The Daytime MODVOLC Algorithm

Initially the MODVOLC algorithm only operated on the nighttime MODIS data stream. Since January 2002 we have also been analyzing daytime MODIS images for volcanic thermal anomalies.

Daytime data pose a slightly different set of problems. The important 4 μm wavebands of MODIS, 21 and 22, measure not only thermally emitted radiation but also reflected light from the Earth's surface. As a result highly reflective surfaces (such as sandy beaches, deserts, snow and ice) can appear 'hot' in the raw 4 μm data. To try to account for these complications we have implemented a slightly different version of the nighttime algorithm described by Wright et al. (2002; 2004).

Correcting for solar reflection

We look at the spectral radiance detected from each pixel at 1.6 μm (MODIS band 6). During the day, radiance detected at 1.6 μm is dominated by reflected sunlight (see Wooster and Rothery, 1997a, 1997b). Based on the top-of-the-atmosphere solar irradiance we assume that the reflected component of the 4 μm radiance is equivalent to 4.26% of the energy reflected by the pixel at 1.6 μm. We then subtract this value from the raw band 21 and 22 data and use the result (L21corr, L22corr) to compute the daytime Normalised Thermal Index. L21corr and L22corr are assumed to represent the thermally emitted radiance from each pixel (equivalent to L21 and L22 in the nighttime data stream, but see note below). The daytime MODVOLC algorithm can therefore be written as:

  • NTI = (L22corr - L32)/(L22corr + L32),

    or, when band 22 is saturated,

  • NTI = (L21corr - L32)/(L21corr + L32).

For example, the following image subsets show band 21 daytime data of Lake Tana, in Ethiopia.

On the left, the raw MODIS band 21 data. Elevated levels of 4 μm radiance within the lake are clearly visible. In fact, these pixels were sufficiently radiant to saturate the band 22 detectors. As a result, calculation of an NTI using the raw band 21 or 22 data would result in the erroneous detection of 'hot-spots' within the lake; false positives. On the right we show the band 21 data corrected using the procedure outlined in the preceding paragraph. The correction we apply (A) confirms that the highly radiant pixels within the lake were indeed reflection anomalies and not real hot-spots and (B) suppresses these reflection anomalies in L21corr and L22corr, the corrected radiances we use for the daytime NTI. In this example, the thermally emitted radiance from the highly reflective lake pixels was very low, and insufficient to trip the MODVOLC daytime NTI threshold of -0.60 (see below). Targets that are very reflective but also very cold (such as clouds, snow, and ice) have low or even negative L21corr and L22corr values, meaning that they also fail to trip the daytime alert threshold.

During the day the 4 μm radiance signal is contaminated by sunlight for all MODIS pixels. The ideal result of the correction we apply is to 'remove' the reflected component from the 4 μm radiance signal, leaving a purely thermal component for use in calculating an NTI. However, it is important to note that it is IMPOSSIBLE to perfectly isolate the thermal radiance component, particularly in coarse spatial resolution data; you always over- or under-correct the raw signal to a certain degree. The correction we apply is merely intended to facilitate the detection of real hot-spots and reduce the scope for false positives (see 'Note on the contents of the MODVOLC text files', below).

As the 4 μm channel is sensitive to thermal emission from surfaces at magmatic AND ambient temperatures, daytime solar heating means that we also have to use a higher threshold during the day. An empirically derived NTI threshold of -0.60 was determined. Hot-spots are identified as those pixels for which the NTI > -0.60

The image below is a subset of the same daytime image of Ethiopia shown above (Lake Tana is visible at the top) showing the corrected band 21 data (L21corr).

Here, active fires appear as bright pixels. Seven, with a range of intensities, are circled. To the right, the NTI, computed using the methods described herein, is shown for each. All have NTI thresholds > -0.60, the daytime MODVOLC threshold and are automatically classified as real hot-spots. Also shown for each fire is the 4 μm brightness temperature calculated using both the raw 4 μm radiances (L21 and L22) and the sunlight-corrected 4 μm radiance (L21corr and L22corr, in brackets). Also shown are the ΔT values for each pixel. The ΔT values for background pixels in this image are in the range 1 to 7 K, much lower than the fire-pixel values.

Screening for sun-glint

Sun-glint is a specular reflection anomaly that occurs in daytime data, predominantly over oceans and standing water bodies (you have seen it whenever you have been blinded looking across a body of water at sunset). We screen the daytime MODIS data for sun-glint using the techniques described in Giglio et al. (2003), on a pixel-by-pixel basis using the solar and satellite zenith and azimuth angles contained in the MODVOLC files. Pixels for which the sun-sensor geometry approaches this specular angle are categorised as potential sun-glint pixels and excluded from the hot-spot maps. Our sun-glint threshold angle is 12 degrees (see Giglio et al., 2003 for details of the calculation).

Note on the contents of the MODVOLC text files

Although we correct the raw 4 μm data for use in the daytime algorithm we report the uncorrected 4 μm values in the MODVOLC ascii files. We also include sun-glint pixels in this file (although the sun-glint pixels will not appear in the map window). We do this to avoid forcing our own corrections on users. So, if you do not agree with the sunglint correction threshold we use, you can implement your own. If you do not agree with the daytime radiance correction we use, you can devise a different one.

References

  • Giglio et al., (2003). An enhanced contextual fire detection algorithm for MODIS. Remote Sensing of Environment, 87, 273-282.
  • Wooster, M.J., and Rothery, D.A. (1997a). Thermal monitoring of Lascar volcano, Chile, using infrared data from the along-track scanning radiometer. Bulletin of Volcanology, 58, 566-579.
  • Wooster, M.J., and Rothery, D.A. (1997b). Time series analysis of effusive volcanic activity using the ERS along track scanning radiometer: the 1995 eruption of Fernandina volcano, Galapagos Islands. Remote Sensing of Environment, 62, 109-117.