Field Reference // Precipitation Measurement Physics

Gauge Bias Correction &
Systematic Undercatch

Primary source: Adam & Lettenmaier (2002/2003) · WRS Tech. Report No. 171, Univ. of Washington
WMO Intercomparison: Goodison et al. (1998) · Modern update: WMO-SPICE / Kochendorfer et al. (2017)
⚠ CAUTION — Reference period 1979–1998. Equations remain valid for manual gauge types (NWS 8", Tretyakov, etc.). Not suitable for climate trend detection. For automated weighing gauges use WMO-SPICE transfer functions.

Plain English: What This Paper Is Actually About

11.2%
global precip increase after correction
95%+
max undercatch, 80°N DJF
7,878
weather stations analyzed
30
countries receiving snow correction

The Core Problem

Every rain gauge on Earth reads too low. Not randomly — systematically. The biggest reason by far is wind. When snow falls in wind, the air flowing over the gauge orifice deflects snowflakes away before they can fall in. At 6 m/s, some unshielded gauges miss more than half of all snowfall. You can't see this error in the recorded number — it just looks like less snow fell.

Smaller problems: water sticking to gauge walls that never gets counted (wetting loss), and water evaporating before observation (evaporation loss). Rain in wind is also affected, just far less severely than snow.

Why It Matters

Global precipitation datasets are built from these gauges. Hydrological models, Arctic water balance studies, river runoff forecasts — they're all working with numbers that are systematically too low. In snowy high-latitude regions, which cover roughly half the Northern Hemisphere land area, the bias is severe enough to fundamentally distort water budget calculations.

The WMO Intercomparison (1986–1993)

The WMO ran a massive field experiment using a special "ground truth" reference gauge called the Double Fence International Reference (DFIR) — a gauge protected by concentric wind-fence rings that catch nearly everything. Each country put its standard gauge next to a DFIR, measured thousands of snow events side by side, and developed equations describing how much each gauge type misses as wind increases. These equations (21 of them, covering the US, Russia, Norway, Japan, China, Finland, etc.) form the backbone of this paper.

What Adam & Lettenmaier Built

They took those equations and applied them to 5 years of daily weather station data (1994–1998) from 7,878 stations worldwide. For each station, each day: given the wind speed, temperature, and gauge type here, how much precipitation was missed? They summed these daily corrections into monthly factors, then gridded them globally at ½° resolution (~55 km). The result: 12 monthly maps showing how much to multiply any precipitation record by to correct for known biases.

The Three Correction Types

CorrectionAnnualDJF (Winter)JJA (Summer)Priority
Wind + Snow+4.5%+8.6%+1.0%★★★ Dominant
Wind + Rain+3.9%+3.7%+3.6%★★ Moderate
Wetting Losses+2.8%+2.9%+2.7%★ Uniform
All Combined+11.2%+15.2%+7.3%

Honest Limitations Acknowledged

  • Only 5 years of wind data available (1994–1998); corrections are climatological averages
  • USA gets the lowest reliability score (30/70) due to gauge heterogeneity and airport wind bias
  • Zero snow depth assumed — ignoring the effect of accumulated snow raising effective gauge height
  • All US stations assumed unshielded — some have Alter shields, causing slight overcorrection there
  • All anemometers assumed at 10m, fully exposed — not always true
  • Cannot be used for climate trend detection — bias corrections are climatological, not year-specific

Key Numbers & Results

11.2%
Mean annual global increase
15.2%
DJF (boreal winter)
7.3%
JJA (boreal summer)
95%+
Max correction at 80°N DJF

Comparison vs. Yang et al. Station Studies

Yang et al. performed station-specific corrections with full metadata (actual gauge heights, wind sensor heights, shielding status) for Alaska, Siberia, and Greenland. Adam's gridded estimates are consistently slightly higher (less corrected) because assumptions rather than exact metadata were used.

RegionStationsMean Diff.Std. Dev.Key Driver
Siberia58+1.6% (Adam higher CR)4.5%Liquid undercatch eq. difference
Greenland12+2.5%6.0%Wind sensor height assumption
Alaska (all)9+3.5%12.0%Shield status unknown — 2 Alter-shielded
Alaska (unshielded only)7+7.9%5.8%Gauge/anemometer height assumptions

Comparison vs. Legates & Willmott (1990)

Key Finding

Adam's dataset shows less warm-season precipitation than Legates & Willmott (evaporation loss not included here) but significantly more cold-season precipitation — because the WMO NWS 8" equation (Eq. 2-17) produces catch ratios up to 140% lower at 6.5 m/s than Legates & Willmott's pre-WMO equivalent.

Spatial Patterns

  • Catch ratios generally increase north→south (less snow = less undercatch)
  • Lower in Alps and Tibetan Plateau (high wind)
  • Lower in US Midwest (high mean wind speeds)
  • North America shows discontinuity at US/Canada border due to different measurement methods
  • Southern South America anomaly (+20–80% liquid correction) — artifact of very few stations representing extreme winds

Modern Updates: WMO-SPICE & Post-2002

Current Status of This Paper

Adam & Lettenmaier (2003) is still widely cited and its equations remain the standard for correcting manual gauges. The WMO-SPICE experiment (2010–2015) extended the framework to modern automated gauges. These two frameworks are complementary, not competing — use Goodison et al. (1998) for NWS 8", Tretyakov, Nipher, Hellmann etc.; use SPICE/Kochendorfer for Geonor T-200B, OTT Pluvio2, and tipping-bucket automated gauges.

WMO-SPICE (2010–2015)

The WMO Solid Precipitation Intercomparison Experiment compared automated precipitation gauges across 8+ sites during the 2013/14 and 2014/15 winters. The new reference gauge is the Double Fence Automated Reference (DFAR).

Kochendorfer et al. (2017) Universal Transfer Function (UTF)

Two functional forms derived from 8 sites, applicable to single Alter-shielded and unshielded automated gauges:

SPICE Eq. 1 — Temperature-Continuous Form (recommended)
CE = exp(-a · U_gh) · (1 - b · exp(c · T_air))

CE = collection efficiency (equivalent to CR, 0–1) · U_gh = wind speed at gauge height (m/s) · T_air = air temperature (°C)
a, b, c = site-specific or universal coefficients from Table 1 of Kochendorfer et al. (2017b)

SPICE Eq. 2 — Phase-Partitioned Form
CE = exp(-a · U_gh^b) [for solid precipitation only; CE=1 for rain (T > 2°C)]

Requires user-determined precipitation phase. Coefficients differ by shield configuration and wind measurement height (gauge height vs. 10m).

Key SPICE Finding

SPICE Performance

Application of UTF reduced bias in unshielded gauges from −33.4% to +1.1% on average. However, less effective at the windiest sites, and some over-adjustment at calm sites — reinforcing that good wind shielding remains preferable to large software corrections.

Current Global Products and Their Methods

ProductCorrection ApproachPhase DeterminationNotes
GPCPFixed monthly climatologies (Legates & Willmott basis)Monthly empiricalAdam & Lettenmaier lineage
GPCC MonitoringDynamic (Fuchs-Schneider model, CF-F)Event-based RH + TUses real-time GTS observations
ECCC CanadaSPICE UTF (Kochendorfer 2017b)Temperature-continuous397 automated stations, 2001–2019

Ehsani & Behrangi (2022) found a ~4% difference in global land precipitation between GPCP and GPCC correction approaches. The community debate is ongoing.

What Remains Valid from Adam & Lettenmaier (2003)

ElementStill Valid?Notes
WMO 1998 catch ratio equations (Table 2-1)✓ YesNot superseded for manual gauge types
Log wind profile scaling (Eq. 3-4)✓ YesStandard boundary-layer similarity
Temperature-based phase partitioning~ AcceptableWet-bulb methods more accurate if available
Wetting loss values (Table 3-3)~ Use with cautionBogdanova & Mestcherskaya suggest some values too high
Reference period 1979–1998✗ Update neededUse ERA5 (1940–present) for modern applications
Trend detection use✗ Not applicableAuthors explicitly prohibit this use case

Core Bias Adjustment Model

Full Adjustment Equation (Adam & Lettenmaier Eq. 3-2)

MASTER BIAS ADJUSTMENTEq. 3-2
Pa = (1−R)·(κr·Pg + ΔPwr) + R·(Pg/CRs + ΔPws)
Evaporation ignored (set to 0) Daily time step
Based on Legates (1987) framework, modified to use WMO catch ratio (CR) rather than correction factor (κ) for solid component.
SymbolMeaningUnits
PaBias-adjusted precipitationmm
PgGauge-measured precipitationmm
RSolid precipitation fraction (0–1)dimensionless
CRsCatch ratio for solid precip — from WMO equations0–1 (or %)
κrCorrection factor for liquid undercatch (≥1)dimensionless
ΔPwrWetting loss, liquid eventsmm/day
ΔPwsWetting loss, solid events (= 0.5 × liquid loss)mm/day
Notation Note

CR and κ are inverses. The WMO intercomparison uses catch ratio CR = gauge/truth (≤1 for undercatch). Legates (1987) used correction factor κ = truth/gauge (≥1). Adam & Lettenmaier use CR for solid, κ for liquid, following the conventions of their respective source literature. Don't mix them up.

Mean Monthly Aggregate Catch Ratio (Eq. 3-3)

AGGREGATE CATCH RATIOEq. 3-3
CR_all = Pg_monthly / Pa_monthly
This is what gets gridded. To apply: P_adjusted = P_raw / CR_all (since CR_all ≤ 1, this increases precipitation).

Snow/Rain Partitioning

Daily precipitation is split into solid (snow) and liquid (rain) fractions using the US Army Corps of Engineers (1956) method based on daily temperature extremes.

ConditionPrecipitation TypeSolid Fraction R
T_min > +1.5°CAll liquid (rain)0.0
T_max < −0.5°CAll solid (snow)1.0
−0.5°C ≤ T_max and T_min ≤ +1.5°CMixed — interpolate(1.5 − T_min) / 2.0
SOLID FRACTION (TRANSITIONAL ZONE)Derived
R = (1.5 − T_min) / 2.0 [where T_min ∈ [−0.5, 1.5]°C]
Yang et al. Threshold Comparison

Yang et al. used −2°C and +2°C as thresholds instead of −0.5°C and +1.5°C. This methodological difference contributes to slight catch ratio divergence between the two approaches — Adam's estimates are consistently 1.6–7.9% higher than Yang's. For transitional climates (Pennsylvania spring/fall), threshold choice meaningfully affects how much precipitation gets the snow correction.

Pennsylvania Context

For central PA (MDT, CXY, LNS), solid precipitation occurs primarily November–March. The transitional period (October, April) is where partitioning assumptions matter most. At MDT, the documented TMIN warm bias means mixed events may be over-classified as liquid — reducing the solid undercatch correction applied to MDT relative to a truly unbiased temperature record.

Wind Profile Scaling to Gauge Height

The WMO catch ratio equations require wind speed at the gauge orifice, not at the standard anemometer height. The logarithmic similarity profile is used.

