A radar colour labelled 45 dBZ is not already a surface rain rate. It first corresponds to linear reflectivity Z, then a chosen relationship or algorithm produces an estimated R.
The central example converts 45 dBZ to Z≈31,623 mm^6/m^3. With Z=200R^1.6, the estimate is about 23.7 mm/h. A different justified relation produces another rate from the same reflectivity.
National Weather Service guidance grounds the explanation of Z–R ambiguity, drop-size distributions, bright band, hail, beam geometry and dual-polarization evidence. No live warning or flood advice is inferred.
The changed-context checks keep dBZ fixed while precipitation type, coefficients, range or attenuation changes. They show why correct arithmetic cannot make the underlying conversion unique.
This specialist guide belongs to the Weather root. It also connects to The Gold Standard of Education | Learning That Transfers into Life and the How X Works hub. The worked cases are original teaching examples; their numbers are stipulated so that the reasoning can be checked.
eduKateSG · Weather
Find your next learning step
Radar reflectivity describes returned electromagnetic power through a reflectivity factor. Rainfall rate describes water-depth flux. An empirical or polarimetric retrieval connects them, with assumptions about drops, phase, sampling height and quality control.
ROUTE 1 · CHAPTERS 1–6
Define the quantities
Name the objects, units and first complete worked example.
ROUTE 2 · CHAPTERS 7–12
Trace the mechanism
Follow scoping, spectra, radar physics or laboratory preparation.
ROUTE 5 · CHAPTERS 24–28
Transfer and continue
Repair shortcuts, change context and follow verified routes.
Chapter contents
Define the quantities · Chapters 1–6
1. Name reflectivity and rain rate
Trace the mechanism · Chapters 7–12
8. Track beam height with range
9. Account for partial beam filling
Control the method · Chapters 13–18
13. Separate instantaneous rate from accumulation
14. Align radar and gauge clocks
15. Align radar and gauge locations
16. Preserve gauge limitations
Audit the evidence · Chapters 19–23
19. Calibrate without erasing structure
21. Protect extreme-rain decisions
Transfer and continue · Chapters 24–28
24. Work the changed-context example
26. Connect to the gauge-undercatch route
Begin with the object being measured. Radar reflectivity factor Z describes returned power under radar-scattering assumptions; rainfall rate R describes liquid-water depth flux at the surface. Preserve reflectivity factor Z, logarithmic dBZ and rain-rate units so the conversion does not disappear inside a colour palette.
Work the numbers in a visible order. One radar pixel reports 45 dBZ while a gauge records millimetres over an interval. State the chosen Z–R coefficients before solving for R.
Check the claim against the method. Keep unit, altitude, scan time and surface window. Compare radar volume, time and location with the surface reference rather than calling them the same observation.
Now change one consequential condition. Hold dBZ fixed and change the drop-size distribution. Change the drop-size distribution or hydrometeor class and recompute.
Keep the limit explicit. The two quantities are connected by a model, not an identity. A radar-derived rate is an estimate under a relationship, not a direct bucket measurement at the ground.
Separate the two quantities before comparing them. dBZ is logarithmic: dBZ=10 log10 Z when Z uses the conventional millimetre-six per metre-three unit. Remember that reflectivity weights large drops strongly, while rainfall rate depends on liquid volume flux and fall speed.
Use this original teaching case. At 45 dBZ, Z is 10^4.5, about 31,623. Convert from dBZ to linear Z before applying a power law.
A reproducible check asks for more than the final number. Show the exponent before applying any Z–R relation. Check bright band, hail, attenuation, beam height and clutter flags.
Test whether the reasoning transfers. Increase to 48 dBZ and recompute. Hold dBZ fixed while changing the Z–R relation.
Do not extend the conclusion past its evidence. Adding three dB roughly doubles Z, not R. One reflectivity value can correspond to more than one physically plausible rainfall rate.
Name the mechanism before interpreting the appearance. A common power-law form is Z=aR^b, so R=(Z/a)^(1/b) after coefficients are chosen. Separate instantaneous rate, scan-volume estimate and accumulated depth over an interval.
Trace this stipulated example. Using a=200 and b=1.6 with Z=31,623 gives about 23.7 mm/h. Keep scan time and accumulation window visible.
Preserve the settings that control the result. Preserve coefficients and exponent. Use gauges as a different instrument with their own siting and undercatch limits.
