Washington, DC - Studio photographers may be familiar with the 1,000-watt quartz halogen lamps known as “FELs.” Scientists use them too—specially calibrated ones, at least—to test the performance of light sensors that monitor Earth’s weather, plant life and oceans, often from space.

A researcher at the National Institute of Standards and Technology (NIST) has recently made an improved mathematical model of the light output of FEL lamps. The new model, developed by NIST theorist Eric Shirley, will make the lamps more useful research tools, the scientists say, particularly for calibrating a relatively new class of cameras called hyperspectral imagers.

Rainbow Vision

Hyperspectral cameras are used for a wide range of monitoring applications, including biomedical, defense, and ground-based, air-based and space-based environmental sensing. While ordinary cameras only capture light in three bands of wavelengths—red, green and blue—hyperspectral imagers can be designed to see all the colors of the rainbow and beyond, including ultraviolet and infrared. Their increased range allows these cameras to reveal the distinctive signatures of processes that are invisible to the naked eye.

Some of these effects are subtle, however—such as when researchers are trying to tease out changes in ocean color, or to monitor plant growth, which helps them predict crop productivity.

“These are both examples where you’re looking at an extremely small signal of just a couple percent total,” said David Allen of NIST’s Physical Measurement Laboratory (PML). In cases like this, achieving low uncertainties in the calibration of their detectors is essential.

Of particular interest to Allen and his colleagues was a calibration technique called the “lamp-plaque” method, popular with scientists because it is relatively inexpensive and portable. For this calibration procedure, researchers use a standard FEL lamp. Incidentally, FEL is the name designated by the American National Standards Institute for these lamps. It is not an acronym.

First, the lamp light shines onto a white, rectangular board called a reflectance plaque, made of a material that scatters more than 99 percent of the visible, ultraviolet and near-infrared light that hits it. Then, after bouncing off the plaque, the scattered light hits the camera being calibrated.

The method has been used for decades to calibrate other kinds of sensors, which only need to see one point of light. Hyperspectral imagers, on the other hand, can distinguish shapes.

“They have some field of view, like a camera,” Allen said. “That means that to calibrate them, you need something that illuminates a larger area.” And the trouble with the otherwise convenient lamp-plaque system is that the light bouncing off the plaque isn’t uniform: It’s brightest in the center and less intense toward the edges.

The researchers could easily calculate the intensity of the light in the brightest spot, but they didn’t know exactly how that light falls off in brightness toward the plaque’s edges.

To lower the calibration uncertainties, researchers needed a better theoretical model of the lamp-plaque system.

Counting Coils

Shirley, the NIST theorist who took on this task, had to consider several parameters. One major contributor to the variations in intensity is the orientation of the lamp with respect to the plaque. FEL lamps have a filament that consists of a coiled coil—the shape that an old-fashioned telephone cord would make if wrapped around a finger. All that coiling means that light produced by one part of the filament can be physically blocked by other parts of the filament. Setting the lamp at an angle with respect to the plaque exacerbates this effect.

To model the system, Shirley took into account the diameter of the wire and both coils, the amount of space between each curve of the coils and the distance between the lamp and the plaque.

“These are all things that were obvious,” Shirley said, “but they were not as appreciated before.”

NIST scientists tested the actual output of some FEL lamp-plaque systems against what the model predicted and found good agreement. They say the uncertainties on light intensity across the entire plaque could now be as low as a fraction of a percent, down from about 10 to 15 percent.

Moving forward, NIST will incorporate the new knowledge into its calibration service for hyperspectral imagers. But researchers are preparing to publish their results and hope scientists will use the new model when doing their own calibrations. The work could also serve as a foundation for creating better detector specifications, potentially useful for U.S. manufacturers who build and sell the cameras.

“There’s an emerging market for hyperspectral sensors in general,” Allen said. “They’re becoming more sophisticated, and this is a component to help them be a more robust product in an increasingly competitive market.”