Nowcasting
CSIR NET Part C aviation meteorology: nowcasting with Doppler radar, satellites and lightning, Z-R relations, Nyquist velocity, extrapolation and verification.
- 19 min
On this page · 16 sections
- Prerequisites
- The Hook: Why the Forecast Model Is Too Slow
- What Nowcasting Is
- The Observing Tools
- Doppler Weather Radar: The Numbers Part C Uses
- Satellites and Lightning
- Nowcasting Methods
- Aviation Nowcast Products
- Verifying a Nowcast
- Memory Tricks
- Don’t Confuse
- Common Mistakes
- Practice
- Quick Revision Table
- Key Takeaways
- What’s Next
Prerequisites
- Weather Hazards at Take-off, Cruise and Landing — the thunderstorm life cycle and aerodrome products such as TREND.
- Visibility and Wind Shear — microbursts and gust fronts, the main targets of terminal nowcasting.
- Scattering regimes (Rayleigh vs Mie) and brightness temperature, from Radiation, Scattering, Radiative Transfer and the Greenhouse Effect.
- Convective instability and CAPE, from Vertical Stability and Tropical Convection.
- Doppler effect and wave basics, from Physical Sciences.
The Hook: Why the Forecast Model Is Too Slow
What Nowcasting Is
Nowcasting is classically defined as a detailed description of the current weather together with forecasts obtained by extrapolation for a period of about 0–2 hours ahead (the WMO usage; some publications extend it to about 6 hours); very short-range forecasting extends to about 12 hours. In practice the boundary is fuzzy: many services issue thunderstorm nowcasts valid for a few hours.
Why aviation needs it:
- Time and space scales. Microbursts last minutes and span a few kilometres; an ordinary thunderstorm cell lasts about an hour or less; fog onset and clearance happen within tens of minutes. These are smaller and faster than routine forecasts resolve.
- Predictability. Errors at the convective scale grow quickly, so deterministic skill for individual storms is limited to short lead times. Within that window, the best predictor of the next hour is often a careful extrapolation of the last hour.
- Decisions are immediate. Runway changes, holding, diversion, ground operations during lightning, de-icing, and wind-shear alerts all need minutes-to-hours guidance.
graph LR
O["Observe
radar, satellite,
lightning, surface,
profilers, aircraft"] --> A["Analyse
identify cells,
boundaries, signatures"]
A --> T["Track and extrapolate
motion vectors,
growth and decay trends"]
T --> B["Blend
with rapid-update NWP
as lead time grows"]
B --> W["Warn and disseminate
aerodrome warnings,
wind-shear alerts,
TREND, SIGMET"]
W --> V["Verify
POD, FAR, CSI"]
V --> AThe Observing Tools
| Tool | What it measures | Nowcasting use |
|---|---|---|
| Doppler weather radar | Reflectivity, radial velocity, spectrum width; dual-polarisation variables | Storm location, intensity, motion, rotation, divergence, hail, rainfall rate |
| Geostationary satellite | Visible, infrared and water-vapour radiances at high time frequency | Cloud growth, cloud-top cooling, overshooting tops, fog and low cloud, moisture boundaries, convective initiation over data-sparse areas and sea |
| Lightning detection networks | Cloud-to-ground and (with some systems) total lightning | Storm electrification, intensification (rapid increases in flash rate), ground-safety warnings |
| Surface automatic weather stations | Wind, temperature, humidity, pressure, rainfall at high frequency | Gust fronts, pressure jumps, temperature drops, fog onset |
| LLWAS, lidar, wind profilers | Low-level winds | Wind shear and microbursts, low-level jets |
| Aircraft reports | Automated wind, temperature, turbulence (EDR) and pilot reports | Turbulence, icing and wind shear where they are actually encountered |
| Radiosondes and model soundings | Vertical profiles | Instability and shear: what kind of convection to expect |
India operates a growing network of Doppler weather radars and uses its INSAT-3D series geostationary imagery, together with lightning networks and automatic weather stations, for thunderstorm nowcasting.
