico following Hurricane Maria? 

Regarding recovery efforts,  I ask:
  a) how does recovery look like in the top 10 most damaged counties?
  b)  What are the demographic characteristics of those counties?
  c) Is there any correlation between recovery and demographic characteristics, for both the top 10 most damaged counties and for all counties?

In other words, are economically worse off or more vulnerable counties experiencing a slower path to recovery comparing with other Puerto Rican counties?             

I utilize four datasets, each vary in source types and time scales, apply remote sensing and geospatial techniques and triangulate the results to conclude that the top 10 most damaged counties are: Rinco, Cabo Rojo, Yabucoa, San Sebastian, Vieques, Jayuya, Ciales, Utuado, Culebra and Maricao. I construct a linear regression model and discover that among the top 10 most damaged counties, there is a positive correlation between recovery and percent of highschool graduates, holding all else constant. The same relationship is found between recovery and poverty rate. For all Puerto Rican counties, there is a positive correlation between recovery and median income, holding all else constant. The regression results, nonetheless, should be interpreted with caution and it is not within the scope of this project to discuss policy implications for disaster relief.

Grace Reville - Assessing the Impact of Hurricane Irma on Barbuda Land Cover Using Sentinel 2 Imagery

A change detection analysis of three land cover classification methods for the island of Barbuda was conducted on two Sentinel 2 Multi Spectral Images. The first image, from January 12, 2017, and the second image, from January 22, 2018, represent near “anniversary date” images before and after a severe hurricane (Irma) directly hit Barbuda. The change detection analysis employed a supervised and unsupervised land classification as well as a Normalized Difference Vegetation Index (NDVI) calculation. All three methods yielded differing results, with the supervised and unsupervised classifications both indicating a loss in urban pixels and an increase in vegetation pixels, while NDVI change detection indicated a net loss in NDVI value for over 40% of the same image’s pixels. Future efforts to investigate hurricane impacts on Barbuda’s land cover should focus on cloud free images with the highest possible spatial and temporal resolution to improve the accuracy of the classifications and subsequent change detections.

Theo Kuhn - Volcanic Hazard Assessment of Mount Sundoro, Java, Indonesia

Central Java is both one of the most volcanically active and densely populated places in the world. Despite this dangerous convergence, maps describing Java’s volcanic hazards in detail have only been created for the most well-known volcanoes. This project uses a digital elevation model and geological maps of Sundoro volcano in Java, Indonesia to produce the first detailed map of volcanic risk in the surrounding area. Threats posed by pyroclastic flows, tephra falls, lava flows, lahars and landslides are all accounted for in the risk map. Land surface classification of an ASTER VNIR image from 2003 is also carried out to identify the inhabited areas most threatened by future eruptions. Over 20,000 people are found to live in areas of relatively high volcanic risk that are well within the range of eruptions from the last thirty-thousand years.

Helen Siegel - Process for Mineral Identification in the Mawrth Vallis, Mars

Minerals preserve a wealth of information about the environment and dominate processes during their formation. Aqueous minerals in particular serve as a time capsule, preserving information about the abundance of water, temperature, pH, and parent materials present during formation. Identification of particular mineral species on the surface of Mars is thus an important step towards understanding the paleoenvironments of Mars and the potential for life sustaining conditions. Additionally, certain mineral species are better indicators of environments capable of preserving such biosignatures. Mapping locations with an abundance of such minerals can help target future research efforts. Hyperspectral data can aid in the remote identification of mineral species by matching characteristic reflectance signatures with known endmember spectra. The Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) aboard the Mars Reconnaissance Orbiter (MRO) was designed with this specific task in mind and thus can be used in conjunction with the CRISM Analysis Package in ENVI to classify and map the distribution of mineral species on the surface of Mars. In this project a single image of a region in the Mawrth Vallis of Mars is selected and analyzed using ENVI to produce a classified and linearly unmixed map of dominant mineral species that inform an understanding of paleoconditions.

