Thunderstorms pose high risks for landings, departing and flying aircraft. This collaborative research effort between the Massachusetts Institute of Technology Lincoln Laboratory (MIT-LL), the University of Alabama in Huntsville (UAH), the National Center for Atmospheric Research (NCAR), the National Aeronautics and Space Administration (NASA), and Cooperative Institute for Meteorological Satellite Studies (CIMSS) at the University of Wisconsin-Madison is to incorporate satellite-derived information into systems designed to nowcast atmospheric convection and its first-time initiation. Developing improved means of monitoring and characterizing convective clouds to nowcast convection initiation (CI) is a key goal. Data sets currently being processed include GOES visible, infrared and sounder-based satellite imagery (from GOES-10, -12, -14, etc.). New research involves the use of MODIS, MSG SEVIRI and eventually GOES-R data.
This CI work supports a component of the NASA Advanced Satellite Aviation Products (ASAP) initiative with the primary focus of improving aviation safety through the use of satellite-based products. This project supports the National Convective Weather Diagnostic Forecast Product developed in part by the NCAR Convective and Oceanic Weather Product Development Teams (PDTs), as operated by the FAA. The plan at UAH is to process several types of satellite information into "Interest Fields" and "Pattern Fields" that can be used to describe convective initiation across large geographical regions.
Ongoing research on 0-1 hour CI is in several areas:
- Assessing how to use MSG SEVIRI data (advance of GOES-ABI) in more sophisticated ways when nowcasting CI, especially toward diagnosing in-cloud processes.
- Evaluating how derived cloud-top properties, such as effective radius and cloud optical thickness, may be used to diagnose in-cloud processes.
- Developing methods to quantify boundaries and low-level moisture from satellite.
- Assessing likely storm intensity and mode/behavior (see below).
- Enhancing an object-based tracking approach to identifying new CI events.
- Using artificial intelligence methods to nowcast CI and new storm growth (new in 2010).
- Enhance the use of statistical approaches to objectively adjust the usefulness of infrared thresholds for growing convective clouds, ahead of CI (new in 2010)
- Applied work for the Convective Initiation Algorithm Working Group (AWG).
Once cumulus cloud tracking is established, eight predictor fields based on Lagrangian trends in IR data are used to characterize cloud conditions consistent with CI. Cumulus cloud-pixels for which ≥7 of the 8 CI indicators are satisfied are labeled as having high CI potential, assuming an extrapolation of past trends into the future. Comparison to future WSR-88D radar imagery then measures the method's predictive skill. CI predictability has been demonstrated using several convective events occurring over a variety of synoptic and mesoscale forcing regimes.
This research focuses on identifying the precursor signals of CI from sequences of 5- and 15-min time resolution 1 km VIS and interpolated IR imagery from GOES. Presently, CI may be forecasted up to ~45 minutes in advance through the monitoring of key IR temperatures/trends for convective clouds. Based on analysis results, we surmise that the current predictability limitation of this algorithm is ~1 hour, as cumulus clouds evolving for longer periods often do not grow to initiate rainfall. CI nowcasting is made possible by first interpolating all IR data to the VIS resolution and projection, second by locating only the clouds capable of initiating rainfall within GOES data through using a cumulus cloud mask at 1 km resolution, third by performing several multi-spectral IR channel differencing techniques to identify cumulus in a pre-CI state, and finally by utilizing combined VIS and IR satellite-derived mesoscale atmospheric motion vectors (AMVs) as a means of tracking individual cumulus clouds in sequential imagery to estimate cloud-top trends. The VIS and IR AMVs contain both balanced and divergent flow components due to modifications made to AMV processing algorithm. In effect, these techniques isolate only the cumulus convection in satellite imagery, track moving cumulus convection, and monitor their IR cloud properties in time. CI is predicted through the accumulation of information within a satellite pixel that is attributed to the first occurrence of a ≥35 dBZ radar echo as obtained from WSR-88D mosaic data.
Given the satellite tracking of moving cumulus for the monitoring of CI, this work represents an advance in the ability to predict CI in routinely available, real-time data streams. The processing methods presented are fully capable of operating in realtime (every ≤15 mins) over large geographical regions [O(106) 1 km pixels], or about 1/4th-1/3rd of the continental U. S. The large-scale processing as part of this algorithm is made possible through use of a cumulus mask that isolates only the 10-30% of convective clouds within a GOES VIS image.
Once cumulus cloud tracking is established using MAMVs, six IR properties (creating eight interest fields) of the clouds are monitored at 1 km resolution for their relative importance with respect to CI occurrence. The method achieves ~60-70% accuracy when applied to three case events that comprise a range of synoptic and mesoscale forcing regimes. In addition to adopting and incorporating results from other research, this study is unique in its first use of several IR multi-spectral methods for monitoring convective clouds, namely the 13.3–10.7 µm, ¶(6.5 (or 6.7)–10.7 µm)/¶t, and ¶(13.3 (or 12)–10.7µm)/¶t interest fields.
