GOESR3 Periodic Reporting

advertisement

GOESR3 Periodic Reporting

Project Team: Kristopher Bedka (NASA LaRC), Chris Velden (CIMSS), Sarah Griffin (CIMSS), and

Konstantin Khlopenkov (SSAI at NASA LaRC)

Reporting Period: July 2013 - June 2014

Team Leads: Kristopher Bedka and Chris Velden

Team Members: Sarah Monette and Konstantin Khlopenkov

Project Title : Research and Development of Satellite-Detected Overshooting Top (OT) Products

Project Number: 101

Executive Summary

Within GOES-R Proving Ground evaluations, it has been demonstrated that Overshooting Tops (OTs) within vigorous convection can be accurately detected within geostationary and polar-orbiting satellite data using the

GOES-R Future Capability OT detection algorithm (Bedka et al. 2010). The OT detection algorithm provides a yes/no binary mask based upon fixed criteria derived from analysis of a large sample of OT-producing storms in

LEO and GEO satellite imagery. Selected analysis and prediction centers have been provided various "flavors" of an objective OT detection algorithm utilizing different criteria to meet their respective forecast needs, which can cause confusion within the user community. It had become clear that a binary product utilizing fixed criteria and a relatively simple pattern recognition scheme is not an optimal long-term solution for GOES-R.

The Geostationary Operational Environmental Satellite (GOES)-14 Imager operated in a 1-min “super-rapid scan” observations for GOES-R (SRSOR) mode during summer and fall 2012 and 2013 to emulate the high temporal resolution sampling of the GOES-R Advanced Baseline Imager (ABI). The current GOES operational scan interval is 15 to 30 mins which is too coarse to capture details important for severe convective storm forecasting including 1) when severe weather signatures such as OTs first appear, 2) how the satellite-observed cloud tops evolve over time, and 3) how well were OTs detected using the GOES-R algorithm. The SRSOR periods offer the most comprehensive set of high-temporal resolution imagery of convective storms ever collected, providing a tremendous opportunity to address these uncertainties.

This GOES-R Risk Reduction project seeks to refine the GOES-R overshooting cloud top (OT) detection algorithm based on user/reviewer feedback, develop new ideas to optimize product applications, and assess future product performance using GOES-R ABI proxy data. Primary objectives for FY13 included: 1) adaptation of the current yes/no binary OT detection mask into a probabilistic product augmented by an improved pattern recognition approach and 2) examination of OT behavior and detection characteristics and other severe weather indicators within GOES-14 1-minute Super Rapid Scan Operation for GOES-R (SRSOR) data.

A derivative of the OT algorithm for specific applications in tropical cyclone analysis, the Tropical

Overshooting Top (TOT) product, will be investigated in Year2 of this project.

FY13 Milestones

1) Identify OT events over diverse geographic regions, seasons, convective regimes, and land/ocean surfaces that will serve as the test database for a probabilistic OT detection product

2) Develop and perform preliminary validation of a probabilistic OT detection product

3) Identify and analyze additional long-lived, OT-producing convective cells within GOES-14 SRSO imagery in a similar manner as the single event analyzed for a separate GOES-R Risk Reduction project

“Integrated GOES-R GLM/ABI approaches for the detection and forecasting of convectively induced turbulence” (PI: L. Carey, UAH). Identify other signatures associated with severe weather.

04/10/2020

GOES-R3 Status Report Template NESDIS STAR GOES-R

Accomplishments & Plans

Accomplishments from September 2003 – June 2014

MODIS Overshooting Top Pixel Database and Probabilistic OT Detection Model Development

Development of a next-generation probabilistic OT detection algorithm involved a two prong approach, 1) identification of a statistically significant sample size of OT and non-OT regions so a statistical analysis can discriminate between the two classes of pixels to provide an OT probability field and 2) improvement of the pattern recognition scheme to identify candidate OT regions without the use of fixed IR brightness temperature thresholds.

OT regions and non-OT anvil clouds were manually identified using 100 granules of daytime MODIS 0.25 km visible imagery. OT pixels were required to have the characteristic distinct cauliflower-like texture commonly present in OTs in visible imagery. Non-OT anvil cloud pixels were randomly selected throughout the granule.

The non-OT pixel was required to be in an optically thick region of the anvil that was relatively smooth in texture and did not appear anything like an OT. The latitude/longitude coordinate of the OT and non-OT pixels were recorded. Typically only one pixel was recorded per individual storm anvil, but if a storm was particularly large as is the case with Mesoscale Convective Systems or tropical cyclones, more than one pixel could have been selected. Attempts were made to sample nearly all anvils present within each MODIS granule, regardless of the intensity of the storm. This yielded as many as 49 non-OT pixels from a single granule. The coldest IR brightness temperature (BT) within 3 pixels of the OT or non-OT region was recorded in addition to the mean BT of the anvil at a ~10 km radius from pixel.

