Most weather products present a single number or colored category representing the chance of severe weather in your area. That single figure compresses two fundamentally different questions into one answer: How bad could it get? and How likely is it to happen? When they are blended together, critical information is lost.
TorCAST separates them. The TorCAST Potential Index (TPI) answers the first question — it describes the severity of what the atmosphere is capable of producing across all three primary severe weather hazards: tornado, hail, and damaging wind. The Confidence indicator answers the second — it tells you how likely that potential is to be realized. Together, they give a complete picture of what you are actually facing.
Consider two contrasting scenarios:
| Day A — Classic Outbreak | Day B — Loaded Gun |
|---|---|
| Strong instability, intense low-level rotation, storms already firing. A confirmed tornado is on the ground 30 miles away. | Equally strong instability and rotation, but a stubborn atmospheric cap is holding. No storms yet — but if the cap breaks, any storm that forms could be immediately violent. |
| TPI: High | Confidence: High | TPI: High | Confidence: Conditional |
A conventional probability approach treats Day B as low risk because storms have not yet formed. TorCAST treats it as high potential with conditional confidence — and communicates that distinction clearly so that emergency managers, storm chasers, and the public understand exactly what the atmosphere is capable of and what depends on whether the cap ultimately breaks.
The TPI — TorCAST Potential Index — is a 0–10 score representing the maximum severity of conditions that severe convective storms could produce at a given location and time. It is calculated across all three primary severe weather hazards (tornado, hail, and damaging wind), continuously across the contiguous United States, and updated every operational model cycle.
TorCAST runs two parallel TPI tracks simultaneously:
| Track | What It Measures | Update Cadence |
|---|---|---|
| Live TPI | Scores the atmosphere as it currently exists using real-time surface observations, active NWS warnings and watches, confirmed storm reports, live radar rotation data, and per-storm severe weather probabilities from NOAA's operational systems. | Continuous — updated as new data arrives |
| Predicted TPI | Scores the atmosphere as it is expected to evolve using high-resolution and extended-range numerical weather prediction model output. Covers forecast windows from one hour to three days ahead. | Each model cycle (hourly short-range; 6-hourly extended) |
The two tracks are compared each cycle to produce a Trend indicator — Rising Fast, Rising, Steady, Falling, Falling Fast — that tells you at a glance whether the threat environment is building, holding, or diminishing. A rising TPI heading into the peak afternoon heating window is operationally very different from a falling TPI as an event winds down, even if the absolute numbers look similar.
Not all severe thunderstorm environments are equal — and more importantly, not all severe thunderstorm environments are dominated by the same threat. A classic supercell outbreak in the southern Plains is driven by tornado potential. A squall line pushing east ahead of a cold front may be primarily a damaging wind event. A high-CAPE, low-shear summer afternoon may favor large hail over either.
TorCAST evaluates each of the three primary severe weather hazards through its own independent scoring track. The dominant hazard drives the displayed TPI value, and the hazard type is shown alongside the score so users understand what kind of threat is being communicated.
Weighted heavily toward low-level storm-relative helicity, updraft organization, storm mode, and radar-detected rotation. The most kinematically demanding of the three tracks — a high score requires multiple independent physical indicators to be simultaneously elevated.
Driven by mid-level instability, updraft intensity proxies, hail-growth-layer depth, and storm structure. Environments with high CAPE but weaker low-level shear may rank high on the hail track without scoring comparably on tornado or wind.
Accounts for deep-layer shear, downdraft energy, storm mode (MCS vs. discrete), and radar-confirmed bow signatures or mesocyclone circulations. Tropical and sub-tropical environments with high moisture but weak shear are scored conservatively on this track to prevent overforecasting of organized wind corridors.
The dominant threat label — Tornado, Hail, or Wind — is displayed alongside every TPI score. This prevents the common misread of seeing a high TPI in a wind-dominated environment and assuming tornado risk, or conversely, of discounting a moderate TPI in a tornado-dominant setup because the overall number looks lower than neighboring hail-dominant areas.
