1 New Developments in GOES-12 and GOES-R Advanced Baseline Imager Convective Initiation Detection Wayne F. Feltz*, Kristopher Bedka^, Lee Cronce*, and Jordan Gerth* John Mecikalski # and Wayne Mackenzie # *Cooperative Institute for Meteorological Satellite Studies (CIMSS) ^ Science Systems and Applications, Inc, Hampton, VA # University of Alabama - Huntsville UW-Madison

2 Overview Satellite CI methodology Satellite CI methodology History of transition from research to operations History of transition from research to operations Examples of near real-time CI nowcast in AWIPS and N-AWIPS Examples of near real-time CI nowcast in AWIPS and N-AWIPS Researcher End-user (NWS, FAA, SPC) iterative feedback Researcher End-user (NWS, FAA, SPC) iterative feedback

3 CI Definition 3 Roberts and Rutledge (2003) and Mecikalski and Bedka (2006) define Convective Initiation as the “first occurrence of a 35 dBZ radar reflectivity.”

-40° C -25° C -10° C +5° C Time X-Z Plane View Satellite View Time

Satellite Observations of Convective Initiation Rapid IR window cloud-top cooling can precede significant rainfall (> 35 dBZ) by ~30-60 mins, imager-based (not sounder) Rapid IR window cloud-top cooling can precede significant rainfall (> 35 dBZ) by ~30-60 mins, imager-based (not sounder) Roberts and Rutledge 2003 (WAF) Roberts and Rutledge 2003 (WAF) The first cloud-top ground strikes were most often observed at IR window BTs of ~-40° C in a study over S. Africa The first cloud-top ground strikes were most often observed at IR window BTs of ~-40° C in a study over S. Africa

GOES-12 IR WindowCloud Typing (Pavolonis/Heidinger – NOAA)

UWCI: Algorithm logic – an overview 1. Computing Box Averaged Brightness Temperature Example using the previous time (1545 UTC)

UWCI: Algorithm logic – an overview 1. Computing Box Averaged Brightness Temperature Algorithm loops through every pixel within an image For each pixel, all pixels within a 7x7 pixel box centered upon the pixel of interest are investigated For pixels within the 7x7 pixel box that are classified by the GOES Cloud Typing product as water, supercooled water, mixed phase, thick ice or cirrus; their 11 micron BTs are summed and then averaged to determine the box-averaged 11 micron BT. The box averaged BT is assigned to the center pixel of the 7x7 pixel box

UWCI Box-Average CI Nowcast Product Description Current and future imagers are operating in 5-min rapid scan, so cumulus do not move very far between images -For 50 CI cases over CONUS, average cloud movement=5 km/5 mins, 1 SD=2 km/5 mins One can disregard cloud motions by time-differencing “box-averaged” cloud top properties to determine CI Compute mean IRW BT for cloud categories identified by day/night cloud type product over 7x7 and 14x14 pixel SEVIRI IR pixel boxes Employ rules to eliminate situations where cirrus anvil moves into box with developing cumulus Use 13 km NWP stability analysis to minimize CI nowcast false alarm from cooling of non-convective cloud Compute cloud-top cooling rate, minimum threshold at -4 K/15 mins Use cloud typing information with cooling rates to further minimize false alarms and develop confidence indicators for convective initiation -Filter out isolated noisy nowcast pixels -Cat 1: Cooling liquid water clouds -Cat 2: Cooling supercooled/mixed phase -Cat 3: Cooling with recent transition to thick ice cloud Though the method is optimized for 5-minute imagery, the examples and validation to the left show results when applied to 15-minute SEVIRI imagery M. Pavolonis (NOAA/NESDIS) Day/Night Cloud Microphysical Typing 15-Min Cloud Top Cooling CI Nowcast Using 15-min Data 45-min Accumulated Cooling 45-min Accumulated CI Nowcast Channel 9 BT at End of 45-min Period Channel 9 BT 1 hour Later SEVIRI Channel 9 IR Window BT Lightning Initiation POD (133 LI cases): 76% Lightning Nowcast FAR (10214 pixels): 29% Lead time decreases with cloud glaciation Product Validation

UWCI Algorithm Description Day/Night UW Cloud typing product (Pavolonis et al, uses 3.9, 6.7, 10.7, and 12.0/13.3 um channels) Day/Night UW Cloud typing product (Pavolonis et al, uses 3.9, 6.7, 10.7, and 12.0/13.3 um channels) Monitor microphysical properties Monitor microphysical properties Infrared window (10.7 um) box-averaging conducted (monitoring mean 10.7 um cooling rate over area) Infrared window (10.7 um) box-averaging conducted (monitoring mean 10.7 um cooling rate over area) This algorithm for convective initiation phase only This algorithm for convective initiation phase only Two primary algorithm products are cloud top cooling (CTC) rate and CI nowcast Two primary algorithm products are cloud top cooling (CTC) rate and CI nowcast

