Fergal McGinley Portfolio
An overview of the work I have done as part of my Geospatial Data Analysis MSc at University College Dublin and beyond
About me
My name is Fergal McGinley. I'm 23 and have recently completed an MSc in Geospatial Data Analysis at University College Dublin. I finished from my undergraduate degree, a BA International in Geography & Linguistics in UCD, in May 2021 and started the MSc the following September. I am passionate about addressing the climate emergency and believe that earth observation and spatial data will be essential tools in reaching this goal.
I have over two years of experience working with GIS software and Earth Observation technology and data. I have experience working with the full suite of ESRI softwares, ESA's SNAP software, Google Earth Engine, QGIS, Agisoft Metashape, and Pix4D Mapper. I have completed projects using both passive and active (Synthetic Aperture Radar) satellite data, as well as bathymetric imagery for seabed mapping. I also have substantial experience in both conducting Unmanned Aerial Vehicle surveys and processing the resultant data. My MSc thesis project involved capturing and processing UAV data to estimate the effect an invasive disease had on a forest plantation's carbon sequestration capacity.
I recently received the final results for my MSc and received a first-class honours degree with a GPA of 3.89.
GIS
Urban Heat
The synergistic effects Urban Heat Islands and longer, hotter, more frequent heat waves is one of the greatest health-related challenges facing the UK today as record high temperatures are set year upon year. I have written extensively on this issue. The first paper linked below is a 30-page report I wrote on heat waves in cities - why their effects are more severe in urban environments, their detrimental and inequitably distributed health effects, and various potential methods for mitigation. The second paper focuses on heat in the London Underground - why the network gets so hot and potential mitigations for this heat.
As a follow-up to these reports, I conducted a GIS-based analysis of heat vulnerability in the city of Detroit, MI, USA. Detroit is a city that experiences very hot summers, while also having a relatively low GDP, an ageing population, and low prevalence of air conditioning - all factors that are indicative of a population highly vulnerability to heat. I combined satellite imagery with census data to construct a Heat Vulnerability Index (HVI) of Detroit's June 2016 population. The purpose of the HVI is to identify areas that are particularly vulnerable to heat so targeted public health interventions can be made. A 10-slide summary of the project is available below.
Given all governments have limited resources, tools like the HVI are essential to identify areas of high priority for targeted public health interventions.
Mapping Australia's Wildfires
One of the first GIS projects I undertook this year was an assessment of the wildfires that took place in Australia in January 2020.
The first map (left) shows the number of active fires in Australia during one week in January 2020. There were as many as 9,440 active fires in south-eastern regions of Australia at this time.
The other maps shows the number as active fires as red points on the map, as well as population per square kilometre. There is also a 100km buffer around each active fire to simulate a 'danger zone' for poor air quality caused by the fires.
Click on each image to take a closer look. The full project is also linked below. I discuss the link between climate change and wildfires, as well as their negative health impacts for humans and ecosystems.
Multi-criteria analysis
I conducted a multi-criteria analysis for the reintroduction of the Little Grebes bird species in county Mayo as part of my Advanced GIS module. I created a map that displayed the number of reintroduction criteria that were met by location within the county.
The target areas had to meet the following criteria:
- Proximity to water - 50m from rivers or 100m from lakes and marshes
- Area - Only lakes and rivers of at least 1km squared, or
- Peatbogs of at least 5km squared
- Slope of less than 10%
- At least 300m away from roads
- At least 250m away from urban areas
Automation - Geoprocessing package
Automation is an important GIS skill that I am keen to develop. I created a geoprocessing package for the above multi-criteria analysis.
The first slide below shows the geoprocessing package. I ran the analysis for county Mayo again through the package (slide 2). I then ran it again after changing the county to Kilkenny (slide 3). All I had to do was change this one variable to run an entire analysis, which really highlighted to me how automating tasks is a valuable skill for GIS specialists.
Python
A great way to automate tasks is through coding, and as Python is the language most widely used in ArcGIS, I chose to learn it. I only have a basic level of competency so far but I am actively working to improve this.
We covered an introduction to Python in our Advanced GIS module and learned how to use Python scripts within the ArcGIS/ArcPy environment.
I converted my geoprocessing package to a python script and practiced changing variables within the script, and was able to run the analysis for all counties in Ireland in a fraction of the time it would take to conduct the analysis manually
I also completed the following short GIS project where I mapped fire stations, ambulances, and voting station in Toronto, Canada by electoral district and fire zone. I used this project to familiarise myself with using the Python window within ArcMap. I wanted to show communities in the city that were isolated from fire stations, ambulances, and voting stations and used Python to do my analysis. I wrote a script in the ArcPy window that created a buffer of 1km around the amenities, converted the buffer shapefiles to raster layers, and reclassified the raster layers to represent the areas within and outside 1km from each amenity. I then used the raster calculator tool to show the areas that were isolated from all 3 amenities (shown in the map on the right below). An example of some of the code I wrote is shown in the centre image below.