Doppler Weather Radar: The Numbers Part C Uses
Reflectivity
A radar sends short pulses and measures the power scattered back. For precipitation-size drops much smaller than the wavelength (Rayleigh regime), the returned power is proportional to the reflectivity factor
$$Z = \sum_i D_i^6 \quad (\mathrm{mm^6\,m^{-3}}), \qquad \mathrm{dBZ} = 10\log_{10} Z$$and falls off as \( 1/r^2 \) with range for a beam-filling target. The sixth-power dependence means a few large drops dominate: doubling a drop’s diameter multiplies its contribution by 64. Hail gives very high reflectivity.
Rain rate from reflectivity. An empirical Z–R relation \( Z = aR^b \) converts Z to rain rate R (mm h\(^{-1}\)). The classic Marshall–Palmer relation is
$$\boxed{Z = 200\,R^{1.6}}$$| dBZ | Z (mm\(^6\) m\(^{-3}\)) | R from Marshall–Palmer (mm h\(^{-1}\)) |
|---|---|---|
| 30 | 1,000 | 2.7 |
| 40 | 10,000 | 11.5 |
| 45 | 31,600 | 23.7 |
| 50 | 100,000 | 48.6 |
| 55 | 316,000 | 99.9 |
Each 10 dBZ is a factor of 10 in Z, and therefore a factor of \( 10^{1/1.6} \approx 4.2 \) in R. The coefficients vary with drop-size distribution (convective vs stratiform, tropical vs mid-latitude), so the same dBZ can mean different rain rates; values above about 55 dBZ usually indicate hail contamination, where Z–R fails.
Doppler velocity and its limits
A Doppler radar measures the phase shift between successive pulses from the same target, giving the radial velocity \( v_r \) (positive away from the radar by the usual convention). The Doppler frequency shift is
$$f_D = \frac{2v_r}{\lambda}$$Pulsed radars sample at the pulse repetition frequency (PRF). This sets two limits:
$$\boxed{R_\text{max} = \frac{c}{2\,\mathrm{PRF}}}, \qquad \boxed{V_\text{max} = \frac{\lambda\,\mathrm{PRF}}{4}}$$- \( R_\text{max} \) is the maximum unambiguous range: an echo from beyond it returns after the next pulse has left and is plotted at the wrong (shorter) range (second-trip echo).
- \( V_\text{max} \) is the Nyquist velocity: radial velocities beyond \( \pm V_\text{max} \) fold (alias) into the interval, so a true 32 m s\(^{-1}\) with \( V_\text{max} = 25 \) m s\(^{-1}\) appears as \( 32 - 2\times25 = -18 \) m s\(^{-1}\).
Multiplying them gives the Doppler dilemma:
$$R_\text{max}\,V_\text{max} = \frac{c\,\lambda}{8}$$You cannot have both large range and large Nyquist velocity at one PRF. Longer wavelengths ease the dilemma; radars use dual-PRF and staggered-PRT schemes and dealiasing algorithms to work around it.
| Band | Wavelength (approx.) | Attenuation in heavy rain | Typical role |
|---|---|---|---|
| S | ~10 cm | Small | Long-range surveillance; least affected by heavy rain |
| C | ~5 cm | Moderate | Common national-network radars |
| X | ~3 cm | Large | Short-range, high-resolution, cheaper; gap-filling |
Beam height
The beam travels in a nearly straight line while the Earth curves away. With standard refraction modelled by an effective Earth radius \( k a \) (\( k = 4/3 \), \( a = 6371 \) km), the beam-centre height at range \( r \) and elevation \( \theta \) is
$$h = \sqrt{r^2 + (ka)^2 + 2 r\,ka\sin\theta} - ka \approx \frac{r^2}{2ka} + r\sin\theta$$For a 0.5° beam: about 0.6 km at 50 km, 1.5 km at 100 km, 2.6 km at 150 km. A microburst outflow, only a few hundred metres deep, is invisible to a radar 50 km away. That is why terminal radars sit near airports and why surface networks (LLWAS) are still needed.