Maddie Shankle - Cloud Phase Properties and Structure in the Indian Monsoon

Determining cloud phase is an important first step in determining other optical properties of clouds; if a cloud’s composition is misidentified as ice or liquid, the differing spectral properties of the two materials will lead to substantial errors in subsequent calculations and products. It is therefore very important to understand the phase of clouds being studied. India and its monsoon is one context in which cloud phase has been studied very little. This project adapts a preexisting ice index to address troublesome pixels in images of monsoon clouds over India. While the method employed here may not be directly transferable to other images, it demonstrates the importance of exploring one’s data thoroughly and only applying modifications with a full knowledge of the data’s characteristics. The monsoon clouds are found to vary considerably in there phase composition over the course of the monsoon season, but in no particular pattern or trend. Remote sensing seems to be a powerful way to get at spatial patterns of cloud phase, useful for many climatological and meteorological studies.

Maggie Yao - Assessing Temporal and Spatial Variations in Dust Storms in Northeastern China

This study analyzes the temporal and spatial variations of dust storms in northeastern China using MODIS images. The Region of Interest (ROI) includes five adjacent provinces including Beijing, Tianjin, Hebei, Shandong, and Jiangsu. Three reported dust storm dates are chosen for analysis, and three other baseline dates in the same calendar years are selected for comparison. The datasets used in this project are MODIS Aqua Level 1B Calibrated radiance and MODIS Aqua Level 2 Aerosol Optical Depth. Two methods are employed in this study to assess the scale of dust storms - The Normalized Difference Dust Index (NDDI) and Aerosol Optical Depth (AOD) data are calculated and retrieved from these two datasets respectively.

Indra Acharja - Land Use Change and urbanization in Punatsangchu Valley, Bhutan

Land use change and urbanization assessment in Punatsangchu valley in Bhutan was done using Landsat 4 8: 5 (TM), and Landsat 8 (OLI) images, 30m spatial resolution, at a temporal interval of 4 and six years; 2007, 2011 and 2017. The study covered a total area of 900sq. Km, 56km north-south and 16.2 km east- west between Phochu and Taksha area, within Punakha and Wangduephodrang districts. The objective of the study was to assess the total change in agriculture land, forest cover, infrastructure development and expansion of towns and urbanization due to hydropower development in the area. The assessment was done in ENVI 5.4 software using image differencing, NDVI comparison, supervised classification and post-classification change detection analysis. This study found that there is an overall decrease in agriculture land, forests area, and surface water bodies while there is a significant increase in barren soil area and settlement and urban area over the eight years of hydropower development period. Overall there is 30% (31.76 sq.km) decrease in agriculture land, 4.5% (30.78sq.l<m) decrease in forest cover, there is also 28% (10.11 sq.km) decrease in surface water bodies. Infrastructure development and settlement has increased by 230% (40.49 sq.km), and barren soil area has increased by 110% (36.22sq.km) along the Punatsangchu valley due to ongoing construction of two hydropower projects.

Younten Phuntsho - Study on the impacts of ban on shifting cultivation through landcover change analysis

The main purpose of this study was to not only assess the impacts of policy change of the Royal Government of Bhutan (RGoB) on shifting cultivation but also to implement and master the skills and knowledge gained from the course F&ES726 01- Observing Earth from Space.

The various tools and techniques that were taught in this course were used that included – preprocessing techniques of image analysis, visual change detection, NDVI change detection and classification methods that included both unsupervised & supervised system of classification. All the methods employed for this study indicated that there has been real change in terms of landcover from 1990 – 2016.
The supervised classification method was observed to be the best method for this study vis-à-vis other methods. The change statistics were generated by comparing the landcover statistics of two years – 1990 and 2016. The study points out that forest cover in the study area – Samdrup Jongkhar district has increased from 1990 to 2016 by roughly 7.27%, which indeed resemble the ground realities. However, the agriculture land use dropped by 1.67% between 1990 and 2016, which too corresponds well with the field realities.