LOGARITHMIC WIND PROFILEEq. 3-4
w_h = w_H · [ln(h/z0) / ln(H/z0)] · m
SymbolDefinitionValue Used
w_hWind speed at gauge orifice heightComputed
w_HWind speed at anemometerObserved (ASOS or archive)
hGauge orifice height above groundGauge-specific (see Table)
HAnemometer heightAssumed 10 m (WMO standard)
z0Roughness length0.01 m (Oct–Mar) / 0.03 m (Apr–Sep)
mExposure coefficientAssumed 1.0 (fully exposed)

Snow Depth Effect on Effective Heights

Accumulated snow raises the effective ground level, reducing both h and H. This is ignored in this study (zero depth assumed). Sensitivity analysis shows:

Snow DepthCR Change (high wind)Primary Driver
0 m (assumed)Baseline
0.5 m+~10% (less undercatch)Gauge height reduction
1.0 m+~30%Gauge height reduction (anemometer height change is negligible)
PA Application Note

NWS 8" gauge orifice height: 1.1 m. At heavy snow depths (>0.5 m), the effective orifice is much closer to the actual snow surface, and the wind speed at orifice becomes substantially higher per the log profile — increasing the correction. This is a known underestimation in the Adam & Lettenmaier approach for deep-snowpack environments.

WMO Solid Precipitation Catch Ratio Equations

Source: Goodison et al. (1998), Table 2-1. All equations give CR in %. Variables: w_h = gauge-height wind (m/s), T_max, T_min, T_mean in °C.

Nipher — Canada2-1 / 2-2
Snow: CR = 100 − 0.44·w_h² − 1.98·w_h
Mixed: CR = 97.29 − 3.18·w_h² + 0.58·T_max − 0.67·T_min
N=241/177 r²=0.40/0.38 Shield: Nipher
SMHI — Norway, Sweden2-3
Snow: CR = 99.81 − 10.8·w_h
N=89 r²=0.80 Shield: Nipher
Tretyakov (Shielded) — USSR, Mongolia2-4 / 2-5
Snow: CR = 103.11 − 8.67·w_h + 0.30·T_max
Mixed: CR = 96.99 − 4.46·w_h + 0.88·T_max + 0.22·T_min
N=381/433 r²=0.66/0.46 Shield: Tretyakov
Tretyakov (Unshielded) — N.Korea variant2-6
Snow: CR = 101.11 − 25.88·w_h + 2.12·w_h²
N=89 r²=0.74 Shield: None
Hellmann (Unshielded) — Poland, Switzerland2-7 / 2-8
Snow: CR = 100 + 1.13·w_h² − 19.45·w_h
Mixed: CR = 96.63 + 0.41·w_h² − 9.84·w_h + 5.95·T_mean
N=172/285 r²=0.75/0.48 Shield: None
Hellmann (Nipher-Shielded) — Greenland, Iceland*2-9
Snow: CR = 100 − 11.95·w_h + 0.55·w_h²
N=43 r²=0.50 Shield: Nipher
*Iceland uses surrogate (Icelandic gauge not studied). Romania surrogate also uses 2-7.
Norwegian Gauge2-10
Snow: CR = 98.18 − 11.27·w_h
N=89 r²=0.79 Shield: Yes
Max diff vs. SMHI (2-3): only 4.5% at 6 m/s. Treat as interchangeable.
RT-4 (Cylindrical Shield) — Japan (primary)2-13
Snow: CR = 100 / (1 + 0.14·w_h)
N=23 Shield: Cylindrical
RT-1 (2-11): 1/(1+0.17·w_h) | RT-3 (2-12): 1/(1+0.24·w_h). Max error RT-4 vs RT-3: 13.4% at 6 m/s.
Chinese Standard (Unshielded) — China, S.Korea2-14
Snow: CR = 100 · exp(−0.056·w_h)
N=38 r²=0.56 Shield: None
NWS 8" (Unshielded) — USA ← PRIMARY2-17 / 2-18
Snow: CR = exp(4.61 − 0.16·w_h^1.28)
Mixed: CR = 100.77 − 8.34·w_h
N=55/59 r²=0.77/0.37 Shield: None
⚠ This eq. produces CR up to 140% LOWER than Legates & Willmott at 6.5 m/s. Primary reason this study shows more cold-season correction over the US. Alter-shielded eq: CR = exp(4.61 − 0.04·w_h^1.75) [Eq. 2-15].
H&H-90 (Tretyakov Shield) — Finland2-20
Snow: CR = 99.36 − 8.49·w_h
N=33 r²=0.64 Shield: Tretyakov
METRA 886 (Unshielded) — Czech, Slovakia2-21 → then 2-4
Step 1: CR = 100 · exp(−0.1046·w_h)
Step 2: Apply Eq. 2-4 (Tretyakov shielded)
N=24 r²=0.19 Shield: None
⚠ Two-step required: METRA calibrated against Tretyakov (not DFIR). Low N and r² indicate high uncertainty.
Wild (Nipher Shield) — Finland pre-1982, Austria surrogate2-19
Snow: CR = 93.52 − 12.68·w_h
N=88 r²=0.08 ← very low Shield: Nipher
Very low r². High scatter. Austria uses this as surrogate for Kostlivi/Mountain gauges.
Wind Speed Threshold

If computed gauge-height wind exceeds threshold, use threshold value. This prevents extrapolation beyond regression range and avoids overcorrection when snow may be blown INTO the gauge. Default: 6.5 m/s. Exceptions: North Korea and Austria → 6.0 m/s (regression behaves unrealistically beyond this).

Wind-Induced Liquid Precipitation Undercatch

The correction factor κ_r (≥1) multiplies gauge-measured liquid precipitation. All from Legates (1987), compiled from multiple sources. Variables: μ = transfer coefficient, w_hp = wind speed during precipitation at gauge height, w_p = at anemometer height.

Transfer Coefficient μ (Eq. 3-5)

TRANSFER COEFFICIENTEq. 3-5
μ = (p/100) · (273 + T_a) / (273 + 0.378·e_a)
With p=100 kPa (approximation): μ = (273 + T_a) / (273 + 0.378·e_a)

Vapor Pressure Estimator (Eq. 3-6)

MONTHLY MEAN VAPOR PRESSUREEq. 3-6
e_a = 0.2 · exp(19.0629 + 0.138952·ln(Pg) − 4798.05/(273+T_a))
Pg = monthly gauge precipitation (mm), T_a = monthly mean temp (°C), e_a in kPa

κ_r Equations by Gauge Type

500 cm² Nipher-shielded3-7/3-8
w_hp ≤ 5: κr = 1 + 0.012·μ²·w_hp²
w_hp > 5: κr = 1 + 0.007·μ²·w_hp²
500 cm² unshielded3-9
κr = 1 + 0.013·μ²·w_hp²
200 cm² Tretyakov-shielded3-10
κr = 1 + 0.008·μ²·w_hp²
200 cm² unshielded3-11
κr = 1 + 0.011·μ²·w_hp²
127 cm² unshielded3-12
κr = 1 + 0.008·μ²·w_hp²
324 cm² NWS 8" — USA3-13
κr = 100 / (100 − 2.12·w_hp)
203 cm² Australian3-14
κr = 100 / (100 − 2.67·w_hp)
200 cm² SMHI (uses anemometer-height wind)3-15
κr = 1 + 0.004·w_p²

Wind Speed During Precipitation (Bogdanova 1969, via Sevruk 1982)

PRECIP-PERIOD WIND SPEED3-16/3-17
w_p = L_r · w_monthly
L_r = 1.12 + 0.295 · (0.826)^M
M = number of precipitation days in the month (precip > 1 mm threshold)

Wetting Losses

Added for every day precipitation occurs. Reflects water adhering to gauge interior walls during and after precipitation. Solid precipitation wetting losses are half of liquid losses.

WETTING LOSS PER EVENT
Liquid: ΔPwr = a (mm/event)
Solid: ΔPws = 0.5 · a (mm/event)
Gaugea (mm/day)Source
NWS 8"0.15Legates (1987)
Australian0.02Legates (1987)
Wild0.20Legates (1987)
R.M.O. Mk 20.20Legates (1987)
Chinese/Japanese0.20**Interpolated
L'Association0.20Legates (1987)
Nipher0.25Legates (1987)
R.M.O. Mk 10.25Legates (1987)
South African0.25**Interpolated
0/200 cm² generic0.25Legates (1987)
Kostlivi0.25**Interpolated
SMHI0.30Legates (1987)
Hellmann0.30Legates (1987)
Metra 8860.30Legates (1987)
Tretyakov (full)0.30Legates (1987)
Tretyakov (Former USSR only)0.10 additionalFSU applies 0.20/0.10 operationally; only 0.10 more added
Caution: Recent Literature Revisions

Bogdanova & Mestcherskaya (1998) found that increasing measurement frequency in Former USSR (2×/day → 4×/day in 1966) did NOT increase daily wetting losses — on humid days containers don't fully dry between measurements. This suggests some historical wetting loss estimates may be too high. Groisman & Rankova (2001) confirmed. Use Table values as order-of-magnitude estimates; site-specific studies are preferable when available.

Country Gauge Parameters Table

Table 3-1: Parameters applied when computing daily CRs from WMO equations.

Country/RegionGauge Corrected ForOrifice Ht (m)Wind Threshold (m/s)Eq. #
Former USSRTretyakov (shielded)2.06.52-4
USANWS 8" (unshielded)1.16.52-17
MongoliaTretyakov (shielded)2.06.52-4
SwedenSMHI (shielded)1.56.52-3
NorwaySMHI (shielded)1.56.52-3
GreenlandHellmann (shielded)3.06.52-9
IcelandHellmann (shielded) [surrogate]2.06.52-9
JapanRT-4 (cylindrical shielded)3.56.52-13
FinlandH&H-90 (Tretyakov shielded)1.56.52-20
PolandHellmann (unshielded)1.56.52-7
RomaniaHellmann (unshielded) [surrogate for IMC]1.56.52-7
SwitzerlandHellmann (unshielded)1.56.52-7
Czech./SlovakiaMETRA 886 → then Tretyakov1.06.52-21 → 2-4
North KoreaTretyakov (unshielded)1.56.02-4
ChinaChinese (unshielded)0.76.52-14
South KoreaChinese (unshielded)0.26.52-14
AustriaWild (shielded) [surrogate]1.06.02-19
CanadaHandled separately — see Canada section

Country Reliability Scores

Scoring system on a 70-point scale assessing accuracy of the wind-induced solid precipitation undercatch adjustment. Points: Gauge Representation (0–20) + Equation Application (0–30) + Interpolation/Station Density (0–20).

