Analyzing and forecasting weather conditions is a difficult task. There are many uncertainties that complicate these interpretive processes, yet repetitive patterns can emerge from the chaotic structure of meteorology. It stands to reason that access to the successes and failures of past forecasts, and knowledge of relevant theories and conceptual models, can assist the meteorologist in the forecast process. These principles are being put into practice at the NWSFO in Tulsa with appreciable success at the synoptic and mesoscale levels in a software package called Coach. It is my belief that these principles may also be applied to the storm-scale, in a sort of Warning Coach. A primary purpose of Warning Coach is to help the warning meteorologist use the tools available in the most efficient manner possible, given the type of severe weather situation.
Without duplicating the functionality of other tools at the disposal of the meteorologist, Warning Coach will seek to integrate crucial pieces of information to make decisions/suggestions and give advice about how to use the array of products to best deal with the scenario at hand.
In the early stages of the Warning Coach design process, the system was scheduled to perform three main functions:
Provide heads up preparation for warning operations after a forecast is complete (pre-warning)
Alert the user to a range of possibilities of storm types and evolutions
Provide quick concise reference to cases and detailed conceptual models supporting a given situations interpretation (during warning operations)
These goals would have been met through the implementation of three separate, but related packages.
Package 1: Radar Sampling Limitations
Package 2: Forecasting Aid-compare real time radar products to a similar case from storm database; both severe and non-severe cases will be available
Package 3: Conceptual Models
In addition to these graphics based Packages, a quick reference guide for operational forecasting papers will be included with entries sorted according to subject matter and author. Where available, link to papers that are on-line will be provided.
After approximately three months, I realized that a full design and implementation of Warning Coach is well beyond the scope of a thesis. However, defining the primary "battlefield" purpose of Warning Coach, and implementing an application that moves towards the attainment of this purpose, makes an excellent thesis topic. In general, I will address the task of increasing the efficiency with which a meteorologist can interrogate a thunderstorm. There are a number of approaches that could be used to attain this goal. I wish to increase efficiency by reducing the cognitive load on the meteorologist through added "quality automation" in thunderstorm interrogation. By simultaneously assessing the importance of several parameters known to be related to severe potential in thunderstorms, the amount of cognitive processing required by the human is reduced, and decisions can be made faster, yet more informed.
After considering which part of the warning operations problem I would like
to tackle, I have chosen to concentrate my efforts on supercell thunderstorms,
which account for a small percentage of thunderstorms, and yet a disproportionate
amount of severe weather damage. I have confidence that, ignoring time constraints,
even the most most junior forecasters in the field can recognize an isolated,
mature supercell thunderstorm, and can interrogate this storm with some effectiveness.
However, in high pressure warning situations, with time constraints, and several
potentially severe thunderstorms in the radar domain, even some of the more
veteran forecasters can be forced to sacrifice
thoroughness for speed, and even miss some dangerous storms. It is this situation that would be tamed by the addition of "quality automation." For my thesis project, I am designing a SuperCell Identification and Assessment Algorithm (SIAA), not only to identify and assess the severity of existing mature supercells, but ALSO (and more importantly) to identify those cells which are adopting characteristics of supercells, and hence pose a developing threat.
The SuperCell Identification and Assessment Algorithm (SIAA):
SIAA is designed to ingest multiple types of data, process each type of data, and then integrate those results to provide the meteorologist with an enhanced contextual output. In order to reduce processing complexity, facilitate organization, and handle data and output in an intuitive fashion, SIAA is constructed as a cell-based algorithm. Storm cells are meteorologically significant because they represent discrete updrafts. Cells also define a discrete object of interest, significant for software engineering.
A very brief and high level description of SIAA follows. SIAA searches for the defining characteristics of a supercell. If a storm possesses these attributes to a high degree, then it may be classified as a supercell. These characteristics are the presence of: a rotating updraft, a significantly strong updraft, an inflow notch, and a hook echo.
1. Process (ScitTable): Use the output of the Storm Cell Identification and Tracking (SCIT) algorithm (from the NSSL Warning Decision Support System) to allow SIAA to be cell-based. Go through the SCIT Table, and throw out all storms that have an empty "CIRC" field, meaning that there is no Mesocyclone Detection Algorithm (MDA, from the NSSL Warning Decision Support System) identified circulation associated with that cell.
2. Process (MesoTable): For each storm that has an MDA identified circulation, pull out the 3D meso rank and run a Fuzzy Membership function on it (simply measures how much this meso belongs to the "meso family." Also, store the location of this meso.
3. Process (BWER): For each storm with a circ, store the marginal BWER confidence value from the Bounded Weak Echo Region Detection Algorithm (BWER from WDSS). I use the marginal value because it is more representative of simple Weak Echo Regions, indicative of a strong updraft.
