Assimilation of Pacific Lightning Data into a Mesoscale NWP Model Antti Pessi, Steven Businger, and Tiziana Cherubini University of Hawaii K. Cummins, N. Demetriades, and T. Turner Vaisala Thunderstorm Group Inc. Tucson, AZ Outline Long-range lightning detection PacNet Relating lightning rates to rainfall rates Assimilating lightning data into MM5 Case studies of impact of lightning data Why lightning data? Very little conventional data (soundings, SYNOPs) over the Pacific Geostationary satellites: IR images - difficult to distinguish Radiosonde sites between areas of active convection and anvil cloud. Low-orbiting satellites: Passive and active sensors provide high resolution data (AMSR-E, TMI, SSM/I) but only ~twice daily coverage at lower latitudes. Ground based radars: limited range (~300 km) Lightning data: continuous, available in ~ real time, long-range (~3-5000 km) Long-range Lightning Detection Ionosphere-earth wave guide allows VLF (5-25 kHz) emissions (sferics) to propagate thousands of km Best propagation during night and over ocean LF/VLF Lightning Waveforms at Various Distances Vertical electric field waveforms for cloud-to-ground return strokes at three different distances. Note the increased complexity and lower frequency content of the waveforms at longer distances. The amplitude scale is not calibrated. The time scale is in microseconds post digitizer trigger. Pacific Lightning Detection Network (PacNet) Motivation and goals PacNet is a network of long-range lightning detectors in the Pacific Continuous, real-time monitoring of convective storms over the Pacific Investigate the impact of data assimilation of lightning derived products on forecast accuracy in regional NWP modeling (MM5, WRF). In particular forecast improvements for: Midlatitude, subtropical and tropical cyclone intensity and track Rainfall patterns and intensity Flash flood events Antti Pessi and installation at Dutch Harbor, AK PacNet Sensor Sites IMPACT ESP Sensor in Lihue, Kauai Currently 4 sensors installed at Dutch Harbor, Lihue, Kona and Kwajalein. Sensors in North-America and Japan contributing 5 Days of Pacific Lightning Activity PacNet Performance Projections Detection Efficiency (DE) with NALDN Day Time Night Time Diurnal Variation of PacNet Lightning Average number of lightning strokes observed at each hour over the N. Pacific (45 day average). 10 UTC is midnight HST. Lightning - Rainfall Ratio Warm Season Normalized Rain-Yields The lightning rainfall relationship may vary significantly, depending on air-mass characteristics and cloud microphysics Over a particular climatic regime and a limited geographic region, lightning is well correlated to convective rainfall (Zipser 1994). Specify typical lightning-rainfall ratio for various storm systems over the Pacific: extraropical cyclones, kona lows, TUTTs, tropical cyclones... Warm season normalized CG flash density vs rainfall Sloping black lines are contours of constant rain yield (kg/fl) (Petersen and Rutledge 1998) Methodology to determine lightning vs rainfall ratio Domain divided into 0.5˚ x 0.5˚ grid Lightning rates from LongRange Network Rainfall rate from AMSR-E and TMI sensors Lightning strokes occurring within ±15 min of satellite overpass time are counted Lightning count and average rainfall are computed over each square Extratropical storm in the northeast Pacific December 2002 Satellite rainfall measurements Aqua’s AMSR/E- Advanced Microwave Scanning Radiometer-EOS Orbiting at 705 km, 70 degrees inclination Cloud properties; precipitation (total, convective); radiative energy flux; land surface wetness; sea ice; snow cover; SST; SS wind fields 12 channels at six discrete frequencies in the range of 6.9 to 89 GHz Goddard Profiling Algorithm (GPROF) is used to calculate rainfall rates using brightness temperatures Convective part of total rainfall: measures of the local horizontal gradient of brightness temperatures polarization of 85.5 GHz scattering signatures TRMM’s TMI: lower resolution, inclination 40˚. Aqua with its AMSR-E on top left Cumulative Probability Distribution (Satellite Rainfall) 1 0.9 Cumulative Probability Matching Technique 0.8 0.7 Cum. prob. 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 Convective Rainfall (mm/h) Cumulative Probability Distribution (Lightning Strokes) 1 0.9 0.8 0.7 Cum. prob. 0.6 0.5 0.4 0.3 0.2 0.1 0 0 5 10 15 20 Strokes 25 30 35 40 Take cumulative probability for rainfall every 0.2 mm The corresponding number of flashes can be found taking the same probability for lightning strokes 6 Lightning - Convective Rainfall Composite analysis of 15 storms in the central Pacific. Blue line is fitted function R 2.2L0.52 where R is rainfall rate and L lightning rate. Rainfall Assimilation into MM5 o o o Alexander et al. (1999) found relatively good correlation between convective rainfall and lightning rates during the 1993 Superstorm. They used a normalized parabolic heating profile, with heating max at ~500mb, to vertically distribute the total latent heating from the observed rain through the temperature tendency equation Resulted in improved numerical forecasts by assimilating latent heating rates derived from lightning and satellite data. (rainfall from SSM/I, lightning from NLDN and VLF networks). p * T L p * g liq R 0 Nh t Cp R0 convective rainfall rate N normalized parabolic heating function h T model predicted temp. from latent heating p*=p sfc-ptop MM5 Model Description PSU/NCAR Mesoscale Model (MM5) Limited-area, nonhydrostatic, terrain-following, sigma-coordinate