Change one input and recompute. Use another justified relation and compare. Move the same storm farther from radar and examine beam sampling.
State what this example cannot establish. The result is a radar estimate under that relation. The example does not authorise flood or safety decisions without current official products and local context.
Keep cause, observation and decision in different ledger columns. Reflectivity weights drop diameter strongly, while rain rate also depends on liquid volume and fall speed; different distributions can share Z. Treat dual-polarization variables as additional evidence about target shape and phase, not as a magical direct rain gauge.
Follow the worked decision. A few large drops and many smaller drops produce the same stipulated reflectivity but different flux. Show which hydrometeor assumption drives the selected algorithm.
Audit the evidence trail. Use disdrometer or dual-pol evidence where available. Inspect spatial gradients and quality-control masks before averaging pixels.
Run the same decision in a changed context. Change only the assumed distribution. Replace liquid rain with melting snow in the model and predict the bias.
Protect the boundary of the claim. No algebra can recover information that the single reflectivity measurement did not contain. A better classifier reduces some ambiguity but does not remove sampling uncertainty.
Use a short answer only after its denominator and scope are fixed. Operational algorithms may select different relations or polarimetric methods because microphysics differs across precipitation regimes. Use the observation chain—pulse, returned power, reflectivity, classification, rate relation, accumulation—so each transformation remains auditable.
Calculate with every unit exposed. The 45 dBZ pixel receives one estimate in stratiform classification and another in convection. Carry coefficient and measurement uncertainty into the reported range.
Verify the intermediate evidence. Record classification source and transition rule. Compare like time windows and areas.
Repeat the test after one controlled change. Move the classification boundary one pixel. Test a convective and a stratiform case separately.
A useful explanation also records uncertainty. A regime label can be uncertain near mixed areas. Radar imagery can be operationally useful even when no single pixel is an exact surface rainfall truth.
Begin with the object being measured. Hail can produce high reflectivity that a liquid-rain Z–R relation converts into implausibly large rain rates. Preserve reflectivity factor Z, logarithmic dBZ and rain-rate units so the conversion does not disappear inside a colour palette.
Work the numbers in a visible order. A 60 dBZ core contains hail signatures in dual-pol fields. State the chosen Z–R coefficients before solving for R.
Check the claim against the method. Check hydrometeor classification and warnings. Compare radar volume, time and location with the surface reference rather than calling them the same observation.
Now change one consequential condition. Apply the rain-only relation deliberately as a failure demonstration. Change the drop-size distribution or hydrometeor class and recompute.
Keep the limit explicit. High dBZ is not a direct surface rain-rate measurement. A radar-derived rate is an estimate under a relationship, not a direct bucket measurement at the ground.
Separate the two quantities before comparing them. Melting snow can create enhanced reflectivity near the freezing level and overestimate surface rainfall when sampled as liquid rain. Remember that reflectivity weights large drops strongly, while rainfall rate depends on liquid volume flux and fall speed.
Use this original teaching case. A distant beam intersects the melting layer and displays a bright ring. Convert from dBZ to linear Z before applying a power law.
A reproducible check asks for more than the final number. Use vertical profile and melting-layer products. Check bright band, hail, attenuation, beam height and clutter flags.
Test whether the reasoning transfers. Lower the beam or compare a closer range. Hold dBZ fixed while changing the Z–R relation.
Do not extend the conclusion past its evidence. The bright band is an atmospheric layer effect, not a circular storm boundary. One reflectivity value can correspond to more than one physically plausible rainfall rate.
Name the mechanism before interpreting the appearance. Earth curvature and beam geometry make a radar sample progressively higher above ground with distance. Separate instantaneous rate, scan-volume estimate and accumulated depth over an interval.
Trace this stipulated example. The same elevation angle samples low rain nearby and ice aloft far away. Keep scan time and accumulation window visible.
Preserve the settings that control the result. Calculate beam centre and width for the site. Use gauges as a different instrument with their own siting and undercatch limits.
Change one input and recompute. Move the target farther from radar. Move the same storm farther from radar and examine beam sampling.
State what this example cannot establish. Surface gauges and elevated radar volumes need not observe the same hydrometeors. The example does not authorise flood or safety decisions without current official products and local context.
