MEarth data release notes ========================= Document version 2.1 2013 September 11 Prepared by Jonathan Irwin This file contains a summary of information that we hope is useful when working with the data, and a description of the contents of the release files (at the end of the document). For additional release documentation, including a more detailed description of how the data were processed, summary tables of objects observed, and finding charts, please see the following HTTP URL: http://www.cfa.harvard.edu/MEarth/Data.html Summary of changes since release 1.0 ==================================== Data from the 2011-2012 season are now included, according to the release schedule. Before the start of this season, the detectors were completely rebuilt in new housings, featuring a higher level of cooling and pre-flash capability, and this has been used to eliminate the persistence effect. The operating temperature was reduced to -30C. Please note that the filter configuration was changed, and uses a slightly different total thickness of RG715 glass as well as other changes to the configuration of glass elements near the focal plane intended to reduce scattered light. This, combined with the change in operating temperature, means the bandpass is probably different when compared to data from the 2008-2010 season using the RG715 filters. Scattered light is much lower in the 2011-2012 season, after the source of the majority of the effect (a reflective black anodized surface in the Cassegrain tube) was identified and eliminated. A change in observational strategy was made at the start of the 2011-2012 season to obtain observations of all MEarth targets at a cadence of approximately 10 days (in addition to the sample being actively searched for transiting planets at high cadence). These data may be useful for long-term monitoring of the photometric behavior of the target stars. Time stamps are now given as Barycentric Julian Date (BJD) in the TDB time system. This was mainly done to eliminate the discontinuities in the original time stamps due to leap seconds in UTC, of which there have now been two during the time MEarth has been operating. Please note that the full accuracy potentially offered by this method still cannot be achieved due to the time stamps from the commercial data acquisition software only being written out to one second precision, uncertainties in exactly what moment in the exposure-taking process the time refers to, and remaining computer clock synchronization errors, which we note are larger than readers familiar with UNIX systems may be accustomed to due to the peculiarities of the Windows NTP implementation used. The main effect of the change to the time stamps is not BJD versus HJD itself, but rather the approx. 65-67 second TDB-UTC offset, which may be significant for some applications. (Addendum 2013 September 11: tests of the data acquisition system performed on site show that the time stamp recorded in the FITS header for data taken after the detector upgrade corresponds to the start of the pre-flash and not the time the shutter opens, which is approximately 9s later. We plan to investigate the stability of this offset and will apply corrections in future releases if possible. Since this issue was discovered after the present release, it has not been corrected in the data files, and for applications requiring precise timing, an offset of 9 seconds should be applied.) The error in the bias subtraction affecting very short exposures in the 2008-2010 season from 2008 October until 2009 May 7 discussed in the release notes for the first data release has been addressed by reprocessing these images. Sky background estimation for the 2008-2010 data was changed to use the same sky annulus method as later seasons. It is not immediately clear this treatment is always an improvement, but it should reduce concerns about systematics due to contamination of the aperture fluxes from scattered light, particularly around bright Moon. A number of master (reference) images were updated to better quality frames. In order to address systematic offsets seen in some light curves, an additional set of zero-point coefficients were added to the 2008-2010 season at the point where the change to operating with the telescopes in focus was made. A small number of errors in the alternate (non-LSPM) identifiers given in the tables have been found and addressed. These were nearly all members of close multiples where the components were confused. Remaining known issues ====================== This data release contains automatically generated "working copies" of the light curves. These have been subject to minimal human intervention. There are a number of known flaws in this processing, which are detailed below. Stellar parameters assumed in the observations and provided in the release materials are known to be flawed, and are based on outdated and incomplete literature data. A greatly improved compilation of literature data for the full target list, including revised stellar parameters, is in preparation, but has proved to be an immense task and is not ready at the time of writing. Handling of close multiple sources is in need of improvement. A "blend" flag is provided in the light curve header, but it should be noted that the flag is neither necessary nor sufficient for identifying potential problems, and sometimes flagged objects have perfectly usable photometry. Please see the description of the "blend" flag, below. When multiple neighboring sources were identified in the master frame, an attempt is made to generate separate light curves for them. This often produces poor results for the fainter members of very close pairs. Conversely, if a source was unresolved in the master frame (e.g. due to this frame being taken in very poor conditions) only a single aperture will be used, but it is possible for this to shift toward or onto the brighter source in frames with better image quality. Systematics were much worse in the 2010-2011 season with the I-band like interference filter than in previous seasons. This was due to a strong humidity dependence in the red end interference cutoff of the filter bandpass, discovered after operations began. The "common mode" method used to mitigate the second-order extinction effects due to telluric water vapor absorption in the original filter successfully corrects these systematics to a large degree, so the data can be dealt with using similar methods to the remainder, but the residuals are often larger. From the start of 2011 May to the end of the 2010-2011 season, tel02 experienced intermittent problems with stuck shutter blades, which caused severe (and variable) vignetting in the corners of the field. Most targets are unaffected, with the notable exception of the field containing LSPMJ1157+7927 and LSPMJ1203+7907, where the targets were placed near the corners. The latter object occasionally experienced large dropouts as a result of this problem. We have not attempted to remove the affected frames, but these data points are spurious. During the 2011-2012 season, it was unfortunately necessary to remove all of the detectors from the telescopes at least once during the season. These events were typically due to detector repairs and were usually accompanied by the loss of approximately one month of data, and likely also changes in flat-fielding error due to the crudeness of the mechanism used to realign the detector on the telescope after return. A complete list of the occasions when this was done is given below. Several other instrumental features and related information