The TPI integrates a broad set of atmospheric parameters. No single variable determines the score — they are combined and weighted based on how strongly each one is associated with the production of significant severe weather under different regional and seasonal conditions.
| Parameter | Why It Matters |
|---|---|
| Surface Moisture (Dewpoint) | Warm, moist air near the surface is the primary fuel for severe convection. Higher dewpoints mean more latent energy is available for storm development and that any storms which form will sustain stronger updrafts. Surface moisture quality is evaluated independently from model-derived instability to catch cases where CAPE is underestimated but moisture is anomalously deep. |
| Atmospheric Instability (CAPE) | Convective Available Potential Energy measures how much buoyancy energy the atmosphere has stored. High CAPE supports explosive storm development and powerful updrafts. TorCAST evaluates both mixed-layer and most-unstable CAPE to account for surface-based versus elevated convective scenarios, and cross-checks model CAPE against multi-model ensemble output to reduce single-model bias. |
| Convective Inhibition (CIN) | The atmospheric "cap" that must be overcome before storms can develop. A moderate cap can be beneficial — it allows instability to build to extreme levels before convection erupts. A very strong cap may hold all day regardless of how favorable everything else appears. CIN is central to the Confidence score: high CIN drives Conditional or Low Confidence classifications even when potential is high. |
| Cloud Base Height (LCL) | The Lifted Condensation Level is the altitude at which condensation begins and clouds form. Lower cloud bases reflect deeper low-level moisture, which is associated with stronger near-surface rotation and a higher probability that tornado vortices survive the journey from the storm base to the ground. |
| Downdraft CAPE (DCAPE) | A measure of the energy available to drive downdrafts. High DCAPE supports cold-pool intensity, which in turn governs how effectively an MCS can generate widespread damaging winds and how aggressively bow echoes can accelerate. The hail and wind tracks weight DCAPE more heavily than the tornado track. |
| Parameter | Why It Matters |
|---|---|
| Low-Level Wind Rotation (0–1 km SRH) | Storm-Relative Helicity in the lowest kilometer of the atmosphere is among the most reliable indicators of tornado potential. It measures how much ambient horizontal spinning motion is available to be tilted into the vertical by a storm's updraft and stretched into a tornado. Strong 0–1 km SRH with a well-organized storm is one of the clearest pre-tornado signals in operational meteorology. |
| Deep-Layer Wind Rotation (0–3 km SRH) | Helicity integrated through a deeper layer of the atmosphere supports mesocyclone formation and persistence. The 0–3 km layer is the foundational kinematic ingredient for supercell thunderstorms — the storm type responsible for the overwhelming majority of significant and violent tornadoes. |
| Deep-Layer Wind Shear (0–6 km) | The change in wind speed and direction between the surface and mid-troposphere governs whether convection organizes into discrete rotating supercells or collapses into less-structured squall lines. Strong 0–6 km bulk shear is required for discrete supercell development. It also discriminates organized MCS wind threats from pulse-storm setups that may have high instability but lack the kinematic structure to sustain a damaging wind corridor. |
| Composite Indices (STP, SCP) | The Significant Tornado Parameter and Supercell Composite Parameter combine instability, rotation, shear, and cloud base height into single discriminants calibrated specifically for distinguishing significant-tornado environments from routine severe setups. TorCAST uses these as check variables against the raw parameter scores rather than primary inputs, preventing single-parameter exceedances from inflating the TPI in environments that fail composite quality checks. |
| Input | How It Influences the Live TPI |
|---|---|
| NWS Warnings & Watches | Active tornado warnings, severe thunderstorm warnings, tornado watches, and Particularly Dangerous Situation (PDS) designations issued by local NWS offices are processed in real time. A confirmed tornado warning in proximity to a scored grid point provides direct uplift to the Live TPI and ensures the confidence classification reflects that storms are already on the ground. |