Keep cause, observation and decision in different ledger columns. A beam that only partly intersects an intense narrow cell averages precipitation and clear space within its volume. Treat dual-polarization variables as additional evidence about target shape and phase, not as a magical direct rain gauge.
Follow the worked decision. A convective shaft fills one quarter of the beam. Show which hydrometeor assumption drives the selected algorithm.
Audit the evidence trail. Compare higher-resolution scans or neighbouring radars. Inspect spatial gradients and quality-control masks before averaging pixels.
Run the same decision in a changed context. Widen the cell while keeping local intensity fixed. Replace liquid rain with melting snow in the model and predict the bias.
Protect the boundary of the claim. The pixel value is a volume sample, not the cell maximum. A better classifier reduces some ambiguity but does not remove sampling uncertainty.
Use a short answer only after its denominator and scope are fixed. Heavy rain can weaken a radar signal along the path, especially at shorter wavelengths, causing downstream underestimation without correction. Use the observation chain—pulse, returned power, reflectivity, classification, rate relation, accumulation—so each transformation remains auditable.
Calculate with every unit exposed. A strong near-radar cell shadows a second cell behind it. Carry coefficient and measurement uncertainty into the reported range.
Verify the intermediate evidence. Inspect attenuation correction and differential phase. Compare like time windows and areas.
Repeat the test after one controlled change. Remove the foreground cell in a simulation. Test a convective and a stratiform case separately.
A useful explanation also records uncertainty. Correction introduces its own assumptions and failure limits. Radar imagery can be operationally useful even when no single pixel is an exact surface rainfall truth.
CHAPTER 11 OF 28 · Trace the mechanism
11. Remove clutter and anomalous propagation
Begin with the object being measured. Ground returns, birds and refracted beams can create reflectivity without rainfall. Preserve reflectivity factor Z, logarithmic dBZ and rain-rate units so the conversion does not disappear inside a colour palette.
Work the numbers in a visible order. A stationary echo persists on clear-air scans near terrain. State the chosen Z–R coefficients before solving for R.
Check the claim against the method. Use velocity, correlation and clutter filters. Compare radar volume, time and location with the surface reference rather than calling them the same observation.
Now change one consequential condition. Compare with a gauge and satellite context. Change the drop-size distribution or hydrometeor class and recompute.
Keep the limit explicit. A coloured radar echo is not automatically precipitation. A radar-derived rate is an estimate under a relationship, not a direct bucket measurement at the ground.
CHAPTER 12 OF 28 · Trace the mechanism
12. Use dual polarization as additional evidence
Separate the two quantities before comparing them. Differential reflectivity, correlation coefficient and specific differential phase help infer target properties and improve precipitation estimation. Remember that reflectivity weights large drops strongly, while rainfall rate depends on liquid volume flux and fall speed.
Use this original teaching case. Two 45 dBZ pixels differ strongly in correlation coefficient and KDP. Convert from dBZ to linear Z before applying a power law.
A reproducible check asks for more than the final number. Inspect the combined product set. Check bright band, hail, attenuation, beam height and clutter flags.
Test whether the reasoning transfers. Hold reflectivity fixed while changing polarimetric variables. Hold dBZ fixed while changing the Z–R relation.
Do not extend the conclusion past its evidence. Dual-pol reduces ambiguity but does not eliminate sampling mismatch. One reflectivity value can correspond to more than one physically plausible rainfall rate.
CHAPTER 13 OF 28 · Control the method
13. Separate instantaneous rate from accumulation
Name the mechanism before interpreting the appearance. Rain rate is a flux at a time; accumulation integrates rate over a period with scan and interpolation choices. Separate instantaneous rate, scan-volume estimate and accumulated depth over an interval.
Trace this stipulated example. A 24 mm/h rate lasting ten minutes contributes about 4 mm if constant. Keep scan time and accumulation window visible.
Preserve the settings that control the result. Integrate over the actual duration. Use gauges as a different instrument with their own siting and undercatch limits.
Change one input and recompute. Make the peak last two minutes. Move the same storm farther from radar and examine beam sampling.
State what this example cannot establish. A peak rate cannot be reported as a daily total. The example does not authorise flood or safety decisions without current official products and local context.
Keep cause, observation and decision in different ledger columns. A radar scan time and a tipping-bucket interval must be matched before residuals are interpreted. Treat dual-polarization variables as additional evidence about target shape and phase, not as a magical direct rain gauge.
