Dual-polarization differential phase data contain propagation, scattering and instrumental contributions. Processing them to estimate KDP requires more than taking the derivative of an unexamined display.
The gold standard retains raw phase, sign convention, system offset, quality masks, unwrapping choices and independent checks when reporting a derived physical gradient.
eduKateSG · Weather Radar Differential Phase Quality Control
Find your next learning step
Follow the physical phase, calculate the KDP slope, test data quality and examine the validity of a chosen algorithm.
Full chapter index · Practice and transfer · Education root · How X Works
Full chapter index
Define radar phase · Chapters 1–2
Work the calculations · Chapters 3–4
Check the signal · Chapters 5–7
Validate the interpretation · Chapters 8–10
Continue the radar route · Chapters 11–12
CHAPTER 1 OF 12 · Define radar phase
1. Why the recorded phase is not yet a rainfall measurement
Dual-polarization weather radar transmits and receives horizontally and vertically polarized fields. A difference in the received phases can reveal how precipitation affected their propagation, but the directly recorded total differential phase includes more than the desired precipitation-path contribution.
Raw observed differential phase can contain a system offset and a backscatter-phase term associated with the scatterers. The quantity useful for some rain algorithms is the change in propagation differential phase with distance rather than one absolute raw phase number.
NOAA’s dual-polarization guidance explains that KDP is helpful in heavy rain, but this does not mean every numerical phase change has a unique rainfall-rate interpretation. Radar band, hydrometeors and processing assumptions influence the result.
The gold standard saves raw source field, scan geometry, sign convention, wrapped range, quality flags, processing version and corrected phase. A visually smooth map is not independent proof that phase was separated correctly.
Before doing calculations, define the weather region and decide whether observations refer to warm rain, mixed precipitation or nonmeteorological echoes. The physical scope determines which retrieval method may be justified.
In a common two-way radar convention, specific differential phase KDP is one half the derivative of propagation differential phase with respect to range. With phase measured in degrees and range in kilometres, units are degrees per kilometre.
The factor of two belongs to the stated two-way propagation convention. A numerical formula copied into a differently defined phase field may give a physically inconsistent answer even if its arithmetic looks plausible.
A stable constant instrument phase offset shifts the overall phase baseline without changing the ideal derivative. A drifting offset, backscatter differential phase or poor sample can, however, disturb an estimated gradient.
The differential signal is often sampled at many range gates and processed with filtering or regression. Differentiating one noisy adjacent-gate difference amplifies noise; this is why the processing path is part of the measurement.
Instrument and software documentation should identify whether a named PHIDP field is raw total phase, unfolded total phase or corrected propagation phase. A missing processing label can invalidate comparisons across sites and years.
CHAPTER 3 OF 12 · Work the calculations
3. Worked example: phase change gives an average KDP
Assume a fictional corrected propagation phase is 20 degrees at 10 km and 50 degrees at 20 km from the radar. Further assume the phase increases approximately linearly along the valid 10-km interval.
The change is 50 − 20 = 30 degrees and the slope is 30/(20 − 10) = 3 degrees per kilometre. Under the stated two-way definition, the mean KDP estimate is ½×3 = 1.5 degrees/km.
The calculation does not establish rainfall rate because it lacks a wavelength-specific precipitation model and independent validation. It also cannot tell whether the unobserved intermediate gates contain a sharp jump.
If a changing backscatter contribution accounts for part of the 30-degree difference, the result is not a pure propagation KDP. The endpoints must be qualified as appropriate physical measurements first.
Reproducibility requires the two ranges, source phase field, sign, correction status, fit interval, gate quality and uncertainty in the gradient. A named ‘1.5 KDP’ with none of that information is a weak scientific claim.
CHAPTER 4 OF 12 · Work the calculations
4. Worked example: unwrap a 350-to-10-degree phase transition
An instrument may represent differential phase modulo 360 degrees. Take invented adjacent valid values of 350 degrees and 10 degrees where nearby context supports a continuous increasing phase across the wrap boundary.
The raw difference is 10 − 350 = −340 degrees. Under the justified unwrapped interpretation, the second reading becomes 370 degrees and the continuous difference is +20 degrees.
If those gates are 5 km apart, the simple average propagation gradient is 20/5 = 4 degrees/km and KDP is ½×4 = 2 degrees/km under the same two-way convention.
Unwrapping adds an integer multiple of the angular period; it does not prove that the true physical phase evolved smoothly. A backscatter-phase jump or a noisy gate could also explain an abrupt reported transition.
Retain both the original and unfolded values, because the correction is model-based. The appropriate period, sign and phase representation are instrument-dependent metadata that should not be assumed.