are summarized below. Instrument description ====================== MEarth uses 8 near-identical telescope/detector systems housed in a single roll-off roof enclosure at the Fred Lawrence Whipple Observatory, Mount Hopkins, Arizona. Approximate site coordinates are: Latitude: +31d 41' 03.1" N Longitude: 110d 52' 41.0" W Elevation: 2384 m (Note: these coordinates are for the center of the telescope array, and were updated compared to release 1, based on nearby National Geodetic Survey markers; we thank the NIST group who share the building with us for bringing the availability of much more accurate site coordinates to our attention) Telescopes are a 0.4m f/9 Ritchey-Chretien Cassegrain design, used on a German equatorial mount. The science detectors are midband coated e2v CCD42-40 operated with thermoelectric cooling at approximately -15C for 2008-2011 (although the operating temperature has varied, see below), and -30C for 2011-2012. Three of the telescopes have a slightly different focal length from the rest. The approximate values of the pixel scale are 0.757 arcsec/pix for tel01, tel02, tel04, tel05, tel07, and 0.764 arcsec/pix for tel03, tel06, tel08. The filter is fixed, and was RG715 Schott glass in 2008-2010 and 2011-2012. In 2010-2011, a custom I-band-like interference filter was used, built on the same Schott glass substrate, with the interference coating used to define the red cut-off at approximately 8950A in an attempt to mitigate systematics due to atmospheric water vapor. The interference cutoff was found to vary substantially with ambient humidity and this experiment was abandoned at the end of the season. The choice of mounting design has a significant impact on the photometry. The telescope must be used on the west of the pier when the target is east of the meridian, and vice versa, with only a 5 degree window during which the target can be observed on the "wrong" side of the meridian before the telescope must be flipped over the pier. Flipping the telescope over the pier rotates the focal plane relative to the sky by 180 degrees, so any target off the optical axis will sample two regions of the detector during the night. This leads to "meridian offsets" in light curves, which require special treatment (see below). We derive the information about which side of the pier the telescope was on from astrometric solutions on the data taken during target acquisition or from the science images as necessary. It is not sufficient to merely assume images with positive hour angle have one orientation, and negative hour angle have another, because it is possible to track slightly across the meridian, as mentioned above. Dedicated flags and an angle measurement are included in the light curves and should be used to detect "meridian flips". Also note that the pointing can be far off in the science images if a flip happened after the target acquisition exposures but before the science image was taken, because there is no way to detect this condition until it is too late. During the 2008-2009 MEarth season, the telescopes were operated intentionally out of focus, adjusted to achieve FWHM of approx. 5 pixels (where a normal in-focus FWHM is approx. 2.5 pixels) in order to be able to collect more photons per exposure before saturation. It was found to be very difficult to maintain a stable defocus, and the FWHM varied substantially with changes in ambient temperature. Stabilizing this was hindered by failure of the focus drive electronics on many of the telescopes (see below). As a result, these data have rather unique systematics and many other related problems. For the 2009-2010 season and onwards, the telescopes were operated in focus. The telescopes are extremely prone to wind shake as used in the present enclosure (which has minimal protection from wind). This results in distorted PSFs in moderate wind, becoming severely distorted in high winds approaching the close-down limit for the enclosure. Due to the design of the building, this affects each telescope differently, and also depends on wind direction, with east generally being more problematic than west due to the lack of shelter. Winds from the northeast and occasionally northwest can sometimes also result in extremely poor seeing. This is noticeable in MEarth images even though the intrinsic PSF FWHM is approx. 2 arcsec, which normally renders them relatively insensitive to seeing variations, and can often be distinguished from wind shake by lower winds and a lack of high image ellipticities, as well as the detailed PSF shape, which tends to be much smoother when the cause of the large images is poor seeing. Images tend to show visible PSF distortions in winds exceeding 25 km/h, and it is likely there are astrometric distortions at even lower wind speeds. The effect on image quality also depends on the nature of the wind with gusty conditions tending to produce a mixture of undistorted and highly distorted frames. Because aperture photometry is used with large apertures, the light curves are remarkably insensitive to these problems, but there is a noticeable trend of large photometric scatter (often caused by a poor zero-point solution) in high wind. These images can usually be detected by examining the FWHM and ellipticity parameters, although the most distorted frames are known to fool the source classification procedure, which can result in no estimates of FWHM or ellipticity being made because it didn't think there were any stellar sources on the frame. This appears in the output as FWHM = -1.0. It should also be noted that the achieved pointing is significantly degraded in high wind. The mounts do natively (before correction) show a small amount of periodic error, which can cause image elongation in the y (RA) direction for long exposures. The worm period is 2.5 minutes, so these effects are most pronounced for exposures longer than 30 seconds or so. However, the effects can also show up in high-cadence continuous sequences of exposures taken for transit followup or similar, where it leads to excess scatter in the y coordinate time-series. It is possible for the periodic error to interact badly with the pointing stabilization software loop used during these observations, and this can increase the size of the error since the feedback attempting to fix the worm error occurs out of phase with the error itself. Periodic error correction has been used since 2010 January to address this problem. The periodic error curves are based on measurements of the error averaged over approx. 