Follow the worked decision. The radar volume ends at 14:05 while the gauge total covers 14:00–14:10. Show which hydrometeor assumption drives the selected algorithm.
Audit the evidence trail. Aggregate both to a common window. Inspect spatial gradients and quality-control masks before averaging pixels.
Run the same decision in a changed context. Shift the window five minutes. Replace liquid rain with melting snow in the model and predict the bias.
Protect the boundary of the claim. Timing mismatch can mimic conversion bias. A better classifier reduces some ambiguity but does not remove sampling uncertainty.
Use a short answer only after its denominator and scope are fixed. Wind advects falling hydrometeors between the sampled radar volume and the surface. Use the observation chain—pulse, returned power, reflectivity, classification, rate relation, accumulation—so each transformation remains auditable.
Calculate with every unit exposed. Aloft rain above the gauge drifts two kilometres before landing. Carry coefficient and measurement uncertainty into the reported range.
Verify the intermediate evidence. Use wind and fall-time context. Compare like time windows and areas.
Repeat the test after one controlled change. Reverse the wind direction. Test a convective and a stratiform case separately.
A useful explanation also records uncertainty. A point gauge is not perfect truth for an entire radar pixel. Radar imagery can be operationally useful even when no single pixel is an exact surface rainfall truth.
Begin with the object being measured. Gauges have siting, wind, blockage, splash and mechanical errors that differ from radar errors. Preserve reflectivity factor Z, logarithmic dBZ and rain-rate units so the conversion does not disappear inside a colour palette.
Work the numbers in a visible order. One gauge undercatches in strong wind while neighbouring gauges agree better with radar. State the chosen Z–R coefficients before solving for R.
Check the claim against the method. Check maintenance and quality flags. Compare radar volume, time and location with the surface reference rather than calling them the same observation.
Now change one consequential condition. Shelter the gauge according to the comparison design. Change the drop-size distribution or hydrometeor class and recompute.
Keep the limit explicit. Disagreement does not identify radar as the only faulty instrument. A radar-derived rate is an estimate under a relationship, not a direct bucket measurement at the ground.
Separate the two quantities before comparing them. Averaging dBZ values directly differs from averaging linear Z because the logarithm is nonlinear. Remember that reflectivity weights large drops strongly, while rainfall rate depends on liquid volume flux and fall speed.
Use this original teaching case. Two subpixels are 30 and 50 dBZ. Convert from dBZ to linear Z before applying a power law.
A reproducible check asks for more than the final number. Convert to linear Z before physically meaningful averaging. Check bright band, hail, attenuation, beam height and clutter flags.
Test whether the reasoning transfers. Change their area weights. Hold dBZ fixed while changing the Z–R relation.
Do not extend the conclusion past its evidence. The correct aggregation depends on the quantity and decision. One reflectivity value can correspond to more than one physically plausible rainfall rate.
Name the mechanism before interpreting the appearance. Estimated a and b values vary across storms and datasets, so a single rate should not hide model sensitivity. Separate instantaneous rate, scan-volume estimate and accumulated depth over an interval.
Trace this stipulated example. Three justified relations produce 21,24 and 29 mm/h from the same Z. Keep scan time and accumulation window visible.
Preserve the settings that control the result. Report a range or method sensitivity. Use gauges as a different instrument with their own siting and undercatch limits.
Change one input and recompute. Refit after excluding hail-contaminated samples. Move the same storm farther from radar and examine beam sampling.
State what this example cannot establish. Coefficient precision does not make the physical relation unique. The example does not authorise flood or safety decisions without current official products and local context.
Keep cause, observation and decision in different ledger columns. Gauge adjustment can reduce broad bias but may degrade local fields when gauge sampling and storm gradients differ. Treat dual-polarization variables as additional evidence about target shape and phase, not as a magical direct rain gauge.
Follow the worked decision. One multiplicative factor corrects the domain mean yet worsens an isolated cell. Show which hydrometeor assumption drives the selected algorithm.
Audit the evidence trail. Validate on held-out gauges and cases. Inspect spatial gradients and quality-control masks before averaging pixels.
Run the same decision in a changed context. Apply spatially varying adjustment with constraints. Replace liquid rain with melting snow in the model and predict the bias.
