ADS Capstone Chronicles Revised

9

In general, the correlation between numerical features (off_nadir_angle, pan_resolution, gsd) is consistent between these plots ● Off_nadir_angle and pan_resolution are highly correlated, suggesting that as the angle of the satellite increases from nadir (directly overhead), the spatial resolution of the image may change. ● The angle of the satellite from directly overhead (off_nadir_angle) and the sun’s position above the horizon (sun_elevation) are moderately linked, meaning these factors tend to increase together but not as strongly as above ● gsd (Ground Sample Distance) is the distance on the ground that each pixel in a satellite image represents. gsd has positive correlation between gsd and off-nadir angle. As the off-nadir angle increases, so does gsd. While gsd and pan_resolution are likely negatively correlated, indicating that as the ground sample distance decreases (higher resolution), the panchromatic resolution increases The numerical features do not have a strong correlation with disaster_type_encoded, and in fact, exhibit a negative correlation. Conversely, off_nadir_angle shows a strong correlation with status_encoded (Pre/Post), but displays a negative correlation with the target_azimuth feature. In addition, when there's high correlation between features in features_status, it might be worth considering feature selection or dimensionality reduction techniques to avoid issues like multicollinearity in modeling. 4.3.3 Density Analysis The purpose is to group similar buildings together based on their latitude/ longitude. It first identifies the shape and location of each building. Then, it calculates the distance between buildings and groups nearby buildings into clusters (1-6). The visualization above in Figure 3 shows the spatial distribution of buildings. Areas with dense clusters

of points indicate higher building density, likely corresponding to urban or suburban areas. The darker clusters in the plot highlight specific areas with significant building concentration. These areas may be of particular interest for first responders on disaster response strategies. Conversely, areas with scattered points suggest less developed or rural regions.

Figure 3 Building Clusters Using DBSCAN

275