Astroinformatics and Astrostatistics (hereafter AIAS) has received an excellent and growing response during the last few years. This indicates the importance and timeliness of a much needed interface between astronomy and various branches of applied mathematics and computer science. As the number of large area surveys grow, the diversity in wavebands, apertures, areas covered and cadence grows manifold. Time-domain methodologies are crucial to make sense of variability, for a wide variety of targets, such as binary stars, exoplanets, or distant supernovae and blazars. Unlike the far more regular time-series of the financial markets, astronomers often have to contend with sparse, irregular, heteroskedastic time-series. The methodologies being developed as a response lie on the interface of mathematics/statistics, domain knowledge, and computer science. Through this special session we hope to spur further discussions on availability of methods to a wider audience, and increased applicability to combined datasets.
The three talks showcase three different aspects of the field, and the panel discussion brings forth the important issue of funding as well as better showcasing of AIAS issues.
Organizers: Ashish Mahabal (Caltech), Aneta Siemiginowska (Harvard-Smithsonian Center for Astrophysics, Cambridge, MA), E Feigelson, E Ford, P Yanamandra-Fisher, and the members of the Steering Committee of the Working Group on A & A (AAS WGAASC).
The first generation of large synoptic survey archives, such as CRTS, PTF and Pan-STARRs, are now (or soon will be) available to the community, enabling unprecedented systematic searches and studies of variable astrophysical phenomena. These range from moving objects in the Solar System to extreme quasars in the distant universe. However, much of the analyses of these data sets conducted so far have aimed at providing statistical descriptions of the variability. Whilst such parameterizations are useful for feeding classification algorithms, they are not effective at describing the underlying type of variability in the sources or the physical mechanism(s) for it. In this talk, we will discuss new approaches, such as wavelet variance, random matrix theory and echo state networks, that can provide insight into the science of variability rather than just statistically characterizing it. We will pay particular attention to sources exhibiting stochastic variation and how much information about the host system can be determined from their time series. For example, characteristic restframe timescales have been identified in quasars, potentially related to the size of coherent noise fields in the accretion disk. Finally, we will also consider the potential limitations of the next generation surveys, such as LSST and SKA.
The synoptic surveys enabled by the LSST will be able to support a wide range of time domain astronomy investigations, but maximizing the science yield through careful design of the observing strategy presents a challenge to the community. I will describe the preparation work of the strong lensing arm of the LSST Dark Energy Science Collaboration, as we seek to understand the response of our measurements of time delays in lensed quasars to the LSST observing strategy. By quantifying time delay accuracy in the Metric Analysis Framework and documenting it in the LSST Observing Strategy White Paper, I will demonstrate the mechanism by which the LSST Observing Strategy can be influenced. Our accuracy quantification was made possible by experiments on simulated light curves as part of an open, blinded, "Time Delay Challenge." This exercise served to stimulate the development of a number of different algorithms for time delay measurement, which I will discuss in the context of subsequent planned challenges leading up to LSST commissioning.
Over the past couple decades, thousands of extra-solar planets have been discovered orbiting other stars. Most have been detected and characterized using transit and/or radial velocity time series, and these techniques have undergone huge improvements in instrumental precision. However, the improvements in precision have brought to light new statistical challenges in detecting and characterizing exoplanets in the presence of correlated noise caused by stellar activity (transits and radial velocities) and gaps in the time sampling (radial velocities). These challenges have afflicted many of the most interesting exoplanets, from Earth-like planets to planetary systems whose orbital dynamics place important constraints on how planetary systems form and evolve. In the first part of the talk, I will focus on the problem of correlated noise for characterizing transiting exoplanets using transit timing variations. I will present a comparison of several techniques using wavelets, Gaussian processes, and polynomial splines to account for correlated noise in the likelihood function when inferring planetary parameters. I will also present results on the characteristics of correlated noise that cause planets to be missed by the Kepler and homegrown pipelines despite high nominal signal-to-noise. In the second part of the talk, I will focus on the problem of aliasing caused by gaps in the radial-velocity time series on yearly, daily, and monthly timescales. I will present results on identifying aliases in the Fourier domain by taking advantage of aliasing on multiple timescales and discuss the interplay between aliasing and stellar activity for several habitable-zone "planets" that have recently been called into question as possible spurious signals caused by activity. As we push toward detecting and characterizing lower mass planets, it is essential that astrostatistical advances keep pace with advances in instrumentation.
Ashish Mahabal (aam at astro dot caltech dot edu) Aneta Siemiginowska (asiemiginowska at cfa dot harvard dot edu)