Modeling / Analytics

Urban Heat Island Effects:

Land Cover’s Influence on Surface Temperatures

Problem Description:

Urban Heat Islands (UHIs) represent a significant environmental challenge where built environments retain heat, leading to higher surface temperatures compared to rural surroundings. This phenomenon impacts energy consumption, public health, and ecosystem stability.

This project aimed to quantify the relationship between specific land cover types and land surface temperatures (LST) in two contrasting North Carolina environments: the highly urbanized city of Raleigh and the semi-urban town of Smithfield. The objective was to move beyond simple mapping and statistically model how different intensities of development contribute to temperature variability, providing data that could support evidence-based urban planning and mitigation strategies.

Analysis Procedures:

The workflow utilized a hybrid approach, integrating ArcGIS Pro for spatial data management and R for statistical modeling. I began by acquiring MODIS 8-day composite thermal imagery to derive mean surface temperatures and the 2019 National Land Cover Database (NLCD) for land use classification. In ArcGIS Pro, I defined a 3-mile buffer around both study areas and generated a standardized grid system. To overcome projection mismatches between the datasets, I resampled the rasters to a consistent resolution and performed Zonal Statistics to aggregate the mean temperature and calculate the proportion of eight simplified land cover classes (such as High Intensity Development, Forest, and Wetlands) within each grid cell.

Once the spatial variables were extracted, I exported the dataset to R to perform the statistical heavy lifting. I merged the datasets and normalized the land cover proportions to ensure comparability. Using bivariate correlation matrices and linear regression analyses, I was able to quantify the strength and direction of the relationship between specific land cover percentages and mean surface temperature, effectively isolating the thermal impact of different urbanization levels.

Smithfield Temperature Map w/ Grid

Raleigh Temperature Map w/ Grid

Results:

The analysis confirmed distinct thermal patterns driven by land use, establishing that Raleigh exhibited consistently higher mean temperatures than Smithfield, a trend directly correlating with its higher density of impervious surfaces.

The regression models demonstrated a statistically significant positive correlation between High- and Medium-Intensity Development and increased surface temperatures. However, the study also revealed critical data constraints. Surprisingly, forested areas showed a minimal statistical cooling effect in this specific model.

This counter-intuitive finding regarding vegetation was likely not a physical reality, but a result of the 1-km resolution of the MODIS data, which was too coarse to accurately capture neighborhood-level microclimates. In the fragmented semi-urban landscape of Smithfield, this low resolution resulted in "mixed pixels" that diluted the statistical signal of smaller green spaces.

Mean temperatures for Raleigh and Smithfield

Regression coefficients: LC impacts on Raleigh's temperature

Developed open space vs. temperature for Raleigh

Reflection:

I designed a workflow that bridged the gap between GIS processing and statistical software. I attempted to correlate land surface temperature with land cover composition using a grid-based sampling method, moving data from ArcGIS Pro into R to run regression models that quantified the impact of urbanization on local heat.

The most valuable lesson from this project came from its limitations. While the workflow was technically sound, the mismatch between the coarse resolution of the thermal data (MODIS) and the fine resolution of the land cover data (NLCD) introduced aggregation bias. I realized during the analysis that while resampling raster data makes pixels align visually, it does not increase the actual granularity of the data. The lack of a strong cooling signal from forests was a data artifact caused by the thermal sensor averaging temperatures over too large an area, hiding the benefits of smaller urban parks.

This project sharpened my ability to select the "right tool for the job" regarding data resolution. I learned that for local-scale urban planning, satellite data like Landsat (30m) or Sentinel is superior to MODIS (1km), even if MODIS offers better temporal frequency. Furthermore, I gained confidence in R-ArcGIS interoperability, learning that GIS is often best used for preparing data, while dedicated statistical software is better suited for analyzing the complex relationships within that data. Future iterations of this work would incorporate higher-resolution thermal imagery to better isolate the cooling benefits of urban vegetation.

Flood Vulnerability Analysis of Buncombe County, NC:

Using GIS to Identify At-Risk Areas

Problem Description:

The devastation wrought by Hurricane Helene in Western North Carolina transformed flood vulnerability from a theoretical risk into an urgent crisis. In mountainous regions like Buncombe County, the convergence of steep terrain, narrow valleys, and rapid urbanization created a funnel effect that catastrophic rainfall exploited with lethal force.