Figure 1 shows the locations of the 2155 OT identifications across the globe. An additional 985 non-OT pixels were also identified as of the current time, with an additional ~1000 planned for the final database. Efforts were made to include a diversity of convective storm types, size, and intensity, geographic regions, and seasons in the database. For example, MODIS observations of storms that produced 4+ inch hail and several EF3+ intensity tornadoes were included in addition to over 10 tropical cyclones of various strength (including Hurricane Katrina) throughout the world. In contrast, OT-producing storms over Mongolia and Alaska were also included. Events over the U.S. were emphasized because an FY15 milestone will be to determine a probability of severe weather associated with each OT detection and we only have severe weather reports over the U.S. 75% of the OT and non-OT pixel database was used to train the logistic regression and the remaining 25% was used for validation.

For each OT and non-OT identification, a set of parameters were extracted from NASA Modern Era

Retrospective Analysis for Research and Applications (MERRA) fields. MERRA provides six-hourly vertical thermodynamic profiles with 42 vertical levels and hourly surface analyses, both at a 0.5° x 0.66° spatial resolution. The MERRA data were time interpolated to the MODIS image time. The MERRA-based fields were combined with the MODIS IR BT-based data for 75% of the OT and non-OT pixels and were input to a logistic regression analysis (See the following for a basic description:

(http://luna.cas.usf.edu/~mbrannic/files/regression/Logistic.html). The logistic regression analysis determines the statistical significance of each parameter toward objectively discriminating between OT and non-OT regions. An intercept term and coefficients for each statistically significant predictor are applied to the various MODIS and

MERRA-based parameters, yielding a probability that ranges from 0 to 1 with 1 meaning that the pixel is certainly an OT. Another output is the statistical significance of each parameter. Only parameters that had > 95% significance were included in the final probabilistic model.

Parameters initially evaluated in the logistic regression analysis included: 1-2) mixed-layer and most unstable

CAPE values, 3-4) the difference between the MODIS OT/non-OT IR BT and both the mixed-layer and most unstable Equilibrium Level temperatures, 5) the difference between the MODIS OT/non-OT IR BT and the

MERRA tropopause temperature, 6) the minimum MODIS IR BT value within the OT or non-OT region, 7) the difference between the minimum MODIS IR BT and surrounding mean anvil BT, 8-10) MERRA wind shear within the 0-1, 0-3, and 0-6 km above-ground-level layers.

Of this list of parameters, only the MODIS IR BT – Most Unstable Equilibrium Level, the MODIS IR BT –

Tropopause Temperature, and MODIS OT – anvil BT difference, and MERRA 0-6 km wind shear were statistically significant at the >95% level. The fact that these particular parameters were significant confirms our knowledge of satellite OT observations and the ambient environment of OT-producing storms. For a region to be

04/10/2020

GOES-R3 Status Report Template NESDIS STAR GOES-R

“overshooting”, the cloud top must be above or colder than the local equilibrium level (or level of neutral buoyancy). Analysis of the OT database shows that the majority of OTs also penetrate through the tropopause.

The large OT-anvil BT difference confirms the presence of a high cloud top and a locally strong convective updrafts that caused the top to be higher than the surrounding cloudy pixels. Studies have also shown (e.g.

Brooks et al. (Atmos. Res. 2003)) that strong convective storms are often present in environments with high instability and greater 0-6 km wind shear, which explains why the wind shear parameter was statistically significant in the logistic regression analysis.

Figure 1 shows the results when the logistic regression coefficients are applied to the 803 pixel validation database. 10% of true OT pixels received < 50% detection probability and 26% of the non-OT anvil pixels received > 50% probability. While the 26% “false alarm rate” may seem relatively high, it is highly likely that the pattern recognition analysis (described below) would not have selected many of these pixels as candidate OTs because many were not in a coherent area of cold pixels that would have been recognized using the algorithm described below.

Development of a New Satellite-Based OT Pattern Recognition Approach

A new pattern cognition algorithm was developed to improve upon many of the deficiencies of the original

Bedka et al. (JAMC, 2010) OT detection method. Two of the primary deficiencies were the reliance on fixed BT and BT difference thresholds and use of a relatively simple BT spatial analysis scheme. For example an OT must have a minimum IR BT is < 217.5 K, a pixel is considered part of an anvil only if its IR BT is < 227.5 K, and the

OT must be at least 6 K colder than the surrounding anvil mean IR BT at a fixed ~8 km radius. The biggest noted source of false alarm of the Bedka et al. (2010) method occurs near the edges of anvils where strong spatial IR BT gradients often occur that can trigger the algorithm tests.