Radar signatures for aviation hazards
| Signature | Variable | Meaning |
|---|---|---|
| High-reflectivity core aloft, then descending | Reflectivity (vertical cross-section) | Precipitation core about to fall: precursor of a downburst |
| Divergent couplet near the surface | Radial velocity | Microburst or downburst outflow |
| Fine line ahead of the storm | Reflectivity (weak, thin) | Gust front, marked by insects and dust lifted along the boundary |
| Bow echo | Reflectivity | Line bowing outward under a strong rear-inflow jet; damaging straight-line winds |
| Rotational couplet (adjacent inbound and outbound maxima) | Radial velocity | Mesocyclone; at small scale, tornado vortex signature |
| Hook echo, bounded weak-echo region | Reflectivity | Supercell with a strong rotating updraft |
| Very high Z with low differential reflectivity and reduced correlation | Dual-polarisation | Hail: large tumbling stones look nearly spherical to the radar |
| High echo tops, high vertically integrated liquid (VIL) | Volume products | Intense deep convection, hail potential |
For a rotational couplet with maximum inbound \( V_\text{in} \) and outbound \( V_\text{out} \) separated by distance \( D \), the rotational velocity is \( (V_\text{out} - V_\text{in})/2 \) and the azimuthal shear is \( (V_\text{out} - V_\text{in})/D \). For solid-body rotation, the vertical vorticity is twice the azimuthal shear.
Satellites and Lightning
Geostationary satellite nowcasting
- Rapid scanning (every few minutes over selected regions) lets forecasters follow cumulus growth before radar detects precipitation.
- Cloud-top cooling rate in infrared: a rapidly cooling top means a vigorous updraft; this gives some lead time for convective initiation before the first radar echo.
- Overshooting tops and cold “enhanced-V” patterns signal very strong updrafts and severe storms.
- Fog and low cloud at night: the difference between two infrared channels (near 3.9 µm and 10.8 µm) separates water-droplet fog from clear ground, because small droplets emit less efficiently at the shorter wavelength. In daytime, visible imagery shows fog directly, and its edges’ retreat gives clearance timing.
- Water-vapour imagery shows upper-level dry intrusions, jet streaks and troughs, useful for CAT and storm environment.
Lightning
- Ground-based networks locate cloud-to-ground flashes from their radio emissions; some also detect intra-cloud flashes (total lightning).
- A sudden rise in total flash rate (a lightning jump) often precedes severe weather at the surface, because it tracks the growth of the updraft and the ice-phase charging zone.
- Aerodromes suspend ramp activity and refuelling when lightning is detected within a set radius; nowcasts of lightning onset and end protect ground staff.
Nowcasting Methods
1. Extrapolation
The simplest and, for the first hour or so, often the best. Assume the echo pattern moves without changing (Lagrangian persistence) and move it forward.
| Method | How it works | Limitations |
|---|---|---|
| Cell (object) tracking | Identify storm cells as regions above a reflectivity threshold; match them in successive scans; extrapolate centroids along fitted tracks | Fails when cells split, merge, or are embedded in large systems |
| Area tracking (cross-correlation) | Find the shift that best correlates sub-areas of successive images; build a motion-vector field | Smooths motion; cannot capture growth and decay |
| Optical flow | Compute a dense motion field from image intensity changes and advect the field | Same basic assumption of persistence |
Timing a hazard is a speed–distance calculation: if a gust front moves 10 km between scans 10 min apart (60 km h\(^{-1}\)) and lies 50 km from the airport, it arrives in about 50 min.
Extrapolation cannot forecast initiation of new storms or the growth and decay of existing ones; its skill falls quickly beyond an hour or two.
2. Conceptual models and expert systems
Forecasters and automated systems combine:
- Boundaries: convergence lines (gust fronts, sea-breeze fronts, dry lines) where new storms form, especially where two boundaries collide.
- Environment: CAPE, CIN and vertical wind shear decide whether storms will be single cells, multicells or supercells, and whether outflows will trigger new cells.
- Life cycle: a cell whose reflectivity core is descending is about to produce a downburst; a cell whose top is cooling is intensifying.
3. Rapid-update NWP
High-resolution, convection-permitting models updated every hour or so, with radar reflectivity and radial velocity assimilated, can represent storm initiation and evolution. They need time to adjust (spin-up) and their exact storm placement is uncertain at short lead times.