UWCI Data Flow Overview UWCI Algorithm (Fortran 90) – Processing time 1-2 minutes CIMSS McIDAS ADDE Server NSSL N-AWIPS Image -> SPC Regridded Overlays WDSS-II and Google Earth Over 2000 GOES images processed and delivered from 27 April – 1 June 2009

UWCI/ACCUMCTC - 60 minute accumulated cloud-top cooling rate UWCI/ACCUMCTC - 60 minute accumulated cloud-top cooling rate UWCI/ACCUMCI - 60 minute accumulated CI nowcast UWCI/ACCUMCI - 60 minute accumulated CI nowcast UWCI/INSTCTC – Most recent single image CTC UWCI/INSTCTC – Most recent single image CTC UWCI/INSTCI - Most recent single image CI nowcast UWCI/INSTCI - Most recent single image CI nowcast For the INSTCI field, there will be values ranging from 0 to 3: For the INSTCI field, there will be values ranging from 0 to 3: Value 0: No CI nowcastValue Value 0: No CI nowcastValue 1: "Pre-CI Cloud Growth" associated with growing liquid water cloudValue 1: "Pre-CI Cloud Growth" associated with growing liquid water cloudValue 2: "CI Likely" associated with growing supercooled water or mixed phase cloudValue 2: "CI Likely" associated with growing supercooled water or mixed phase cloudValue 3: "CI Occurring" associated with cloud that has recently transitioned to a thick ice cloud top 3: "CI Occurring" associated with cloud that has recently transitioned to a thick ice cloud top

UTC Instantaneous CI Nowcast Possible CI Possible CI Occurring CI Likely Dryline CI Case SPC HWT Proving Ground

KAMA 2103 UTC Base Reflectivity KAMA 2018 UTC Base Reflectivity CI nowcast at 2015 UTC CI likely CI occurring Non-detection due to cirrus KAMA 2035 UTC Base Reflectivity First echo >= 35 dBZ, 17 minutes after nowcast 29 April 2009

SD/NE Border Warm Frontal 0545 UTC 0546 UTC 0615 UTC 0606 UTC UTC Nocturnal Instantaneous CI Nowcast

KFSD 0616 UTC Base Reflectivity 6 May 2009 CI nowcast at 0545 UTC KFSD 0546 UTC Base Reflectivity Nothing on radar at this time KFSD 0601 UTC Base Reflectivity First echo >= 35 dBZ, 16 minutes after nowcast

University of Wisconsin Convective Initiation (UWCI) High-level algorithm overview Compute IR-window brightness temperature cloud top cooling rates for growing convective clouds using a box- average approach Combine cloud-top cooling information with cloud-top microphysical (phase/cloud type) transitions for convective initiation nowcasts Example from June 17, 2009 over northern KS First UWCI cooling rate signal precedes NEXRAD 35 dBz signal by 37 minutes 1545 UTC – first cloud top cooling signal 1610 UTC - Continued cooling signal 1732UTC - Severe t-storm First NEXRAD 35+ dBz echo at 1622 UTC NEXRAD at 1735 UTC

Hastings, NE NEXRAD Radar Reflectivity from 06/17/2009 First significant cooling (<-4K/15min) at 1545 UTC No echo on radarReflectivity echo >35 dBZ First signs of convection on radar 33 min after significant cooling detected Resulting strong convection ~3.5 hrs after significant cooling detection Radar Reflectivity at 1544 UTC Radar Reflectivity at 1618 UTC Radar Reflectivity at 1826 UTC

Eastern CO No radar reflectivity at 1745Z CTC indicated at 1745Z Radar reflectivity at 1908Z Visible image at 1745Z

Nebraska : Nocturnal Case IR4 image and CTC at 0245Z Radar reflectivity at 0246Z IR4 image and CTC at 0315Z Radar reflectivity at 0311Z

24 AWIPS CI/CTC Interaction with Sullivan (MKE) NWS Office 04:30 UTC 06:30 UTC 05:02 UTC Forecaster generated screen captures from AWIPS at MKE CI likely CI occurring Lightning

