Abstract
We present a novel method for extracting the radiance from High Temperature Events (HTEs) recorded by geostationary imagers using Independent Component Analysis (ICA). We use ICA to decompose the image cube collected by the instrument into a sum of the outer products of independent, maximally non-Gaussian time series and images of their spatial distribution, and then reassemble the image cube using only sources that appear to be HTEs. Integrating spatially gives the time series of total HTE radiance emission. In this study we test the technique on a number of simulated HTE events, and then apply it to a number of volcanic HTEs observed by the SEVIRI instrument. We find that the technique performs well on small localised eruptions and can be used to correct for saturation. The technique offers the advantage of obviating the need for a priori knowledge of the area being imaged, beyond some basic assumptions about the nature of the processes affecting radiance in the scene, namely that (i) HTE sources are statistically independent from other processes, (ii) the radiance registered at the sensor is a linear mixture of the HTE signal and those from other processes, and (iii) HTE sources can be reliably identified for the reconstruction process. This results in only five free parameters — the dimensions of the image cube, an estimate of the data dimensionality and a threshold for distinguishing between HTE and nonHTE sources. While we have focused here on volcanic HTEs, the methodology can, in principle, be extended to studies of other kinds of HTEs such as those associated with biomass burning.
Highlights
For the simulated volcanic eruptions, we test the effectiveness of the procedure in two ways: Firstly we find the coefficient of determination (r2) between the simulated and extracted source and spatial contribution to see if the ‘shape’ of these vectors are recovered accurately
To give an overview of the decomposition into sources stage, we present a selection of the sources extracted by the FastICA algorithm showing the range of High Temperature Events (HTEs), diurnal cycle and cloud process dominated sources found
In this study we have presented an approach for extracting the radiance from High Temperature Events (HTEs) from time series of geostationary satellite images using Independent Component Analysis
Summary
Satellite images have been used to observe thermal emissions from High Temperature Events (HTEs) for over four decades, including terrestrial volcanic activity (e.g. Gawarecki, Lyon, & Nordberg, 1965; Glaze, Francis, & Rothery, 1989; Hanel et al, 1979; Harris et al, 1997; Oppenheimer, 1991), wildfires (e.g. Justice et al, 2002; Kaufman et al, 1998; Roberts & Wooster, 2008), burning fossil fuels (Casadio, Arino, & Minchella, 2012; Kwarteng & Bader, 1993) and eruptions on the Jovian moon Io (e.g. Carr, 1986; Davies, 1996; McEwen et al, 1998). Satellite images have been used to observe thermal emissions from High Temperature Events (HTEs) for over four decades, including terrestrial volcanic activity Isolating the HTE radiance from other sources, such as reflected sunlight and thermal emission from ground, clouds and atmosphere as well as instrument response effects such as stray light image. A relationship between neighbouring pixels is frequently used, for example mean neighbour subtraction or the band ratio method, these methods are prone to large errors (Wooster & Kaneko, 2001). Geostationary imagers are suited to the time series approach due to their consistent acquisition geometry, which excludes complicating factors such as changes in view angle, pixel size and irregular acquisition intervals which make analysing data from Low Earth Orbit (LEO) imagers more difficult
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have