20-25 worm cycles, using a dedicated set of very high-cadence short exposures on a bright, equatorial star taken during low wind. However, it is found that the solutions need to be updated quite frequently (every few months) as there is some drift in the worm error over time. This has occasionally lagged due to the special conditions (extremely low wind) needed to obtain a good calibration or because it was not noticed quickly. The periodic error can change dramatically when the mounts are re-lubricated, so the calibration is always re-run at these times (annual, at startup after the monsoon). During the 2011-2012 season, telescope 6 began to show extremely large periodic error, and there is some residual effect even after correction. The possibility of replacing the worm drive is under investigation. The original detectors showed residual images ("persistence"), which affects all 2008-2011 data. The time-dependence of the residual images is is found to be well approximated by an exponential decay with a time constant of approximately 20 minutes, although this varies from one telescope to the next and is likely to be dependent on operating temperature. We discuss the effects this has on the photometry and strategies used for mitigation, below, and provide more details of the residual image problem itself on the web pages. During the 2011 summer shutdown, the detector housings were re-built to achieve a lower operating temperature, and a preflash system was added to properly address the problem, so persistence should not affect 2011-2012 data. Data organization ================= The philosophy followed in the data processing has been to start over with a completely separate set of calibration files and light curves whenever major changes were made to the instrument that alter the bandpass of the photometry or other key photometric properties. So far, this has been done on three occasions: 1. September 2008: move from commissioning to the start of the survey. At this time, the full suite of telescopes were bought on-line, and the filter was changed from the RG9 filter used during commissioning to the custom-made RG715 filters for the main survey. 2. October 2010: changed from the RG715 filter to a custom interference I-filter (hereafter called I_715-895). 3. October 2011: CCD cameras were completely re-built in new housings with higher cooling performance and a preflash system. Filter was changed back to RG715, although note that it is not from the same manufacturer, has a slightly different thickness, was used in conjunction with different glass elements near the focal plane, and at a different CCD operating temperature, so the bandpass is not identical. Numerous other small improvements were also made. The present release concerns data taken in all three of these periods, and there are therefore three sets of files. These are named "2008-2010", "2010-2011", and "2011-2012" after the range of calendar years over which they were taken. We refer to these as "seasons", even though 2008-2010 contains two observing seasons (years). Please note that although quite major changes have been made to the hardware, the same physical CCD chip has always remained associated to the same telescope throughout the survey, and the same set of stars have been observed on each telescope. After the detector housing re-build, the detectors were re-identified using cosmetics (defects) and put back with their original telescope, even though the serial numbers had changed. It was possible to do this unambiguously, so these associations should continue. Other hardware and observational changes ======================================== Active development and improvement of the survey hardware, observational strategy and data processing have continued throughout operations. We have attempted to minimize impact on data taken during a season, preferring to consolidate major changes at the ends of seasons around the times of the summer monsoon when the observatory is shut down. However, hardware failures have meant it has occasionally been necessary to make larger than ideal disturbances to individual telescopes during seasons. This is particularly the case for 2011-2012. The most important changes that are known or suspected to affect the photometry are summarized in this and the following sections. A number of large focus changes occurred during the 2008-2010 season, mostly due to failure of the focus drive electronics on the telescopes, a situation which has gradually improved and was largely under control by the 2010-2011 season. These are readily identified by examining the FWHM in the light curves (see below). We note that telescopes were operated intentionally out of focus in 2008-2009, but it was found that this was very difficult to maintain at an adequate level of stability, and the decision was made to operate in focus from 2009-2010 onwards. An automatic focus routine was implemented early during this year of operations to maintain focus, and automatic temperature compensation was used to follow focus changes during the night. The 2009-2010 season continued to be plagued by focus electronics failures, leading to seasonal focus drifts as the ambient temperature changed, and mount mechanical failures (due to an error in our maintenance procedures that was addressed during the 2010 summer shutdown). During the 2008-2009 season, the CCD temperature setpoint was -20C, which proved to be too optimistic, and became very difficult to maintain during the summer. The setpoint was changed to -15C for the 2009-2010 and 2010-2011 seasons. During the 2010-2011 season, problems were experienced with the cooler on telescope 2, so the setpoint for this telescope was changed to -10C at the start of the season. The setpoint on telescope 5 was adjusted to -10C on 2010 December 10 due to concerns about condensation in the CCD chamber. For the 2011-2012 season, the temperature setpoint was -30C. Variations in the device operating temperature were a particular concern due to persistence in the earlier seasons (see below) although temperature likely also has a small effect on the quantum efficiency, and thus the bandpass. Changes to individual telescopes ================================ The following table summarizes the dates when a detector was replaced on a telescope after having been removed, with a brief description of the reason for the change. Removal and replacement of the detector is likely to cause changes in flat-fielding error, and sometimes collimation or focus - telescopes are always re-focused after re-installing a detector, so during 2008-2009 this caused changes in the amount of de-focus. Each of these events increments the "instrument version number" (see "V" in the light curves) by one, and this triggers the creation of a new per-star magnitude and meridian offset (where data on both sides of the meridian exist) when producing in the light curve, corresponding to the numbered "segment"s (see "S" in the light curve table). Tel. V 1stnight Description ----- - -------- ----------- tel01 2 20081001 Start of 2008-2010 season tel01 3 20090901 Changed to operate in focus tel01 4 20091115 Repair (condensation). tel01 5 20101028 Filter changed to I tel01 6 20111011 Rebuilt camera installed, RG715 filter tel01 7 20120101 Camera removed to extract leaf. tel01 8 20120131 Repair (shutter driver). tel02 2 20081001 Start of 2008-2010 season tel02 3 20090901 Changed to operate in focus tel02 4 20091119 Repair (condensation; controller replaced). tel02 5 20101028 Filter changed to I tel02 6 20111011 Rebuilt camera installed, RG715 filter tel02 7 20111223 Repair (shutter driver). tel03 1 20080929 Commissioned tel03 2 20090901 Changed to operate in focus tel03 3 20091115 Repair (condensation). tel03 4 20101028 Filter changed to I tel03 5 20111011 Rebuilt camera installed, RG715 filter tel03 6 20120206 Repair (shutter driver). tel03 7 20120621 Repair (shutter driver). tel04 2 20081001 Start of 2008-2010 season tel04 3 20090901 Changed to operate in focus tel04 4 20101028 Repair (condensation); filter changed to I tel04 5 20111011 Rebuilt camera installed, RG715 filter tel04 6 20120425 Repair (shutter driver). tel05 1 20080605 Commissioned tel05 2 20090901 Changed to operate in focus tel05 3 20091115 Repair (condensation) tel05 4 20100208 Repair (stuck shutter) tel05 5 20100525 Repair (stuck shutter) tel05 6 20101028 Filter changed to I tel05 7 20111011 Rebuilt camera installed, RG715 filter tel05 8 20120107 Repair (shutter driver). tel06 1 20080928 Commissioned tel06 2 20090901 Changed to operate in focus tel06 3 20101028 Repair (condensation); filter changed to I tel06 4 20111011 Rebuilt camera installed, RG715 