| Local Storm Reports (LSR) | Verified reports of tornadoes, large hail, and damaging winds submitted by NWS spotters and emergency managers are ingested as they arrive. LSRs provide ground-truth confirmation that the atmosphere has delivered on its potential, and active reports near a grid point directly boost the Live TPI to reflect the ongoing threat. |
| Radar Rotation (Azimuthal Shear) | Live dual-polarization radar-derived rotation rates from the NEXRAD network are processed to detect actively rotating storms. Azimuthal shear detects the tight wind speed differences across a storm that indicate a developing or mature mesocyclone — the most direct precursor to tornado development available in real time. |
| Per-Storm Probabilities (ProbSevere) | NOAA's NWS ProbSevere system tracks individual convective cells and issues continuously updated per-storm probabilities of tornado, hail, and damaging wind. Where available, ProbSevere data provides the highest-fidelity real-time severe weather signal and is given significant weight in Live TPI calculations during active storm events. |
| Lightning Jump Detection (GLM) | Rapid increases in lightning flash rate — "lightning jumps" — are a well-documented leading indicator of severe weather intensification. TorCAST monitors flash rate data from NOAA's Geostationary Lightning Mapper to detect storms undergoing rapid intensification before that signal appears on conventional radar products. |
| NWS Area Forecast Discussions (AFD) | Official narrative forecast discussions written by local NWS meteorologists for each Weather Forecast Office are analyzed to extract qualitative situational context — including severe weather concerns, moisture quality assessments, and explicit risk language — that complements and checks the quantitative model-derived score. |
Not all severe thunderstorm environments are equally capable of producing tornadoes, regardless of how impressive the instability and shear look. The type of storm that develops — its internal organization, structure, and mode — is as important as the raw atmospheric ingredients.
| Storm Mode | Tornado Capability | TPI Adjustment |
|---|---|---|
| Discrete Supercell | Highest. A single organized rotating storm with a sustained mesocyclone is the most tornado-capable storm structure in nature. Produces the overwhelming majority of EF3+ tornadoes. | Full kinematic weighting applied. No mode penalty. |
| Marginal Supercell | Moderate. Some supercellular characteristics present but storm organization is less complete. Capable of producing brief tornadoes. | Moderate reduction to tornado track score. |
| Mixed Modes | Variable. Storm interactions and mergers common. Tornado-capable supercells possible but may be short-lived. | Blended scoring; tornado track discounted, wind track elevated. |
| Linear MCS / Bow Echo | Lower for significant tornadoes. Squall lines produce brief spin-up tornadoes but rarely long-track events. Primary hazard shifts to damaging straight-line wind. | Significant tornado track reduction; wind track elevated. Bow echo detection adds direct wind score uplift. |
| Disorganized / Pulse | Minimal. High-instability, low-shear pulse storms can produce brief microbursts and hail but rarely sustain organized rotation. | Tornado track substantially reduced. Wind and hail tracks scored conservatively — high moisture alone without kinematic organization does not support T5+ scoring. |
TorCAST estimates the likely storm mode from the wind shear profile, model updraft helicity, composite reflectivity, and instability structure. This estimate is continuously updated as model cycles advance and, during active events, as radar data reflects what storms are actually doing.
A key application of storm mode assessment is in the Gulf Coast and Southeast United States, where warm-season environments frequently produce very high instability with relatively weak deep-layer shear. These setups can generate locally intense pulse storms capable of damaging winds and hail in the strongest cores — but they are not organized wind or tornado corridors. TorCAST applies a conservative ceiling to wind threat scoring in high-moisture, low-shear environments without radar or model confirmation of organized storm structure, preventing broad overforecasting in these climatologically challenging regions.