A coherent range profile, supporting polarimetric quantities and an appropriate instrument reference give the unwrapping inference its credibility.
CHAPTER 5 OF 12 · Check the signal
5. System phase offset and radar field conventions
A receiving system can introduce a phase offset unrelated to the rainfall along a beam. A stable offset may be estimated and accounted for under a documented calibration or data-processing procedure.
Some radar files store raw differential phase in a signed interval, while other fields may represent the same signal using another direction or angular range. Already-unfolded or corrected products may coexist with raw ones in one dataset.
The radarx technical documentation specifically warns that multiple phase fields can be present and advises explicitly selecting the raw input field when estimating KDP. Processing an already corrected field as raw can corrupt results.
An offset may be harmless to an ideal slope only while it is constant over the interval. An incorrect near-radar baseline or a changing offset can interfere with identifying the valid start of propagation phase and with distinguishing other phase contributions.
Keep timestamps, antenna and receiver calibration history, field name, unit metadata, folding convention and every transformation. Algorithm convergence without clear input provenance is not a guarantee of a physical result.
CHAPTER 6 OF 12 · Check the signal
6. Quality masking uses correlation and signal strength
Copolar correlation coefficient, signal-to-noise and the behavior of related polarimetric fields help decide which range gates contain phase information suitable for analysis.
Low correlation can occur in mixed hydrometeors, clutter or poor measurements. A sharp isolated phase spike at a gate with weak signal should not be interpreted as a narrow zone of extreme rain without corroboration.
NCAR LROSE’s KdpFilt documentation describes identifying valid phase regions through correlation and optional signal-quality conditions, then unfolding, filtering and evaluating possible backscatter-phase effects.
An excessively permissive mask makes noisy measurements dominate derivatives. An excessively conservative mask removes weak but potentially real precipitation and increases gaps, so quality selection must be evaluated for the actual observation and purpose.
Preserve rejected gates and their reasons rather than filling them silently with zeros. A missing estimate and a physically verified zero KDP are different states that should not share the same display code.
CHAPTER 7 OF 12 · Check the signal
7. Range smoothing balances noise against real structure
Estimating a range derivative from raw phase data can amplify random error. Regression, low-pass filtering, regularization and other procedures can estimate a more stable gradient by using information from multiple adjacent gates.
A long smoothing window may suppress random spikes but also flatten a short intense rain region. A short window may preserve sharper detail but return a large noisy KDP estimate when individual phase samples are uncertain.
The radarx KDP notebook describes adapting the range fitting window to reflectivity conditions, using shorter windows in stronger precipitation and longer windows in weaker conditions when enough valid gates are available.
Comparing a raw derivative with a smoothed result is useful but not definitive. A very smooth result could be physically biased by oversmoothing; a detailed one could contain statistical artifacts.
Report gate spacing, window length, weighting, number of valid samples, interpolation rules and algorithm version. Changes to these settings can change the derived result even when raw weather data are identical.
The gold standard tests the chosen processing scale on representative valid cases rather than selecting it after seeing a favorable rainfall map.
CHAPTER 8 OF 12 · Validate the interpretation
8. Backscatter phase and melting layers complicate gradients
Total phase can include a backscatter differential-phase contribution caused by target scattering properties. Hail, mixed-phase particles and the melting layer may create phase features that are not equivalent to a smooth propagation effect.
A step in total phase may yield a large apparent derivative if treated as ordinary propagation. The location should therefore be checked against correlation coefficient, reflectivity, differential reflectivity and atmospheric temperature profiles.
NOAA research on differential phase discusses the importance of melting-layer scattering and instrument phase effects. Those influences matter when inferring a physical rain-specific gradient from an observed total signal.
Some rain-only retrieval methods explicitly assume monotonic increasing propagation phase. The Py-ART Maesaka method documents that its formulation is limited to suitable warm cloud or below-melting-layer rainfall cases.
Applying those assumptions indiscriminately to strongly mixed or hail-bearing samples may create spurious estimates, however well the numerical optimization runs. Report where the model was valid and where the phase was left uncertain.
CHAPTER 9 OF 12 · Validate the interpretation
9. Negative values and constrained numerical methods
Some KDP retrievals can return small negative values because of measurement noise, profile filtering, offset issues or a different phase convention. Their meaning depends on the instrument and precipitation physics.
A particular variational technique may constrain the retrieved quantity to be nonnegative under its chosen warm-rain model. Py-ART’s Maesaka implementation documents a squared internal variable and a monotonic phase assumption.
A nonnegative output from such a model is therefore not independent proof that every measured location has positive propagation differential phase. It is partly the result of the model’s allowed solution set.