filter tel06 5 20120108 Repair (shutter driver). tel07 1 20080605 Commissioned tel07 2 20090901 Changed to operate in focus tel07 3 20091115 Repair (condensation) tel07 4 20101028 Filter changed to I tel07 5 20111011 Rebuilt camera installed, RG715 filter tel07 6 20120417 Repair (shutter driver). tel08 1 20090106 Commissioned tel08 2 20090901 Changed to operate in focus tel08 3 20101028 Repair (condensation); filter changed to I tel08 4 20111011 Rebuilt camera installed, RG715 filter tel08 5 20120101 Repair (shutter driver). When detectors were repaired, changes in the shutter shading, non-linearity and gain can result, so these calibrations were regenerated and used to reduce all subsequent data when any evidence for a difference from past behavior was found. This has been noted below in a separate table where appropriate, detailing which calibrations were changed at each occasion. No changes of this nature were made mid-season for the 2010-2011 or 2011-2012 data, so this table applies only to the 2008-2010 data. Tel. 1stnight Calibrations changed ----- -------- --------------------- tel01 20091112 gain, nonlin, shutter tel01 20100901 gain, nonlin, shutter tel02 20091112 gain, nonlin, shutter tel03 20091112 gain, nonlin, shutter tel03 20100901 gain, nonlin, shutter tel04 20100303 nonlin tel04 20090901 shutter tel04 20100901 gain, nonlin, shutter tel05 20091112 gain, nonlin, shutter tel05 20100201 shutter tel05 20100524 gain, nonlin tel05 20100901 gain, nonlin, shutter tel06 20090901 shutter tel06 20100901 gain, nonlin, shutter tel07 20091112 gain, nonlin, shutter tel07 20100901 shutter tel08 20090901 shutter tel08 20100901 gain, nonlin, shutter The following table gives the approximate reciprocal gain in e-/ADU, readout noise in e-, and saturation level in ADU appropriate for each telescope before and after the detector upgrade. The values before the upgrade are from the 2010-2011 season, but should also be close for the 2008-2010 season. 2010-2011 2011-2012 Tel. Gain RON Sat. Gain RON Sat. ----- ---- ---- ----- ---- ---- ----- tel01 1.4 11.3 61000 1.8 8.2 48000 tel02 1.6 9.5 58000 1.7 6.9 41000 tel03 1.5 11.6 52500 1.5 9.2 46000 tel04 1.5 11.5 58000 1.7 6.4 48000 tel05 1.1 10.7 59000 1.8 9.5 53000 tel06 1.5 10.4 50000 1.5 7.9 48000 tel07 1.7 9.7 44000 1.4 8.4 45000 tel08 1.2 10.7 63000 1.4 8.8 51000 Target and field selection ========================== Our target selection procedures are detailed in Nutzman & Charbonneau (2008), and the survey target list was drawn directly from this work. Note that the target stars selected for observation were exclusively those with estimated radii < 0.33 Rsol, but see the following discussion for cases where there were other objects in the "LSPM-North M-dwarfs" list present in the field-of-view. It is possible (but fairly unusual) for multiple M-dwarfs to appear within a single field-of-view (approx. 26x26 arcmin). In these cases, we observe a single field, with the position adjusted to contain as many of these M-dwarfs as possible (any left over are assigned their own field as normal). This was done using the full "LSPM-North M-dwarfs" list, i.e. without the < 0.33 Rsol criterion. Exposure times were set to avoid saturation on the brightest M-dwarf in the field with estimated radius < 0.33 Rsol, and only fields containing at least one M-dwarf meeting this estimated radius criterion were observed. Field centers chosen for each target field were dithered within a 1x1 arcmin box about the expected positions of the target M-dwarfs in order to mitigate the effects of persistence. This procedure tries to ensure each target never "sees" the persistent image of a previous target in the photometric aperture, which could severely disturb the photometry (especially since the other targets under observation might change during the night or from night to night). We try to return a given target to the exact same detector pixels every time it is observed on a given side of the meridian. This means that the target does see its own persistent image. We have found that this can lead to ramps (mostly for the highest cadence continuous observations, such as transit follow-up), and magnitude offsets depending on the cadence for objects where the cadence has been changed during a season. It is obviously extremely undesirable to attempt high-precision photometry using detectors with residual image problems, and this was properly addressed by implementing a pre-flash system at the 2011 summer shutdown in time for the start of the 2011-2012 season. During photometry, apertures are placed on all objects in the "LSPM-North M-dwarfs" list and light curves extracted. This means there is occasionally usable survey photometry for an earlier star than our notional cut-off. Observational strategy and cycling of the targets list ====================================================== The observational strategy has been summarized in several publications (Nutzman & Charbonneau 2008; Irwin et al. 2009; Berta et al. 2012), but we repeat the important features here. We also provide time-lapse movies of selected nights on our web pages, which illustrate many aspects of the discussion visually. Except for cases of bad weather, or where adjustments were made by the operator, targets were observed for the entire time they were at a zenith distance of less than 60 degrees (approximately) and the Sun was more than 12 degrees below the horizon (the hours between nautical dusk and dawn). All targets are assigned a scheduling priority, which are mostly the same, but this can be used to ensure a given target is observed at the desired cadence when the schedule becomes busy during the night (e.g. due to realtime triggers in later seasons). Any target within 40 degrees of the Moon at full Moon, scaling down as the fraction of lunar disk area illuminated away from full, was not observed on the night in question. Target observations are sometimes interspersed with other observations, e.g. follow-up activities, and in later seasons, realtime triggers, both of which can cause lengthy interruptions, but occur infrequently. Otherwise, we aim to return to each target every 20 minutes. This is called a "visit" below, to distinguish it from an individual exposure - multiple exposures might be taken at each "visit". We attempt to maintain the same set of "well-observed" targets for as long as possible in a given season in order to prevent small amounts of data being accumulated on many targets, which is an occupational hazard with a dynamic scheduling algorithm. Exposure times were modified as infrequently as possible for stability reasons, due to the large shutter shading corrections and concerns about persistence. This was generally done annually during summer shutdown, and the quality of the exposure time estimates has been progressively refined throughout survey operations. Mid-season changes were generally reserved for objects where saturation was found to be a problem. For the 2008-2009 season, exposures were set to obtain approx. 