Skip to main content

Two heads are better than one

By [email protected] - 23rd August 2016 - 11:17

Hyperspectral and multispectral remote sensing are tools for detailed mapping of materials. Well-developed application areas include geology and mineral exploration; forestry; marine, coastal zone, inland waters and wetlands; agriculture; ecology; urban land use; and snow and ice. There are also numerous military applications in camouflage, littoral zone mapping and landmine detection.

Hyperspectral sensors pose an advantage over multispectral sensors in their ability to identify and quantify molecular absorption. The high spectral resolution of a hyperspectral imager allows for the detection, identification and quantification of surface materials, as well as the inferring of biological and chemical processes.

For all these applications – but particularly for those using hyperspectral remote sensing – ground truth signatures collected in the field and indexed in spectral libraries are critical for many methods of analysis. While image processing packages often include basic spectral libraries, application-specific libraries collected from the area being mapped and containing field reflectance spectra of the specific materials occurring in the target area greatly improve the accuracy of generated maps. In particular, spectra of vegetation and soils are influenced by such a wide range of environmental conditions that it is difficult to adequately represent this variability without the collection of site-specific field reflectance spectra.

To best match remotely sensed imagery, field-collected reflectance spectra must be collected from materials in-situ using natural solar illumination. Spectra collected in the laboratory using materials collected in the field are often not directly comparable to remotely sensed imagery, since the collection process typically disrupts or modifies the sample, particularly for vegetation and bare soil surfaces. In the case of vegetation, the structure or architecture of the plant is a significant factor in the characteristics of the observed reflectance spectrum and the collection process alters this architecture. As soils typically have significant near-surface stratification, a reflectance spectrum measured using a collected sample is unrepresentative of the reflectance signature observed in a remotely sensed image.

Rapidly changing atmospheric conditions pose one of the greatest obstacles to the collection of accurate field-collected reflectance spectra using solar illumination. Ground truth reflectance spectra measured in the field are computed as a ratio of two measured radiance spectra: the radiance of a reference panel of known reflectance and the radiance of the target surface. Implicit in this calculation is the assumption that the spectral irradiance illuminating the reference panel and target surface are identical. Thus, field measurements collected under variable illumination conditions often cannot be reproduced or correlated to imagery with any degree of certainty.

Using a single field portable spectroradiometer, collection of field reflectance spectra is limited to days with stable atmospheric conditions. Even a nearly cloud-free day can have time-varying atmospheric vapour that precludes the measurement of accurate field reflectance spectra. This then results in spending extended time in the field waiting for good weather.

Overcoming limitations

This limitation is overcome by using two field portable spectroradiometers to simultaneously measure both the reference panel and target surface. The FieldSpec Dual software package provides the means to wirelessly link two field portable FieldSpec spectroradiometers and precisely intercalibrate the wavelength and radiance scales of the two instruments, to ensure the collection of accurate reflectance spectra. Since the FieldSpec spectroradiometers use fibre optics to collect reflected light and deliver it to the internal spectrometers, the fibre optics of both instruments are easily aligned to simultaneously view the intercalibration targets.

One of the two instruments – the fixed base unit (FBU) – is mounted on a tripod so that it continuously views the reference panel. The second unit – the mobile unit (MU) – is carried by an operator who moves around the field area measuring the various target materials. Whenever the operator commands the MU to collect a field reflectance signature, the FBU is wirelessly signalled to collect a simultaneous reference panel measurement. This reference measurement is then wirelessly transmitted to the MU and the reflectance spectrum is computed using MU and FBU measurements as well as the intercalibration spectra previously collected. The computed reflectance spectrum is then stored and displayed for the operator.

Small differences in spectral resolution and wavelength calibration between the two synchronised instruments cause spikes and other errors in the computed spectra. Matching of the wavelength resolution of the two instruments is carried out at the time of manufacture by careful focusing and alignment of spectrometer components while viewing a set of atomic emission line sources. Calibration software is used to monitor in real-time the instrument’s resolution at each atomic emission line, which enables the calibration technician to tune the spectrometer’s wavelength resolution to predetermined target values.

Just before performing field measurements, the two systems are set up side by side to perform intercalibration measurements. The two instruments first make a simultaneous measurement of a wavelength inter-calibration tile to perform any adjustment necessary to match their wavelength calibrations. Next, the reflectance scales of the two instruments are matched by view a reference panel of known reflectance. This reflectance scale inter-calibration is periodically repeated during the collection period.

Proof of concept

As a proof of concept both the dual system and a standard single spectrometer system were used to collect field reflectance spectra of several geologic materials (see Figure 1). The dual system collected the test spectra about 15 minutes after initial configuration. The single spectrometer system collected the test spectra about 5 minutes after performing the reference panel scan. These measurements were made on a day with very unstable atmospheric conditions that would not normally allow for field collection of reflectance spectra: overall sunlight brightness varied by over 30% in a five minute period while the atmospheric water vapour varied by over 45% in the same time period.

Variability of water vapour is of particular concern since absorption by water vapour has a strong influence on over 70% of the wavelengths used by most remote sensing imaging systems. The influence of changing atmospheric water vapour is seen in the spectra collected with the single spectrometer system (dotted lines in Figure 1) where the water vapour changes that occurred in the few minutes between the collection of the reference panel scan and the sample scan resulted in spectral artefacts at wavelengths of significant water vapour absorption. These artefacts aren’t present in the spectra collected with the dual system (solid lines in Figure 1) since the reference panel and sample scans are collected simultaneously. This data clearly shows that, without the FieldSpec Dual system of a pair of linked field spectrometers, it would not be possible to collect accurate field reflectance spectra under these conditions.

The use of application specific spectral libraries greatly improves the accuracy of maps generated from hyperspectral and multispectral imagery. The use of the FieldSpec Dual system enables the field collection of reflectance spectral signatures under much less favourable sky conditions than is possible with a single spectrometer. This results in much less wasted time in the field waiting for suitable atmospheric conditions resulting in less time spent collecting the application specific spectral libraries that are key to maximising information extraction from remotely sensed imagery.

Dr Brian Curtiss is chief technology officer for ASD.

www.panalytical.com

Download a PDF of this article

Download