Alongside the TPI, TorCAST displays a Confidence/Realization indicator that communicates how likely the displayed potential is to be realized. This is not a probability in the strict statistical sense — it is an operational assessment of whether the atmospheric setup is primed to deliver on what the TPI indicates.
| Level | Display | What It Means |
|---|---|---|
| High | Solid fill | The cap is weak or absent, a synoptic trigger is in place, moisture is deep and well-mixed, and the timing is favorable. Storms are expected to develop and the TPI potential shown is likely to be realized. If storms are already active, this reflects ongoing confirmed severe weather. |
| Conditional | Diagonal hatching | The environment is loaded with severe weather potential, but storm initiation depends on a catalyst — a passing short-wave, an outflow boundary, afternoon heating sufficient to erode the cap. The potential is real and should not be ignored, but it requires the atmosphere to cooperate before it is released. |
| Low | Dotted pattern | A strong cap, a significant moisture deficit, or unfavorable timing makes storm development unlikely despite otherwise impressive-looking parameters. The TPI is shown to communicate what is at stake if the cap were to break unexpectedly — but under current assessment, the atmosphere is unlikely to release it. |
Confidence is evaluated using cap strength (CIN), synoptic-scale forcing (approaching fronts, short-wave troughs, jet stream dynamics), surface moisture depth, low-level jet timing, and diurnal cycle. The presence of active NWS tornado or severe thunderstorm watches automatically anchors confidence at Conditional or higher — official watch issuance from the Storm Prediction Center represents the highest-quality operational severe weather assessment available, and TorCAST defers to it accordingly.
The TPI runs from 0 to 10 and maps to six operational risk categories. Thresholds have been established by calibrating against the multi-decade public archive of verified severe weather reports — NWS/SPC and Iowa Environmental Mesonet storm reports — evaluated by retrospective hindcast, not by a live track record. They are not arbitrary round numbers.
| TPI Range | Level | Operational Meaning |
|---|---|---|
| 0 – 2 | Very Low | No meaningful severe weather potential. Clear skies, cold air, or post-frontal scour conditions. No action warranted. |
| 2 – 3.5 | Low | Typical spring background environment. Some atmospheric moisture or shear may be present but no active trigger and no organized threat. Routine monitoring only. |
| 3.5 – 5 | Guarded | Atmosphere is primed but not yet active. A trigger mechanism is needed for storms to develop. Worth monitoring, particularly when Confidence is Conditional and the trend is Rising. Awareness appropriate for outdoor activities. |
| 5 – 6.5 | Elevated | Multiple significant severe weather ingredients are in place. Isolated significant storms producing tornadoes, large hail, or damaging winds are possible. Active weather awareness is appropriate. Know your sheltering plan. |
| 6.5 – 8 | High | Significant tornado risk. Supercell or multi-cell storm mode expected. A regional severe weather outbreak is possible. Days verified at this level to date have produced confirmed tornadoes. Heightened situational awareness and preparedness warranted. |
| 8 – 10 | Extreme | Major outbreak conditions. The atmospheric ingredients for long-track, violent (EF4–EF5) tornadoes are in place. These days are historically rare. At this level the environment supports significant, long-track tornadoes. Immediate protective action may be warranted in highlighted areas upon storm initiation. |
TorCAST ingests multiple real-time and model data streams. All primary sources are government-operated or publicly available scientific data feeds operated by NOAA, the National Weather Service, and the FAA.