250,000 photons per exposure regardless of stellar properties. Only a single exposure was taken per visit. In 2009-2010 and onwards we adjusted the exposure times as discussed in Nutzman & Charbonneau (2008) to achieve a 3 sigma detection of a 2 Earth radius planet in each "visit", given the assumed stellar properties (see below), although some exceptions were made for bright, late-type objects, where we simply continued to gather as many photons as possible before saturation (the observing time for these objects is dominated by the slew and field acquisition overheads so the extra time spent integrating comes at negligible cost). Telescopes were operated in focus, and it was found that it was often not possible to obtain enough photons before saturation in one exposure, particularly for earlier-type targets, so visits for these objects were split into multiple sub-exposures to make up the required total integration time. The exposure time computations try to account for all of the noise sources included in our noise model (see below, under e_Mag). Exposures longer than 120s were found to sometimes exhibit poor tracking (there are no auto-guiders), so we also split these to keep the maximum single-exposure time below 120s. These groups of observations are presented individually, rather than stacking or binning them, because this would result in loss of information. The scatter within a visit can be a useful noise estimator, and having all of the data permits the use of robust (outlier-resistant) averages. It is not possible to observe all survey targets on the same night at the desired cadence. In practice, approx. 120-240 targets are observed at the 20 minute survey cadence per night (the number of targets observed per night depends on the length of the night / time of year, but has also decreased each survey year; there are a number of reasons for this, including increases in the time per visit due to increasing numbers of sub-exposures or more realistic exposure times, and poor scheduling decisions in the initial implementation leading to the targets with the shortest time taken per visit tending to be observed early in the survey). In order to complete the survey (which intended to target 2000 stars), we must stop observing targets after sufficient observations have been obtained to reach the phase coverage needed to detect planets in our desired size and orbital period ranges (see Nutzman & Charbonneau 2008). The implementation of this has varied, but real-time transit detection performing at the level required to reach habitable zones for small planet sizes was not available during many of the seasons included in the release, so observations were extended to partially compensate for this. Except for cases where the data were clearly unusable (which were removed immediately), this "cycling" of the targets list occurred annually during the summer shutdown period. Targets with phase coverage deemed sufficiently good at this point were dropped, allowing new targets to take their place, although exceptions were made for objects of particular scientific interest, which therefore have much longer time-series. Many of these criteria were quite subjective. Data processing =============== For full details of our data processing procedures, please refer to the documentation on the release web page. The light curve files present aperture photometry derived using a series of concentric, circular apertures, starting from a "base aperture radius" of r = 5 pixels for the 2008-2010 data, and 4 pixels for the 2010-2011 data. Three larger apertures are also computed at factors of two in area, i.e. sqrt(2)*r, 2*r, and 2*sqrt(2)*r in radius. Sky background estimation was performed using a circular annulus between 6*r and 8*r. Light curves are generated for all four apertures, and the one showing the smallest RMS scatter was chosen. For most targets, this was the 2*r or 2*sqrt(2)*r aperture. The methodology followed in producing differential light curves is mostly standard (e.g. Honeycutt 1992), with one important exception. In order to avoid diluting the empirical inverse variance weights used to solve the individual frame zero-points, we account for the various "instrument versions" and the "meridian flips" by deriving a suite of N_S per-star magnitudes (or equivalently, one magnitude and N_S-1 offsets). These are called "segments" in the discussion in this document, and each distinct "S" value in the light curve tables (see below) was allowed its own set of per-star magnitudes for every star on the frame. These are all tied to the reference (master) frame, so in doing so the meridian offsets and other sudden flat-fielding changes at instrument modifications are removed, on assumption of a constant intrinsic stellar magnitude. It is important to be mindful that this has been done for some applications of the data (i.e. where the assumption of a constant stellar magnitude is not correct). In practice, these coefficients are always re-fit in any use of the data where this is important (e.g. the variability searches). Known systematics and techniques for their mitigation ===================================================== The dominant systematic effect that is (deliberately) not corrected in the MEarth light curves is due to "second-order" (color-dependent) atmospheric extinction, resulting from telluric water vapor absorption redward of 9000A in the bandpass of the RG715 filter (combined with CCD quantum efficiency), or variation of the filter bandpass itself (long wavelength interference cut-off) for the I-band like filter used in 2010-2011. This effect appears in the differential photometry due to the extreme spectral mismatches between the targets and comparison stars, combined with the nature of the targets (M-dwarf spectra rise steeply through the bandpass toward the red end, which is where the problems are). In order to address this systematic, the observations of all M-dwarf targets are combined to make a second, lower cadence (binned) comparison light curve, which we call the "common mode". We then use this to correct the data for the individual targets. This "double differential" photometry procedure is found to work well, but with one complication. Due to differences in the target spectral types (and in some fields, the comparison stars, e.g. at low galactic latitude) it is necessary to scale this "common mode" by a factor which varies from object to object. It has proved difficult to calculate this quantity a-priori, perhaps in part due to limited available data (e.g. photometry and spectral types) for many targets, so instead we fit for it using the target light curves to derive a multiplicative coefficient applied to the common mode before subtracting it (for the equations, see the description of the "null hypothesis model" below). This coefficient is usually well-constrained, but needing to derive it empirically by fitting the light curve brings with it the highly undesirable feature of necessitating making some kind of assumption about the intrinsic photometric behavior of the target. The present implementation of the "common mode" derivation takes a simple median of all target M-dwarfs in 0.02 JD (approx. half-hour) time bins, which appears to be adequate, although it should be noted that this procedure could potentially be improved, e.g. by accounting for the different target spectral types. We give the "common mode" itself in the original bins as a separate data file (commonmode.txt), but the light curves include the "common mode" interpolated to the dates of observation for straightforward application. We have also detected some systematics correlated with the FWHM of the images. This is not unusual for light curves derived from aperture photometry, and is thought to result from contamination of the aperture fluxes by neighboring stars in crowded fields. While it is not completely correct to do so, we address this issue by also performing a "decorrelation" against FWHM in light curves where there are strong trends. Finally, we repeat the mention of the "meridian offsets". The light curve generation allows a meridian offset for each star in each "instrument version" (these appear as