The product is a 2.5 deg x 2.5 deg gridded composite of daily time-series of total (IC+CG) lightning bulk production, expressed as a flash rate density (fl/km²/day).  Separate gridded time series from the 5-yr OTD (4/95-3/00) and 8-yr LIS (1/98-12/05) missions are included, as well as a combined OTD+LIS product.  Lowpass temporal filtering (110-day for OTD, 98-day for LIS, 110-day for combined) and spatial moving average filtering (7.5 deg) have been applied, as well as best-available detection efficiency corrections and instrument cross-normalizations, as of the product generation date (9/01/06).

The product is distributed in HDF (Hierarchical Data Format).  HDF is accessible from C or Fortran using interfaces provided by NCSA (http://hdf.ncsa.uiuc.edu), or through various commercial software packages, such as IDL or Noesys (http://www.rsinc.com).  All data in this product are stored in HDF Scientific Data Sets.  Note that inline (HDF internal) GZIP compression has been applied.  Users should verify that their NCSA HDF library distribution or third-party software application versions are current enough to transparently decompress inline-compressed HDF grids.

The v2.2 gridded time series satellite lightning data were produced by the NASA LIS/OTD Science Team (Principal Investigator, Dr. Hugh J. Christian, NASA / Marshall Space Flight Center) and are available from the Global Hydrology Resource Center (http://ghrc.msfc.nasa.gov).

• NOTE: Version 2.2 release included new documentation in PDF format, which can be viewed and downloaded here:
  ftp://ghrc.nsstc.nasa.gov/pub/doc/lisotd_gridded_products/lis_otd_gridded_products_documentation_v2.2.pdf
  
  
• Approach

Bulk lightning production, expressed in fl/km²/day, is calculated as a “counting experiment”.   The calculation procedure is as follows:

1. For each day of each mission, the viewing of each sensor is recorded (in km² sec) on a 0.5 deg grid.  The number of flashes observed by each sensor, weighted by the inverse of the instantaneous instrument detection efficiency [see below] is also recorded on a 0.5 deg grid.  These daily high resolution grids are generated for each day from 5/4/95 – 2/28/0
2. “Combined” grids are generated for each day by adding the flash totals and viewtime totals for each mission.  The “combined” grids only contain joint data from 12/97-04/00.
3. A boxcar moving average of duration [110, 110, 98] days is applied to the flash and viewtime grids for the [combined, OTD, LIS] products, respectively.   This reduces aliasing of the local diurnal cycle due to precession of the instruments’ host platform orbits, which would otherwise fatally bias the time series with spurious high frequency signals.  A window of 110 days is used for the combined product since the OTD sampling dominates over LIS (4x greater).
4. The data are resampled at 2.5 deg resolution to limit product size.  Each 2.5 deg daily grid cell value is thus the average of 25 0.5 deg grid cell values over +/- [55, 55 or 49] days.
5. A further spatial moving average of +/- 1 2.5 deg grid cell (7.5 deg total) is applied to reduce sample variance
6. 2.5 deg flash rate estimates are calculated as weighted flash sums divided by total viewing time.
7. A further lowpass filter with cutoff [110, 110, 98] days is applied to each map location’s time series in order to eliminate residual high frequency noise.
8. The results of step 7 occasionally introduce negative flash rates into the time series, typically near the beginning/end of the missions, the “top” and “bottom” of orbits, and in very low lightning production regions (cold ocean gyres, deserts) or regions with extreme seasonality.  These are reset to “0” flash rate.

The final products are thus daily, 2.5 deg resolution maps of estimated flash rate density, whose estimates contain 7.5 deg spatial moving average and [110, 110 or 98] day lowpass filtered observations (approximately 3-month windows).  Due to LEO instrument undersampling, instrument and/or host platform operational dropouts, etc., higher spatial or temporal resolution can not yet be recommended to users for quantitative analysis.

