How does it impact Clemson students?
Through this recognition, students at Clemson using GIS for their research are allowed special access to Esri software for their projects, get additional training, and the possibility to attend the Esri Developer Summit in Palm Springs, California.
Additionally, students are encouraged to participate in the Annual Outstanding EDC Student of the Year award.
Student of the Year Award
The EDC Student of the Year is an annual award open to all Clemson students. The winner will be recognized as Student of the Year for outstanding achievements in development of innovative tools, applications or techniques using ArcGIS platform. Each year the winner will be awarded with a plaque, cash prize, and one Esri Developer Summit registration.
All undergraduate and graduate students at Clemson University in any discipline are eligible for the award. Students demonstrate innovative use of the ArcGIS platform in their application.
2019 Clemson Student of the Year: Justin Dowd
In 2019, the Clemson Center for Geospatial Technologies was designated an official Esri Development Center.
This special recognition is only awarded to a small number of leading university entities across the world that have exemplary programs focused on educating students to design and develop innovative applications based upon the ArcGIS platform.
Justin is a senior in Clemson University’s School of Computing as well as a Senior Intern in the Clemson Center for Geospatial Technologies.
He has been been working with researchers in the Department of Environmental Engineering and Earth Sciences who are exploring the potential for rooftop solar photovoltaic (PV) deployment in South Carolina. The project aims to identify suitable locations for rooftop solar across the state using remotely sensed imagery (e.g. aerial and/or satellite imagery) and machine learning coupled with CyberGIS.
An initial assessment of Clemson's main campus has already been conducted, which identified a number of high-value locations. Justin is expanding on that analysis using known building footprints on campus and other training data to conduct supervised image classification to identify and calculate suitable rooftop area across the entire state of South Carolina.
Justin has developed and tested workflows for parallel computation to scale the analysis from the small test region to the entire state while exploring several GIS platforms: Esri ArcGIS Pro, GRASS GIS, and Google Earth Engine. Using a variety of imagery and additional data sets, he is testing each of these platforms to:
Evaluate the accuracy of the machine learning algorithm implementations
Identify the best combination of input data for most accurate rooftop identification
Determine the most efficient method of scaling the computation on each platform
The imagery data are the United States Department of Agriculture National Agriculture Imagery Program (NAIP) 1-meter resolution data, both a 3-band true-color image from 2017 and a 3-band multispectral image from 2009, the most recent year available for the area.
Justin is incorporating 1-meter resolution elevation data derived from the South Carolina Department of Natural Resource’s LiDAR database, where possible, to improve the classification accuracy, although this is not supported in GRASS GIS.
The Support Vector Machine (SVM) algorithm is utilized in ArcGIS Pro and Google Earth Engine, whereas the Maximum Likelihood algorithm is used in GRASS GIS as the SVM method is not implemented in the base software package.
The results presented in the table below show the percentage of buildings on Clemson University’s main campus correctly classified in each platform and with different combinations of imagery and elevation data. The classification accuracy is only considered for buildings present in all input data sets to remove the impact of construction and demolition over time.
Justin is discovering that each platform offers its own advantages and challenges. Google Earth Engine scales automatically without any change in workflow, though it is limited by the computational resources allocated to the free accounts. ArcGIS Pro gives the highest overall accuracy but is limited by the computing resources of a desktop computer. It does not operate natively in a Unix environment typical of high performance computing (HPC) clusters and has many limitations which reduce potential efficiency in a high throughput computing environment. GRASS GIS is producing the lowest overall accuracy, but runs on UNIX systems, meaning the workflow can be parallelized and scaled efficiently on an HPC cluster, such as Clemson University’s Palmetto Cluster.
Justin will continue to work on this project through and hopes to incorporate deep learning methods to improve the classification results. Other projects he has worked on at the Center for Geospatial Technologies include writing scripts to deploy Structure from Motion (SfM) software in parallel on the Palmetto Cluster to efficiently process photos from UAVs into a variety of geospatial data.