model 27 km grid spacing, 39 vertical levels Kain-Fritch convective parameterization FDDA QuickTime™ and a TIFF (LZW) decompressor are needed to see this picture. Four Dimensional Data Assimilation (FDDA) Lightning observations are mapped to vertical moisture profiles (e.g. Papadopoulos et al. 2004) Vertical moisture profiles are assimilated using MM5 FDDA Newtonian nudging (or relaxation) nudges the model state toward the observations by adding artificial tendency terms to prognostic equations based on the difference between modeland observed states. The following were defined: - Obs nudging radius of influence in horizontal and vertical - Time window of influence - Nudging coefficient G FDDA N p q * F(q, x j ,t) Gq p t * [W i (x j ,t) i (qo qˆ ) i ] 2 i1 N W i(x j ,t ) i1 • • • • • • F : model's physical forcing terms G : nudging coefficient (relative magnitude of term) W : 4-D weighting function (used to determine G) : observational quality factor (0-1) qo : locally observed mixing ratio qˆ : model mixing ratio interpolated to observation location The model equations are written in "flux" form, where the prognostic variables for horizontal wind, temperature, and mixing ratio are mass weighted by p*. p* = ps - pt where ps is surface pressure and pt is constant pressure at the top of the model Experiment Design Initial conditions from GFS-model Boundary conditions every 6 hours Initialization 00Z or 12Z Model integration 60 h Assimilation 8 h Horizontal radius of influence R=54 km, vertical 0.001 sigma Time window of influence ±15 min Model timestep 81 sec, nudging every second timestep Nudging only if observed value is higher than model computed value R Obs=lightning gridpoint Construction of Moisture Profiles Seven vertical moisture profiles typical for a range of rainfall rates constructed using MM5 data: - Go through each grid point over the storm - Bin rainfall and corresponding moisture profile into one of 7 categories - Make a composite of all gridpoint values which results in 7 rainfall and moisture profile categories Compute lightning rates over 0.25º x 0.25º squares and 30 min time window during the whole assimilation period Moisture profiles Mixing Ratio (kg/kg) 0 0.002 0.004 0.006 0.008 0 5 10 Sigma Level 15 20 25 30 35 40 No rain 1-3 mm/h 3-6 mm/h >6 mm/h Convert of Lightning Rate to Moisture Profile Use lightning-rainfall relationship to relate lightning rate with moisture profile. The relationship has been derived by comparing lightning rates with rainfall rates from TRMM and Aqua Lightning rate => rainfall rate => moisture profile Moisture profiles Mixing Ratio (kg/kg) 0 0.002 0.004 0.006 0.008 0 5 10 Sigma Level 15 20 25 30 35 40 No rain 1-3 mm/h 3-6 mm/h >6 mm/h Model Nudging Nudge MM5 initial and computed moisture profiles Nudging only if observed value is higher than model computed value Model area 5 strokes btw 0:15 and 0:45=> obs. time 0:30 2 strokes btw 0:30 and 1:00=> obs. time 0:45 Model integration 3 strokes btw 0:00 and 0:30=> obs. time 0:15 Case Studies Impact of PacNet Lightning Data Timing of Squall Line Over Hawaii * * * Lightning strikes between 5-7 UTC on 28 Feb. 2004. Six-hour MM5 control forecast for rainband position was off by ~150 km at 06UTC, 28 February 2004. Timing of Squall Line Over Hawaii L1000 Six-hour MM5 FDDA forecast improved surface pressure and wind forecasts for 06UTC, 28 February 2004. North-East Pacific Low 19 December 2002 983 972 Observed Sea-level Pressure (left) and ETA 24-hr SLP and rainfall forecasts valid at 12 UTC 19 December 2002 (middle), show a 11mb forecast error in storm central pressure (12 hr forecast shows 9mb error). Lightning observations 09-12Z 12/19/2002 Reducing Forecast Error over the Eastern Pacific Assimilation of lightning data results in a significantly improved forecast of storm central pressure. 972 L 972 L 983 North-East Pacific Low 19 Dec. 2002 Lightning Strokes 06-09Z (last assim. time 08Z) Central sea-level pressure of simulated storm with lightning nudging is 10 mb deeper than control run. North-East Pacific Low 19 Dec. 2002 Discussion and future work MM5 FDDA was used to assimilate vertical moisture profiles derived from lightning observations Assimilating moisture profiles resulted in correct surface pressure vs. 11mb error in CTRL run for Dec. 2002 storm. Assimilating moisture profiles also improved simulation of squall line over Hawaii. Results are sensitive to nudging coefficient and to moisture profiles Increasing G (nudging coefficient) has the same effect as increasing moisture values but is numerically unstable if G≥1/∆t. This experiment proved to be unstable even at larger G (G=0.01) FUTURE WORK Uncertainties remain in lightning-rainfall-moisture relationship Refine methods for finding and assimilating more realistic moisture profiles Investigate latent heating profiles using MM5 3/4D-Var Method can be made operational relatively easily by allowing 8 hours of assimilation in the beginning of the model run Acknowledgements The authors would like to thank ONR and NASA for support of PacNet. Questions? Moving Toward Operational Assimilation MM5 forecast initialized at 00Z gets its initial conditions from ETA-model at 03-04Z LAPS is run to initialize the model and integration is started ~08Z Lightning data has only ~1/2 hour delay => it can be assimilated 00Z - 07Z operationally MM5 Control/FDDA Squall Line over Hawaii 28 Feb. 2004 MM5 Control/FDDA Squall Line over Hawaii 28 Feb. 2004 MM5 Control/FDDA NE Pacific Low 19 Dec. 2002