new "segments" and are numbered with unique values in the "S" column, below), so this effect is already somewhat mitigated, but the process makes the assumption of no stellar variability, which is likely to be false for most targets, and in some cases the meridian offset removal can corrupt astrophysical signals. Therefore, due to the treatment made in processing the light curves, it is necessary to re-fit these "meridian offsets" when modeling the long-term stellar behavior, e.g. variability. Since the flat fields also change slightly over time, for some uses of the data it may be necessary to consider separate meridian offsets for each night, e.g. as done in Berta et al. (2012). Other "external parameters" are included in the file, and many of these are helpful in identifying systematics. We note particularly that large pointing offsets can introduce systematics caused by flat-fielding errors, and in many analyses we perform on the data (particularly automated transit searches) we filter out points with large pointing offsets (see the Delta_X and Delta_Y columns) in order to avoid false alarms. Release files ============= Each season (as defined above) has its own directory in the release. Within each of the directories, there is one light curve file per star per telescope. No attempt has been made to merge data taken on the same star with multiple telescopes. Data files are named: LSPMJhhmm+ddmm_telnn_yyyy-yyyy.txt The components of the filename (separated by _ characters) are as follows: LSPMJhhmm+ddmm is the designation of the star in the Lepine-Shara proper motion catalogue (LSPM-North; Lepine & Shara 2005) used to select the MEarth targets. Please note that the LSPM designation is always used, even when the star has other, more common, names in the literature. Note that the LSPM designations sometimes include additional letters after the abbreviated position. telnn telescope number (nn = 01 to 08). Telescopes are identified by a running number, following the layout of the telescopes in the building. yyyy-yyyy season (see above). Finding charts are named: LSPMJhhmm+ddmm_telnn_yyyy-yyyy_aperture.png these reside in a separate directory, and show the location of the photometric aperture and sky annulus (where applicable) on the MEarth master image, and the first and second epoch Palomar sky survey scans from the Digitized Sky Survey (DSS; accessed via the ESO archive). Where possible, the second epoch image used the red (R_F) plate, but there are a few cases where the blue (B_J) plate was used because the red was unavailable. These charts are intended for assessing which sources contribute light inside the photometric aperture, but can be useful for other purposes, such as assessing the quality of the master image (they are probably too small to be used as conventional finding charts, however). The charts were generated with SAOImage DS9. The astrometric registration used in this version of the charts relies on the world coordinate system (WCS) in the DSS FITS headers, which we find can sometimes be off (particularly poor quality MEarth master images can also be off). We intend to perform our own astrometric registration on the charts for future releases to address this issue. When there are other stars visible, this problem is usually quite obvious to the eye, but it is very rare for it to be large enough to significantly impact use of the charts. NOTE: the DSS is subject to copyright, and acknowledgments are requested in any publication based on DSS data. Please see: http://archive.stsci.edu/dss/acknowledging.html There are also some other "global" files included in the release: README.txt This file. commonmode.txt The "common mode" in the original 0.02 JD bins. summary.txt Summary information on the light curves (see below). Light curve file contents ========================= The light curve files are space-separated, formatted ASCII, with newline (line feed) record terminators. Each record presents one photometric measurement. This format has been used to ensure human- as well as machine-readability and maximize compatibility across different software platforms. In the files, # is used as a "comment" character. The files start with a header (all as "comments", i.e. with # at the start of the line) containing a set of name = value pairs with information about the target drawn from the tables of Lepine & Shara (2005), Lepine (2005), and Nutzman & Charbonneau (2008). Some parameters derived from the light curve are also included. See below for a full description of the quantities in the header. Two lines then give the column titles, also in a comment. The table follows these lines until the end of file, and is guaranteed not to contain any comments. The table uses space separated, formatted ASCII records, with each record terminated by a newline character (line feed). There are no blanks, and the column spacing is also consistent throughout all files in the release, so extraction of bytes within each record (as with the common CDS/Vizier ASCII catalogues), or splitting on whitespace, such as the default behavior of the awk(1) utility, will parse these records into columns correctly. The files can also be read in popular spreadsheet programs, by setting the record delimiter to space, and switching on the option to treat multiple consecutive delimiters as a single delimiter. The following columns are present in the light curve table: Column Format Unit Description BJD D14.6 days Barycentric Julian date of mid-exposure, in the TDB time-system. The full JD is given without truncation or subtraction to avoid ambiguity, but please note that as a result, it is essential to read these in double precision if they are left formatted as-is. Timing accuracy is only approx. a few seconds, due to limitations of the data acquisition software and the network time protocol implementation on the Windows computers used to perform the data acquisition. Mag F11.6 mag Differential magnitude. e_Mag F11.6 mag Uncertainty in the differential magnitude. Computed using a standard CCD noise model, which includes Poisson noise in the target photon counts, sky background noise (this is estimated empirically, so it includes readout noise), scintillation noise using the formula of Young (1967), and the uncertainty in the magnitude zero-point correction (derived from the comparison stars) that was applied to produce the differential light curve. This noise model is known to significantly underestimate the uncertainty for data taken in poor conditions, e.g. thick clouds. See below for information useful to detect these and other conditions that can degrade data quality or lead to systematic errors. tExp F5.1 sec Exposure time. DMag F7.4 mag The magnitude zero-point correction that was applied to the frame to produce the differential light curve. THIS HAS ALREADY BEEN APPLIED in the "Mag" column. It is included for the purpose of detecting frames with large light losses, e.g. due to clouds, which might need to be treated with suspicion or possibly discarded, depending on the nature of the analysis. More negative numbers correspond to less light, i.e. greater losses. Typically, we find that points with DMag < -0.5 should be treated with suspicion. FWHM F6.3 pix Full width at half maximum estimated from the stellar images on the frame. This quantity can be spurious in cases of extreme image distortion due to wind shake, or if no stars were seen on the frame. A value less than or equal to zero indicates that no estimate could be made, usually due to there being too few sources the code thinks are stellar on the frame. It is essentially impossible for the FWHM of the images to be less than 2.0 pixels, so any such value should be treated as spurious. Ellip F5.3 - Ellipticity of the stellar images on the frame. This quantity can also be spurious in cases of extreme image distortion due to wind shake, although is more reliable than the FWHM. A value of exactly zero in conjunction with FWHM <= 0 indicates that no estimate could be made (see FWHM, above). Airmass F7.5 - Airmass at mid-exposure. This quantity is computed for the position of the target star, as measured from MEarth astrometry. Delta_X F9.3 pix Displacement in X, Y coordinates, and angle Delta_Y F9.3 pix from the reference (master) frame, measured Angle F7.2 deg using the comparison stars. This is intended for detecting pointing errors, and similar issues, which might introduce systematics due to flat fielding error into the photometry. The angle can also be used to detect a "meridian flip", where it changes by approx. 