(Table 2) Picked maximum velocities from 3 radars (RAY, SHK, and SUM), showing 10 explosions at Erebus volcano, Antarctica, and their respective directivity vectors

We used a novel system of three continuous wave Doppler radars to successfully record the directivity of i) Strombolian explosions from the active lava lake of Erebus volcano, Antarctica, ii) eruptions at Stromboli volcano, Italy, and iii) a man-made explosion in a quarry. Erebus volcano contains a convecting phonolite lava lake, presumably connected to a magma chamber at depth. It is one of the few open vent volcanoes that allow a direct observation of source processes during explosions. Its lava lake is the source of frequent violent Strombolian explosions, caused by large gas bubbles bursting at the lake surface. The exact mechanism of these bubble bursts is unclear, as is the mechanism of the creation of the infrasound signal accompanying the explosions. We use the Doppler radar data to calculate the directivity of Strombolian eruptions at Erebus. This allows us to derive information about the expected type of infrasound source pattern (i.e. the role of a dipole in addition to the monopole signature) and the physical structure of the volcano. We recorded 10 large explosions simultaneously with three radars, enabling us to calculate time series of 3D directivity vectors (i.e. effectively 4D), which describe the direction of preferred expansion of the gas bubble during an explosion. Such directivity information allows a comparison to dipole infrasound radiation patterns recorded during similar explosions only a few weeks later. Video observations of explosions support our interpretation of the measurements. We conclude that at Erebus, the directivity of explosions is mainly controlled by random processes. Since the geometry of the uppermost conduit is assumed to have a large effect on the directivity of explosions, the results suggest a largely symmetrical uppermost conduit with a vertical axis of symmetry. For infrasound recordings, a significant dipole signature can be expected in addition to the predominant monopole signature.

Minimum description length, regularisation and multi-modal data

Conventional feed forward Neural Networks have used the sum-of-squares cost function for training. A new cost function is presented here with a description length interpretation based on Rissanen's Minimum Description Length principle. It is a heuristic that has a rough interpretation as the number of data points fit by the model. Not concerned with finding optimal descriptions, the cost function prefers to form minimum descriptions in a naive way for computational convenience. The cost function is called the Naive Description Length cost function. Finding minimum description models will be shown to be closely related to the identification of clusters in the data. As a consequence the minimum of this cost function approximates the most probable mode of the data rather than the sum-of-squares cost function that approximates the mean. The new cost function is shown to provide information about the structure of the data. This is done by inspecting the dependence of the error to the amount of regularisation. This structure provides a method of selecting regularisation parameters as an alternative or supplement to Bayesian methods. The new cost function is tested on a number of multi-valued problems such as a simple inverse kinematics problem. It is also tested on a number of classification and regression problems. The mode-seeking property of this cost function is shown to improve prediction in time series problems. Description length principles are used in a similar fashion to derive a regulariser to control network complexity.