The lightning flashes used in this work were mapped using data from the LO-FAR (LOw Frequency ARray) radio telescope. Due to its effective lightning protection system, LOFAR is able to continue to operate during thunderstorm activity[1]. The Dutch LOFAR stations consist of 38 (24 core + 14 remote) stations spread over 3200 km 2 in the northern Netherlands. The largest baseline between core stations is about 3 km, the largest baseline between remote stations is about 100 km. From each station we use 6 dual-polarized low band dipole antennas (LBA), sampled at 200 MHz, to observe the 30-80 MHz band. The raw time series data were saved to the transient buffer boards, which continuously buffer the last 5 s of data from a maximum of 48 dual-polarized antennas per station. The resulting relative timing accuracy is better than 1 ns. See [2] for more details on LOFAR. When a lightning flash occurs within the area enclosed by the Dutch LOFAR stations, as observed by www.lightningmaps.org, the transient buffer boards are stopped and the data is read to disk. The method we used to map each lightning flash has three major steps. In the first step we fitted plane-waves to the time of pulses received by individual LOFAR stations. Note that the LOFAR stations are less than 100 m in diameter and the lighting is many kilometers from the closest LOFAR station, so that a plane-wave approximation is very good for individual LOFAR stations. These plane-waves were used to identify non-functional antennas, and the intersection of their arrival directions gave a rough first estimate of the flash location, accurate to a few kilometers. Since each station has its own clock and cable delays, in the second step we found the clock offsets between the different LOFAR stations by simultaneously fitting the locations of multiple events and station clock offsets to the measured times of radio pulses, with a Levenberg-Marquardt minimizer. In order to achieve the highest precision, we chose to fit locations of 5 events that created pulses that were strong but not saturating, had a simple structure, and did not change shape significantly across different stations. After fitting, the root-mean-square difference between the modeled and measured arrival times of the radio pulses was around 1 ns. The resulting station clock offsets are consistent with LOFAR station clock calibrations, which are known

We study the link between an expanding coronal shock and the energetic particles measured near Earth during the ground level enhancement of 2012 May 17. We developed a new technique based on multipoint imaging to triangulate the three-dimensional (3D) expansion of the shock forming in the corona. It uses images from three vantage points by mapping the outermost extent of the coronal region perturbed by the pressure front. We derive for the first time the 3D velocity vector and the distribution of Mach numbers, M FM , of the entire front as a function of time. Our approach uses magnetic field reconstructions of the coronal field, full magnetohydrodynamic simulations and imaging inversion techniques. We find that the highest M FM values appear near the coronal neutral line within a few minutes of the coronal mass ejection onset; this neutral line is usually associated with the source of the heliospheric current and plasma sheet. We illustrate the variability of the shock speed, shock geometry, and Mach number along different modeled magnetic field lines. Despite the level of uncertainty in deriving the shock Mach numbers, all employed reconstruction techniques show that the release time of GeV particles occurs when the coronal shock becomes super-critical (M FM > 3). Combining in situ measurements with heliospheric imagery, we also demonstrate that magnetic connectivity between the accelerator (the coronal shock of 2012 May 17) and the near-Earth environment is established via a magnetic cloud that erupted from the same active region roughly five days earlier.