| Source | Operator | How TorCAST Uses It |
|---|---|---|
| HRRR (High-Resolution Rapid Refresh) | NOAA / NCEP | The primary high-resolution short-range forecast model for the contiguous United States. Updated hourly. Provides the detailed atmospheric parameter data — CAPE, CIN, SRH, shear, updraft helicity, reflectivity — that drives TorCAST's 0–18 hour potential scoring. |
| CAM Ensemble (HREF) | NOAA / SPC / EMC | The High-Resolution Ensemble Framework combines multiple convection-allowing models to provide probabilistic short-range guidance. TorCAST uses CAM ensemble output to cross-check single-model HRRR values, identify model-to-model spread, and improve cap and moisture assessments in convectively uncertain environments. |
| GFS (Global Forecast System) | NOAA / NCEP | The US global forecast model, updated four times daily. Provides extended-range (Day 1–3) potential scoring beyond HRRR's domain, covering synoptic-scale environment assessment for multi-day severe weather outlooks. |
| Surface Observations (ASOS / METAR) | NWS / FAA | Automated weather stations at airports and other surface sites across the country report surface conditions continuously. Provides real-time dewpoint, temperature, wind, and pressure data used in Live TPI scoring and to validate model surface analyses. |
| NWS Warnings, Watches & Outlooks | National Weather Service / SPC | Active tornado warnings, severe thunderstorm warnings, tornado and severe thunderstorm watches, and SPC convective outlooks are processed in real time. SPC categorical risk levels serve as a foundational anchor for TPI scoring in their respective areas — TorCAST will not issue scores inconsistent with official operational guidance without strong physical justification. |
| Local Storm Reports (LSR) | NWS / Iowa Environmental Mesonet | Verified reports of tornadoes, large hail, and damaging winds from NWS spotters, emergency managers, and trained observers, archived in real time by the Iowa Environmental Mesonet. Used for live scoring uplift when reports are active nearby, and — alongside issued NWS warnings — as one of two primary verification sources for long-term calibration of scoring thresholds against confirmed severe weather outcomes. |
| NEXRAD / Dual-Pol Radar (MRMS) | National Weather Service | The NOAA Multi-Radar Multi-Sensor (MRMS) system fuses data from the national NEXRAD radar network into seamless national mosaics. TorCAST uses MRMS-derived azimuthal shear and reflectivity mosaics to detect rotating storms, track bow echoes and MCS structures, and provide direct radar-based input to the Live TPI. |
| NWS ProbSevere | National Weather Service | A real-time probabilistic guidance system that tracks individual convective cells and issues continuously updated per-storm probabilities of tornado, hail, and damaging wind. ProbSevere integrates radar, satellite, and environmental data at the storm scale and represents the highest-fidelity real-time severe weather signal available in the TorCAST data stream. |
| GLM Lightning Data | NOAA / GOES-East/West | Total lightning data from NOAA's Geostationary Lightning Mapper, carried aboard the GOES-East and GOES-West satellites, provides full-disk lightning flash rate and density every 20 seconds. Rapid increases in flash rate are a well-established leading indicator of severe storm intensification. |
| NWS Area Forecast Discussions (AFD) | National Weather Service (122 WFOs) | Narrative forecast discussions produced by meteorologists at each NWS Weather Forecast Office are analyzed for qualitative severe weather context. Geographic WFO assignment is performed spatially — each scored grid point is matched to its nearest WFO centroid — ensuring that local meteorologist assessment is geographically accurate rather than applied region-wide. |
TorCAST's scoring engine starts with a physics-based foundation. The atmospheric parameters described in Section 4 are scored against thresholds derived from peer-reviewed severe weather research and from operational forecasting practice. The physics layer is interpretable — any score it produces can be traced back to specific physical reasons — and it works on day one without needing historical training data.
The physics layer has a well-known limitation: the hand-tuned thresholds and weights that work well in one regime (say, a classic southern Plains supercell outbreak in late April) may systematically over- or under-estimate risk in a different regime (say, a Gulf Coast overnight MCS event in late August). No fixed set of parameters captures the full range of severe weather environments across the continental US, and honest acknowledgment of that limitation is built into how TorCAST is designed.
On top of the physics foundation, TorCAST trains a set of per-hazard machine learning correction models on verified severe weather outcomes. These models do not replace the physics scoring — they learn where it is systematically wrong and by how much, then apply a targeted residual correction to reduce that bias.