Protect the boundary of the claim. A lower average error can conceal worse extremes. A better classifier reduces some ambiguity but does not remove sampling uncertainty.
Use a short answer only after its denominator and scope are fixed. Reflectivity changes with height as precipitation grows, melts, evaporates or changes phase. Use the observation chain—pulse, returned power, reflectivity, classification, rate relation, accumulation—so each transformation remains auditable.
Calculate with every unit exposed. The lowest usable tilt is blocked, so an elevated profile is projected downward. Carry coefficient and measurement uncertainty into the reported range.
Verify the intermediate evidence. Use vertical-profile correction with uncertainty. Compare like time windows and areas.
Repeat the test after one controlled change. Compare warm rain and melting-layer cases. Test a convective and a stratiform case separately.
A useful explanation also records uncertainty. A column correction is not universal across storm types. Radar imagery can be operationally useful even when no single pixel is an exact surface rainfall truth.
Begin with the object being measured. Flash-flood decisions need current warnings, terrain, drainage, soil and accumulation context beyond a single radar-derived rate. Preserve reflectivity factor Z, logarithmic dBZ and rain-rate units so the conversion does not disappear inside a colour palette.
Work the numbers in a visible order. A high rate over a tiny area differs from persistent basin-wide rain. State the chosen Z–R coefficients before solving for R.
Check the claim against the method. Use official multi-source products. Compare radar volume, time and location with the surface reference rather than calling them the same observation.
Now change one consequential condition. Keep rate fixed but extend duration and area. Change the drop-size distribution or hydrometeor class and recompute.
Keep the limit explicit. This teaching calculation is not a warning service. A radar-derived rate is an estimate under a relationship, not a direct bucket measurement at the ground.
Separate the two quantities before comparing them. Record radar, wavelength, scan, tilt, range, quality masks, hydrometeor class, Z–R relation, gauge window and aggregation. Remember that reflectivity weights large drops strongly, while rainfall rate depends on liquid volume flux and fall speed.
Use this original teaching case. The 45 dBZ example carries every transformation to 23.7 mm/h. Convert from dBZ to linear Z before applying a power law.
A reproducible check asks for more than the final number. Retain intermediate Z and coefficients. Check bright band, hail, attenuation, beam height and clutter flags.
Test whether the reasoning transfers. Reprocess with another relation. Hold dBZ fixed while changing the Z–R relation.
Do not extend the conclusion past its evidence. Without the receipt, two equal rates may come from incomparable methods. One reflectivity value can correspond to more than one physically plausible rainfall rate.
Name the mechanism before interpreting the appearance. “45 dBZ means 24 mm/h” hides the empirical relation, altitude and precipitation type. Separate instantaneous rate, scan-volume estimate and accumulated depth over an interval.
Trace this stipulated example. Rewrite it as a conditional estimate with named coefficients. Keep scan time and accumulation window visible.
Preserve the settings that control the result. Check bright band, hail and range. Use gauges as a different instrument with their own siting and undercatch limits.
Change one input and recompute. Place the same dBZ in a melting layer. Move the same storm farther from radar and examine beam sampling.
State what this example cannot establish. The repaired sentence is longer because the evidence has conditions. The example does not authorise flood or safety decisions without current official products and local context.
Keep cause, observation and decision in different ledger columns. At constant 45 dBZ, a different relation or hydrometeor class changes R even though displayed reflectivity is unchanged. Treat dual-polarization variables as additional evidence about target shape and phase, not as a magical direct rain gauge.
Follow the worked decision. Compare the baseline 23.7 mm/h with a second stated model. Show which hydrometeor assumption drives the selected algorithm.
Audit the evidence trail. Show both equations. Inspect spatial gradients and quality-control masks before averaging pixels.
Run the same decision in a changed context. Keep relation fixed but add attenuation correction. Replace liquid rain with melting snow in the model and predict the bias.
Protect the boundary of the claim. Changed output does not mean the original arithmetic was wrong; the model changed. A better classifier reduces some ambiguity but does not remove sampling uncertainty.
Use a short answer only after its denominator and scope are fixed. Students can convert dBZ, apply two relations and explain why neither is direct surface truth. Use the observation chain—pulse, returned power, reflectivity, classification, rate relation, accumulation—so each transformation remains auditable.
Calculate with every unit exposed. Use three pixels with quality flags. Carry coefficient and measurement uncertainty into the reported range.