4. Blending and newer methods
Operational systems blend extrapolation (weighted heavily at the start) with NWP (weighted more as lead time increases). Probabilistic nowcasts use ensembles of perturbed extrapolations. Machine-learning models trained on radar sequences are an active area of research and are entering operations at some services.
graph TD
S["Lead time"] --> E["0 to about 1 h:
extrapolation and
observation-led methods lead"]
S --> M["About 1 to a few hours:
blending; skill of extrapolation
falls, NWP catches up"]
S --> N["Beyond a few hours:
NWP leads"]The exact crossover depends on the phenomenon, region and system; for organised, long-lived systems extrapolation stays useful longer than for scattered, short-lived cells.
Aviation Nowcast Products
| Hazard | Nowcast inputs | Product |
|---|---|---|
| Thunderstorms and squalls at the aerodrome | Radar tracking, satellite, lightning, surface stations | Aerodrome warnings; TREND forecasts on METARs; updated TAFs |
| Microburst and wind shear | Terminal Doppler radar, LLWAS, lidar | Automated alerts to air traffic control and crews (location and expected headwind loss) |
| Fog onset and clearance | Surface temperature and dew point, visibility and RVR trends, night-time fog satellite product, inversion strength | Low-visibility procedures, TREND, warnings |
| En-route convection | Satellite, radar composites, lightning | SIGMETs, convective nowcast charts for route planning |
| Turbulence and icing | EDR reports, pilot reports, satellite, model diagnostics | SIGMETs and AIRMETs, graphical nowcasts |
| Volcanic ash | Satellite detection, dispersion models | Volcanic ash advisories and SIGMETs |
Verifying a Nowcast
Yes/no forecasts (thunderstorm at the aerodrome in the next hour, or not) are verified with a 2 × 2 contingency table:
| Observed yes | Observed no | |
|---|---|---|
| Forecast yes | a (hits) | b (false alarms) |
| Forecast no | c (misses) | d (correct negatives) |
| Score | Formula | Perfect | Meaning |
|---|---|---|---|
| Probability of detection (POD) | \( a/(a + c) \) | 1 | Fraction of events that were warned |
| False alarm ratio (FAR) | \( b/(a + b) \) | 0 | Fraction of warnings that were false |
| Critical success index (CSI, threat score) | \( a/(a + b + c) \) | 1 | Hits over all forecast-or-observed events; ignores d |
| Frequency bias | \( (a + b)/(a + c) \) | 1 | > 1 over-forecasting; < 1 under-forecasting |
CSI is used for rare events because the many correct negatives (d) would make simple accuracy look misleadingly good. There is a trade-off: warning more often raises POD and FAR together.
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| Nowcasting: about 0–2 h, extrapolation-led | vs | Very short-range forecasting: up to about 12 h |
| Reflectivity factor Z: \( \sum D^6 \), mm\(^6\) m\(^{-3}\) | vs | dBZ: \( 10\log_{10}Z \) |
| Maximum unambiguous range: \( c/2\,\mathrm{PRF} \) | vs | Nyquist velocity: \( \lambda\,\mathrm{PRF}/4 \) |
| Second-trip echo: range folding | vs | Velocity aliasing: velocity folding |
| Divergent couplet: microburst outflow | vs | Rotational couplet: mesocyclone |
| Radial velocity: component along the beam only | vs | Actual wind: needs multiple radars or VAD assumptions |
| POD: a/(a + c) | vs | FAR: b/(a + b), not b/(b + d) (that is the false alarm rate) |
| Cell tracking: objects | vs | Cross-correlation / optical flow: fields |
Common Mistakes
Practice
All questions below are original practice questions in CSIR NET style. They are not previous-year questions.
Level 1: Foundation
- Write the Marshall–Palmer Z–R relation.
- Give the formula for the maximum unambiguous range of a pulsed radar.
- What does CSI stand for, and what is its formula?
Answers
- \( Z = 200R^{1.6} \) (Z in mm\(^6\) m\(^{-3}\), R in mm h\(^{-1}\)).
- \( R_\text{max} = c/(2\,\mathrm{PRF}) \).
- Critical success index: \( a/(a + b + c) \).