Score Criteria

CategoryMax PointsCriteria
Gauge Representation20How well does assumed national gauge represent actual network?
Equation Application30−5 if N<100; −5 if r²<0.70; −5 if equation is a surrogate
Station Density20+10 if <5,000 km²/gauge; +10 more if <2,500 km²/gauge
CountryDensity (km²/gauge)Gauge Rep.Eq. Appl.InterpolationTotal / 70
Canada1,491202515
60
Sweden1,601152520
60
South Korea1,824202020
60
Norway2,014102520
55
Poland4,667153010
55
Romania1,389152020
55
North Korea4,305202510
55
Switzerland1,00753020
55
Former USSR13,79120250
45
Mongolia30,68620250
45
Japan1,74952020
45
China12,52920200
40
Former Czech./Slovakia2,241101020
40
Greenland10,09515200
35
Iceland2,711151010
35
Austria65551020
35
USA ← LOWEST9,8165250
30
Finland6,24110200
30

Canada: Special Case

Why Canada Is Different

Canada uses a snow ruler (not a gauge) to measure fresh snowfall at ~1,800 stations, converting depth to SWE using an assumed density of 100 kg/m³. A shielded Nipher gauge is used at only ~125 stations. This means the standard catch ratio framework doesn't apply — Canadian "catch ratios" are actually adjustment ratios (archived measurement over ground truth), not traditional gauge-to-DFIR catch ratios.

Two Source Datasets Blended

FeatureGroisman (1998b)Mekis & Hogg (1999)
Stations6,692495 (best quality)
TimestepMonthlyDaily
Snow correctionClimatological density + CR=0.90 assumedWMO equations + observed wind
Trace correctionNone0.1 mm per trace event
AccuracyLower; wider networkHigher; gold standard

Blending Approach

  1. Find 485 stations common to both datasets
  2. Compute monthly ratio: Groisman/Mekis for 1979–1990
  3. Grid ratios to ½° using SYMAP
  4. Scale full Groisman network to Mekis quality using gridded ratios
  5. Derive catch ratios: CR = unadjusted / scaled-Groisman
Mountain Rockies Note

Canadian catch ratios can exceed 100% in the Rocky Mountains. This isn't an error — it means the snow ruler was overestimating SWE because fresh mountain snow is less dense than the assumed 100 kg/m³. The bias correction actually reduces the archived value in these regions.

Catch Ratio Calculator

Compute the solid precipitation catch ratio for a given gauge type and wind conditions, following the WMO equations. Results are in %.

Input Parameters

Results

Catch Ratio (CR)
Undercatch / Correction Factor

CR vs. Wind Speed Curves

Interactive visualization of catch ratio degradation with increasing wind speed for major gauge types. Wind speed shown at gauge height. Assumes pure snow conditions.

Solid Precipitation Catch Ratio (%) vs. Wind Speed at Gauge Height (m/s)

Symbol Reference

SymbolMeaningUnitsEquation(s)
PaBias-adjusted precipitationmm3-1, 3-2
PgGauge-measured precipitationmm3-1, 3-2, 3-3
RSolid precipitation fraction0–13-1, 3-2
CRCatch ratio (gauge/truth)% or 0–12-1 through 2-21
CRsCatch ratio for solid precipitationfraction3-2
CR_allAggregate mean monthly catch ratiofraction3-3
κCorrection factor (truth/gauge) = 1/CR≥13-1
κrCorrection factor, liquid precipitation≥13-7 through 3-15
w_hWind speed at gauge orifice heightm/s3-4, 2-1–2-21
w_HWind speed at anemometer heightm/s3-4
w_pWind during precipitation (anem. height)m/s3-15, 3-16
w_hpWind during precipitation (gauge height)m/s3-7–3-14
hGauge orifice height above groundm3-4
HAnemometer height (assumed 10 m)m3-4
z0Roughness length (0.01/0.03 m)m3-4
mExposure coefficient (assumed 1.0)3-4
T_maxDaily maximum temperature°C2-2, 2-4, 2-5
T_minDaily minimum temperature°C2-2, 2-5
T_meanDaily mean temperature°C2-8
T_aMonthly mean air temperature°C3-5, 3-6
μTransfer coefficient (thermodynamic)3-5, 3-7–3-12
e_aMonthly mean vapor pressurekPa3-6
ΔPwrWetting losses, liquidmm3-2
ΔPwsWetting losses, solid (= 0.5·a)mm3-2
aGauge wetting loss per eventmm/dayTable 3-3
LrEmpirical coeff., wind during precip3-16, 3-17
MPrecipitation days per monthdays3-17
NObservations in WMO regressioncountTable 2-1
Coefficient of determination0–1Table 2-1
DFIRDouble Fence International ReferenceWMO reference standard
DFARDouble Fence Automated ReferenceWMO-SPICE reference standard
UTFUniversal Transfer Function (SPICE)Kochendorfer et al. (2017)

NEXRAD Beam Geometry

Surface precipitation gauges and NEXRAD radar measure precipitation in completely different volumes of the atmosphere. Understanding where the radar beam is relative to the precipitating cloud is essential for interpreting any gauge–radar comparison, and directly affects how bias corrections from Adam & Lettenmaier should be applied alongside radar-derived fields.

The Standard Beam Height Equation

The height of the center of the radar beam above ground at range r is:

BEAM CENTER HEIGHTRadar Geometry
H(r) = sqrt(r² + (ke·Re)² + 2·r·ke·Re·sin(θ)) − ke·Re + h_s
Standard 4/3 Earth model Used in all NWS NEXRAD products
SymbolMeaningValue
H(r)Beam center height above ground at range rkm
rSlant range from radarkm
ReTrue Earth radius6,371 km
keEffective Earth radius factor (standard atmosphere)4/3 ≈ 1.333
θElevation angleradians
h_sRadar site elevation above MSLkm

Beam Width

BEAM HALF-WIDTH (1° beamwidth)Derived
half_width(r) = r · tan(0.5°) ≈ r · 0.00873
At 100 km range: beam is ±0.87 km = ~1.74 km total diameter. At 200 km: ~3.5 km diameter.

Beam Bottom and Top

BEAM BOTTOM / TOP HEIGHTSDerived
H_bottom(r) = H(r) − half_width(r)
H_top(r) = H(r) + half_width(r)
Beam bottom height is what matters for overshooting — if H_bottom > melting level, the beam is entirely above the bright band and sees ice crystals, not rain.

Key Physical Consequences

Range from RadarApprox. Beam Center (0.5° tilt)Consequence
25 km~0.5 km AGLBeam is in rain layer — good QPE
50 km~1.0 km AGLStill in rain for most events
100 km~2.0 km AGLNear or above bright band in weak events
150 km~3.3 km AGLFrequently above melting layer in winter
200 km~5.0 km AGLAlmost always overshooting in PA winter events
250 km+>7 km AGLSevere beam overshoot — QPE unreliable
Critical Interaction with Gauge Bias

Gauge undercatch (corrected by Adam & Lettenmaier methods) and radar beam overshoot produce opposite biases in the same direction — both cause the apparent surface precipitation to be lower than actual. When comparing gauge records to NEXRAD QPE at long ranges, the radar may itself be underestimating, masking or compounding the gauge undercatch signal. Never apply gauge bias corrections to improve gauge-radar agreement without first verifying the beam is actually sampling the precipitating layer.

Beam Zone Classification

Precipitation observations can be classified by which portion of the radar scanning geometry they fall in. This determines how much confidence to place in radar QPE for a given station, and whether gauge undercatch corrections are the dominant error source or whether beam geometry effects dominate.

Four-Zone Framework

ZoneCriteriaRadar QualityDominant Bias Source
Zone 1: Beam Below ML Beam center < melting level; beam bottom < ML − 500m High — sampling liquid precipitation directly ZR relationship error; gauge undercatch
Zone 2: Bright Band Beam intersects melting layer (ML ± ~500m) Degraded — bright band enhancement inflates Z Bright band contamination; mixed-phase ZR
Zone 3: Partial Overshoot Beam bottom above ML; beam still sampling precipitation Poor — ice crystal sampling, Z-to-snow relationship Ice-phase ZS relationship; beam overshoot
Zone 4: Complete Overshoot Beam entirely above precipitation top None — no signal or noise floor only Radar cannot detect event at all

Melting Layer Height Estimation

MELTING LAYER HEIGHT (0°C ISOTHERM)Environmental Lapse Rate Approx.
ML_height ≈ (T_surface − 0) / Γ_env + h_station
Γ_env = 6.5 °C/km (standard environmental lapse rate). More accurately: use RAP/RUC soundings or nearest upper-air data. IEM provides BUFKIT sounding archives for all NEXRAD sites.

Seasonal Patterns for Pennsylvania

SeasonTypical ML Height (AGL)Zone 1 Radius (0.5° tilt)Notes
Summer (JJA)3.5–5.0 km~200–250 kmBeam usually in rain layer for all PA coverage
Spring/Fall1.5–3.5 km~100–175 kmTransitional; check soundings case-by-case
Winter (DJF)0.3–1.5 km~25–75 kmBeam overshoot dominates beyond ~75 km
Cold precip events<0.5 km (surface)~0–25 kmNEXRAD nearly useless for QPE; gauge is only truth
Winter Implication

In Pennsylvania winter events, NEXRAD stations (KCCX, KPBZ, KDOX, KBGM, etc.) are frequently operating in Zone 3 or 4 for large fractions of their coverage area. The gauge record — even with its wind-induced undercatch — is often the more accurate of the two measurements. This is why bias-corrected gauge networks (CoCoRaHS, COOP, ASOS) remain the authoritative truth for winter precipitation QPE verification in PA.

Five-Point Spatial Integration Method

A single beam height calculation at a station's exact coordinates is insufficient for representing the spatial variability of radar beam geometry across a ½° grid cell. The five-point integration method samples the beam geometry at five locations within a grid cell and averages the result, providing a spatially representative beam height estimate consistent with the resolution of the bias correction grids.

Sampling Pattern

FIVE-POINT SAMPLING LOCATIONSGrid Cell Integration
Points: center, N-offset, S-offset, E-offset, W-offset
offset = grid_resolution / 4 (= 0.125° for ½° grid)
The four cardinal offset points are placed at ¼ of the grid spacing from the center, sampling the corners of the inner half of the grid cell. This captures the gradient in beam height without going to the cell edges.
SPATIALLY INTEGRATED BEAM HEIGHT5-Point Average
H_spatial = (H_center + H_N + H_S + H_E + H_W) / 5
Each H value computed from the full beam height equation at that point's slant range and azimuth to the radar. H values include terrain elevation at each point when a DEM is available.

Why Five Points Instead of One

ScenarioSingle-Point ErrorFive-Point Error
Near-range, flat terrain<50 m<20 m
Mid-range, ridge-valley200–800 m100–300 m
Far-range, complex terrain500–2000 m200–600 m
Transition zone (beam near ML)May misclassify zoneBetter zone boundary estimate

Terrain Correction

In mountainous or ridge-valley terrain (Appalachians, Ridge & Valley province), ground elevation varies significantly within a single ½° grid cell. The beam height above ground at each of the five points must account for local terrain elevation:

BEAM HEIGHT ABOVE GROUND (TERRAIN-CORRECTED)Derived
H_AGL(r, az) = H_MSL(r, az) − DEM(lat, lon)
DEM = Digital Elevation Model value at sampling point. SRTM 30m or NED 10m recommended. For PA Ridge & Valley, elevation variation of 300–600m within a single ½° cell is common.