This project utilized a GIS-based hydrological modeling approach to evaluate regional flood vulnerability. A key objective was to demonstrate how open-source technologies can be deployed to assess these risks without the barrier of expensive software licensing. By integrating terrain, hydrology, and land cover data, the analysis aimed to identify the specific geomorphological factors that amplify flood destruction in the Buncombe basin and simulate water levels comparable to the surge events witnessed during Helene.

Analysis Procedures:

The analysis generated a comprehensive profile of the region’s hydrological instability, beginning with the terrain itself. Visualizations derived from the digital elevation model and slope maps revealed pronounced elevation gradients, particularly in the uplands surrounding the French Broad River. These steep slopes confirmed that the county acts as a rapid-collection funnel, where high-velocity runoff leaves almost zero time for infiltration or evacuation once peak flow begins. Furthermore, the flow direction and accumulation analyses successfully delineated the county’s complex watershed boundaries. By validating these modeled stream networks against known vector drainage patterns, the study confirmed that the open-source processing workflow was robust enough to accurately predict hydrological connectivity even in complex mountain topography.

The flood simulations produced two sharply contrasting scenarios that illustrated the "tipping point" of the region's capacity. The moderate, 5-foot flood event indicated localized inundation, primarily contained within the immediate banks of the French Broad River and specific low-lying urban pockets. However, the 15-foot "extreme" scenario demonstrated a total failure of natural containment. This model showed widespread inundation extending far beyond the riverbanks, closely mirroring the devastation footprint of Hurricane Helene. The simulation highlighted that once the river breaches the initial 5-foot threshold, the water spreads rapidly across the flat valley floors where the majority of the population resides.

Finally, the integration of these flood extents with land use data revealed a critical conflict between development and geography. The analysis showed that agricultural lands and developed urban zones were consistently the most affected categories in both scenarios, while forested areas remained largely largely untouched due to their higher elevation. This pattern underscores a historical planning failure: the county’s economic engines, housing density, and food systems are disproportionately located in the exact zones the model identified as high-risk. These findings highlight the urgent need for sustainable floodplain protection and green infrastructure to mitigate the impacts of future storms.

Buncombe County DEM (10m)

Buncombe County Slope Derived from DEM

Buncombe County Flow Direction Derived from Slope

Results:

The analysis generated several key outputs that collectively illustrate flood vulnerability patterns across the study area. Terrain visualizations derived from the digital elevation model and slope maps revealed pronounced elevation gradients, particularly in mountainous regions where rapid runoff is most likely to occur. These visualizations provided the foundation for subsequent hydrological modeling.

Flow direction and accumulation analyses established the county’s hydrological connectivity, accurately delineating watershed boundaries and validating the modeled stream networks against known drainage patterns. These results confirmed the reliability of the terrain preprocessing and hydrological modeling phases.

Flood simulations produced two contrasting scenarios. The moderate, 5-foot flood event indicated localized inundation along the French Broad River and within low-lying urban areas, while the catastrophic 15-foot scenario demonstrated widespread flooding across agricultural and developed zones. These findings underscored the disproportionate exposure of built environments and farmland to severe flooding.

Finally, the integration of land use and flood extent data revealed that developed and agricultural areas were consistently the most affected in both modeled scenarios. This pattern highlights the critical role of sustainable land management, floodplain protection, and green infrastructure in mitigating future flood impacts.

Derived Wetness Index Map

Flood Risk Map

Catastrophic Flood Extent

Reflection:

I built a complex hydrological model using GRASS GIS to simulate flood scenarios in a region recently devastated by a historic natural disaster. I purposely utilized open-source tools to demonstrate that high-level hazard modeling does not require proprietary, expensive software stacks.

The project took on a new, sobering weight in the aftermath of Hurricane Helene. Seeing my "extreme" 15-foot model—which looked abstract on a screen—match the news footage of underwater neighborhoods was a stark lesson in the reality of geospatial data. Technically, using GRASS GIS highlighted the power of command-line geospatial processing; it processed large terrain datasets faster and more efficiently than many commercial alternatives, proving that open-source tools are viable for emergency response scenarios.

This project taught me that software accessibility is a component of disaster resilience. If we rely solely on expensive, proprietary software for risk assessment, we exclude underfunded communities and non-profits from understanding their own risks. I learned that while GIS can map the extent of a disaster like Helene, the impact is defined by where we choose to build. Moving forward, I advocate for using open-source modeling not just for academic analysis, but as a tool for democratizing data—empowering local communities to see, understand, and prepare for their own vulnerabilities before the next storm hits.

Page updated

Google Sites

Report abuse