25 "The UWCI performed very well in Iowa last night! These thunderstorms fired up along an existing boundary and are coincident with the leading edge of 700mb moisture transport and weak 850mb warm air advection.” - Marcia Cronce NWS Forecaster 25 AWIPS CI/CTC Interaction with Sullivan (MKE) NWS Office

UWCI – Ongoing Algorithm Improvement Algorithm weaknesses were noted during the Spring 2009 Storm Prediction Center Hazardous Weather Testbed Field Experiment The weaknesses are currently being addressed by UW/CIMSS scientists Areas of improvement Decrease false alarms associated with thin cirrus moving across scenes Decrease false alarms associated with rapid anvil expansion Increase POD and/or POD lead-time by updating to latest GOES-R AWG Cloud Mask and Cloud Type algorithms (at the same time monitor POD and FAR to ensure latest version of cloud algorithms do not adversely impact algorithm) Two examples of improvements Lower FAR from thin cirrus movement across a scene Increase POD lead-time for storms using latest cloud mask/cloud typing algorithms

Expanding Cloud Edges Thin Cirrus CI: False Cooling Rates Due to Anvil Expansion and Overriding Cirrus

UWCI – Decreased FAR with thin cirrus False alarms over the Central Plains at 1915 UTC May 20, 2009 associated with thin cirrus

UWCI – Decreased FAR with thin cirrus False alarms eliminated with improved UWCI algorithm over the Central Plains at 1915 UTC May 20, 2009 associated with thin cirrus

UWCI – Increased Signal Lead-Time Missed CTC signal at 1945 UTC 01 May 2009 over north central Texas; not captured until 2002 UTC Also note false alarms over SE NM/central TX

UWCI – Increased Signal Lead-Time CTC signal at 1945 UTC 01 May 2009 over north central Texas is now captured; 15 minutes sooner than old version of UWCI algorithm Better lead-time due to better cloud detection with latest version of GOES-R AWG Cloud Mask (and Cloud Typing algorithm); able to calculate a 1945 UTC – 1932 UTC cloud-top cooling rate because cloud was detected at 1932 UTC with new version of cloud mask; not the case with previous version of cloud mask False alarms over SE NM/central TX have been eliminated

32 UWCI Algorithm (Very new methodology, still learning) Advantages Advantages – Computationally fast – Day/night independent – The code is flexible and operates on both rapid-scan (5-min) and operational (15 to 30-min) scanning patterns – Heritage – Algorithm already being used/evaluated by operational centers (SPC, SAB, local NWSFOs) using current GOES imager – Product offers up to minute lead-time before significant radar echoes/cloud to ground lightning is present Disadvantages Disadvantages – Algorithm does not operate in mostly cloudy/cloudy scenes or in areas dominated by ice clouds (including thin cirrus debris cloud) – Algorithm performs more as a diagnostic than prognostic tool in areas of air mass driven convection (e.g.-southeastern US) – Thin cirrus moving across scene may result in false alarms

Exploring the Use of Object Tracking for CI Nowcasting UW-CIMSS and the University of Alabama in Huntsville (UAH) are working toward development of object- based methods for CI nowcasting in the GOES-R ABI era, using current GOES-12 and MSG SEVIRI as proxies for GOES-R UW-CIMSS is experimenting with the Warning Decision Support System-Integrated Information (WDSS-II, Lakshmanan et al. (J. Tech., 2009), (WAF, 2007)) to compute cloud-top cooling rates, which can be used with cloud-top microphysical trends to produce CI nowcasts Radar reflectivity can be remapped to the satellite resolution/projection and carried along with the satellite- derived objects for reliable product validation. This capability is not available with current pixel-based CI nowcast methods which causes significant difficulty in evaluating current product accuracy over large scenes and numerous cases MSG SEVIRI Example GOES-12 Example

Why do we need GOES-R ABI for convective initiation nowcasts? Temporal resolution of imager needs to be 5 minutes or less operationally to match convective initiation time scale! (30 second tendency information can and will also be used) Temporal resolution of imager needs to be 5 minutes or less operationally to match convective initiation time scale! (30 second tendency information can and will also be used) Improvement in spatial resolution of infrared imagery from 4 to 2 km (this a factor of 4 improvement!) will improve cloud detection and box averaging cooling rate signal Improvement in spatial resolution of infrared imagery from 4 to 2 km (this a factor of 4 improvement!) will improve cloud detection and box averaging cooling rate signal Additional spectral channels will improve microphysical cloud typing and sensitivity of thin cirrus detection Additional spectral channels will improve microphysical cloud typing and sensitivity of thin cirrus detection