Level 2: Part B Style
Q1 (practice). An S-band radar (λ = 10 cm) operates at PRF = 1,000 Hz. Its maximum unambiguous range and Nyquist velocity are:
- (a) 300 km and 50 m s\(^{-1}\)
- (b) 150 km and 25 m s\(^{-1}\)
- (c) 150 km and 50 m s\(^{-1}\)
- (d) 75 km and 25 m s\(^{-1}\)
Answer and explanation
(b). \( R_\text{max} = 3\times10^8/(2\times1000) = 150 \) km; \( V_\text{max} = 0.1\times1000/4 = 25 \) m s\(^{-1}\). Check: product = \( c\lambda/8 = 3.75\times10^6\ \mathrm{m^2\,s^{-1}} \).
Q2 (practice). With the radar of Q1, a target has a true radial velocity of +32 m s\(^{-1}\). The displayed velocity is:
- (a) +32 m s\(^{-1}\)
- (b) +7 m s\(^{-1}\)
- (c) −18 m s\(^{-1}\)
- (d) −7 m s\(^{-1}\)
Answer and explanation
(c). The Nyquist interval is ±25 m s\(^{-1}\), width 50. Folding: 32 − 50 = −18 m s\(^{-1}\). A strong outbound wind appears as moderately inbound, a classic aliasing pattern that dealiasing must correct.
Q3 (practice). Using \( Z = 200R^{1.6} \), a reflectivity of 45 dBZ corresponds to a rain rate of about:
- (a) 5 mm h\(^{-1}\)
- (b) 12 mm h\(^{-1}\)
- (c) 24 mm h\(^{-1}\)
- (d) 100 mm h\(^{-1}\)
Answer and explanation
(c). \( Z = 10^{4.5} \approx 31{,}600 \); \( R = (31600/200)^{1/1.6} = 158^{0.625} \approx 23.7 \) mm h\(^{-1}\).
Q4 (practice). A thunderstorm nowcasting scheme over one season gives 30 hits, 10 false alarms, 20 misses and 940 correct negatives. Its POD, FAR and CSI are:
- (a) 0.60, 0.25, 0.50
- (b) 0.75, 0.25, 0.60
- (c) 0.60, 0.33, 0.50
- (d) 0.60, 0.25, 0.97
Answer and explanation
(a). POD = 30/50 = 0.60; FAR = 10/40 = 0.25; CSI = 30/60 = 0.50. Accuracy (a + d)/total = 970/1000 = 0.97 looks excellent only because of the many correct negatives, which is why CSI is preferred for rare events.
Level 3: Part C Style
Q1 (practice). A C-band radar (λ = 5.3 cm) must observe winds up to 30 m s\(^{-1}\) without aliasing.
(i) What minimum PRF is needed? (ii) What is the maximum unambiguous range at that PRF? (iii) What would an S-band radar (λ = 10.6 cm) achieve for the same Nyquist velocity? (iv) What does this say about the choice of wavelength?
Answer and worked solution
- \( \mathrm{PRF} = 4V_\text{max}/\lambda = 4\times30/0.053 \approx \) 2,264 Hz.
- \( R_\text{max} = 3\times10^8/(2\times2264) \approx \) 66 km.
- S-band: PRF = \( 4\times30/0.106 \approx 1{,}132 \) Hz; \( R_\text{max} \approx \) 132.5 km (double, because λ doubled).
- From \( R_\text{max}V_\text{max} = c\lambda/8 \), a longer wavelength relaxes the Doppler dilemma. S-band also attenuates less in heavy rain, but needs a larger antenna for the same beam width and costs more. C-band networks therefore rely on dual-PRF techniques to extend the velocity interval.
Q2 (practice). A terminal radar at the airport shows, at 0.5° elevation, a storm 40 km to the north-west moving towards the airport. Two scans 6 min apart show the leading edge of a thin-line echo moving from 40 km to 34 km. Separately, under a small new cell that has developed about 8 km from the radar, the velocity field shows two adjacent maxima along one radial: −14 m s\(^{-1}\) (towards the radar) on the near side and +16 m s\(^{-1}\) (away) on the far side, 2.5 km apart, at very low altitude.
