The Cooperative Institute for Research in the Atmosphere (CIRA) in Fort Collins, Colorado, together with the NOAA/NESDIS RAMM Branch is developing and distributing the Volcanic Ash Principal Component Image Product.
The Volcanic Ash Principal Component Image product is created at CIRA and made available online. It will be implemented at the NWS Alaska Region Headquarters in Anchorage.
The size of one Volcanic Ash Principal Component Image product is less than 0.5 MB, with updates available every 30 minutes. Plans are to improve the temporal resolution, by generating image products every 15 minutes.
The Volcanic Ash Principal Component Image product, developed at CIRA, is demonstrated on the RAMSDIS Online web page for GOES-West, because the “split-window” bands are not available on GOES-East. (The Product is also available from GOES-10 over South America, and the technique is applicable to GOES-East, but with an altered interpretation.) The product displays standard GOES Imager data in a unique way using linear sums and differences of multi-spectral imagery, and enhancement can be applied. Inputs are all 5 bands of the GOES Imager.
The Volcanic Ash Principal Component Image product demonstrates a unique kind of imagery that is primarily available from older GOES Imager instruments (through GOES-11) and all GOES Sounder instruments (although at lower, 10 km, spatial resolution). The “split-window” bands, that are the best inputs for this product, will reappear in the GOES-R era. GOES-R will feature the Advanced Baseline Imager (ABI) sensor which will be able to produce both higher spatial resolution (2 km) and higher temporal resolution (5 min) version of the Volcanic Ash Principal Component Image product.
Here, is a brief description of how the Volcanic Ash Principal Component Image product is created from GOES-11 Imager data:
The Volcanic Ash Principal Component Image product is computed from all 5 bands of GOES imagery, but is heavily weighted by the two longwave infrared window bands available on older operational geostationary satellites (up through GOES-11). (On GOES-12 and beyond, the less-transparent longwave band was replaced by a 13.3 um band used to better detect the levels of clouds. This does not prevent the product from being generated, but results in an altered interpretation).
The key to the image differencing is the use of Principal Component Analysis on the multi-spectral GOES (or other) imagery. This analysis sorts out the image content according to explained variance, with the first component containing the most variance, and with decreasing variance in successive components. This process relegates most of the redundant image content to the first component and image difference information into the other components. And, because the volcanic ash signal is often best detected by an image difference, the split-window difference in particular, that difference appears as one of the higher-order Principal Components. Image differences seen in other components may show other features, such as cloud type: water vs. ice, etc…
The use of Principal Component Imagery for the detection of various atmospheric phenomena, including volcanic ash, has been documented in the published literature. See Hillger and Clark1 for volcanic ash applications, and Hillger and Ellrod2 for other applications, including smoke and blowing dust.
The Principal Component Analysis results in new images, equal in number to the number of spectral bands that were analyzed. Thus 5-band GOES data results in 5 Principal Component Images (PCIs). Four of those images (PCI-1, 2, 3 and 5) are displayed in Figure 2. A brief explanation of each component is provided on each image. The missing component, PCI-4, was the least useful with respect to the detection of volcanic ash, and was therefore excluded from the 4-panel display.
The Volcanic Ash Principal Component Image product is generated day or night, whether or not visible imagery is available. However, at night the components will change due to the absence of reflected solar radiation
The Volcanic Ash Principal Component Image product is even beter when viewed as an image loop. A real-time example of this 4-panel product as a loop is available on the RAMSDIS Online web page for GOES-West. The center point for analysis can be changed as needed to focus on volcanic events of interest.
An additional step that can be applied to the Volcanic Ash Principal Component Image product is to combine some of the components using RGB (Red, Green, and Blue) color combinations. An example of this is shown in Figure 3, where three of the PCIs are given colors and combined. The number of possible color combinations is large, so some prior knowledge of useful combinations has been applied. A criteria for such combinations, is that the volcanic ash be a desired “warning” color, such as red, and that background colors be more subdued. The combination in Figure 3 uses Red for PCI-3, Green for PCI-5, and Blue for PCI-2.
A real-time example of this colored volcanic ash product is available as a loop on the RAMSDIS Online web page for GOES-West. And, the accompanying 1 km visible image loop is also available from GOES-West. An alternate two-color combination is shown in Figure 4. In this case two of the RGB colors (Green and Blue) are assigned to the same component (PCI-3). This reduction in the number of PCIs employed, results in higher-level cirrus cloud surrounding the volcanic ash, to appear more like the denser parts of the ash plume, an undesirable effect. Therefore this RGB combination, although less noisy, has its tradeoffs.
The Volcanic Ash Principal Component Image product provides a visually powerful display of areas of volcanic ash, and other associated cloud features that might interfere with the detection of volcanic ash. The components show volcanic ash as dark in the split-window difference in PCI-5. Other cloud features are seen in other components. The user can learn to distinguish between image features with minimal training, and through the use of examples of this product produced during past volcanic eruptions.
The main disadvantage of the product is that the Volcanic Ash Principal Component Image product, when generated to include visible imagery, will change at night due to the absence of reflected solar radiation. One option is to generate the product on infrared spectral bands only, so that the product will not change from day to night. Alternately, the user can be trained to alter the PCI interpretation caused by changes in solar radiation from day to night. This is aided by observing the Volcanic Ash Principal Component Image product over time. Also, with new images available, potentially every 15 minutes, a relatively rapid evolution of the volcanic ash plume is possible.
The higher-order PCIs, such as PCI-5, are very noisy, being a highly-stretched image difference. Users need to be aware that small volumes of volcanic ash may not be readily detected, especially considering the relatively-low 4 km spatial resolution of the GOES infrared imagery. (Even though the GOES visible/reflective band is available at 1 km resolution, it’s necessary to combine the images at the same resolution, resulting in images at the lower 4 km resolution of the GOES infrared imagery.)
Interactions with NWS users via the Proving Ground will assist algorithm developers in improving this product, in particular the color combinations that may be most useful for particular applications. The development of future, improved products will also benefit from user feedback.
1) Hillger, D.W., and J.D. Clark, 2002: Principal Component Image analysis of MODIS for volcanic ash, Part-1: Most important bands and implications for future GOES Imagers, J. Appl. Meteor., 41(10),985-1001.
2) Hillger, D.W., and G.P. Ellrod, 2003: Detection of important atmospheric and surface features by employing Principal Component Image transformation of GOES imagery, J.Appl.Meteor., 42(5), 611-62