Application to PA NEXRAD Sites

NEXRAD SiteElev. (m MSL)LocationKey Coverage Challenge
KCCX733State College (Bald Eagle Mtn)High elevation; near-range beam very low — good winter coverage
KPBZ361PittsburghWestern PA; good coverage but ridge blockage to east
KDOX15Dover DE (coastal)Low elevation; overshoot over central/western PA
KBGM490Binghamton NYNorthern PA coverage; moderate terrain effects
KBUF211Buffalo NYNW PA; lake-effect snow geometry issues
KAKQ34Norfolk VASE PA only; severe overshoot over interior
KENX557Albany NYNE PA fringe; complex terrain
KRLX381Charleston WVSW PA tip only; limited impact

Pennsylvania Station Network Context

Pennsylvania's geography creates systematic spatial gradients in both gauge undercatch and radar beam quality. Understanding these gradients is essential for interpreting any PA precipitation climatology.

Physiographic Provinces and Precipitation Bias

ProvinceKey StationsUndercatch RiskRadar Quality (Winter)Notes
Ridge & ValleyLNS, SEG, IPTModerate–High (wind channeling)Variable (terrain blockage)Valley stations miss upslope snow; ridge gauges experience high winds
Allegheny PlateauJST, DUJ, IPTHigh (elevation + wind)Poor (far from KCCX at high elevation)Heaviest snowfall region; highest undercatch corrections expected
Piedmont / SEPHL, ABE, RDGLow (mild, less snow)Good (near KDOX/KPBZ range)Least affected by solid undercatch
Susquehanna ValleyMDT, CXY, THVModerate (mixed precip season)Moderate–good (KCCX coverage)River moisture enhancement real (MDT vs CXY ~11% bias confirmed physical)
Lake Erie CorridorERI, BFDVery High (lake-effect snow)Poor (lake-effect shallow, high wind)Lake-effect snow is notoriously difficult for both gauge and radar

MDT vs. CXY — Gauge Bias Decomposition

The documented ~11% mean annual precipitation bias between Harrisburg International (MDT) and Capital City Airport (CXY) illustrates how multiple bias sources stack:

Bias SourceDirection at MDTMagnitude (estimated)Mechanism
Susquehanna River moisture enhancement↑ Real precipitation higher at MDT~4–6%River evaporation augments low-level moisture convergence in SW flow
Valley convergence↑ Real precipitation higher at MDT~2–4%Susquehanna Valley funnels low-level flow toward MDT on synoptic days
Frozen precipitation measurement↑ MDT catches more SWE (Alter shield)~1–3%MDT has Alter-shielded gauge; CXY unshielded → CXY misses more snow
TMIN warm bias at MDT↓ Reduces solid fraction estimate<1%Airport heat island slightly suppresses solid classification for MDT
Net observed biasMDT > CXY~11% annuallyPhysical, not artifactual — confirmed by spatial analysis
Key Finding

The MDT/CXY bias is larger than can be explained by wind-induced gauge undercatch differences alone. The Adam & Lettenmaier framework would predict only 1–3% gauge catch difference between two nearby ASOS stations with similar wind exposure. The additional 8–10% is real physical precipitation difference driven by mesoscale geography — exactly the kind of signal that spatially averaged gridded bias corrections cannot resolve at ½° resolution.

PAPRISM500 Implications

The 500-point Pennsylvania climatological reconstruction presents a spatial interpolation problem where gauge undercatch corrections interact with the PRISM topographic regression. Key considerations:

  • PRISM already partially corrects for elevation-precipitation relationships — but it uses raw gauge data as its training input, which includes undercatch bias. PRISM precipitation in high-wind ridgeline areas may be systematically low.
  • ERA5's ~27 km resolution smooths out the valley-ridge precipitation gradient that the five-point beam integration is designed to capture.
  • Monthly bias corrections applied after PRISM interpolation are preferable to correcting at the station level before interpolation — the corrections are already on a gridded climatological basis from the Adam & Lettenmaier framework.
  • Cold-season months (Nov–Mar) warrant separate treatment given the nonlinear interaction between wind field, solid fraction, and PRISM's regression weighting of high-elevation gauges.

Gauge vs. Radar Bias: Integrated View

When evaluating any precipitation dataset for Pennsylvania, both gauge undercatch (corrected by Adam & Lettenmaier) and radar beam geometry effects (quantified by NEXRAD beam analysis) must be understood simultaneously. They interact in non-obvious ways.

Error Direction Matrix

ConditionGauge ErrorRadar ErrorNet Comparison
Warm-season rain, near-rangeGauge low (~3–5% liquid undercatch)Radar accurate (Zone 1)Radar > gauge; gauge correction closes gap
Cold-season snow, near-rangeGauge very low (10–50%+ solid undercatch)Radar moderate (Zone 1-2)Radar >> gauge; large correction needed
Cold-season snow, far-rangeGauge very low (10–50%+ solid undercatch)Radar also very low (Zone 3-4 overshoot)Both low; radar-gauge agreement is illusory
Warm-season convection, far-rangeGauge near-accurate (warm, low wind)Radar undersamples cell topsGauge > radar; spatial sampling mismatch
Transition season mixed precipGauge partially biasedBright band contaminationBoth uncertain; ZR algorithm worst here

The Illusory Agreement Problem

Critical Warning

In winter, at long ranges (>150 km from radar), gauge undercatch and radar beam overshoot both push their respective measurements low. This means gauge–radar difference plots may show apparent good agreement while both instruments are simultaneously and substantially underestimating true surface precipitation. A low bias difference does not mean either instrument is accurate — it may mean both are wrong by similar amounts in the same direction. This is the most dangerous error mode in winter QPE verification.

Recommended Analysis Protocol

  1. Classify each gauge–radar comparison point by beam zone (using 5-point integration at appropriate elevation angle)
  2. Apply Adam & Lettenmaier CR corrections to gauge record first
  3. Apply NWS MLQPE or MRMS beam-block corrections to radar field
  4. Stratify gauge–radar comparisons by beam zone — only Zone 1 comparisons are diagnostic of gauge undercatch alone
  5. Zone 2 comparisons require bright-band correction to radar before gauge comparison
  6. Zones 3 and 4 comparisons should be excluded from gauge bias validation entirely
  7. For PAPRISM500 / historical reconstructions: where NEXRAD data predates 1994 (pre-WSR-88D), there is no independent radar truth — gauge corrections are the only available correction pathway

IEM ASOS Pipeline Notes (PA-Specific)

Data SourcePrecip Type ClassificationWind Data AvailableUse for CR Correction?
ASOS 1-minutePresent weather sensor (METAR)Yes — 2-min average at 10mYes — best available daily input
IEM COOP DailyObserver-reported (manual)No coincident windUse climatological wind from nearby ASOS
CoCoRaHS DailySWE only, no phase obsNo windPhase from temperature; wind from gridded reanalysis
ERA5 reanalysisPhase from T2m / Td2m10m wind at ~27 km resModern wind proxy for historical corrections
BUFKIT RAP soundings (IEM)Full sounding — best phase determinationWind profile to determine z_0 layerGold standard for event-based analysis

Dunkerley (2023): Recording Rainfall Intensity — Has an Optimum Method Been Found?

Dunkerley, D. (2023). Water, 15, 3383. https://doi.org/10.3390/w15193383 · Open Access CC BY 4.0 · Monash University

This is a landmark review paper cataloguing every known method for recording rainfall intensity — 15 categories spanning point-based gauges to satellite systems, from the 17th century through 2023. Its central finding is blunt: no standard or optimum method has emerged after 80+ years of systematic effort. New approaches are still actively being explored, and the widely-used tipping-bucket gauge is, in Dunkerley's assessment, poorly suited to the very thing it is most often used to measure.

15
gauge categories reviewed
223
references cited
34
pages, open access
1s
finest achievable time resolution (acoustic)

The Ground Truth Problem

A central theme throughout the paper is what Dunkerley calls the "ground truth problem": identifying what rainfall intensity actually is at a point is itself deeply non-trivial. Radar and microwave link measurements need ground calibration — but the ground-level gauges used for calibration all have their own biases. There is no universally accepted reference for intensity the way the DFIR serves as a reference for total accumulation in the WMO solid precipitation intercomparisons.

Why Intensity Is Hard

Intensity is not the same as accumulation. It requires time-resolved measurement. The fundamental challenge is that rainfall intensity fluctuates continuously, sometimes by ~800 mm/h per minute during convective bursts. Hourly totals hide all of this. Even 1-minute TBRG data misses the structure of intensity fluctuations because tip events are not synchronized to clock time. Sub-minute resolution — ideally seconds — is required to faithfully capture intensity, yet almost no routine networks achieve this.

The 15 Method Categories

#CategoryTypeBest Intensity ResolutionIn Common Use?
1Historical / obsolete (Jardí, timed-entry)Point~15 sNo
2Tipping-bucket (TBRG)Point~1 min (unreliable)Yes — dominant globally
3Drop-forming & countingPoint~6–15 sLimited research use
4Disdrometers (impact, optical)Point1 minYes — research networks
5Weighing gauges (Geonor, Pluvio2)Point1 min (≥7–12 mm/h threshold)Yes — automated networks
6Acoustic gaugesPoint1 s or finerResearch only
7Optical / camera / videoPoint / path~2 sResearch / emerging
8Thermal (hotplate)Point1 minNiche (snow measurement)
9Weighing lysimetersPoint / area10 minResearch sites only
10Other electro-mechanical (piezo, capacitive)PointSecondsResearch only
11Radiation / nuclear (gamma flux)Point1 minExperimental
12Radar (NEXRAD, micro-rain, X-band)Area~5 min, km² resolutionYes — operational
13Microwave attenuation (CML / cellular)Path / areaMinutes, path-averagedEmerging operational
14Seismic methodsArea~6 minResearch only
15Miscellaneous (vehicles, wipers, phones)Mobile / crowdsourceSeconds–minutesEmerging / experimental

Key Cross-Cutting Finding: The TBRG Is Used as Ground Truth But Shouldn't Be

Dunkerley's Core Critique

The tipping-bucket rain gauge is the global standard used to validate disdrometers, radar, microwave links, and almost every other measurement method. Yet TBRGs are poorly suited to recording intensity — they cannot resolve rainfall duration accurately, they miss rain during tipping, they are biased at high rates, and they cannot detect drizzle or light rain that falls between tips. Dunkerley argues a better standard reference is genuinely needed before the field can make confident progress on intensity measurement.

Tipping-Bucket Rain Gauges (TBRG)

The most widespread recording gauge in the world. The see-saw mechanism records rainfall in fixed increments — typically 0.2 mm, less often 0.1 mm or 0.5 mm. The time of each tip is logged, and intensity is estimated from inter-tip intervals.