(i) Identify the thin-line echo and estimate its arrival time at the airport. (ii) Identify the velocity pattern and estimate the horizontal wind change an aircraft would meet flying through it along that radial. (iii) Why must the velocity pattern be measured close to the radar to be reliable for this hazard?
Answer and worked solution
- A thin, weak echo ahead of a storm is a gust front (insects and dust lofted along the outflow boundary). Speed = 6 km in 6 min = 60 km h\(^{-1}\); from 34 km it arrives in about 34 min.
- Air approaching the radar on the near side and receding on the far side, along the same radial, means the air spreads outward from a point between them: a divergent couplet, the signature of a microburst outflow. Velocity difference = 16 − (−14) = 30 m s\(^{-1}\) (about 58 kt) across 2.5 km: an aircraft flying along the radial would go from a 14 m s\(^{-1}\) headwind to a 16 m s\(^{-1}\) tailwind (or the reverse, depending on direction). That is a severe headwind loss.
- The outflow is only a few hundred metres deep. Beam height rises with range (about 0.6 km at 50 km for a 0.5° beam), so only near the radar does the beam sample the shallow outflow; farther away it overshoots it. This is why terminal radars are sited at airports and complemented by LLWAS.
Q3 (practice). Which statements about nowcasting methods are correct?
(A) Extrapolation assumes Lagrangian persistence and so cannot predict new storm initiation. (B) Blending systems give more weight to NWP as lead time increases. (C) A rapid rise in total lightning flash rate can precede severe weather at the surface. (D) Because Z depends on \( D^6 \), a few large drops or hailstones dominate the radar return.
- (a) A and B only
- (b) A, C and D only
- (c) B, C and D only
- (d) A, B, C and D
Answer and worked solution
(d) A, B, C and D.
- (A) Extrapolation moves existing echoes without change; it has no mechanism for new cells. Correct.
- (B) Extrapolation skill falls quickly with lead time while NWP’s skill falls more slowly, so blending weights shift towards NWP. Correct.
- (C) The lightning jump reflects updraft intensification. Correct.
- (D) The sixth-power dependence makes large hydrometeors dominate. Correct.
Quick Revision Table
| Item | Fact |
|---|---|
| Nowcasting (WMO) | Current weather + extrapolation for about 0–2 h; very short range to about 12 h |
| Why | Small, short-lived hazards; limited convective-scale predictability |
| Reflectivity | \( Z = \sum D^6 \) (mm\(^6\) m\(^{-3}\)); dBZ = 10 log Z |
| Marshall–Palmer | \( Z = 200R^{1.6} \); +10 dBZ ≈ ×4.2 in R |
| Doppler shift | \( f_D = 2v_r/\lambda \) |
| Max range | \( c/(2\,\mathrm{PRF}) \) |
| Nyquist velocity | \( \lambda\,\mathrm{PRF}/4 \); aliasing folds by \( 2V_\text{max} \) |
| Doppler dilemma | \( R_\text{max}V_\text{max} = c\lambda/8 \) |
| Bands | S ~10 cm (least attenuation), C ~5 cm, X ~3 cm (most) |
| Beam height | \( \approx r^2/2ka + r\sin\theta \), \( k = 4/3 \) |
| Signatures | Divergent couplet (microburst), fine line (gust front), bow echo, rotational couplet, hook, high Z + low ZDR (hail) |
| Satellite | Cloud-top cooling, overshooting tops, IR fog product, WV imagery |
| Lightning | Lightning jump precedes severe weather; ground-safety stops |
| Methods | Extrapolation (cell tracking, cross-correlation, optical flow), conceptual models, rapid-update NWP, blending |
| Verification | POD = a/(a+c); FAR = b/(a+b); CSI = a/(a+b+c); bias = (a+b)/(a+c) |
Key Takeaways
What’s Next
- Quick Revision — the whole unit on one page.
- Weather Hazards at Take-off, Cruise and Landing — revisit the products (TREND, SIGMET) that nowcasts feed.
- Vertical Stability and Tropical Convection — the environments that decide which storm types to expect.
- Atmospheric Optics — the scattering physics behind radar and visibility.