Extracting indirect information relevant to the Earth & environmental sciences from data is common to almost all of our research, as it is in all of science. We use a wide-variety of machine learning methods, almost all of the regression flavor as our unknown variables are continuous, and while we will use any technique that accomplishes are aims, we tend to favor largely on Bayesian approaches. Development of novel machine learning applications and estimation techniques that transcend a specific problem are done under the πEES banner.
InSAR time-series analysis
Multiscale InSAR Time Series (MInTS) is a technique to extract time series from catalogs of interferometric synthetic aperture radar (InSAR) measurements of ground deformation. MInTS is based on a wavelet decomposition in space and a general parameterization in time. The latter is partly motivated by common techniques for estimating time series of deformation from GPS observations, but also allows MInTS to efficiently deal with subsets of interferograms that may be spatially decorrelated – in essence, MInTS interpolates ground deformation both in space and in time, i.e. over “holes” in the interferometric data. In MInTS the estimation of time-dependent ground deformation is done in the wavelet domain. Development of MInTS was in collaboration with Pablo Musé (U. de la Republica, Uruguay), Mark Simons (Caltech), and Piyush Agram (Caltech). The theory behind MInTS, as well as a processing “tree” and example application to Long Valley, CA, is described in Hetland et al. (2012) (available from JGR as open access).
MInTS was originally developed as a Matlab toolbox, which is available for download at this link MInTS. MInTS has been fully integrated into the InSAR package Generic InSAR Analysis Toolbox (GIAnT; available from http://earthdef.caltech.edu/; also see Agram et al. 2013).
spatio-temporal strain-rate mapping
vulnerability & resilience to natural & environmental hazards