4. Process (RadialSet): For each storm with a circ, use the lowest sweep of reflectivity to
- Extract the storm's inner border (outlining boundary of the storm using 35dBz as a boundary threshold).
- Fit approximating B-Splines to the storms boundary to remove a large amount of the noise.
- Go around the boundary looking for points of positive curvature (potential hook echo)
- When a potential hook is found, test its location. The hook must lie near the meso, on the inflow side of the storm (there are a few details to this location restraint, but this conveys the general idea).
- If these location restraints are met, pass the total curvature value to the Fuzzy Membership function (measures how much this hook belongs to the "hook echo" family). Store the value of the strongest hook echo signature.
- Search down-shear for an inflow notch (large negative curvature) in the same manner as the hook echo search was done
5. Combine results: This is where Fuzzy Logic pays off (but gets slightly complicated). Basically, I use an ordered weighted average of the four fuzzy numbers computed earlier (fuzzyMeso, fuzzyWER, fuzzyHook, fuzzyInflow) as the possible_supercell_confidence. Both supporting and refuting evidence is considered in the final membership calculations of Supercell_Membership, Marginal_Membership, and NonSupercell_Membership. SIAA records each of these final membership values, and uses the max of the three as the final output of "Supercellness".
SIAA output will be shown in the form of a table of all the included intermediate parameters and an addition to the SCIT icon ("D" for developing supercell, "S" for mature supercell). These modes of output convey the greatest amount of information to the user, and do so in the most concise and cognitively efficient manner.
Prospective Operational Benefits of SIAA and Future Work:
SIAA effectively mimics the interrogation process a meteorologist would follow in diagnosing the severe potential of a thunderstorm. I believe the algorithm achieves the goal of reducing the cognitive load, but WITHOUT REMOVING THE HUMAN FROM THE DECISION PROCESS. The user has easy access to the data used by SIAA in making its decision, so it does not have the "black box" feel. Enabling a warning meteorologist to make a more informed decision in a smaller amount of time is the ultimate goal of providing the "quality automation" that I have spoken of. I believe that SIAA is a stride in this direction.
Discovering an efficient and accurate version of SIAA has been a rather tedious iterative process. Several combinations of techniques have been investigated in reaching the current successful implementation. The most difficult problem encountered in SIAA design involved constructing the logic that discriminates among potential hook echoes. A number of "gotcha" cases were discovered with previous versions of the logic; however, the current use of the mesocyclone location as a hook echo position restraint has added a deeper physical basis for the selection of the proper signature, which has resulted in a higher probability of detection (> .95) and a much lower false alarm rate (< .15). Preliminary tests have shown SIAA to possess a high degree of skill in discriminating between supercells and non-supercells. The use of both supporting and refuting evidence has proven valuable in the final decision characterizing the "supercellness" of the storm in question.
More extensive testing will be performed very soon, including running SIAA on a very large database of both tornadic and non-tornadic supercell storm days, which also contain non-supercell storms (for null cases). In addition, this coming convective season, in cooperation with the NWSFO, Norman, and NSSL, SIAA is scheduled to be tested alongside the entire suite of WDSS severe weather algorithms. It is during this testing that the most NWS feedback will occur, hopefully allowing me to tailor SIAA according to the operational warning meteorologist's needs.
In future versions, SIAA will use near-storm environment data, to determine if the storm in question is in, or is moving into, an environment capable of supporting supercells. SIAA has been designed to be the front end to a Warning Coach-like decision aid. Different storm types require a different set of products for effective and efficient interrogation. Therefore, once storm type is determined, interrogation advice may be administered based on this storm type, as well as many other factors (position in radar domain, motion, near storm environmental characteristics, etc.). SIAA's job in this context would be to detect all developing or mature supercells, and pass along appropriate advice on how to efficiently deal with the situation. For the more veteran meteorologists, SIAA could function as a product set generator, automatically assembling the preferred product set a meteorologist uses to interrogate a thunderstorm.
In short, I believe that SIAA represents a high level decision aid of the near future. I am hopeful that the algorithm will gain acceptance operationally, as it simply performs the same tasks that the meteorologist would for the given situation, but in a fraction of the time and at a higher level of completeness. Nevertheless, the human still possess the greater cognitive power in diagnosing severe weather; therefore, SIAA has been designed to include the human in the decision process by informing him/her as to how the storm type decision was made by sharing the intermediate output in the form of a table. "Quality automation" must involve the human, otherwise we are putting our fate entirely in the hands of a machine with no ability to adapt and adjust to unforeseen situations.