How Intensity Is Derived

RAIN RATE FROM TIP INTERVALDunkerley Eq. 1
RR = V / (T2 − T1)
V = bucket capacity (mm), T1/T2 = times of successive tip events (h). Assumes constant intensity between tips — an assumption that is almost never exactly true.

Fundamental Error Sources

Error SourceDirectionMagnitudeNotes
High-rate undercatch (tip during fill)↓ Underestimate>10% at 100 mm/h; larger at higher ratesRain arriving during the finite tip time allocated to already-full bucket
Intensity-dependent bucket capacity↓ UnderestimateVaries by design; needs dynamic calibrationEffective bucket capacity is not fixed — changes with rain rate
Partially filled bucket at rain end↓ Underestimate durationMinutes to hoursLast real rain may occur well before final tip; TBRG cannot detect rain end
Non-clock-synchronised tip events↔ Timing ambiguityLarge in light rainTip events not tied to clock minutes; "1-min rain rates" from TBRGs are unreliable
Funnel wetting lag↓ Delays start detectionMinutesFunnel must wet up before delivering water to buckets; first drops may not be counted
Wind undercatch↓ Underestimate2–50%+ depending on conditionsStandard gauge aerodynamics — addressed by Adam & Lettenmaier framework
Evaporative loss from funnel↓ UnderestimateSmall but real~1346 cm² inclined funnel surface vs. ~700 cm² orifice area available for evaporation

The Intermittency Problem

Rainfall routinely starts and stops multiple times per hour — intra-event intermittency. If a TBRG is used to estimate intensity during a 1-hour period containing 20 minutes of actual rain, dividing total mm by 60 minutes gives a mean rate of one-third the actual intensity. Even if you use RR = V/(T2−T1), you cannot identify the intermittent gaps — the gauge simply appears to tip slowly, indistinguishable from continuous light rain.

Dunkerley's acoustic field data showed a 1h 46min event delivering 5.6 mm at an average rate of 3.2 mm/h produced 924 acoustic voltage readings — but only 28 TBRG bucket tips. The TBRG record had no visibility into the rapid intensity fluctuations visible in the acoustic record.

Modifications and Improvements

ModificationDeveloperImprovementStatus
Weighing TBRG (bucket weight during fill)Lee (2004), Kim & Lee (2004)Eliminates tip-time intensity loss; 0.01 mm resolutionCommercial (Lambrecht "Rain[e]") — limited deployment
Dual-bucket designChoi et al. (2022)Upper 0.1 mm + lower 0.5 mm catches splash; take max of twoResearch
Rotating disc (12 chambers)Mink & Forrest (1976)0.005 mm resolution — 40× standard; stainless steel discNot in common use; requires tight manufacturing tolerances
Solenoid-valve dual-chamberDrabbe (1975)0.1 mm resolution; automatic emptyingNot in common use
Triboelectric self-poweredHu et al. (2022)Float-lift nanogenerator — generates power from rainfallExperimental
Dynamic calibrationSypka (2019), Duchon & Biddle (2010)Treats bucket capacity as variable function of intensityBest current practice for high-rate events
Syphon-Based Gauges (Related)

Syphon gauges (Dines tilting-syphon, R.M. Williams self-siphon) self-empty less frequently than TBRGs. The R.M. Williams gauge triggers at 50 mm accumulated. Syphon emptying takes ~30 s during which rainfall is not recorded — same fundamental deficiency as TBRGs during tip events. Widely used at sea (buoys, ships). Small collecting area (100 cm²) limits accuracy for large drops.

Weighing Rain Gauges

Weighing gauges continuously measure the accumulated mass of collected water. Unlike TBRGs, they can in principle record any intensity from the moment rainfall starts. The two dominant commercial designs are the Geonor T-200B (vibrating wire) and the OTT Pluvio2 (load cell).

Operating Principles

GaugeSensing MethodCollecting AreaResolutionMax CapacityMin Detectable Rate
Geonor T-200B3 vibrating-wire sensors; resonant freq. ∝ load200 cm²0.05 mm (600 mm cap) / 0.1 mm (1500 mm cap)600–1500 mmNot specified by Dunkerley; ~0.1 mm/h practical
OTT Pluvio2Load cell; DC voltage ∝ weight200 or 400 cm²0.001 mm/h (stated)750 mm (400cm²) / 1500 mm (200cm²)≥7–12 mm/h for reliable 1-min data (Saha et al.)
Lambrecht "Rain[e]"Weighing + tipping hybrid200 cm²0.001 mm/hUp to 1200 mm/h operational range

Error Sources for Weighing Gauges

Error SourceEffectNotes
Temperature fluctuationSpurious rainfall recordedThermal expansion/contraction of bucket creates apparent weight change (Knecht et al. 2019)
Wind-induced vibrationNoisy signalWind rock on gauge creates apparent weight changes; post-processing required (Nayak et al. 2008; Ross et al. 2020)
Low-intensity floorMisses light rainBelow ~7 mm/h, 1-min weighing resolution insufficient to reliably detect each minute's accumulation
Oil/antifreeze contaminationPotential calibration driftOil added to reduce evaporation; antifreeze in winter — both affect density and weighing accuracy
Wind undercatch (solid)Underestimate snowSame problem as all catching gauges — addressed by SPICE/Kochendorfer transfer functions (see Modern Updates section)
Manual emptying requiredOperational constraintBuckets must be emptied before reaching capacity; at remote sites this limits deployment duration

WMO-SPICE Connection

The Geonor T-200B and OTT Pluvio2 are the two gauges used in the WMO-SPICE experiments (Kochendorfer et al. 2017). The Universal Transfer Function (UTF) derived from SPICE was specifically developed for these weighing gauge / shield configurations. See the Modern Updates section for the UTF equations.

Lysimeter Note

Weighing lysimeters (isolated soil monoliths on load cells) offer a related approach with zero wind-undercatch — the large surface area at ground level means drops arrive at natural trajectories. Gebler et al. (2015) found lysimeters recorded 16.4% more precipitation than a co-located TBRG at 1 m height, largely due to elimination of wind undercatch and ability to record dew. Lysimeters are research tools but represent a useful upper-bound estimate of true precipitation arrival.

Drop-Forming & Counting Gauges

Drop-counting gauges collect rainfall through a standard funnel, then deliver it to a narrow capillary tube (~3 mm inside diameter) that forms individual drops of known volume. Each drop is counted as it falls — optically (infrared beam) or electrically (contact with two fine electrodes). The key advantage: no waiting for a bucket to fill.

Performance Characteristics

DesignDrop VolumeEquivalent SensitivityResponse TimeUpper Limit
Norbury & White (1971)~0.07 mL~0.006 mm (33× finer than 0.2mm TBRG)10–15 s>100 mm/h (trickle onset)
Stow & Dirks (1998)~3.0 mm diameter drop~0.006 mm~6 s>50 mm/h (trickle onset)
Sansom & Gray (2002) "RIG"Standard capillaryDetects 0.1 mm/h with 1–2 min lag~6 s>100 mm/h

Key Problems

  • Trickle phenomenon: At intensities above ~50–100 mm/h, the capillary tube produces a continuous stream rather than discrete drops — the count-based approach breaks down entirely
  • Drop volume is not constant: Drop volume varies with rain rate; dynamic calibration needed (same issue as TBRGs)
  • Capillary fouling: Detritus, mineral deposits, and biological growth in the capillary tube alter drop volume over time; requires regular cleaning and an "aging-in" period after construction
  • Overestimation at low rates: According to Stagnaro et al. (2021), drop-counting gauges tend to overestimate at low-to-intermediate intensities
  • Still funnel-dependent: All the standard funnel problems apply (wetting lag, wind undercatch, evaporative loss from the large inclined funnel surface)

Applications

Primarily used in research networks studying storm fine-structure, microwave signal attenuation, and orographic rainfall (e.g., 10-gauge Pluvimate network in Tahiti by Sichoix & Benoit). Commercial examples include the Pluvimate drop-counting gauge. Not currently deployed in any routine monitoring network at scale.

Why This Matters for Intensity Climatology

Drop-counting gauges are one of the few instruments that can resolve true rainfall intermittency at the sub-minute scale and capture brief intensity bursts (e.g., a 30-second spike to 450 mm/h within an otherwise 20 mm/h storm) that are completely invisible in TBRG records. Field data from Norfolk Island by Stow & Dirks showed intensity jumps from 50 to 454 mm/h captured at ~15 s resolution — events that would appear as three or four consecutive TBRG tips with no indication of their true character.

Disdrometers

Disdrometers measure the drop size distribution (DSD) of precipitation — the number and sizes of drops per unit volume of air. Some also measure fall speeds. From DSD + fall speed, rain rate can be estimated. They are the primary instrument for understanding precipitation microphysics and for calibrating Z-R relationships used in radar QPE.

Major Types

TypeSensing MethodSensing AreaFall Speed?Key Limitation
Joss-Waldvogel (JWD) RD-80Electromechanical impact; Styrofoam cone + sensing coil50 cm²No — uses Gunn & Kinzer empirical relationshipTiny sensing area; misses rare large drops; no direct fall speed
OTT Parsivel2 (optical)Laser light sheet; shadow dimensions~54 cm²Yes — two beams, fall speed from time-of-flightOblique drop entry in wind causes overestimated diameter, underestimated speed
Thies Laser (optical)Laser light sheet~46 cm²YesConsistently underestimates rain rate by ~16.5% (Fehlmann et al. 2020)
2DVD (2D video)Two orthogonal light sheets; full drop shape imaging~100 cm²YesExpensive; fragile; large data files
Micro Rain Radar (MRR)K-band Doppler radar; vertical pointingVolumetric (~m³)Yes — DopplerLarge drops overrepresented; complex in mixed-phase

Critical Wind Effects on Disdrometers

Parsivel Wind Problem

When wind causes raindrops to enter a Parsivel disdrometer obliquely, they take a longer path through the laser sheet — producing an apparently slower fall speed and apparently larger diameter. Friedrich et al. (2013) showed this leads to anomalously large, slow drops being reported. At wind speeds >14 m/s, Lin et al. (2021) found a 22.4% underestimation of 5-min rainfall vs. a co-located TBRG. Thurai et al. (2019) found gusts reduce fall speeds by 25–30%. This is directly relevant when using Parsivel data for Z-R calibration at windy sites — every Pennsylvania Ridge & Valley station qualifies.

The Small-Area Sampling Problem

One 5 mm drop delivers the same volume as ~126 drops of 1 mm diameter. At a sensing area of 50 cm² and a 1-minute tallying interval, large drops are severely undersampled — they arrive too infrequently relative to their volumetric importance. Tapiador et al. (2017) found an optical disdrometer could underestimate rainfall intensity by up to 70% due to undersampling of large drops. The Marshall-Palmer exponential DSD assumption does not rescue this problem if the actual DSD has a heavier large-drop tail.