Replacing all negative or missing estimates by zero can alter statistical bias and rainfall accumulation. If a user needs rainfall rates, the estimation method and how it handles invalid or model-constrained gates must be documented.
Compare methods on independent physical and numerical test cases with different noise, rain types and phase artifacts. A universal claim of one algorithm’s superiority needs evidence well beyond a visually attractive image.
CHAPTER 10 OF 12 · Validate the interpretation
10. Worked example: independent KDP estimation errors
Imagine four fictional independent reference KDP values [1.0, 2.0, 0.5, 3.0] degrees/km and matched algorithm outputs [1.2, 1.8, 0.6, 2.6].
Errors defined as output minus reference are [+0.2, −0.2, +0.1, −0.4] degrees/km. Their mean is (0.2−0.2+0.1−0.4)/4 = −0.075 degrees/km.
Mean absolute error is (0.2+0.2+0.1+0.4)/4 = 0.225 degrees/km. Root mean squared error is sqrt[(0.04+0.04+0.01+0.16)/4] = sqrt0.0625 = 0.25 degrees/km.
These are constructed results, not measured performance of any operational radar. Truly independent reference KDP is difficult to establish because it is itself an inferred differential quantity.
A proper test may combine calibrated scenes, simulation, paired observations and ground rain gauges, each with representativeness limits. It should report error by rain regime, range, radar band and gate quality.
The fraction of excluded or uncertain gates is also essential. Reporting accuracy only on easy gates can exaggerate reliability.
CHAPTER 11 OF 12 · Continue the radar route
11. Practice, FAQs and a changed-weather transfer
Practice A: Corrected propagation phase rises from 20 to 50 degrees over 10 km. Calculate a 3-degree/km phase slope and KDP = 1.5 degrees/km under the stated convention.
Practice B: A wrapped 350-to-10-degree transition is interpreted as 350-to-370 over 5 km. Confirm 2 degrees/km KDP and identify a competing explanation for the apparent wrap.
Practice C: Four estimates give errors [+0.2,−0.2,+0.1,−0.4] degrees/km. Verify MAE 0.225 and RMSE 0.25, then explain why excluded gates must also be counted.
FAQ — Is raw total phase the same as propagation phase? No. Instrument and scatterer terms can contribute.
FAQ — Is negative KDP always a physically negative rainfall rate? No. Processing, noise, conventions and target type matter.
FAQ — Does smoother KDP imply more accurate radar rain estimates? Not necessarily. Smoothing can remove real gradients as well as noise.
Transfer the methodology from warm rain to mixed hail and melting snow. Reconsider unwrapping, backscatter-phase effects, correlation masks and the valid range of the algorithm.
Verified linked eduKateSG subject routes:
- The Gold Standard of Weather Radar Dual-Polarization Hydrometeor Classification — particle-type context
- The Gold Standard of Weather Radar Rainfall Estimation — rain-rate inference
- The Gold Standard of Weather Radar Attenuation Correction — different propagation correction
- The Gold Standard of Weather Radar Bright-Band Correction — melting-layer effects
- The Gold Standard of Meteorology — physical precipitation
- The Gold Standard of Weather Forecast Verification — comparison metrics
- The Gold Standard of Statistics — error and model selection
CHAPTER 12 OF 12 · Continue the radar route
12. Original sources and connected Gold Standard learning
NOAA’s dual-polarization descriptions establish the physical meaning and meteorological uses of phase-derived products. NCAR’s LROSE, radarx and DOE Py-ART document the operational distinctions among raw phase, quality masks, unfolding and specific differential-phase retrieval.
The linked eduKateSG articles cover hydrometeor classification, reflectivity-related rainfall estimation, attenuation and the broader statistical questions that connect one measurement to a defensible interpretation.
All phase numbers, radar ranges, gradients and verification scores in this article are fictional teaching examples, not operational weather warnings or live rainfall products.
A reproducible independent project should store source field identity, phase representation, system-offset estimate, gate-quality mask, unfolding choices, range filter, scattering treatment, KDP output and valid independent comparison cases.
Official and scientific method references:
- NOAA JetStream — Dual Polarization — specific differential phase in weather
- NOAA NCEI — Radar Products — KDP product meaning
- NCAR LROSE — KdpFilt Parameters — quality masking, unfolding and differentiation
- radarx — Differential Phase and KDP — raw fields and window processing
- DOE Py-ART — Maesaka KDP Retrieval — model-specific limits
- NOAA Library — Transmitted Differential Phase Research — system and scattering phase
The immutable Clementi Mathematics reference is an educational model, not weather radar operational guidance.