University of Wisconsin Convective Initiation (UWCI) Algorithm logic – an overview Justin Sieglaff UW/CIMSS/SSEC September 2009

UWCI: Algorithm logic – an overview 1.Box average 11 micron brightness temperature (BT) is calculated for current time and previous time, using specific classes from GOES Cloud Typing product 2.Unfiltered Cloud Top Cooling (CTC) Rate is calculated by differencing box average 11 micron BT for current time from previous time 3.Large/small box approach eliminates most of false CTC due to cloud motion (and additional checks reduce false cooling further) 4.CI Nowcast is assigned to remaining CTC pixels with cooling rates less than or equal to -4.0K/15 minutes by leveraging GOES Cloud Typing product

UWCI: Algorithm logic – an overview 1. Computing Box Averaged Brightness Temperature

UWCI: Algorithm logic – an overview 1. Computing Box Averaged Brightness Temperature Algorithm loops through every pixel within an image For each pixel, all pixels within a 7x7 pixel box centered upon the pixel of interest are investigated For pixels within the 7x7 pixel box that are classified by the GOES Cloud Typing product as water, supercooled water, mixed phase, thick ice or cirrus; their 11 micron BTs are summed and then averaged to determine the box-averaged 11 micron BT. The box averaged BT is assigned to the center pixel of the 7x7 pixel box

UWCI: Algorithm logic – an overview 1. Computing Box Averaged Brightness Temperature Result of previously described box average BT calculation for all pixels Black areas are where less than 5% of pixels within 7x7 pixel box met cloud type requirements and are assigned to missing

UWCI: Algorithm logic – an overview 1. Computing Box Averaged Brightness Temperature Example using the previous time (1545 UTC)

UWCI: Algorithm logic – an overview 2. Computing Unfiltered Cloud Top Cooling (CTC) Rate To compute unfiltered CTC: unfiltered_ctc = box_average_bt_current – box_average_bt_previous In this example: unfiltered_ctc_1602utc = box_average_bt_1602utc – box_average_bt_1545utc

UWCI: Algorithm logic – an overview 2. Computing Unfiltered Cloud Top Cooling (CTC) Rate Resultant unfiltered box averaged cloud top cooling rate Must have valid box averaged BT for current time and previous time to have valid cooling rate for a given pixel In the unfiltered box averaged CTC, there is false cooling due to cloud motion and true cooling due to vertical cloud growth False Cooling Due To Cloud Motion Accurate Cooling

UWCI: Algorithm logic – an overview 3. Large/small box approach eliminates false CTC To eliminate false cooling due to cloud motion the following double box system is employed For each pixel, the box averaged BT at the current time (plus symbol) is compared to the minimum box averaged BT from the previous time along the perimeter of the large box (13x13 pixels) For each pixel, the box averaged BT at the current time (plus symbol) must be 2.0K (or more) cooler than the minimum box averaged BT from the previous time from the perimeter of the large box This large/small box approach eliminates false cooling due to cloud motion and ensures only pixels with cooling clouds are retained

UWCI: Algorithm logic – an overview 3. Large/small box approach eliminates false CTC Unfiltered cloud-top cooling rate (left) and resultant filtered cloud-top cooling rate (right) The false cooling associated with cloud motion along the top of the image has been eliminated, mainly by large/small box check but possibly by other false cooling elimination tests One effect of large/small box check results in smaller area of legitimate cooling rate without adversely impacting product

UWCI: Algorithm logic – an overview 4. Determine CI Nowcast Class After all cloud top cooling (CTC) false cooling checks are complete, CI Nowcast categories are assigned to remaining CTC pixels less than or equal to -4.0K/15 minutes (aside from isolated CTC pixels) Percentage of cloud types within the small box is computed for each pixel If water cloud percentage > 10% and supercooled, mixed phase, and thick ice < 5% then class 1 is assigned or ‘Pre-CI Cloud Growth’ If supercooled or mixed phase percentage > 5% and thick ice < 5% then class 2 is assigned or ‘CI Likely’ If thick ice percentage > 5% and thick ice percentage from previous time < 5% then class 3 is assigned or ‘CI Occurring’

UWCI: Algorithm logic – an overview 4. Determine CI Nowcast Class Resultant CI Nowcast category assignment ‘CI Likely’ class is assigned to portion of the cloud containing sufficient supercooled cloud type (green) ‘Pre-CI Cloud Growth’ class is assigned to remainder of cloud with only warm water type

UWCI: Algorithm logic – an overview Progression of above steps on entire domain