Cross-Instrument Disagreement

Multiple studies have found significant disagreement even among identical, co-located instruments. Tokay et al. (2005) found significant divergence among six co-located JWDs. Tapiador et al. (2017) found disagreements among 14 co-located laser disdrometers. Chang et al. (2020) compared 2DVD, MRR, X-band radar, and JWD — and found the JWD to be the least accurate despite being the most widely used calibration reference.

StudyInstrumentBias vs. ReferenceNotes
Feloni et al.JWD vs. TBRG−2%Good aggregate agreement — but TBRG itself is biased
Islam (2012)Parsivel vs. TBRG (hourly)−30%Cherrapunji, India — small drops dominate local DSD
Jaffrain & Berne (2011)Parsivel vs. TBRG (15 months)−4.3%Good long-term; location-specific result
Fehlmann et al. (2020)Thies vs. reference−16.5%Consistent systematic underestimate
Lin et al. (2021)Parsivel in wind >14 m/s−22.4%Wind-induced oblique entry bias

Acoustic, Optical, Thermal & Novel Methods

Acoustic Gauges

Acoustic methods detect the sound of raindrops striking a surface — a metal plate, water surface, or soil — and relate the acoustic signal to intensity. Key advantage: no collecting funnel. Drops fall directly onto the sensor. This eliminates wetting lags, evaporative losses from funnel surfaces, and wind undercatch entirely. The first drop is detected with zero delay.

DUBOUT (1969) ACOUSTIC REGRESSIONEmpirical
dB = 7.13 · ln(I) + 47.0
r² = 0.99 · I = rainfall intensity (mm/h) · Calibrated on galvanised steel roof shed. Linear relationship between sound intensity (dB) and ln(intensity). Valid over at least 3 orders of magnitude (0.1–100 mm/h).
Acoustic MethodSensorTime ResolutionKey AdvantageKey Problem
Metal roof/plate microphoneMicrophone under surface10 s (Dunkerley 2023); 44.1 kHz possibleNo funnel; zero lag; instant start/end detectionExtraneous noise (wind, traffic, wildlife); large data files
Submerged marine hydrophoneUnderwater microphoneSecondsWidely proven at sea; Black et al., Ma & NystuenOcean environment only
Acoustic disdrometer (water tank)Submerged microphone in tankSecondsWinder & Paulson: estimates drop sizes AND intensity from bubble acousticsRequires water-filled tank in field
Sodar (sound radar)Acoustic transceiver~minutesBradley & Webb: samples ~20 m³ volume; works at low intensitiesComplex system; outdoor deployment challenges
Sommer RHD sensor160 mm stainless hemisphere, no moving partsFirst-drop detectionCommercial; compact; no moving partsNot yet validated for long-term operational intensity monitoring
Dunkerley's Acoustic Data Insight

Dunkerley's own field data logged at 10 s intervals showed a 1h46m event with 924 voltage readings — vs. 28 TBRG tip events. The acoustic record revealed rapid intensity fluctuations (including a spike to ~55 mm/h) entirely invisible in the TBRG record. This single comparison illustrates the fundamental mismatch between operational gauge records and the actual intensity structure of rainfall.

Optical / Camera / Video Methods

MethodApproachResolutionStatus
Long-path optical (Bradley et al. 2000)Halogen lamps → CCD camera, 2022 m path; OD = 0.86·RR^0.667 (r²=0.99)1 minResearch only; difficult validation over km path
Security / traffic camerasDrop streak length in image frames with known exposure; fall speed estimation~2 s potentialEmerging — millions of potential sites globally; complex processing
Video frame analysis (Dong et al.)Mean of 1000 successive frames estimates rain rateSecondsProblems identifying streaks in intense rain
CNN on smartphone images (Yin et al. 2023)1/200 s exposure images; convolutional neural netNot directly clock-syncableEmerging; potentially huge number of sites

Thermal Methods

DevicePrincipleRangeKey Issue
Battalino & Vonnegut (1978) thermal sensorHeater power required to evaporate incident drops → W = 3.71·RR^1.06 (r²=0.96)0.3–350 mm/h~1 kW needed at 100 mm/h; wind on horizontal cylinder untested
Hotplate precipitation gauge (Rasmussen et al. 2011)Power to maintain hotplate constant temp ∝ precipitation rate0.25–35 mm/hBest for snowfall; 1-min data aggregated to 5 min; low max rate
Raynor rotary detector (1955)Heated rotating cylinder with electrode array; signals single-drop arrivalSingle-drop sensitivityFound 69 min of rain before first TBRG tip; warm updraft prevents dew/drizzle detection

Radiation (Nuclear) Methods

Changes in atmospheric gamma-ray flux during rainfall are measurable. Zelinskiy et al. (2021) found RRm = 0.97·RRo + 0.71 (r² = 0.93) comparing gamma-dose model vs. TBRG + disdrometer at 1-minute resolution. Bottardi et al. (2020) linked ²¹⁴Pb flux increases to rainfall statistically. Potential for long-term monitoring using existing dosimeter networks — though requires wider testing across geology and vegetation types.

Areal Methods: Radar, Microwave Links, Seismic

Point-based gauges measure precipitation at a single location. For catchment hydrology, flash flood prediction, and urban drainage design, spatially distributed intensity data is needed. Three areal approaches are reviewed by Dunkerley.

Radar-Based Methods

Radar detects backscattered energy Z from hydrometeors. The Z-R relationship links this to rain rate R. The approach is indirect — calibration requires ground truth from gauges, and the ground truth problem means this calibration is never fully clean.

MARSHALL-PALMER Z-R RELATIONSHIPStandard
Z = a · R^b
Standard: Z = 200·R^1.6 (stratiform). Convective: Z = 300·R^1.4. Units: Z in mm⁶/m³, R in mm/h. Different a/b values are required for different precipitation types — this is a fundamental source of QPE error. See Kirsch et al. (2019) for stratiform vs. convective Z-R comparison.
Radar TypeCoverageTemporal Res.Key Limitation
WSR-88D NEXRAD (S-band)Hundreds of km radius~5 minBeam overshoot (see Spatial Context section); beam blockage; anomalous propagation
X-band (portable, local)~50–80 km~1 minSignal attenuation in heavy rain; useful for urban applications
Micro-Rain Radar (MRR)Vertical column, ~3 km height~1 minVertically pointing only; K-band attenuation in heavy rain; Chang et al. found MRR most accurate of 4 types
Lufft WS100 (low-power)~pointSecondsVokoun & Moravec (2022): reported rainfall far larger than conventional gauges in mountain tests
Raindrop Drift: A Neglected Error

Dai et al. (2019) showed that between the radar scan altitude and the ground, drops drift laterally in wind. A 5 mm drop can drift several kilometres; a 0.2 mm drop can drift up to 14 km. Drop diameter also declines during fall due to evaporation, reducing the intensity below what the radar scan suggested. This means even a perfectly accurate radar scan at beam height does not represent surface precipitation at the point directly below — a problem that grows with beam height and wind speed.

Microwave Attenuation (Cellular Links — CML)

Rain attenuates microwave signals along tower-to-tower cellular links. The greater the path-integrated attenuation, the higher the rainfall rate along the link. Global CML networks already exist in most countries — the infrastructure is in place. Urban areas are most densely covered.

AdvantageLimitation
Vast existing infrastructure — no new gauges neededWet antennas cause attenuation independent of rain — overestimates when rain film forms on antenna
High spatial density especially in urban areasPath-averaged, not point — cannot be directly compared with gauge data
Can detect rainfall onset / cessationRain may only cover part of the path — how much is unknown
Continuous real-time availabilityLess effective at quantifying exact rainfall amounts vs. detecting occurrence

Seismic Methods

Raindrop impact on soil creates seismic energy detectable by geophones. Bakker et al. (2022) showed that 90% of seismic power arises from drops >3 mm — making seismic monitoring best suited to intense rainfall with large drops. Diaz et al. (2023) used a high-density seismic network to track storm cell passage across Spain, collecting data at 6-min intervals. Like CML, seismic methods can provide spatial coverage — but validation against ground truth remains limited.

Miscellaneous / Crowdsourced

MethodSignal UsedKey Finding
Windscreen wipers (Rabiei et al. 2013)Wiper speed (manual or automatic sensor)Relationship established in lab simulation; 1% of South Korea's 20M vehicles would yield enormous dataset even with wide uncertainty
Smartphone microphone (Gaucherel & Grimaldi)Acoustic signal from rain on phone"Pluviophone" — promising but requires careful acoustic design
Smart umbrella + phone (Guo et al. "Chaac")1 s and 10 s audio clips from umbrella attachmentDemonstrates concept; field validation limited
Citizen rain gauges (Mapiam et al. 2022)Manual gauges read by citizens via appImproved hourly radar bias correction in Thailand using two-step Kalman filter
Dashcam wiper activity (Bartos et al. 2019)Wiper on/off from vehicle camera visionHigh-accuracy binary rainfall maps (rain/no rain) from connected vehicles in US cities

Comprehensive Method Comparison Matrix

Synthesized from Dunkerley (2023). Ratings reflect capability for rainfall intensity measurement specifically, not just accumulation. All methods perform better for accumulation than for intensity.

Method Time Res. Wind Undercatch? Low Intensity High Intensity Start/End Detection Spatial Coverage Ops. Cost
TBRG (standard) ~1 min unreliable Yes — significant Poor (misses between tips) Undercatch >10% at 100 mm/h Poor (± minutes to hours) Point only Low
Weighing gauge (Geonor/Pluvio2) 1 min (≥7 mm/h) Yes (solid, correctable with SPICE) Floor ~7 mm/h for 1-min data Good to 1200 mm/h (Lambrecht) Moderate (weight threshold) Point only Moderate
Drop-counting gauge 6–15 s Yes (has funnel) Detects 0.1 mm/h with <2 min lag Trickle onset >50–100 mm/h Good (~minutes) Point only High (maintenance)
Disdrometer (Parsivel/JWD) 1 min Oblique entry in wind — 22% bias Misses small drops Misses rare large drops Moderate Point only Moderate–High
Acoustic gauge 1 s or finer None (no funnel) First-drop detection Linear response across wide range Excellent — first/last drop Point only Moderate (noise filtering)
Hotplate gauge 1–5 min Reduced (no funnel, but geometry untested) 0.25 mm/h lower limit 35 mm/h upper limit Good Point only High (power consumption)
NEXRAD radar ~5 min N/A — not a surface gauge Misses very light rain; clutter Z-R saturation in heavy convection Moderate Hundreds of km² Infrastructure exists
Micro-rain radar (MRR) 1 min N/A K-band sensitivity threshold Attenuation in heavy rain Good Vertical column only Moderate
Cellular microwave links (CML) Minutes N/A Wet antenna noise obscures light rain Good for heavy rain detection Moderate Urban network coverage Essentially free (data already collected)
Lysimeter 10 min None (ground level, natural trajectory) Can detect dew; high sensitivity Surface runoff risk in heavy rain Moderate (weight threshold) Point (large footprint) Very high
Smartphone crowdsource Seconds–minutes N/A Detection threshold uncertain Saturates acoustic signal Moderate Potentially millions of points Near-zero

Dunkerley's Conclusions

The Field's Current State

No optimum method has been found. The TBRG remains dominant despite its known deficiencies because of cost, simplicity, and the value of the continuous historical record it has accumulated globally. The most promising paths forward, per Dunkerley, are: (1) acoustic methods for true high-resolution point intensity; (2) cellular microwave links for distributed urban coverage; (3) crowdsourced visual/acoustic data from smartphones and vehicles as a supplement to sparse gauge networks. None of these yet provides a validated replacement for the TBRG in the routine collection of long-term climatological data.