The new pattern recognition approach attempts to analyze a satellite image in the same manner as the human eye would to objectively recognize OT regions. Convective cloud tops with OTs exhibit many similarities in satellite imagery regardless of storm location, intensity, or season. An OT typically appears as a small region of cold IR BT and highly textured cloud top surrounded by a warmer and smooth anvil cloud. The region of cold

BTs is usually comprised of more than one pixel and is often circular. The IR BT value within an OT varies depending on the intensity of the convective updraft and the equilibrium level and tropopause temperature of the ambient storm environment which varies seasonally and regionally. The human mind identifies a convective anvil by recognizing coherent circular or elliptical regions of relatively cold clouds with a distinct edge, marking the transition between cloud and clear sky. While the human mind can quickly recognize these patterns, to a computer, a satellite image is a matrix of individual pixel values that must be processed using a sophisticated pattern recognition algorithm to identify convective clouds and OTs.

The first step of the algorithm develops a parameter called the “BT Score”. The objective of the BT Score is to normalize an IR satellite image in a way that highlights pixels that may be within convective clouds. The BT

Score is based upon 1) the IR BT of the pixel, 2) how a pixel’s MODIS IR BT compares with the mean BT in a large window (> 300 pixels) surrounding it and 3) the standard deviation of brightness temperatures within the window. Parameters #1 is used to derive the weighting for parameter #2, i.e. a pixel significantly colder than the window BT mean that is 190 (255) K will get a very high (low) score. Parameter #3 enhances pixels that may be within deep convection as cold clouds surrounded by warm clear sky regions will induce a high standard deviation within the window. From this BT Score field, a set of possible candidates for OT locations is selected by searching for pixels having locally high BT Score relative to their neighbors. For each of these candidates, an

OT Detection Rating is calculated where higher Rating means higher confidence of an OT occurrence. The OT

Detection Rating is modified with the following series of tests that can either increase or decrease the Rating depending on the degree to which a region “looks like” an OT.

The first test involves developing a score based on analysis of the IR BT histogram within a 7x7 pixel window around a given pixel. If a pixel was an OT within a deep convective cloud, the histogram would peak at the anvil IR BT. If the pixel were broken cirrus, the histogram would be broader and the OT Detection Rating would be penalized.

The next step computes an additional score based on the correlation between the original BT Score field and a pre-calculated synthetic 5x5 matrix of pixels with a model distribution that appears like an OT. If the vicinity of a candidate OT pixel were to have a typical bulge-like shape, then the OT Detection Rating would be increased.

04/10/2020

GOES-R3 Status Report Template NESDIS STAR GOES-R

If the BT Score filed near the candidate pixel has an irregular shape, then that candidate’s Rating would be penalized.

The next step involves determining the spatial boundaries of the anvil cloud surrounding the candidate OT.

The BT Score along rays in 16 directions is analyzed to look for abrupt changes that would signal the anvil edge.

The standard deviation along each ray is computed, and its high/low value can also decrease/increase the total

Rating. Once the anvil is mapped in all 16 directions, the anvil shape is analyzed to ensure its rounded appearance which is rewarded with an OT Detection Rating increase. If the mapped anvil were amorphous, it is likely that the cloud isn’t truly deep convection and the Rating is penalized. Lastly, the difference between the BT Score and anvil mean BT Score is computed and if the difference is large, the Rating is increased. The total OT

Detection Rating can be as high as 100, and pixels with a Rating greater than 10 are considered OT Detections and the OT would be evaluated with the logistic regression model.

MODIS imagery is averaged and remapped to a 4 km resolution grid to smooth out subtle BT gradients that appear as “noise” to the methodology described above and could trigger false detections. OTs are typically cold enough that they still remain distinct after the image averaging and remapping. This also allows the algorithm to be easily applied to geostationary imager and AVHRR Global Area Coverage data as each of these datasets have

~4 km resolution. The pattern recognition software takes approximately 2 seconds to process the domain covered by a MODIS granule (~2000x1600 km region). The impact of averaging on the 2 km ABI resolution will be assessed in FY15. Figure 2 shows GOES-13 IR BT, BT Score, and OT Detection Rating for one scene, illustrating that the IR BT Score isolates the deep convective clouds and the pattern recognition scheme and OT

Detection Rating identifies only the intense cells in the image. As each of the tests/steps described above can be weighted individually, more work is required to optimize the impact of each test on the final OT Detection

Rating.