180 degrees when the frame in question was taken on the opposite side of the meridian from the reference frame. Sky F8.2 ADU Local sky background level used in the aperture photometry. Peak F5.0 ADU Peak pixel counts in the object, including sky. Used for detecting saturation and/or identifying points where non-linearity might be a concern (see under "Changes to individual telescopes", above, for a table of saturation levels appropriate for each detector). S I1 - "Segment number". Identifies which points of in the light curve were solved together with one set of per-star magnitudes in order to produce the light curve. This numbers from 1 in each light curve. Points with the same value in this column are essentially those believed to share the same flat-fielding error. V I1 - "Instrument version number". This integer (numbered starting from 1, over the whole lifetime of the telescope) is incremented every time the detector was removed and replaced on the telescope, or other changes that are likely to cause offsets or systematics in the photometry were made. The software uses separate per-star magnitudes on either side of the change to account for this. R I1 - Realtime status flag. Non-zero if the data point was taken in response to a real-time trigger. Values are currently 1 or 2, corresponding to two stages of the real-time detection process: confirmation (R=1), and high-cadence followup (R=2). F I1 - Flags. The following values are used, combined with a bitwise OR operation: 2 Aperture contains known bad pixels. 4 Possible saturation detected. F != 0 means the magnitude may be corrupted, but not necessarily. Usable measurements can still be obtained (depending on the requirements) when either flag is set. CM F9.6 mag Common mode, interpolated to the time of observation. Corr_Mag F11.6 mag Differential magnitude with a simplistic model applied to correct for the common mode, "segment" zero-point offsets, and FWHM correlated systematics. This provides a very simple-minded "detrended differential magnitude" with the major known systematic effects removed. However, it assumes the star is constant, which is likely to be false for a large majority of the targets, and is intended only to get up and running quickly (e.g. for making a cleaned plot of the data). We strongly advise against using "Corr_mag" for scientific purposes. This is particularly true for studies of stellar variability, where we consider it mandatory to simultaneously model the "common mode" systematic with the variations. The following parameters are included in the header. Any parameter that is not available (null) is excluded. REMINDER: the original source publications must be cited when reproducing any data drawn from the literature (i.e. all fields below except the first five: "telescope", "season", "filter", "aperture", and "deblend"). Parameter Unit Description telescope - MEarth telescope that observed this light curve (see above). season - Season the light curve was observed in (see above). filter - Filter used. This can take two possible values: RG715 the RG715 long-pass filter I the "I_715-895" interference filter aperture pix Photometric aperture radius used to generate the light curve. deblend - Flag to indicate that the detection of the target was de-blended. Can indicate the presence of a nearby star that may contaminate the aperture photometry. Note however that it is still possible for this to occur even when the flag is not set, if the source detection software could not resolve the blend in the MEarth master image! It is also sometimes possible to obtain entirely usable (or indeed, relatively uncontaminated) photometry when the flag is set. It is recommended to examine the finding charts to assess the contamination. lspmname - Lepine-Shara Proper Motion catalogue (LSPM) identifier (Lepine & Shara 2005). The following cross-identifications and parameters are also from this catalogue, except where noted. gliese - Gliese (1969) or Gliese & Jahreiss (1979) catalogue of nearby stars number. Retrieved by cross-matching (via the 2MASS identifications) with the catalogue of updated positions for the Gliese stars by Stauffer et al. (2010). These identifiers have been manually verified wherever there was any ambiguity, and are believed to be correct, but there may be errors remaining in cases of close multiples. lhs - Luyten Half-Second catalogue number. See the revised LHS (Bakos et al. 2002). nltt - Revised New Luyten Two Tenths catalogue number (Salim & Gould 2003). hip - Hipparcos catalogue number. tycho - Tycho-2 catalogue identifier. ascc - ASCC-2.5 catalogue number. ucac - UCAC-2 catalogue number. twomass - Two-Micron All-Sky Survey all-sky data release identifier. usnob - USNO-B1.0 catalogue identifier. kic - Kepler input catalogue number. ra hr Right ascension and declination in : dec deg delimited sexagesimal (base-60) format. Positions are equinox J2000 and epoch 2000.0, see "aflag" for the source of the astrometry. pmra arcsec/yr Sky-projected absolute proper motion. See pmdec arcsec/yr "aflag" for the source of the astrometry. aflag - Astrometry source flag, from Lepine & Shara (2005). The following values are used (repeated from Lepine & Shara 2005): T Tycho-2 catalogue A ASCC-2.5 catalogue S SUPERBLINK proper motion, 2MASS position O Other source plx arcsec Trigonometric parallax and uncertainty, e_plx arcsec from the compilation by Lepine (2005). r_plx - Reference for trigonometric parallax. The following values are used (most of which are ADS bibliographic codes): 1991adc..rept.....G The Preliminary Third catalogue of nearby stars, Gliese & Jahreiss (1991). 1995GCTP..C......0V The Yale General Catalogue of Trigonometric Parallaxes; van Altena et al. (1995). 1997HIP...C......0E The Hipparcos catalogue (Perryman et al. & ESA 1997). 1998A&A...333..882D Ducourant et al. (1998). HIP-CPM Hipparcos parallax of a common proper motion companion was used. 2002AJ....124.1170D Dahn et al. (2002) or references therein. 