Remote Sensing
3-D model and flood simulation
An assignment as part of my Advanced GIS module required us to create a 3D model of a town of our choosing. I made a video where I flew through the 3D model, which you can watch over on the right.
I decided to create a model of the town of Zarautz in the Basque Country. I processed LiDAR datasets freely available from the Basque government to create a 3D model of the landscape. I combined this with aerial imaergy (also from the Basque government) and infrastructure data (roads, railways etc.) from OpenStreetmap to create the final model.
I chose the town of Zarautz for a couple of reasons. We had to create a flood simulation as part of the assignment, and Zarautz's town centre is at extremely low elevation so I thought the flood simulation would be particularly striking there. There's also a nice contrast in the landscape between urban and rural, as well as low and high elevation.
Supervised vs unsupervised classifications
I conducted unsupervised and supervised landcover classifications of the area around Cahore Point, Co Wexford using Sentinel-2 imagery. I then converted these classifications to vector datasets and added them as hosted layers on ArcGIS Online. Both classifications are embedded below as interactive maps.
Cahore Point - Unsupervised Classification
Cahore Point - Supervised Classification
Change Detection Analysis
This project aimed to assess the spatio-temporal variation in the area covered by Lough Ennell in Co Westmeath in response to drought conditions.
I used both passive/optical and active (Synthetic Aperture Data) remotely sensed data from Sentinel-2 and Sentinel-1 respectively to see if I would find different results.
The Sentinel-1 data was processed primarily through SNAP, where I used a random forest classification method to distinguish between water and land, while the Sentinel-2 data was processed using Arcmap.
The table to the right shows the area covered by the lake based on each analysis.
Each method produced a slight variation in results, particularly in 2019. Ireland experience drought conditions in April 2018, which is reflected in the lower areal coverage in both analyses. The results highlighted to me that the most reliable insights from EO data can be derived when multiple data sources and processing methods are used in tandem.
Forest Landcover Classification - Combing Optical & SAR Imagery
I attended a NASA ARSET training programme called 'Forest Mapping and Monitoring with SAR Data'. One of the four sessions focused on using Optical (Landsat 8) and SAR (Sentinel-1) imagery in tandem to create a landcover classification map of a forested area in Rodonia Brazil using Google Earth Engine.
This image shows an example of some of the code I used during the session. I used a random forest algorithm with 64 decision trees to classify both the SAR and optical imagery individually. I then combined the SAR & optical imagery and ran the same classification algorithm.
The individual SAR & Optical classifications both performed well in accuracy assessments, with 0.9856 and 0.9988 respectively.
As expected, the combined classification performed marginally better than the individual classifications with an accuracy of 0.9994.
The final landcover classification. The section in light green refer to forested landcover.
A future project could modify the change detection analysis workflow from the Lough Ennel project above to assess how much additional forested area has been added over a specified period of time. This information could be used to determine the success of one of the many reforestation and afforestation projects currently taking place globally.
Thesis Project
The world's forests are an invaluable asset in the fight against the climate emergency, extracting and storing vast amounts of carbon from the atmosphere. Therefore, threats to the world's forests are threats to world's climate. Extensive reforestation and afforestation projects are being carried out against the backdrop of an extremely globalised trade system, where tree and plant seedlings are exported across vast distances with increasing regularity. This globalised plant-trading system facilitates the spreads of invasive species and diseases which pose substantial threats to the world's forests.
Ash Dieback (hymeniscyphus fraxineus) is an invasive fungal pathogen originating from Asia that is projected to kill 70-90% of Europe's ash trees. Using the example of an ash plantation of ~1000 trees in Ireland severely affected by dieback, I set out to create a framework for how low-cost, off-the-shelf unmanned aerial vehicles (UAVs) can be used in concert with a minimal amount of ground-surveying to accurately assess the extent to which invasive diseases/species affect the capacity for carbon sequestration in forest plantations. A summary of the project is included below, as well as a link to the full thesis.
Materials & Methods
The study site on the right was identified as part of a UCD Forestry project trialling mitagory methods against ash dieback. The ash plantation is outlined in purple and consisted of approx. 1,000 trees. As part of this project, a sample of tree diameters at breast height (DBH), tree heights, and precise tree locations were taken. The project team also rated the severity of dieback in each tree, ranging from 1 (unaffected) to 5 (dead).