Verify the intermediate evidence. Grade the reasoning ledger, not only numeric answers. Compare like time windows and areas.
Repeat the test after one controlled change. Swap a rain pixel for hail. Test a convective and a stratiform case separately.
A useful explanation also records uncertainty. The task should not simulate live severe-weather advice. Radar imagery can be operationally useful even when no single pixel is an exact surface rainfall truth.
Begin with the object being measured. Radar-conversion error and gauge undercatch are different parts of a comparison. Preserve reflectivity factor Z, logarithmic dBZ and rain-rate units so the conversion does not disappear inside a colour palette.
Work the numbers in a visible order. A radar estimate appears high partly because the gauge misses wind-driven precipitation. State the chosen Z–R coefficients before solving for R.
Check the claim against the method. Keep instrument errors separate. Compare radar volume, time and location with the surface reference rather than calling them the same observation.
Now change one consequential condition. Correct only the gauge side. Change the drop-size distribution or hydrometeor class and recompute.
Keep the limit explicit. Do not tune the radar model to compensate blindly for gauge bias. A radar-derived rate is an estimate under a relationship, not a direct bucket measurement at the ground.
CHAPTER 27 OF 28 · Transfer and continue
27. Transfer the method across storm phases
Separate the two quantities before comparing them. A single event can move through snow, melting precipitation, stratiform rain and embedded convection, so one retrieval assumption need not remain valid. Remember that reflectivity weights large drops strongly, while rainfall rate depends on liquid volume flux and fall speed.
Use this original teaching case. A time series shows a bright-band enhancement before surface rain intensifies. Convert from dBZ to linear Z before applying a power law.
A reproducible check asks for more than the final number. Classify each interval with the available polarimetric and temperature evidence before integrating. Check bright band, hail, attenuation, beam height and clutter flags.
Test whether the reasoning transfers. Recompute the total after excluding a hail-contaminated interval and state both results. Hold dBZ fixed while changing the Z–R relation.
Do not extend the conclusion past its evidence. Do not label every reflectivity change as a change in surface rain rate. One reflectivity value can correspond to more than one physically plausible rainfall rate.
Name the mechanism before interpreting the appearance. NWS guidance describes reflectivity-to-rate conversion, non-unique drop distributions, bright band, hail and dual-polarization evidence. Separate instantaneous rate, scan-volume estimate and accumulated depth over an interval.
Trace this stipulated example. Read the operational explanations with their limits. Keep scan time and accumulation window visible.
Preserve the settings that control the result. Keep source date and product context. Use gauges as a different instrument with their own siting and undercatch limits.
Change one input and recompute. Continue to forecast horizons or cloud cover only when that is the reader question. Move the same storm farther from radar and examine beam sampling.
State what this example cannot establish. This guide owns reflectivity versus rain rate, not forecast skill. The example does not authorise flood or safety decisions without current official products and local context.
Continue through the verified reading route
Rain-gauge undercatch and true precipitation change
Rain-gauge undercatch and true precipitation change. Use this when the surface reference instrument is biased by wind or siting.
Forecast update frequency and forecast horizon
Forecast update frequency and forecast horizon. Use this when the reader asks about model cycles and lead time rather than observation physics.
Cloud cover and sunshine duration
Cloud cover and sunshine duration. Use this for sky fraction and recorded bright sunshine.
Frequently asked questions
Is 45 dBZ a fixed rain rate?
No. The rate depends on the retrieval relationship, precipitation type and sampling conditions.
Why can hail look like extreme rain?
Hail can return strong radar power, so a liquid-rain relationship can produce an invalidly large rate.
Is a gauge perfect ground truth?
No. Gauges measure a different sample and can have wind, siting, timing and mechanical errors.
Do dual-polarization products remove every ambiguity?
They add evidence about target shape and phase, but sampling geometry and algorithm uncertainty remain.
Authoritative sources and further reading
NWS: WSR-88D Radar Rainfall Estimation. Reflectivity, Z–R conversion, drop-size ambiguity and operational limitations.
NWS: Dual-Polarization Radar. Additional variables, hydrometeor classification, melting-layer detection and precipitation products.
NWS Glossary: Bright Band. Why melting snow can enhance reflectivity and bias rainfall estimates.
Return to the Weather subject guide, The Gold Standard of Education, or the How X Works hub.