The training dataset is built by retrospective hindcast — re-running the TorCAST physics engine over archived model analyses from prior seasons — and pairing each historical score against two independent verification sources: confirmed NWS Local Storm Reports for tornadoes, large hail, and damaging winds, and issued NWS tornado and severe thunderstorm warnings. LSRs provide ground-truth confirmation that an event reached the surface and was observed. Issued warnings provide a complementary signal — they reflect the operational judgment of trained NWS meteorologists responding to real-time radar and environmental data, and they capture developing threats that may not yet have generated spotter reports. Using both sources together reduces the risk of training on an incomplete picture of when and where severe weather actually occurred. For each scored grid point and time, the system knows what the physics engine predicted and what the combined verification record shows. The correction models learn the residual — the difference between the physics output and reality — and try to explain it using the full set of available atmospheric inputs. The result is a score that inherits the physical interpretability of the rule-based layer while having the empirical accuracy of a model trained on real outcomes.
TorCAST trains three independent correction models — one per hazard track (tornado, hail, wind). This matters because the biases are different. The physics engine may underestimate tornado risk in a high-shear, low-CAPE Dixie Alley setup while simultaneously overestimating wind risk in a high-CAPE, low-shear Gulf Coast pulse-storm environment. A single blended correction model would trade one off against the other; separate per-hazard models correct each independently.
The correction models are gradient-boosted statistical models trained on a feature set that includes all of the physical scoring inputs plus time-of-day, geographic coordinates, seasonal context, and the live observational state at scoring time. Practically, they tend to learn patterns like:
The models do not learn to chase historical events in isolation. The training includes regularization to prevent overfitting to specific past cases, and verification on held-out event years confirms that corrections generalize to new events rather than just memorizing the training data.
The correction models are retrained periodically as new verified severe weather outcomes accumulate. Each verified event — tornado, hail report, damaging wind — adds to the training archive. Over time this improves regional calibration, particularly for rare but high-impact event types that take years to accumulate sufficient training cases.
When a new model version is deployed, it is run in shadow mode alongside the current operational model for a period before going live, and its per-hazard verification statistics are compared against the prior version across multiple event types and regions. A new version is not deployed if it degrades performance in any major regime even while improving overall accuracy.
Machine learning corrections operate within bounds set by official operational guidance. The Storm Prediction Center's convective outlooks — Marginal, Slight, Enhanced, Moderate, and High Risk — are treated as a foundational ceiling and floor for TPI scoring in their respective areas. TorCAST will not issue Extreme-level scores in areas the SPC treats as sub-marginal without strong, multi-indicator physical support that the ML layer independently agrees with. The SPC outlook is the best operational severe weather assessment in the world; TorCAST treats it as a boundary condition, not just one more input.
Because TorCAST scores each hazard track independently, verification is also conducted per-hazard. A tornado-dominant TPI of 6.5 is verified against confirmed tornado reports; a wind-dominant TPI of 5.5 is verified against confirmed damaging wind reports. A day that scores high on tornado and delivers a large hail event instead is not a clean "hit" or "miss" — it is a hazard-type error, which is tracked separately from a magnitude error. Per-hazard verification is what makes per-hazard ML correction tractable: you need the right ground-truth label for each training example, and the label has to match the thing you are trying to predict.
Severe weather is not a one-size-fits-all phenomenon. The atmosphere behaves differently in different parts of the country, the parameters that matter most for tornado potential vary significantly by region and season, and a TPI threshold calibrated for a Kansas supercell outbreak is not automatically appropriate for a Southeast winter-tornado event. TorCAST applies regionally differentiated scoring approaches.