I conducted an aerial survey using the DJI Phantom 4 UAV shown in the photo on the right, equipped with a standard RGB optical camera. I created a flight plan in the DJI Fly app that specified the area to be flown. The UAV flew in a cros-grid patter, capturing images at a 90° angle with 80% front overlap and 70% side overlap. Ground control points (GCPs) (shown in the image above) were captured using a rimble DA2-BT Global Navigation Satellite System (GNSS) receiver. This high overlap between images and use of GCPs ensured the accuracy of the resultant digital recreations of the plantation.
I used the Agisoft Metashape software to create a point cloud with ~400 million points. Using this point cloud, I created a Digital Surface Model and Digital Terrain Model of the site (see image on the right). Subtracting the DTM from the DSM resulted in a Canopy Height Model - an elevation model illustrating the heightso of trees in the plantation.
I identified individual tree heights in the plot by inverting the CHM and employing a Local Minima algoritm in ArcMap, successfully identifying and measuring the heights of 951 trees in the plot.
I used the 'true tree heights' measured in the ground survey to validate these findings. The UAV-derived heights illustrated a bias, with average heights about 1.4m taller than their 'true' counterparts. A portion of this bias can be accounted for by tree growth, as the ground survey was conducted ~9 months before the aerial survey
I then delineated and measured the diameter of these trees' crowns using the Inverse Watershed Segmentation Method (IWS). IWS involves inverting a Canopy Height Model (CHM) so the tree peaks become ‘ponds’ and the branches become tributaries. The ‘catchment area’ of each tree is then calculated by using the hydrology tools in ArcGIS. ‘Watershed’ refers to a point/line where if a drop of water falls, it is equally likely that it will fall into one of two different ‘catchments’. Thus, the watershed lines can be used as a proxy for the extent of tree crowns.
My hypothesis was that the relationship between these UAV-derived tree parameters (height & crown diameter) and the measured DBHs in a sample of trees could be used to predict the DBH for every tree in the plantation. DBH is the most important measurement in calculating the carbon sequestered by a tree so if it can be predicted accurately based on UAV-derived tree parameters, it would be possible to accurately assess the amount of carbon stored in a forest plantation. I had a sample of 95 trees with measured DBH, UAV-derived height, and UAV-derived crown diameter. I used two linear regression models to predict the DBH of every tree in the plot based on the relationship between measured DBH and (1) UAV-derived height, and (2) UAV-derived crown diameter. Each tree was also classified by how severely it was affected by the dieback disease. I categorised the dataset based on this dieback severity classification, then used both the ground-surveyed ('true') data and the UAV-derived data to calculate the carbon sequestered in the plantation.
Results
The results based on both the ground and aerial surveys showed that carbon sequestration decreased in trees more severely affected by the dieback disease.
The table shown here is based on the sample dataset of 95 trees.
The ground and aerial surveys produced very similar results for the amount of carbon stored in the entire plantation. This illustrates that UAV data can be used to accurately assess the amount of carbon stored in a forest plantation.
The baseline values shown in the table are a surrogate for the amount of carbon stored in the plantation were it never affected by the dieback disease. It is based on the trees deemed 'unaffected' by dieback during the ground survey, i.e. the extrapolated values show how much carbon would be stored in the plnataion if all the trees were unaffected by dieback.
This graph illustrates how the carbon sequestration rates in the plantation have rapidly fallen below the 'baseline' values since the dieback disease was identified in the plot in 2017.
The graph also predicts how the carbon sequestration rates may fall even further below the baseline over the next 5 years. Given the rapid rate of decline, it is possible that the plnatation could become a net carbon emitter within the next 15 years.
These results show that the ash dieback disease has a severe effect on the capacity for carbon sequestration in affected plantations. This project's methodology relies on conducting a ground survey on a small sample of a target forest plantation, followed by a UAV survey of the entire site. This method is much less time consuming and much more cost-effective than conducting a full ground-survey, while delivering highly accurate results. The entire workflow can be completed by one person using relatively inexpensive equipment (e.g. an off-the-shelf drone with an optical camera) and is therefore highly replicable and scaleable. Reforestation and afforestation projects with the primary goal of offsetting carbon emissions can use this framework to assess how invasive diseases/species are affecting the efficacy of these projects in capturing atmospheric carbon. Aside from carbon sequestration, the framework could also be used by governments and forest managers to assess how invasive diseases/species affect timber yields, habitat loss, and other ecosystem services.
The full thesis write-up can be found at the link below.
Thank you for taking the time to read my portfolio. If you have any questions or would like to get in touch with me, you can reach me at fergal.mcginley@ucdconnect.ie