2003AJ....125..354R Reid et al. (2003). NOTE: The trigonometric parallax compilation is known to be significantly out of date. distmod mag Adopted distance modulus. s_distmod - Source of the adopted distance modulus. The following values are used: T Trigonometric parallax. P Photometric distance estimate (see Table 3 of Lepine 2005 for references). S Spectrophotometric distance estimate (see Table 3 of Lepine 2005 for references). L Estimated by Lepine (2005) based on "vmjest". See below and especially "s_vest" for the source of the V magnitude adopted when doing this. Some of these distance moduli are highly uncertain. bjmag mag Photographic B_J magnitude from USNO-B1.0. rfmag mag Photographic R_F magnitude from USNO-B1.0 or SUPERBLINK. inmag mag Photographic I_N magnitude from USNO-B1.0. jmag mag 2MASS J-band magnitude, "combined" e_jmag mag uncertainty, and "qual" flag. Please see the q_jmag - 2MASS explanatory supplement for a detailed explanation of the flags, noting particularly the way upper limits are presented (hint: qual=U). The uncertainties, flags, and the last digit of the magnitudes were recovered using the 2MASS identifiers ("twomass", above) to look up the records in the 2MASS database, and were not present in the original LSPM-North table. hmag mag Likewise for 2MASS H-band. e_hmag mag q_hmag - kmag mag Likewise for 2MASS Ks-band. e_kmag mag q_kmag - vest mag Estimated V-magnitude from Lepine & Shara (2005). Read this reference carefully before using. s_vest - Source of the Lepine & Shara (2005) estimated V magnitude. T Tycho-2 catalogue A ASCC-2.5 catalogue U Estimated from the USNO-B1.0 photographic magnitudes (bjmag, rfmag, inmag). If s_vest=U, vest and vmjest can be highly uncertain and may have substantial systematic errors, particularly for red objects. Exercise extreme caution. vmjest mag Estimated V-J colour from Lepine & Shara (2005). Read this reference carefully before using. mass Msol Estimated stellar mass and radius used to radius Rsol compute exposure times, from the analysis of Nutzman & Charbonneau (2008) except where noted (see documentation and below). It is important to check the "s_distmod" flag (above) to assess the reliability of the distance assumed in computing these values. NOTE: two modifications were made to the methods described in the Nutzman & Charbonneau (2008) paper when generating the catalogue used for observations, and shown here. Trigonometric parallaxes were required to be measured to better than 10% in order to be used (whereas a value of 15% was used in the original analysis); and the Bayless & Orosz (2006) mass-radius polynomial was used to estimate the stellar radii from the masses. IMPORTANT NOTE: The stellar parameters are known to have (in some cases serious) flaws, and are based in many cases on out-of-date or incomplete data. Exercise extreme caution. The values in the table show what was originally adopted when deciding the observational strategy and exposure times, and are not definitive or complete. We strongly advise re-evaluating the stellar parameters rather than using these estimates. Note particularly that estimated (rather than measured) V photometry (i.e. s_vest = U), and distance, mass and radius values based on s_distmod != T can be highly uncertain. The following parameters show the "null hypothesis" model (corresponding to no variability) from our variability search software. This was also used to compute the "Corr_Mag" column. The model is as follows: Mag = null_dc[S] + null_cm * CM + null_fwhm_k * (FWHM - null_fwhm_0) where Mag, S, CM, and FWHM refer to the columns in the light curve table, and the square bracket notation refers to an array of values (indexed here by S, which numbers from 1). The parameters null_chidof, null_nfit, and null_nparm give the chi squared per degree of freedom (assuming the uncertainties in e_Mag), number of light curve points that were fit, and the number of parameters that were fit, respectively. Common mode file contents ========================= The following columns are present in the common mode file: Column Format Unit Description JD D10.2 days Full Julian date of the start of the bin in the UTC time-system. (NB: not Barycentric) CM F8.5 mag Differential common-mode magnitude. Currently this is derived simply by taking the median differential magnitude in the bin. sigma F8.5 mag Scatter in the common-mode bin using a robust MAD (median absolute deviation) estimator, scaled to Gaussian equivalent standard deviation (i.e. 1.48*MAD). nmeas I4 - Number of measurements in bin. nobj I4 - Number of independent objects in bin. tamb F6.1 C Ambient temperature. humid F6.1 % Relative humidity. press F6.1 hPa Ambient pressure. skytemp F6.1 C Sky-Ambient temperature from cloud sensor. NOTE: the last four columns are -999.0 if not available (e.g. due to failures of the various weather sensors) and are averages of the instantaneous measurements current at the start times of each exposure included in the bin. When pressure is available, the tamb, humid and press measurements are from the Vaisala WXT510/520, otherwise tamb and humid are from the cloud sensor. The cloud sensor is considered a critical device for observatory operation, so the skytemp measurement is normally available, but we note that the sensor started malfunctioning around 2010 June 21, and continued to do so until it was replaced on 2011 March 17. Data during this period show a higher baseline skytemp level, very poor sensitivity to clouds, and dropouts. Summary table contents ====================== The following columns are present in the summary table. Note that many columns are simply repeats of header information, so see also the more verbose descriptions, above. Fields are entered as NULL when not available, except "Ref", which appears as three dashes in these cases due to lack of space. There is one row per light curve. If a star was observed on multiple telescopes, there will be multiple rows for it. Column Format Unit Description LSPM_Name A15 - Lepine-Shara Proper Motion catalogue (LSPM) identifier (Lepine & Shara 2005). Gl/GJ A7 - Gliese (1969) or Gliese & Jahreiss (1979) catalogue of nearby stars number. LHS A5 - Luyten Half-Second catalogue number. rNLTT I5 - Revised New Luyten Two Tenths catalogue number (Salim & Gould 2003). HIP I6 - Hipparcos catalogue number. 2MASS A16 - 2MASS all-sky release identifier (should be prefixed with "2MASS J" to obtain full identifier). RAh I2 hr Right ascension and declination given as RAm I2 min space-separated base-60. Equinox J2000, RAs F5.3 sec epoch 2000.0. DEn A1 - DEd I2 deg DEm I2 am DEs F4.2 as PMRA F7.3 as/yr Absolute proper motions. See above for notes PMDEC F7.3 as/yr regarding the RA proper motions. Pi F6.4 as Trigonometric parallax, uncertainty, and code e_Pi F6.4 as for reference. See above for important notes Ref A3 - and caveats regarding the parallaxes. The following abbreviations are used under "Ref": CNS = 1991adc..rept.....G YPC = 1995GCTP..C......0V HIP = 1997HIP...C......0E HCP = HIP-CPM D98 = 1998A&A...333..882D D02 = 2002AJ....124.1170D R03 = 2003AJ....125..354R B_J F4.1 mag Photographic magnitudes. R_F F4.1 mag I_N F4.1 mag J F6.3 mag Infrared magnitudes from 2MASS. Upper limits H F6.3 mag are denoted by a leading < symbol, and one Ks F6.3 mag less decimal place is printed to make space. Mass F4.2 Msol Estimated stellar mass and radius. See above Rad. F4.2 Rsol and the next column for important notes and caveats regarding the stellar parameter estimates. S A1 - Source of the distance modulus estimate used to obtain the mass and radius estimates. See above for definitions of the abbreviations and important notes. Tel. A5 - MEarth telescope used to observe the light curve. Exptime A11 sec List of exposure times used in the light curve. If there were multiple exposure times, they are all given, ordered from shortest to longest and separated by commas. Nmeas I5 - Number of measurements (exposures) in the light curve. Nni I3 - Number of separate nights on which observations were obtained. Start I8 date Nights on which the first and last exposures End I8 date in the light curve were taken. Uses the common "night of" notation written yyyymmdd - the date on which the start of the night occurred, in LOCAL TIME (U.S. Mountain Standard Time, UTC-7).