A Note About STELLA Data
STELLA is solely an educational and outreach tool designed to help students and the community learn about Landsat and remote sensing. The performance of STELLA has not been scientifically peer-reviewed and data retrieved should be used for educational purposes only.
STELLA spectrometers provide users the ability to capture point-based spectral observations in the visible (VIS) and near infrared (NIR) bands of the Electromagnetic Spectrum. These spectrometers can be used over a variety of land cover surface targets such as urban areas, bare soils, grasses, and other vegetated surfaces.
Calculating Reflectance from STELLA Data
STELLA spectrometers measure incoming light as irradiance in microwatts per centimeter squared (uW/cm²), and must be converted to reflectance when comparing measurements over different targets.
To calculate reflectance, collect coincident calibration measurements using the white reference panel. Ensure the target and reference measurements are collected using the same acquisition geometries, light levels, and distance between the device and the target or panel*. Record the distance from the device to, and the area of, the target.
Using the raw irradiance values collected by STELLA, the distance in cm, and the area, calculate radiance for both the target and the reference observations:
Radiance (uW/cm²) = Irradiance * Distance²/Area
From the calculated radiance of the target and the white reference, calculate reflectance of the observed target:
Reflectance (%) = Radiance of target / Radiance of reference
*An easy way to ensure measurements of the target and reference panel are taken under the same conditions is by following the ‘sandwich’ method, which consists of collecting a sequence of panel-target-panel measurements and using the average of the two reference panel readings for calculating reflectance.
Limitations and Considerations
While it is possible to obtain reliable measurements over short, uniform vegetation canopies and flat surface targets, limitations exist when collecting data over surfaces with diverse Bidirectional Reflectance Distribution Function (BRDF) properties. Surfaces with heterogeneous texture and vertical canopy structure (e.g., tall corn or forest canopies) are associated with an increase in measurement uncertainty. To derive accurate spectral measurements, the BRDF properties of the target land cover must be considered. To enable comparison between observations, it is important to maintain consistent acquisition geometry and light levels between measurements.
STELLA data was evaluated against measurements collected with an Analytical Spectral Devices spectrometer (ASD). An ASD is a common field instrument, calibrated to radiance, that provides stable reflectance estimates in the 400 – 2,500 nm region with a 3 nm spectral resolution. STELLA and ASD data were both collected in situ using a white panel target under consistent light and acquisition (sun-sensor-target) geometry conditions.
While the STELLA reflectance measurements were significantly higher than those collected using the ASD, the two datasets were strongly correlated under these conditions and no significant differences in uncertainty were observed (figure). Generally, the standard deviation of the data were found to vary between 8 – 18% while the standard errors were found in the range of 3 – 6%, with the VIS bands having lower uncertainties than those in the red-edge and NIR bands.
What do the STELLA spectrometers measure?
STELLA-Q2 – Light intensity (uW/cm2), in the visible and near infrared wavelengths (in nanometers): 410, 435, 460, 485, 510, 535, 560, 585, 610, 645, 680, 705, 730, 760, 810, 860, 900, 940
The light intensity is measured in microwatts per centimeters squared, with error bars of +/- 12% of the value. The sensor selects the specific wavelength bands by using a set of silicon thin-film interference filters, to a precision of +/- 5 nanometers. The bands are centered around the wavelengths listed above, and the bandwidth of each band is +/- 10 nanometers full-width half-maximum around the band center, in a Gaussian distribution of sensitivity. Silicon is a hard material with low thermal expansivity, so the sensor characteristics are stable over a broad temperature range, as well as over the life of the sensor. The sensor’s field of view is cone-shaped, with a cone angle of +/- 20° for a total field of view of 40°.
STELLA-Q, 1.0, 1.1 and 2.0 – Light intensity, in the visible wavelengths (in nanometers): 450, 500, 550, 570, 600, 650
The visible light intensity is measured in microwatts per centimeters squared, with error bars of +/- 12% of the value. The sensor selects the specific wavelength bands by using a set of silicon thin-film interference filters, to a precision of +/- 5 nanometers. The bands are centered around the wavelengths listed above, and the bandwidth of each band is +/- 20 nanometers full-width half-maximum around the band center, in a Gaussian distribution of sensitivity. Silicon is a hard material with low thermal expansivity, so the sensor characteristics are stable over a broad temperature range, as well as over the life of the sensor. The sensor’s field of view is cone-shaped, with a cone angle of +/- 20° for a total field of view of 40°.
STELLA-Q, 1.0, 1.1 and 2.0 – Light intensity (in microwatts per centimeters squared), in the near infrared wavelengths (in nanometers): 610, 680, 730, 760, 810, 860
The near infrared light intensity is measured in microwatts per centimeters squared, with error bars of +/- 12% of the value. The sensor selects the specific wavelength bands by using a set of silicon thin-film interference filters, to a precision of +/- 5 nanometers. The bands are centered around the wavelengths listed above, and the bandwidth of each band is +/- 10 nanometers full-width half-maximum around the band center, in a Gaussian distribution of sensitivity. Silicon is a hard material with low thermal expansivity, so the sensor characteristics are stable over a broad temperature range, as well as over the life of the sensor. The sensor’s field of view is cone-shaped, with a cone angle of +/- 20° for a total field of view of 40°.
STELLA-1.0, 1.1 and 2.0 – Surface temperature, in degrees Celsius. This sensor measures a far infrared light spectrum to produce a spectral curve. This curve is fit to a black-body thermal emission curve, to derive the surface temperature. This sensor is calibrated to objects that are emissive (not shiny metal surfaces). The temperature reading is good to +/- 0.5°C. The sensor’s field of view is cone-shaped, with a cone angle of +/- 17.5°, for a total field of view of 35°.
STELLA-1.0, 1.1 and 2.0 – Air temperature, in degrees Celsius. This sensor measures the ambient air temperature to an accuracy of +/- 0.25°C. It measures the semiconductor conduction-band quantum energy band-gap to derive the temperature. This method of measurement is accurate across a wide range of temperatures (-40 to +125°C) and is stable over the life of the sensor.
STELLA-1.0, 1.1 and 2.0 – Ambient conditions: Relative humidity, barometric pressure, altitude, and air temperature. This sensor measures those four parameters, though the air temperature measurement is less accurate than that of the MCP9808, so we do not record this sensor’s air temperature reading. The relative humidity measurement is good to +/- 3% and the barometric pressure reading, in hectoPascals, is good to +/-1 hPa. The altitude measurement is uncalibrated, so the absolute value is not accurate. The precision of the altitude measurement is better than 0.1%, so we include it to allow data marking by altitude excursion (a quick rise and fall) if the STELLA is in use on an aerial drone. In this way, the drone data and the STELLA data can be synchronized.
All STELLAs – Time. The clock on the STELLA is set to Coordinated Universal Time (UTC) to avoid confusion of time zones and daylight savings time. The clock is powered by the backup battery and will continue to keep time, even when the STELLA is off. The accuracy of this clock chip is +/- 2 seconds per day, about +/- 12 minutes per year.