Algorithm adjustments are likely needed before this method may be applied over environments that differ significantly from those presented here (i.e. Midlatitude), over the Tropics in particular, where "warm rain" microphysical processes plays a large role in rainfall initiation. Another area of active work is towards operating this algorithm at night when the convective cloud mask (in its present form) cannot be used, 3.9 mm AMVs replace VIS AMVs, and we are limited to 4 km IR resolution data. This new research will be reported on in subsequent papers.
Validation results are (Mecikalski et al. 2008): 1) measures of accuracy and uncertainty of the Mecikalski and Bedka (2006) algorithm via commonly used skill scoring procedures, and 2) a report on the relative importance of each interest field to nowcasting CI using GOES. It is found that for non-propagating convective events, the skill scores are dependent on which CI interest fields are considered per pixel, and are optimized when 3-4 fields are met for a given 1 km GOES pixel in terms of probability of detection, and threat and Heidke skill scores (HSS). The lowest false alarm rates are found when one field is used, that associated with cloud-top glaciation 30 minutes prior to CI.
The maximum POD for the MB06 method (with 8 of 8 fields in range) is ~99.8%, with FAR skills minimized at ~69%. The Threat Score and HSS maximize at 26.1 and ~38%, respectively, when 3 fields are scored. Use of <8 fields results in "conditional skills," which may be used as well within this algorithm, and which implies considerable variability in per-pixel scores depending on which IR fields are within range. Analysis shows that using the13.3–10.7 µm difference, 10.7 µm TB or 10.7 µm TB Drop Below 0° C in MB06 alone maximize the TS and HSS skills, hence maximizing POD at ~99% and keeping the FARs relatively low.
Describing the CI algorithm terms of cumulus cloud behavior as observed by GOES suggests the following: 1) Incorporation of the 8 km resolution 13.3 µm channel on GOES-12 (and prior) provides high-value in detecting and observing cloud growth of the larger cumulus clouds, i.e., as updraft widths increase there is increased likelihood of CI over the next hour (larger updrafts survive longer); 2) the transition from above to below 0° C as measured by the 10.7 µm channel (i.e. cloud-top glaciation), 3) cloud-top temperature (colder cumulus imply CI), 4) the 6.5–10.7 µm TB difference (i.e. cumulus growing into middle tropospheric levels implies a capping inversion is no longer present), and 4) the time trend in 6.5–10.7 µm TB difference. The 15- and 30-min cloud-top cooling rates measured through the 10.7-µm channel values are important as well for contributing to CI nowcasting value, as suggested by Robert and Rutledge (2003). The statistical results highlight the need to understand when certain IR fields won't add value to the CI nowcast, and that conditional scoring has more value over simply using all 8 fields per pixel. In the MB06 method, when fewer than 8 interest fields are used in the scoring, each pixel possesses a unique set of skill scores, depending on which fields happen to be within range.
Beginning in 2008, the CI algorithm (termed the SATellite Convection AnalySis and Tracking system, SATCAST; Mecikalski et al. 2010a,b), has been developed into the Consolidated Storm Prediction Algorithm (CoSPA)/Corridor Integrated Weather System (CIWS) at MIT-Lincoln Laboratory (Iskenderian et al. 2009).
In 2009, work began to develop an "object tracking" approach to the CI nowcasting algorithm, which will be documented and validated in Walker et al. (2011). SATCAST with object tracking has become the NOAA/GOES-R Algorithm working Group's AWG) official CI algorithm, and it currently nearing the end of its initial development phase as an "Option 2" GOES-R product. In this final stage of maturity, only minor enhancements or additions are planned for adjusting the algorithm before the 100% delivery. However, once this delivery is made, even with fulfilled AWG requirements, critical improvements to the official algorithm are needed, which include: (a) further development of a nighttime component, (b) improved methods for using specific interest fields based on observed weather and airmass variability, and (c) increasing understanding of how environmental factors influence infrared observations made by satellite. The CI algorithm theoretical basis document (ATBD) as part of the NOAA AWG provides a high level description of, and the physical basis for, the assessment of convection initiation derived from the Advanced Baseline Imager (ABI) flown on the GOES-R series of NOAA geostationary meteorological satellites. The CI algorithm provides an assessment of the clouds that may precipitate. The CI algorithm is designed to monitor the growth of non-precipitating clouds, and once a series of spectral and temporal thresholds are met, that cloud is identified as likely to have a radar reflectivity greater than 35 dBZ within 0-2 hours. The AWG CI algorithm produces a binary field at 2 km spatial resolution of areas where CI has a high likelihood of occurring. The product uses an IR TB spectral thresholding technique, which tracks clouds within their early stages of development, and monitor their spectral characteristics. If a large majority of the spectral "interest fields" thresholds are exceeded, then the pixels within the cloud object are flagged for having a high likelihood for CI.
