Ground truthing and validating imagery

Hyperspectral remote sensing data can be used to distinguish vegetation types, water, soil, and different rocks. However, to determine specific attributes such as water quality, subsurface contamination, or vegetative health, researchers collect ground-truth (or validation) data.

This synopsis outlines what is needed for ground truthing and its purpose. More detailed information for the research community is in the tutorial section of CSTARS.

Recommended ground-truthing data to collect include:

• Atmospheric conditions
• Surface water
• Vegetation
• Soil, bare ground, rock outcrops.

The atmosphere intercepts incoming solar radiation and affects the intensity and/or frequency of reflected energy signals. Correct hyperspectral images for different atmospheric conditions including:

• Aerosols/haze
• Humidity
• Incident solar radiation
• Temperature
• Wind direction and speed.

Ideally, take your atmospheric measurements at the same time as the image is sensed – ~2 hours before or after solar noon is standard protocol.

To calibrate the hyperspectral image and to correct for atmospheric influences, collect spectra of uniformly bright and dark calibration targets within your study site as close to the time of the imaging overflight as possible. Choose calibration targets on level ground that is relatively uniform in appearance and as close to black or white as possible. Good targets include asphalt, cement, sand, a flat aluminum roof, or bare soil. Targets should be larger than the resolution of the imaging instrument (preferably > 9 to 25 pixels).

Clear water absorbs most of the solar radiation in the bandwidths imaged by hyperspectral sensors. Not all surface water is clear. Organic and inorganic solids increase reflectivity. Dissolved materials also alter hyperspectral images.

To account for differences in reflectivity and absorption in surface water, sample to determine &/or collect information about:

• Chlorophyll content
• Dissolved chemical analyses (e.g. VOC, SVOC, metals, etc.)
• Georeference point(s) (GPS)
• Incident solar radiation
• Site location and light conditions (field notes)
• Spectral signatures/spectroradiometer readings
  • calibration target
  • different depths in water column
  • distilled water reference (bench test)
  • sample water
  • water surface at ~2m
• Turbidity and Secchi depth

The spectral signature within a pixel of an image consists of an average of reflectances of all materials within that pixel. So, for a spatial resolution of 5x5 pixels, the spectral response for a stand of vegetation will consist of a combination of spectra of all vegetation types, soil, ground litter, etc.

Take representative spectrometer readings above the canopies for as many categories or classes of vegetation as possible within time or access constraints. Examples of categories of vegetation in California include grassy field, salt or freshwater marsh, stands of pine or live oak, and mixed vegetation.

Site conditions and/or project goals likely will dictate sampling methodology – either take many readings in a single location (narrow, deep sampling) or obtain few readings at many locations (broad, shallow sampling). If surface or subsurface contaminants are important, collect appropriate spectral data. Sampling along transects from contaminated areas to background (clean) areas across a plume may prove useful.

Collect actual leaf/soil/litter samples for later reflectance analysis in the lab. If the entire canopy consisted of leaves, presumably, its spectra would resemble the leaf spectra. In reality, the canopy is made up of different materials accounted for accordingly in its spectra.

Vegetation ground-truthing data to collect at each site include:

• Field notes regarding site and light conditions
• Spectroradiometer readings
  • calibration targets (dark and light)
  • vegetation
• Vegetation sample – leaf or leaf cluster

Principles similar to those for vegetation sampling apply to soil, bare ground and rock sampling. Take measurements representative of the types of soils or contamination existing in the remotely sensed site.

Wet soils are darker than dry ones – so, at sites where water content varies, measure both wet and dry soil conditions. Features like wet and dry soils and rock outcroppings may require additional location-specific GPS data. Again, data collected will be somewhat dependant on project goals.

Use a power-stabilized light source and calibration standards for leaf, soil and litter measurements done in the lab.