Relevance to Pennsylvania Precipitation Monitoring

ContextCurrent InstrumentKey GapRecommended Supplement
ASOS (MDT, CXY, etc.)Heated TBRG (FP-5 or equivalent)Sub-1-min intensity; light rain miss; liquid undercatchDisdrometer co-location; dynamic calibration
CoCoRaHS / COOPManual daily accumulation gaugeNo intensity at all; next-day reading introduces timing errorsCheap TBRG dataloggers; citizen acoustic sensors
WeeWX personal station (GW3000B / WS90)Piezo-electric rain sensorIntensity response non-linear; no TBRG-equivalent calibrationEcowitt WH40 bucket gauge as primary; piezo as onset indicator
PAPRISM500 historical reconstructionPRISM (based on COOP/ASOS)All historical data is accumulation-based; no intensity in recordERA5 precip rate fields for modern period; daily → hourly disaggregation with storm-type conditioning
Hershey Crowd IQ / M5StickC WiFi probeOpen-Meteo precip rateERA5 precip rate is ~hourly, ~27 km — cannot resolve convective burstsNearest ASOS 1-min data for intensity-triggered crowd behavior analysis

Dunn et al. (2025): TBRG Undercatch — Error Framework

Dunn, R.E., Fowler, H.J., Green, A.C., Lewis, E. (2025). Tipping-bucket rain gauges: a review of the undercatch phenomenon, and methods for its reduction and correction. Weather, 80(6), 196–205. doi:10.1002/wea.7736 · Open Access CC BY · Newcastle University / Univ. of Manchester

Published June 2025 in Weather (Royal Meteorological Society), this is a tightly focused solution-oriented review — not just cataloguing problems but specifically tracing the evolution of design improvements and correction factor (CF) methodologies for TBRGs. The authors are explicit that previous reviews were problem-based rather than solution-oriented. Their goal: a framework useful for practitioners applying bias corrections today, culminating in three concrete recommendations for the field.

5–40%
typical TBRG underestimate (Sypka 2019)
75%
max underestimate in stormy conditions (Neff 1977)
1300
TBRGs in Great Britain (EA + Met Office + SEPA)
1670s
UK monthly rainfall record start date

Two-Category Error Framework

All TBRG errors fall into two top-level categories (Ciach 2003; Villarini & Krajewski 2008), which subdivide further:

Top CategorySubcategoryNatureTypical MagnitudePrimary Mitigation
Spatial Sampling Errors Network density / representativeness Point measurement applied to area Highly variable — 250m² catchment data covering km² area Denser networks; radar/satellite fusion
Siting bias (exposure, rain shadows, regional inconsistency) Systematic Compounding — see Figure 3 of paper Strategic siting; metadata documentation
Local Errors — Catching Aerodynamic undercatch (wind-induced) Systematic 5–46% liquid; up to 67%+ solid/frozen Aerodynamic design; wind shields; lower mounting
Evaporation, sublimation, out-splash Systematic Evap: 4–11% (warm); Sublim: 0.5–0.75mm/day (cold); Splash: 1–2% Funnel design; deep collector walls
Local Errors — Counting Mechanical / high-rate undercatch Systematic 2–30%; >30–50mm/h: rapid degradation Dynamic calibration; gauge resolution increase
Random (human error, equipment failure) Random Unpredictable Quality control; co-located gauges

The Catching vs. Counting Distinction

This is the key conceptual split in the paper. Catching errors = the gap between rain falling from the sky and rain entering the gauge collector. Counting errors = the gap between rain entering the collector and rain being correctly measured. Wind undercatch is a catching error. High-rate mechanical undercatch (rain during tip) is a counting error. They compound — and their combined effect is what Adam & Lettenmaier's correction framework must address.

Why Undercatch Dominates All Other Errors

Mechanical counting errors are 2–30%. In stormy conditions, wind-induced catching errors alone can reach 75% underestimation (Neff 1977). The TBRG is a well-engineered counter of water volume delivered to it — it is simply that during the most hydrologically important events (storms, heavy rain, snow), it fails catastrophically at catching that water in the first place. This is why the Adam & Lettenmaier framework, which is fundamentally about catching errors, captures the dominant bias.

The Undercatch Phenomenon — Detailed Physics

Aerodynamic Undercatch: The Primary Mechanism

When wind flows around a TBRG, the gauge body creates a bluff-body disturbance — a zone of decelerated airflow above and around the orifice. This is unavoidable physics (Constantinescu et al. 2007). The deformed wind field deflects raindrop trajectories away from the collector (Cauteruccio et al. 2020). Three methods are used to quantify this:

MethodWhat It CapturesStrengthsLimitations
Wind tunnel experimentsFlow deformation around specific gauge geometriesControlled; isolates design variables; particle image velocimetry availableArtificial wind; cannot fully replicate natural turbulence or DSD
Computational fluid dynamics (CFD)Full 3D flow field; embedded particle tracking for drop trajectoriesNo physical experiment needed; any geometry; any wind/DSD scenarioRequires validation; cannot include all real-world complexity
Field intercomparison (pit gauge reference)Real-world integrated undercatch under all conditionsGround truth closest to operational reality; no artificial conditionsCannot isolate individual error sources; pit gauge itself has small errors

Wind Speed Dependency

The magnitude of wind-induced undercatch scales directly with wind speed. Bratzev (1963), Larson (1971), and Larson & Peck (1974) all found approximately 1–2.2% additional undercatch per mph increase in wind speed for an unshielded gauge. At typical UK/US inland storm wind speeds (15–25 mph = 6.7–11.2 m/s), this implies 15–55% wind-induced catching error from wind alone — before any mechanical errors.

Additional Factors Modifying Wind Undercatch

FactorDirection of EffectPhysical ReasonNotes
Wind turbulence (vs. uniform flow)Generally increases undercatchTurbulent fluctuations create asymmetric trajectory deviations not captured in mean-wind equationsCauteruccio 2020 PhD thesis shows turbulence specifically amplifies updraft above collector
Drop size distribution (DSD)Small drops = more undercatchSmall drops have lower terminal velocity; their trajectories bend more in cross-wind; larger drops have more inertiaNešpor & Sevruk (1999) — this is why DSD matters to CF development
Precipitation form (solid vs. liquid)Solid vastly worseSnow crystals have extremely low fall velocity and are easily deflected; can produce <33% catch in exposed sitesBenning & Yang (2005): recorded precipitation can be less than one-third of actual
Gauge heightHigher = more undercatchWind speed increases with height per log profile; higher orifice = more aerodynamic disturbancePollock et al. (2018); UK now installs TBRGs at ground level specifically to address this
Gauge design (shape)Aerodynamic shapes reduce undercatchCalix/champagne-glass profile minimizes flow separation above orificeColli et al. (2018) CFD; Cauteruccio et al. (2024) — current state of the art
Orifice rim geometryRim thickness and profile matterRim creates turbulence wake that affects the final drop trajectory as it crosses the orificeCauteruccio et al. (2021) — even subtle rim changes measurably affect catch efficiency

Evaporation, Sublimation, and Splash Losses

Error TypeMechanismMagnitudeClimate Dependency
EvaporationPrecipitation collected in funnel vaporises before draining to buckets; inclined funnel surface area ~1346 cm² vs. ~700 cm² orifice — nearly double the evaporation-prone surface4–11% in warm regionsNegligible in cool regions (Yang & Ohata 2001); significant in arid/warm climates
SublimationSolid precipitation collected in funnel sublimes directly to vapour0.5–0.75 mm/day where snowfall occurs (Fassnacht 2004)Irrelevant in warm climates; significant for daily snow accumulation in cold climates
Out-splashDroplet fragments on impact with collector rim/walls; not all fragments are captured~1–2%Most modern deep-funnel gauges render this negligible (Legates 1992)
Seasonality of Undercatch (Temperate Climates)

In temperate climates like Pennsylvania or the UK, undercatch exhibits strong seasonality: it peaks in winter (high wind + frozen precipitation + sublimation), is moderate in autumn/spring (transitional — mixed precipitation, variable wind), and is lowest in summer (warm, less wind, all-liquid precipitation). Zhao et al. (2024) documented this pattern specifically for China. This seasonality directly affects the seasonal breakdown in Adam & Lettenmaier's results — DJF corrections are 2× JJA corrections globally (+15.2% vs. +7.3%).

Undercatch Reduction: Design & Installation

Evolution of TBRG Shape

The history of TBRG design is a history of progressively reducing aerodynamic disturbance. Four stages (Figure 7 of paper):

Design StageShapeUndercatch ImprovementStill in Use?
Early TBRGsPlain cylinderBaseline — worst aerodynamics; flat-top rim creates large wakeLegacy networks
Cylindrical with funnel topCylinder + coneMarginal improvement from raised funnelCommon
Aerodynamic (plastic) — Institute of Hydrology designTapered body, reduced frontal areaSignificant reduction in flow separation (Colli et al. 2018)Widely deployed
Calix / champagne-glass shape (aluminium)Inverted flared profileCurrently optimal per Folland (1988) and Pollock et al. (2018)Best practice recommendation

Detailed Design Parameters That Affect Undercatch

Beyond overall shape, the following gauge-level parameters have been quantified to affect catching efficiency:

  • Orifice rim thickness and profile — Cauteruccio et al. (2021): rim geometry alters the turbulent wake over the collector opening. Sharper, thinner rims are preferable.
  • Depth of collector vertical wall — deeper walls reduce splash losses and help retain drops that enter obliquely
  • Slope of the funnel — steeper funnel angles speed drainage, reducing retention losses and funnel-surface evaporation
  • Funnel coating — hydrophobic coatings reduce adhesion and wetting losses (WMO 1996; Mekis & Vincent 2019; Padrón et al. 2020)
  • Funnel size in arid regions — larger funnel improves detection of <1mm events; however, Al-Wagdany (2015) found it introduces underestimation bias for events >10mm at high intensities. Trade-off must be evaluated regionally.

Installation Height: The UK Solution

Ground-Level Installation

TBRGs in the UK are now installed at ground level — orifice flush with the surrounding surface — following Pollock et al. (2018). This eliminates the log-profile wind speed amplification that affects elevated gauges. At ground level, wind speed is near zero, virtually eliminating aerodynamic undercatch. However, this practice increases the risk of debris blown into the orifice and in-splash from rain striking the ground near the gauge. It is not universally replicated — the US NWS 8" gauge is typically at 1.1m, and this discrepancy is exactly why the Adam & Lettenmaier wind profile scaling (Eq. 3-4) matters for US records.