The grids contain 4018 daily maps, one for each day from 1/1/95 to 12/31/05.  Note that grids prior to the start of the OTD mission and after 12/31/05 are included for simplicity, but are empty or contain data using less than a [110 or 98] day sample.  The recommended start/end dates (daily grid indices) of useful data are listed in section 10.

• Contents

The HDF file contains 7 Scientific Data Set (SDS) grids compressed internally (and transparently) using GZIP lossless compression.  These include:

LRTS_COM_FR         Combined Flash Rate [3287x144x72, float, fl/km²/dy]

LRTS_OTD_FR          OTD-only flash rate [3287x144x72, float, fl/km²/dy]

LRTS_LIS_FR            LIS-only Flash Rate [3287x144x72, float, fl/km²/dy]

LRTS_COM_VT        Combined Viewing [3287x144x72, float, km² dy]

LRTS_OTD_VT          OTD-only Viewing [3287x144x72, float, km² dy]    

LRTS_LIS_VT           LIS-only Viewing [3287x144x72, float, km² dy]

LRTS_AREA              Grid cell areas [144x72, km²]

The LRTS_xxx_FR products are the primary data products.   The LRTS_xxx_VT products are included for user reference to demonstrate the variable instrument viewing; users are encouraged to review these grids to understand the relative variability in the input data sampling.  The LRTS_AREA product is included for convenience.

Best-available calibrations for instrument detection efficiency (as of 9/1/06) have been applied in the v2.2 product, as a function of mission date, local time of day, and location relative to the South Atlantic Anomaly.  These are documented in Boccippio et al, 2002 and Christian et al, 2003, and summarized in the documentation for the High Resolution Full Climatology (LISOTD_HRFC) product, also available from the GHRC.  The HRFC product also contains the spatially and temporally variant detection efficiency grids applied to the LRTS data.

Users wishing to cite this calibration procedure may use or modify the following:

Observations in the LIS/OTD v2.2 time series gridded products have been corrected by the LIS Science Team by estimated flash detection efficiency, applied as a function of sensor, local hour, date of mission, and (for the OTD) geographic location.  For the entire dataset, these corrections correspond to average flash detection efficiencies of 47% (OTD) and 82% (LIS).  The adjustments derive from a combination of laboratory calibration, ground-validation and cross-normalization between the two instruments.  Uncertainty in these corrections is estimated as +/-10%.  The calibration procedure is described in the dataset documentation.  The gridded time series products additionally have been averaged spatially (7.5 deg, at 2.5 deg resolution) and temporally ([select 110 or 98 days as appropriate] low pass filters have been applied).

Consistency between modeled and ground-validated detection efficiency suggests that the applied corrections are know within about +/- 10%.  This is thus the minimum uncertainty in the gridded data, arising from possible bias in the correction.  A much higher source of uncertainty in the time series products is undersampling of a given grid location, even with the severe spatial and temporal averaging that has been applied.  Intercomparison of OTD and LIS time series during the missions’ overlap window readily illustrates this undersampling effect.  It is not yet known if the total uncertainty in the time series products will preclude quantitative use of the data.   However, qualitative interannual variability has been observed [e.g., Goodman et al, 2000] which appears consistent with known ENSO effects on local meteorology.

All OTD and LIS orbits undergo both automated and manual quality assurance.  For the preliminary reanalysis, the most stringent orbit rejection criterion was applied: any orbit which was assigned a manual Q/A “warning” flag has been rejected from the reanalysis. 

Each OTD flash is further assigned an automated quality index (the ‘Thunderstorm Area Count’ or ‘Density Index’), indicating its likelihood of being lightning rather than optical or radiation noise.  For this reanalysis, only flashes with values of the metric >= 140 have been included.  This is the same cutoff value used in all validation and science analysis published by the LIS science team to date.  This filter removes most radiation noise from the data; a slight residual “ring” artefact of very low spurious flash rates remains at the periphery of the South Atlantic Anomaly (southeast Pacific and southeast Atlantic).