Spatial Analysis to Accurately Identify Vulnerable Populations Adversely Affected by Climate Change
The Rutgers New Jersey Climate Change Resource Center’s Accelerator grant partnership is advancing the development of a climate hazard data visualization and mapping tool that draws on spatial modeling to more accurately determine the location and needs of populations most vulnerable to climate change-driven health inequities.
by Lucas Marxen
The Rutgers New Jersey Climate Change Resource Center (CCRC) and the NJAES Office of Research Analytics (ORA) have been working as partners in leading the development of data informatics tools under the NJADAPT program for understanding the potential impacts of climate change in New Jersey. These tools have focused on supporting efforts to connect scientific climate change data with other critical information so that stakeholders can better understand the impacts of changing climate conditions on people, assets, and communities in New Jersey. ORA, with its team of data scientists and application developers, has created tools around flood hazard impacts, climate impacts on the forestry sector, data tools for hazard mitigation planning, and municipal level reporting tools on climate change impacts on local infrastructure and populations.
A critical component of the NJADAPT program is stakeholder engagement and working to provide tools to aid in understanding the impacts of climate change and support decision-making at various levels of government. An area of concern that has emerged from these efforts is understanding the potential impacts of climate change on vulnerable populations that may be more adversely affected than others. When we talk about vulnerable populations, this can include socially vulnerable demographics of society, communities facing environmental justice issues, or populations with health risks. While many datasets exist to attempt to quantify these populations geographically, most data sources are reported at geographic representations such as at the county, municipal, or Census Tract level– levels that are not adequate when trying to estimate impacts from more spatially explicit climate change risks such as coastal or in-land flooding. These datasets provide no information as to how vulnerable populations may be concentrated within these geographies.
As part of its mission, ORA is continually striving to advance the analytical and data visualization capabilities of the office to address these types of issues. While our office has expertise in data science, statistics, and programming/application development, our team has not yet had the opportunity to explore machine learning, and high-performance computing techniques to address some of the data challenges our collaborators, and we face. The Data and Society Accelerator Program at the Patrick J. McGovern Foundation (PJMF) provided an ideal opportunity to build capacity in these areas and address an existing climate data issue we were seeing in the planning and policy community.
Once our project was funded, our biggest challenge was educating ourselves on the cloud-based machine learning platform we would be using, understanding the software capabilities it offered, and investigating appropriate analytical models to accomplish our task of generating more refined population data for use in our data visualization tools. There were some initial challenges our team faced such as finding out that the geographic software we anticipated using (ESRI ArcGIS) was not easily compatible with the cloud environment we would be working with. Fortunately, our team is experienced in coding in a multitude of languages and software platforms, and we were able to pivot to using either the R or Python environments that were supported. Other challenges primarily involved a simple lack of experience or knowledge working with cloud computing platforms. An example of this was our team being unsure of which data storage option to use and which were most appropriate for certain workloads. This is where our access to the PJMF technical team through regular meetings and the program’s Learning Network Slack channel proved invaluable. The PJMF team has been a great resource for our team, answering both complex and simple questions as we work to get familiar with the machine learning environment and scaffold the components we will need for our analysis.
The other main challenge facing our team was determining an appropriate model to accomplish our project goal. After some research with our team statistician, we came across a methodology called statistical downscaling, which is similar to the techniques used by climate scientists to create localized climate data from global climate models. Statistical downscaling uses a two-stage approach and begins with attempting to model a relationship between your primary, geographically coarser dataset and a different finer resolution dataset. The second stage then builds finer resolution geography (typically a grid) and predicts the primary dataset values using both the secondary datasets values within the grid cells and the results from the relational model in the first stage. Different software packages can implement this differently, but the general approach is the same.
Having identified a modeling approach, our team focused on developing our tactical roadmap, which provided the opportunity to plan out the specific datasets we were to use in our model, the nuances of those datasets and any challenges they may present, and the pre-processing and data transformations needed to prepare the data for use in the model and machine learning environment. In addition, we evaluated different statistical downscaling packages in both R and Python to determine which would provide us with the greatest chance of success based on the level of documentation provided, the maturity of the package, and the general level of community support. The ‘Downscale’ package for R best met these criteria and allowed us to make decisions regarding our coding and machine learning environment to support its use. The final aspect of our tactical roadmap focused on providing a detailed implementation plan to accomplish our project within the technical and time parameters of the Accelerator Program. This allowed us to perform a critical analysis of our methodology and strategy for the analysis and its deployment. In particular, we ensured we allocated adequate time to perform test runs of our model to allow for any modifications to its design and the potential to run a secondary model (Random Forest Regression, for example) in order to provide a comparison of results for our final methodology document.
Our team is still in the early implementation phase of our project at this point of the program, but we have already benefited from our experience thus far and are excited as we make progress toward our project goals. We’re proud of our partnership with PJMF, which is providing this unique opportunity and support for our work in addressing climate change.
Lucas Marxen is the Associate Director of the Office of Research Analytics at Rutgers University where he provides leadership and direction on the development of analytical capabilities and services to meet the ongoing demand of stakeholders both within and outside the university. Lucas has expertise in developing geographic information systems (GIS), enabling spatial analysis and modeling for research initiatives in a variety of disciplines. In addition, he leads a team that specializes in developing data informatics systems that combine computer science and data visualization techniques into platforms that bring research and data to a broad set of stakeholders. He has also established an in-house capacity for developing custom software, database, and website solutions for a variety of research and extension initiatives. Lucas earned a Master in City and Regional Planning from the Edward J. Bloustein School of Planning and Public Policy, a Master of Science in Computer Science from Rutgers University, and is currently pursuing a PhD in Planning and Public Policy from the Edward J. Bloustein School of Planning and Public Policy.