| Region | Scoring Approach and Key Considerations |
|---|---|
| Tornado Alley (KS, OK, TX, NE, IA) |
The classic high-CAPE, high-shear supercell environment. Storm mode discrimination — discrete supercell versus squall line — is the primary scoring discriminant once instability and shear thresholds are met. Dryline position, boundary layer moisture recovery, and cap erosion timing are weighted heavily in the Confidence assessment. Afternoon and evening severe weather timing is the expected norm. |
| Dixie Alley (AL, MS, TN, AR, LA, KY) |
High-shear, relatively lower-CAPE environments, often with significant capping in place. Surface moisture quality and low-level jet strength are critical — the nocturnal low-level jet can maintain and intensify storms well into overnight hours, requiring TorCAST to score evening and nighttime setups carefully without applying standard diurnal penalties. The highest fatality risk from winter and early-spring tornado events in the US originates from this region. |
| Gulf Coast and Southeast (FL, GA, SC, NC, MS coast) |
Sub-tropical and tropical moisture regimes with frequent convective initiation, but storms often organized in mixed-mode or pulse setups rather than sustained supercell corridors. This region presents the greatest overforecasting risk for wind TPI scores: very high CAPE driven by tropical moisture can suggest T5+ wind potential where only localized pulse-storm gusts are warranted. TorCAST applies conservative wind-track ceilings in high-PWAT, low-deep-shear environments without radar or model confirmation of organized storm structure. |
| Northeast and Great Lakes (NY, PA, OH, MI, IN) |
Less frequent but high-impact events, typically tied to strong synoptic forcing systems. Terrain interactions, lake-effect moisture corridors, and the influence of the Appalachians on low-level wind profiles require localized adjustment. Overnight and early-morning convective events associated with strong jet dynamics are not uncommon and are handled by the system's nocturnal scoring pathway. |
| Southern Plains and Ozarks (MO, AR, western TN) |
A transitional zone between classic Tornado Alley and Dixie Alley environments. Often captures the highest-impact events from both regimes — early-season high-shear setups and late-season subtropical moisture invasions. WFO geographic assignment is managed carefully in this region to ensure the correct local meteorologist perspective is applied to each scored grid point. |
| West Coast and Pacific (CA, OR, WA, AZ) |
Low-CAPE, high-shear environments where standard severe weather composite indices such as STP and SCP are unreliable or irrelevant. TorCAST uses a dedicated kinematic-dominant scoring regime for these regions that gives appropriate weight to low-level helicity and deep-layer shear over raw thermodynamic instability. Brief spin-up tornadoes in organized shear environments are the primary tornado threat; the standard thermodynamic thresholds are explicitly bypassed in this regime. |
| Principle | Explanation |
|---|---|
| Read Potential and Confidence together | A TPI of 8 with Low Confidence and a strong cap in place is a conditional warning — the environment is dangerous, but storms may never form. A TPI of 5 with High Confidence, an active tornado watch, and radar showing rotating storms may be more immediately life-threatening. Both dimensions must be read simultaneously. |
| Watch the trend, not just the number | A TPI of 6 that is Rising Fast into the afternoon heating window on a High Confidence day is a very different operational situation than a TPI of 6 that is Falling Fast after the evening convective maximum has passed. The trend arrow is among the most operationally useful elements of the TorCAST display. |
| Look at the dominant hazard label | A TPI of 7 with a tornado-dominant designation means the atmosphere is loaded for tornado production. The same score with a wind-dominant designation means the primary hazard is organized straight-line wind — a different protective action calculus than a tornado. The hazard label is not decorative. |
| TorCAST is not a tornado track forecast | The TPI describes the severity of environmental potential at a grid location. It does not predict whether a specific tornado will touch down at a specific address, or what path an individual storm will take. Always follow active NWS tornado warnings and local emergency management guidance for immediate life-safety decisions. |
| High TPI does not guarantee a tornado | Many days with TPI scores of 7 or 8 produce significant severe weather without confirmed tornadoes. TorCAST tells you what the atmosphere is loaded with — what actually happens depends on whether storms initiate, how individual cells evolve, and local-scale factors that no regional grid can fully resolve. Maximum potential is not guaranteed outcome. |
| Extreme TPI should always be taken seriously | When TorCAST reaches Extreme levels (8+), the atmospheric ingredients for a major outbreak are in place. Extreme days are historically rare. When the atmosphere reaches this level, the ingredients for significant, long-track tornadoes are present. Even in the absence of confirmed storms at the time, Extreme TPI with High Confidence warrants immediate attention and preparation. |
TorCAST is designed to be honest about what it can and cannot do. The following limitations are inherent to any environmental-potential scoring system: