The following reflection is from a participant in NLT's Summer Research Internship program.
How long does it take to recover from a hurricane?
This was the main research question our team was tasked to answer during a ten-week research internship at New Light Technologies. I had previously held research assistant positions at a university, but working as one with the company was a completely different experience. For the first time, I was no longer just an executor of given tasks, but a rightful member of a team. I participated in all discussions about the research, methodologies, and data collection, and ultimately presented my research and findings. Moreover, the company not only encouraged — but expected — that interns share their results with the world community of scientists, researchers, and policymakers. For my team, that opportunity might be the upcoming Fall Meeting of the American Geophysical Union in Washington, DC.
Our first task was to define what "economic recovery" actually looked like, and to develop the processes we would use to track it before we could answer our question. We debated multiple variables. Is it the gross domestic product (GDP) of a region? The employment rate? The brightness of nighttime lights captured by satellites? Where does policy fit in, and how do we even measure it? After intensive discussion, we decided to focus on nighttime lights and employment data. The combination of the two felt like a suitable proxy for the amount of economic activity present in a region before and after a hurricane hits.
Our first important technical task was to design and build a geodatabase to store all the data we would collect. The design of the geodatabase is a crucial early aspect of research, because that's when you determine what data to collect and when to stop — an important step in facilitating efficient analysis and maintenance. It was tempting to collect as much data as possible from the start, but we had to be cognizant of constraints on both time and resources. Rather than doing it all at once, my team found it far more efficient to proceed thoughtfully, one thing at a time.
We settled on using the Visible Infrared Imaging Radiometer Suite (VIIRS) to collect Nighttime Day/Night Band (DNB) Composites. We used these to quantify and map the amount of light reflected at night, allowing more sensitivity than other, lower-resolution datasets. This data was accessed through Google Earth Engine (GEE), which allowed us to both interact with and analyze it. The first challenge was to extract data in a way that would let us aggregate it to the unit of analysis (the census block). We researched the literature for the most efficient method and found that, although GEE stores a lot of data, we needed to extract it in a specific way rather than simply downloading it. For this we wrote code that calculated the sum of nighttime light per unit of analysis and stored it back in a table together with a unique identifier.
But converting and extracting data was not enough — much work still remained. Once the data was stored, processed, and projected, we found that in many cases it looked different from what we expected; some results came off as outright bizarre. A researcher needs to logically think through a dataset, not simply trust it blindly. My team stepped back to contemplate the nature of the data, to think about the story behind it and why there might be a mismatch between our expectations and what we found.
(Graph 1)
For example, in the image above (Graph 1) we mapped the nighttime light data captured in February 2018, attempting to show only non-zero values (making all zero-value pixels transparent). Theoretically, nighttime lights are visible where there are urban markets. We found, however, many zero values that could not be explained by a lack of economic activity. To find out why, we used an additional band of the VIIRS dataset we hadn't considered before, and quickly understood that we had to adjust for cloud coverage to ensure data quality, as that band consisted only of cloud-free observations.
Our ten-week tenure offered a fully immersive experience in conducting every component of real-world research. We learned a great deal collecting data and collaborating as a team, and ultimately produced something we were all proud of. The maps in Figure 1 are the final result of our analysis, which we present to the team as a base for further and deeper work. As we can see, the effects of hurricanes are lagged, and the nature of a hurricane is not simple and requires further investigation. The discussion about goals and methodology is complicated and constantly evolving, so while we have concluded this portion of our research, it is by no means a conclusion of the research as a whole.
(Figure 1)
It can be frustrating when things don't work out as originally planned, but you have to be flexible and go with the flow. The most important lesson I learned at New Light Technologies was that research is an ever-changing process requiring constant adjustment — and that teamwork is the most valuable tool in a researcher's toolbox. Your team members are not only a resource with immense knowledge, they are also your biggest cheerleaders. A good team will never leave you on your own, and I was lucky to have found a good team during my time with New Light Technologies.
About New Light Technologies
New Light Technologies (NLT) is a Washington, DC-based firm with 25+ years of experience delivering:
- Geospatial systems and enterprise GIS
- Cloud-native data platforms
- AI/ML and advanced analytics
- DevSecOps and cybersecurity solutions
NLT supports federal, state, and international organizations in operationalizing data for mission impact.