Analysis of Convective Cloud Tops and Overshooting Top Detections in GOES-14 SRSOR Data

GOES-14 observations of deep convective cloud tops were analyzed in combination with ground-based lightning and radar datasets for four storms that occurred during the 2012 SRSOR period. An example of this multi-dataset analysis is shown in Figure 3. A fifth storm was analyzed in a separate GOES-R Risk Reduction project “Integrated GOES-R GLM/ABI approaches for the detection and forecasting of convectively induced turbulence” (PI: L. Carey, UAH). The analyses of these five storms were summarized in a recently submitted

Weather and Forecasting article (see Additional Information section below).

The results showed that, despite the relatively coarse GOES BT data, when OT signatures were evident in the visible channel and had an OT-anvil height differential near to or greater than 1 km, they typically were also associated with distinct BT minima that were regularly detected by the Bedka et al (2010) algorithm. OTs were detected during periods of elevated lightning flash activity and detections typically ceased when the storm was in decay, indicating that this product could be used to identify hazardous and potentially electrically active convection in data sparse regions. Visible imagery and the OT detection products indicate that an individual OT persisted without any interruption for over 30 mins within the two Alabama storms and the Pierce City hail storm, so OTs are not always short-lived phenomena.

Three of the five storms produced above-anvil cirrus plumes, prompting a more detailed analysis of plumes since the plume-severe weather relationship has not been examined thoroughly in the literature. Visible channel images were examined for all storms occurring over the CONUS during the 2012 SRSOR period to identify plume-producing storms and the time of initial plume appearance. The time of plume appearance is compared with severe weather reports to determine the average lead-time that could be provided by initial recognition of the plume signature. Above-anvil cirrus plumes appeared during or shortly after periods of relatively rapid cloud top cooling (~5 K/5 min). Though cloud top cooling of equal or greater magnitude occurred at other points during the storm lifetimes, three of the four plumes present within these storms appeared near to times when the BT was coldest and echo tops peaked, suggestive of moisture (either in water vapor or ice form) injection at very high altitudes where thermodynamic conditions were conducive to cloud formation.

58 storms during the 2012 SRSOR period produced plumes and 33 (57%) of the plume-producing storms were severe. Plumes appeared in advance of a severe weather report in SRSOR data for 28 of the 33 (85%) events. For the five events with no lead-time, the plumes emerged a maximum of 10 mins after the severe weather report. For the other 28 events, the plumes appeared an average of 18 mins in advance of severe weather with a standard deviation of 14 mins. The large standard deviation was caused by seven of the plumes providing

04/10/2020

GOES-R3 Status Report Template NESDIS STAR GOES-R

greater than 30 mins lead-time. The lead-time that would be offered by GOES if it were operating in 15-30-min operational scanning mode was computed to compare with the SRSOR-based results. The scan schedule of

GOES-13 is the basis of this analysis (http://www.ospo.noaa.gov/Operations/GOES/east/imager-routine.html). A plume would have been observed prior to the severe weather report for only 48% of the plume events if GOES were in normal operations. For these events, SRSOR provided a 27-min mean lead-time and the operational imagery would have provided an 18-min lead-time.

Additional Information

1. Interaction with operational partners

The Bedka et al. (2010) OT detection algorithm was evaluated during the 2014 GOES-R Proving Ground at the

NOAA Hazardous Weather Testbed. A voice-articulated training module was developed for the Proving Ground using GOES-R3 funding. Proving Ground feedback included “OTD was really useful for me, in busy environment. I can speak for every broadcaster, we’d all love to have this product.” and “Helpful to CWSU cause

I'm looking at a much larger area… OTD quickly shows me where the strongest updrafts are”. A Proving Ground blog post highlighting the utility of the OT detection can be found here: http://goesrhwt.blogspot.com/2014/06/june-45-mcs-and-overshooting-tops.html

.

The OT detection product was also cited in several Storm Prediction Center Mesoscale Discussions. See the following links for examples: 1) http://www.spc.noaa.gov/products/md/2014/md0753.html

, 2) http://www.spc.noaa.gov/products/md/md0401.html

, and 3) http://www.spc.noaa.gov/products/md/md0162.html

2. Conference/workshop participation

Aspects of these results were presented by K. Bedka at the 2014 EUMETSAT Convection Working Group

Meeting that took place from 7-11 April 2014 in Zagreb, Croatia

3. Funding concerns

Only 75% of the requested funding was provided to K. Bedka during FY13. The remaining 25% will be added to the FY14 budget. Funding for CIMSS tropical overshooting top work was withheld until FY14, and as of this reporting period has not yet arrived. Therefore, no progress to report on during this period..