Andry Rajaoberison - Mangrove Cover and Land Use Change in Coastal Environments

This project assess mangrove cover and land use change in southwest coast of Madagascar. I looked specifically at mangrove and terrestrial vegetation to understand where in the landscape are the most threatened by deforestation. My analysis was based on Landsat images from 2002 and 2017. Prior to my analysis I performed sub-setting, radiometric calibration and atmospheric correction. Then, I conducted unsupervised and supervised classification to identify the key landscape features I want to highlight in my study. My results showed a significant decrease in terrestrial vegetation with a deforestation rate of 5% per year. Mangroves were less threatened with a decrease of 0.9% per year. However, mangroves were more threatened in the northern part of the study area.

As I was not able to do any ground-truthing my accuracy assessment was based on changing NDVI values in my land cover. I found out that separating the region of interest between the analysis of mangrove cover and the analysis of terrestrial vegetation could improve the accuracy of the classification. I also found very challenging to explain why the vegetation covers that I classified as terrestrial occurred in intertidal environments. Only mangroves grow in these area and a further analysis using high spectral resolution images could help identify these vegetation which could be mangroves.

Nicholas Fields - Assessing land cover change on a Caribbean small island state: Barbados, 2002 to 2017

Geospatial technologies provide for rapid and reliable assessments of land use and land use change across the globe, which are major priorities for Small Island Developing States (SIDS) where land resources are limited and are undergoing major transitions over time and space from development pressures. Using data acquired from NASA’s ASTER Level 1 Precision Terrain Corrected Registered At-Sensor Radiance (AST_L1T) product, this paper presents the results of an exercise using remote sensing tools and techniques to perform a coarse-level supervised land cover classification and change detection analysis of the northern portion of the island of Barbados in the eastern Caribbean.

Outputs suggest that over a 15-year time-step between 2002 and 2017, the greatest absolute change in land cover detected is the area under agricultural production. Initially the dominant land cover type in the study zone, it has reduced by approximately two-thirds, and may well be in continual decline. This is juxtaposed against a high (relative) increase in infrastructure and related development on the landscape. Challenges with pervasive cloud cover across scenes and subsequent pre-processing rectification efforts affected robustness of results and thus reasonable caution is advised with data interpretation, but general trends in the direction and magnitude of land cover change are considered substantive. Future research efforts could entail ground-truthing and dissecting the current classification scheme into more specific land cover identities.

Jack Singer - Forest Type Change in the Roaring Fork Watershed, Colorado

The intermountain west of North America is increasingly subjected to threats to forest cover, most prominently including fire suppression, suburban development and urban sprawl, shifting land-use patterns, forest health disease, and unstable climate patterns like drought. I sought to utilize historic Landsat datasets to determine land cover change over the last few decades, in order to make interpretations about forest resiliency and shifting patterns of forest type across the landscapes. The Roaring Fork watershed, an area in Colorado of 1,451 square miles on the Western edge of the continental divide, contains an extremely high density of diversity of forest type, tree species, elevation, and climatic regimes, and thereby represents nearly all major habitats and land-uses present throughout the Rocky Mountain region. My investigation determines that analyzing forest type change with Landsat data in this region is unreliable due to the scale of heterogeneity being finer-grained than the Landsat pixel-size (30m x 30m). However, important conclusions about land-use, most notably changes in irrigated area and developed suburban surfaces, are possible with these datasets.

Yookyung Kim - Time Series Analysis of the Grand Canyon National Park Region

The Record of the Climatological Observations suggests that the Grand Canyon National Park’s monthly average temperature has increased 6-10 degrees Fahrenheit since 1977 for the month of January and October. Hence, I wanted to conduct a remote sensing analysis using 4 satellite images to demonstrate what are the changes occurring in the Grand Canyon National Park area in terms of snow covers, NDVI, and landscape since 1977. Two images were acquired from 1997, one from 2017 and one from 2018 and they have been analyzed using maximum likelihood supervised classification, NDVI, and K-means unsupervised classification. The analysis results show decrease in snow coverage, increase in vegetation at the upper left region, decrease in vegetation in the upper right region, but inconclusive result from the lower region. However, overall, it is evident that the increase in average temperature for the month of January and October have caused noticeable change in the Grand Canyon National Park area.