Reading beyond the visible
NOAA coastal imagery / Spatial analysis
This analysis compares natural-color and color-infrared products from the 2004 coastal Connecticut acquisition. It examines band composition, display stretches and differences between vegetation indices calculated from 8-bit and 16-bit digital numbers. Source collections and acquisition metadata
Data selection and sampling
Each view selects a 2004, 0.50 m raster at the map center and locks its service raster ID. Paired comparisons require matching tile names. These checks constrain source selection; they do not establish subpixel registration or radiometric equivalence.
Display views request PNG imagery. Numerical operations request uncompressed TIFFs with an explicit U8 or U16 pixel type and no additional rendering function. Requests use EPSG:3857 and nearest-neighbor sampling. The analytical grid is 128 × 128; reported spacing varies with the view extent and is not necessarily the native ground sample distance. ImageServer export parameters
The 16-bit container does not imply a 16-bit sensor: NOAA documents a 12-bit dynamic range for this ADS40 acquisition. Distributed pixel values also reflect orthophoto production and subsequent service processing. NOAA acquisition record
Map controls
Maps load directly from NOAA. Drag to pan; scroll, pinch or use +/− to zoom. Previous imagery remains visible during navigation and refreshes after release or a pause. Drag the divider to compare the selected left and right images. Enable Inspect pixels for point samples. Retained views are not used for new measurements. Archived imagery is not substituted if NOAA is unavailable.
1. RGB and color infrared
Natural-color rendering maps red, green and blue bands to their corresponding display channels. Conventional CIR maps near infrared to red, red to green and green to blue. Compare vegetation, water and built surfaces using the same acquisition and extent.
The inspector reports source digital numbers and service-specific raster IDs. Display color alone is insufficient for identifying vegetation species or estimating reflectance.
2. Bit depth and delivered values
Unsigned 8-bit data represent integers from 0 to 255; unsigned 16-bit data represent 0 to 65,535. Compare the CIR products, then inspect their sampled histograms.
Similar PNGs can conceal different source distributions. A U16 response can also contain values confined to a smaller range. The histogram reads a bounded TIFF sample, not the display image or a full-collection histogram.
3. Contrast stretch
Compare minimum–maximum, percentile and standard-deviation stretches, then vary gamma. The figure record retains the rendering rule; histogram requests omit that display rule.
Use a consistent stretch when comparing appearance across products. Contrast enhancement does not recover information lost during quantization or prior processing.
4. Near-infrared band inspection
Inspect water, canopy, pavement and wetland vegetation in grayscale NIR and CIR. The NIR band here is extracted from the 2004 CIR product; the separate 2017 infrared mosaics are not used in these comparisons.
5. NDVI from digital numbers
$$NDVI = \frac{NIR - Red}{NIR + Red}$$
The browser calculates floating-point ratios from the requested TIFF and displays them on a fixed −1 to +1 scale. All-zero source pixels and zero denominators are excluded.
Both CIR products supply the required bands. Common multiplicative scaling cancels in the ratio; band-dependent scaling, offsets and quantization do not. These results use processed digital numbers, not calibrated surface reflectance, and should be interpreted accordingly.
6. Product differences
$$\Delta NDVI = NDVI_{16} - NDVI_{8}$$
The calculation requires matching tile names, dimensions and georeferencing, and uses the intersection of valid-pixel masks. The readout reports mean and maximum absolute difference and the percentage within a selected tolerance. Display colors saturate at ±0.25; statistics retain the calculated values.
Differences characterize the delivered products on this sampling grid. They do not isolate sensor quantization from processing, resampling or residual registration error. Missing paired sources or incompatible grids stop the calculation.
7. Threshold sensitivity
Vary the threshold and input bit depth. Compare classified-pixel counts and proportions of the valid area. The area estimate applies a center-latitude correction to the Web Mercator grid; use an appropriate projected CRS for a defensible area calculation.
This threshold is an analytical example, not a validated vegetation or Phragmites classification. An operational classifier requires reference observations, an explicit sampling design and independent accuracy assessment.
Reproducibility and export
Each map's Source, analysis parameters & reusable embed panel provides request URLs, raster identities, bounds, timestamps, response checksums and parameters as JSON. It also exports an interactive HTML map using the current controls. Exported maps require access to NOAA; they do not contain an imagery archive.
For a repeatable comparison, retain the figure record and document the acquisition, sampling grid, mask, stretch and threshold. Repeat at more than one location before generalizing a result. Source revisions and changes in service processing can affect subsequent requests to the same endpoint.