Specifically, one such component of the algorithm that needs enhancement is that which identifies and follows the same satellite-derived cloud "objects" between subsequent images, known as the "object-tracking" routine. The current algorithm employs a simple temporal-overlapping method for this in which identified cloud objects from one image are superimposed onto an array of cloud objects from the next image. Where there is overlap between two objects in the superimposed image, the objects are considered to be the same object from each of the two input image times. Because of the high temporal resolution of the GOES-R ABI (2 km in the infrared), it is reasonable to assume that most objects can be tracked via this method, since object overlap between two closely timed images is relatively easy to achieve.
The work by Mecikalski et al. (2010a) expanded the GOES-focused SATCAST algorithm to operate on Meteosat Second Generation (MSG) SEVIRI data, in advance of the Advanced Baseline Imager on GOES-R. A total of 67 IR CI interest fields are initially assessed in MSG SEVIRI data for containing information on three attributes of growing convective clouds: cloud depth, updraft strength and cloud-top glaciation. Through correlation and principal component analyses, 21 fields out of the 67 are identified as containing the least amount of redundant information. Using between 6 and 8 fields per category, two methods are proposed on how growing convective clouds may be quantified per MSG pixel (with 3 km scaling distance), or per cumulus cloud object, toward monitoring cumulus cloud development. In Table 5, the 6.2--10.8 µm (Schmetz et al. 1997) difference has the highest ranking when estimating cumulus cloud depth, with the 6.2–7.3 µm difference and 10.8 µm TB the second and third on the list, respectively. Table 6 shows the top three fields as the 15-min trend in the tri-spectral difference [(8.7–10.8)–(10.8–12.0)], the instantaneous tri-spectral difference, followed by the 15-min 8.7–10.8 µm time trend, all physically consistent, well-documented indicators of cloud-top glaciation (Baum et al. 2000b). The 30-min 6.2–7.3 µm difference, and the two time trends of 10.8 µm TB (as in Roberts and Rutledge 2003) as highly ranked IR fields for estimating updraft strength, and hence cumulus cloud growth rates in advance of CI. The cited literature highlights the physical interpretation of a given channel difference or time trend. In contrast, several fields not documented in the literature are shown to possess unique value when monitoring cumulus cloud growth and evolution. These include: (a) the 8.7–12.0 µm difference, (b) the 6.2–7.3 µm difference, (c) the 6.2–9.7 µm difference, (d) the 7.3–13.4 µm difference, (e) the 30-min 9.7–13.4 µm trend, (f) the 15-min 7.3–9.7 µm trend, (g) the 15-min 6.2–7.3 µm trend, and (h) the 30-min 6.2–7.3 µm trend.
Mecikalski et al. (2010b) developed understanding on how visible data can be used to monitor growing cumulus ahead of CI. A total of 27 IR CI interest fields were initially assessed for containing reflected and brightness variability (BV) information. The reflectance fields ultimately help determine cloud-top glaciation, related in many cases to changes in particle size and the formation of ice hydrometeors, while BV fields diagnose the presence of active convective clouds that further correlate to updraft vigor. Through correlation and principal component analyses, 11 fields [5 peak detection (PD), 6 reflectance] out of 27 initial fields are identified as containing the least amount of redundant information. The main findings include: (1) Time trends of decreasing Refl.6 and Refl3.9 correlate well to growing cumulus clouds undergoing CI, with 15-min trends of reflectance being near –0.83 to –0.81% for both channels. Thirty-minute trends for Refl1.6 and Refl3.9 are –4.8% and –2.6%, respectively. (2) Cloud-top reflectances at or below ~3.6-3.5% as measured at 3.9 µm appears to be a good indicator that ice hydrometeors are in abundance, and a phase change has occurred. (3) Higher PD indicates the presence of a cumulus field, or the highly non-uniform brightnesses produced by cumuli against an otherwise darker or uniform background. (4) Data in the highly correlated 0.6 and 0.8 µm channels, although indicators of optical depth changes as clouds deepen, appear to be insensitive to cloud development and are not valuable indicators alone of growing convective clouds. Yet, (5) Refl0.6 and its time rate of change, have more value for describing cloud-top conditions for warmer, lower cumulus clouds as compared to clouds more likely to contain significant percentages of ice hydrometeors. Several methods are proposed on how growing convective clouds may be quantified per cumulus cloud object, towards monitoring cumulus cloud growth rates, and to perhaps nowcast CI over 1 hour timeframes.