4. Outside project publicity – None

5. Journal articles

Bedka, K. M., C. Wang, R. Rogers, L. Carey, W. Feltz, and J. Kanak, 2014: Monitoring the Co-Evolution of Total

Lightning, WSR-88D, and GOES-14 Super Rapid Scan Observations Within Deep Convective Clouds.

Submitted to Wea. Forecasting.

Plans for the next Reporting Period:

1) Complete development of the MODIS non-OT pixel database. Incorporate the OT Detection Rating within the logistic regression analysis. Develop a visible channel imagery-based OT Detection Rating that can be used to augment detection probability during the daytime. Finalize the logistic regression analysis.

Optimize pattern recognition approach and finalize algorithm validation

2) Optimize pattern recognition approach and finalize algorithm validation

3) Submit paper for peer-review describing the probabilistic OT detection algorithm

04/10/2020

GOES-R3 Status Report Template NESDIS STAR GOES-R

4) Assess impact of remapping MODIS imagery to 2 km on detection algorithm performance

5) Analyze detection output using multiple years of current GOES imagery and compare against Bedka et al.

(2010) results

6) Begin preliminary studies toward the development of a probability of severe weather field to be associated with OT detections

In collaboration with Chris Velden and Sarah Griffin (UW-CIMSS):

7) Develop methodology to estimate the height magnitude of OT tropopause penetration using co-located

MODIS, CloudSat, and CALIPSO observations. The current OT products are defined in temperature space. For example, the OT, surrounding anvil, OT-anvil difference, and co-located tropopause are all given in units of temperature. However, for some uses such as aviation, the estimated height of the OTs is desirable. We will use the Bedka et al. (JAMC, 2012) CloudSat-observed OT database to conduct an analysis. The results of this analysis will be shared with the GOES-R Cloud Team in the hope that they can implement this knowledge into future versions of ACHA automated cloud height algorithm.

8) Refine and modify the Tropical Overshooting Top (TOT) branch of the OT algorithm for tropical applications in partnership with the NHC/TAFB GOES-R Proving Ground. Furthermore, once refined, the products and diagnostics will be presented and displayed in AWIPS-II to optimize the operational applications at NHC/TAFB.

9) Expand the investigation of TOT parameters, thresholds, and trends for tropical cyclone genesis and rapid intensification prediction tools. Thus far, a moderate correlation has been found between tropical cyclone rapid intensification and TOT trends based on an initial set of empirically-selected parameters (Monette et al. (JAMC, 2012)) in the Atlantic Basin. A lesser correlation was identified with genesis, however much more research is needed to fully explore that relationship. We will incorporate NHC forecaster feedback and evaluations as available.

10) Test detected TOTs and trends as predictors within objective, multi-parameter probabilistic tropical cyclone intensity and genesis prediction models. Initial attempts with established models/schemes have been mixed but promising. Further experiments are warranted, and will include testing within operational schemes now at NHC when initial testing yields successful results. If successful, the TOT algorithm will be prepared for transition into operations for GOES-R readiness.

04/10/2020

GOES-R3 Status Report Template NESDIS STAR GOES-R

Key Graphics

Figure 1: (left) Locations of 2103 OT regions manually identified within 100 granules of MODIS 0.25 km visible channel imagery. (right): OT pixel probability derived from logistic regression analysis and applied to the 25% of the OT and non-OT anvil pixels from the MODIS database. The x-axis represents the

MODIS pixel number which ranges from 1-803. The vertical black line is the separation between the true

OT pixels to the left and non-OT pixels to the right. The red horizontal line identifies the 50% probability.

Figure 2: (left) GOES-13 IR BT on 6 June 2014 at 2215 UTC. (middle) The BT Score derived from the

IR BT. (right) A zoom of the OT Detection Rating for the convective complex across Nebraska and Iowa, illustrating that the Rating only identifies the updraft regions of deep convective cells in this image.

04/10/2020

GOES-R3 Status Report Template NESDIS STAR GOES-R

Figure 4: (upper-left) The GOES-14 BT minimum and objective OT detection time series for a storm that produced a series of severe hail events along the Iowa/Wisconsin border region and an F2 tornado in

Wisconsin on 4-5 September. (upper-right) The ENTLN and NLDN lightning flash rate time series. (lowerleft) WSR-88D multi-reflectivity echo top time series from the La Crosse, WI radar and MESH time series.

Colored vertical dashed lines indicate the times of initial above-anvil plume identifications and severe weather reports. See the legends in the lower-right for a description of the line colors and symbols.

04/10/2020

GOES-R3 Status Report Template NESDIS STAR GOES-R

Download