Ghent Canopy Loss

Norfolk, VA — satellite analysis, 2011–2023

Norfolk's Ghent neighborhood has a mature urban tree canopy that residents value deeply. The city has not updated its tree survey in years. So we looked at the data ourselves.

Using publicly available USDA NAIP aerial imagery at 1-meter resolution, we compared Ghent's tree canopy between June 2011 and October 2023. Drag the slider to compare.

True Color Aerial

June 2011 (left) vs October 2023 (right). Drag the divider.

Color Infrared (CIR)

In color-infrared composites, healthy vegetation appears bright red. Roads and buildings are gray or blue. This makes it easy to spot where trees have been removed.

Drag the divider to compare 2011 (left) vs 2023 (right).

Change Map

Vegetation change detected by comparing each pixel's NDVI rank between 2011 and 2023. Red = relative greenness loss. Green = relative greenness gain. Unchanged areas show the 2023 aerial photo at reduced brightness.

What the data shows

The CIR comparison reveals visible canopy thinning across Ghent's residential blocks. Street trees that formed continuous canopy in 2011 now show gaps. Several parcels that were shaded by mature trees now show exposed rooftops. The change map highlights these losses in red, concentrated along the commercial corridor and scattered through residential streets.

Some caveats: the 2023 image was captured in October vs June for 2011. Deciduous trees in Norfolk (USDA zone 8a) still carry most of their leaves in early October, but NIR reflectance is somewhat lower in fall. Some apparent change may be seasonal rather than permanent. The rank-based comparison mitigates this by comparing each pixel's relative greenness within its own year, but a same-month comparison would be more definitive.

NAIP imagery is available for seven vintages covering Ghent: 2011, 2012, 2014, 2016, 2018, 2021, and 2023. Unfortunately, the USDA changed its radiometric processing between years, making automated NDVI differencing unreliable without cross-calibration against invariant targets (parking lots, rooftops). The visual comparison above remains the most trustworthy evidence.

Method

Imagery was accessed from the Microsoft Planetary Computer NAIP collection via STAC API. Each year's best-overlapping scene was selected for the Ghent bounding box (-76.305, 36.865 to -76.285, 36.880). Red (Band 1) and NIR (Band 4) were read via GDAL's /vsicurl/ directly from Azure Blob Storage. CIR composites map NIR→Red, Red→Green, Green→Blue.

All processing was done in Rust using the gdal crate. Source code: tools/canopy-loss.