Wind Shields

Shield TypeDescriptionUndercatch ReductionPractical Notes
Turf wallLow wall of cut turf around gauge, height matching gauge orificeVery effective — approximates pit gauge conditionsRequires regular maintenance; roots can disturb gauge base
Alter shieldHanging vertically-oriented slats in a ring around gaugeStandard operational shield; also available as double-AlterMost widely deployed operational shield; double-Alter better for snow (WMO-SPICE)
OTT screen styleLouvered cylindrical screenModerate — depends on louvre geometryIntegral with OTT Pluvio2 design; cleaner aesthetically
Tretyakov shieldConcentric rings of metal leavesHigh performance for solid precipRussian standard; used in WMO 1998 intercomparison as primary shield reference
DFIR (WMO reference)Double-fence with inner Alter; 12m outer fence diameterReference standard — near-perfect for snowResearch use only; not operational

Siting Considerations

Modern guidance emphasises careful siting alongside good gauge design. Key principles from Dunn et al. and the WMO:

  • Highly exposed sites (open fields, ridge lines, airports) experience the worst wind undercatch — the aerodynamic equations apply maximally here
  • Obstructions within ~4× their height should be avoided to prevent rain shadows (local precipitation deficit downwind)
  • Terrain slope affects rainfall angle and can systematically bias catch in one wind direction
  • Detailed metadata on gauge type, height, shielding, and siting must be recorded — this is the critical missing input for station-level CF development
  • Network heterogeneity (mixing of gauge types and heights across a national network) is essentially unavoidable in multi-decadal records; CFs must account for this

Correction Factors — Table 1 Complete Reference

Dunn et al.'s Table 1 is the most comprehensive recent summary of TBRG catching error correction methodologies. All 8 methods reproduced below with full detail from the paper.

The Ground Truth Selection Problem

Every correction factor requires a "ground truth" reference. The choice of reference shapes the resulting CF fundamentally. Early CFs used ground-level gauges. Modern CFs use pit gauges (most common), shielded gauges (Kochendorfer et al. 2017 / SPICE), laboratory experiments (Nešpor & Sevruk 1999), or CFD numerical simulation (Cauteruccio et al. 2024). None of these is perfect — pit gauges still slightly underestimate; shielded gauges have their own biases; lab conditions don't replicate real turbulence; CFD cannot capture all atmospheric complexity. The choice of ground truth is the first and most consequential decision in CF development.

1. Allerup & Madsen (1980) — Denmark — Statistical Analysis ModelDaily / Gauge
Method: Analyses ratio of daily precipitation at ground level vs. standard height (1.5m). Tabulated correction values for three exposure classes: well-sheltered, moderately sheltered, unsheltered.
Required inputs: Air temperature, exposure level, wind speed at 10m AGL, precipitation intensity
Considers: Solid and liquid precipitation (binary); aerodynamic effects; wetting loss
Gauge scope: Hellmann gauge only
Key strength: Simple tabulated application; three exposure levels
Key weakness: Assumes ground-level exposure errors are negligible; wind speed often unavailable at TBRG sites; Hellmann-only
2. Legates & Willmott (1990) / CF-L — USA/Global — Gridded MonthlyMonthly / ~238 m²
Method: Irregularly distributed gauge records interpolated to ½° lat/lon grid. Single CF per location/month.
Required inputs: Location, date only
Considers: Wind effects, wetting losses, gauge evaporation
Gauge scope: Not gauge-specific — universal
Key strength: Global coverage; extremely easy to apply; foundation of Adam & Lettenmaier (2003)
Key weaknesses: Based on 1920–1980 data; arid/mountainous/polar underrepresented; ½° grid misses topographic variability; rough meteorological parameter estimates may cause over-correction; Ehsani & Behrangi (2022) found ~4% difference vs. GPCC CF-F method
3. Førland et al. (1996) / Dynamic Correction Model — Norway/FinlandDaily / Gauge
Method: Statistical regression with log-values. DSD incorporated via rainfall intensity assumption. Crystal structure for solid precip via air temperature.
Required inputs: Precipitation intensity, wind speed at gauge level, precipitation amount, gauge type, air temperature at 2m
Considers: DSD (assumed from intensity), crystal structure, site exposure
Gauge scope: 4 specific gauges with specific windshield setups
Key strength: DSD and crystal structure incorporated; exposure in correction formula
Key weakness: Tabulated parameters only valid for the 4 specific gauges/shields; evaporation/wetting provided as fixed monthly values in Nordic context
4. Habib et al. (1999) — USA — Numerically Based Correction1-min to monthly / Gauge
Method: Wind-induced error characterised as nonlinear function of wind speed, rainfall rate, and DSD via numerical simulation.
Required inputs: DSD, wind speed, precipitation intensity
Considers: DSD via rain type (orographic, thunderstorm, showers); multiple timescales
Gauge scope: Specific class of gauges (Mk2 type and similar geometry)
Key strength: Rain type as DSD parameter; applicable at multiple timescales; reduces over-correction risk
Key weakness: Only wind-induced errors; Mk2 gauge class only
5. Allerup et al. (2000) — Denmark — Explicit Mathematical StatisticalDaily / Gauge
Method: CF derived from four independent variables. Key innovation: on-site wind/temperature not required — remote observations (up to 50km for wind, greater for intensity/temp) can substitute.
Required inputs: Precipitation intensity (remote), snow fraction (remote), wind speed (remote ≤50km), temperature (remote)
Key strength: No on-site wind measurement required; remote stations can fill data gaps
Key weakness: Hellmann gauge only; Danish context; only tested on 12 gauges; single exposure level assumed; remote data still must be within 50–100km
6. Fuchs et al. (2001) / CF-F — Germany/Austria — Fuchs Dynamic MethodDaily / Gauge
Method: Based on Førland et al. (1996) framework. Key addition: GPCC phase scheme weights CFs for mixed precipitation across daily increments.
Required inputs: Air temperature, dew point temperature, relative humidity, wind speed at gauge rim, precipitation intensity
Considers: Inconsistent precipitation form within single time increment via GPCC phase scheme
Key strength: Mixed precipitation handled correctly; currently used operationally in GPCC Monitoring product
Key weaknesses: Inherits Førland (1996) limitations; in tropical regions, calculated intensities may be too low → over-correction; wind speed at gauge rim is an unusual measurement; data-intensive
7. Ye et al. (2004) — China — Catch Ratio MethodDaily / Gauge
Method: Catch ratio as function of daily mean wind speed; separate regression equations for snow and rain; mixed precipitation by linear interpolation of daily temperature.
Required inputs: Air temperature, precipitation intensity, wind speed at 10m AGL
Considers: Trace precipitation, wetting losses, wind-induced errors, mixed precipitation
Key strength: Trace precipitation explicitly included; mixed precip handled via temperature-based partitioning
Key weaknesses: Does not incorporate gauge design or exposure; conservative assumptions for trace/wetting; no evaporation correction
8. Mekonnen et al. (2015) — Czech Republic — Calculated Relative DifferencesDaily / Gauge
Method: Simplified equation from logarithmic wind profile and catch ratio relationship. Only additional measurement needed is wind speed.
Required inputs: Wind speed, precipitation intensity
Considers: Wind-induced losses, wetting losses, evaporation losses, trace amounts — all used in CF development though only wind speed needed operationally
Key strength: Minimal data requirements; wind speed alone sufficient to apply
Key weaknesses: Single study site only; specific gauge types and set-ups only; liquid precipitation only; no terrain exposure differentiation
9. Cauteruccio et al. (2024) — Italy — Overall Collection Efficiency via CFDVariable resolution / Gauge
Method: Wind-induced error characterised by nonlinear complex CFD model with embedded liquid and solid particle tracking. Most physically rigorous method available.
Required inputs: Precipitation intensity, DSD, precipitation form, wind speed
Considers: Full gauge geometry effects including orifice rim; both liquid and solid particle dynamics
Key strength: Highly detailed physically-based investigation; explicit gauge design impact; captures nonlinear DSD-wind interactions
Key weaknesses: Complex to apply; developed in controlled laboratory conditions — not designed for direct operational application; only wind-induced undercatch; DSD and rim wind speed not routinely available at operational sites

Cross-CF Comparison: What Each Method Does and Doesn't Cover

CF MethodWindWettingEvap.TraceGauge-SpecificOperational?
Allerup & Madsen 1980Hellmann onlyYes
Legates & Willmott 1990 (CF-L)UniversalYes (GPCP)
Førland et al. 19964 gaugesYes
Habib et al. 1999Mk2 classResearch
Fuchs et al. 2001 (CF-F)Inherited 4Yes (GPCC)
Ye et al. 2004China gaugesYes
Mekonnen et al. 2015Specific siteLimited
Cauteruccio et al. 2024Lab-specificNo
Adam & Lettenmaier 2003✗ (excluded)30 gauge typesYes — gridded

Over-Correction Risk

200% Over-Correction Warning (Ehsani & Behrangi 2022; Cauteruccio et al. 2024)

Universal CFs applied at coarse temporal scales can over-correct by up to 200%. The risk is highest when: (1) a globally-gridded CF is applied to a specific site with very different exposure than the surrounding grid cell average; (2) monthly CFs are applied to events that happened mostly in one extreme day within that month; (3) the CF was derived under conditions not representative of the target site's exposure. Dunn et al. explicitly warn that simple universal CFs "should be used with caution." A categorical approach — different CFs for liquid/mixed/solid, different exposure classes — is a recommended middle ground.

Future Directions: AI, Machine Learning & Emerging Technologies

Why This Matters Now (2025 Context)

Dunn et al. note that TBRGs have become more precise over time, requiring less correction. However, heterogeneous multi-decadal datasets — combining old cylindrical gauges, modern aerodynamic gauges, ground-level UK installations, elevated US NWS 8" gauges, and variable shielding — still require effective CFs for research continuity. The CF problem has not been solved; it has evolved.

Key Statistic (2025)

Ehsani & Behrangi (2022) showed that existing CFs result in significantly different rainfall estimations — with the two major global products (GPCP using CF-L; GPCC using CF-F) differing by ~4% in annual global precipitation. Most global hydrological simulation datasets still do not account for rain gauge catch deficiencies at all (Adam & Lettenmaier 2003 is still cited as justification for not correcting).

Machine Learning and Big Data Approaches

ApproachWhat It EnablesCurrent StatusKey Reference
Spaceborne sensor validationGRACE satellite gravity anomalies → snowfall accumulation estimates → independent validation of high-latitude gauge undercatch corrections without ground truthResearch — Behrangi et al. 2018 applied to ArcticBehrangi et al. (2018)
ML merging of heterogeneous sourcesCombines TBRG, radar, satellite, CML, and other data to remove biases from TBRG datasets that traditional correction factors cannot addressResearch — Guarascio et al. (2020)Guarascio et al. (2020)
Neural network CNN on rainfall imagesSmartphone/camera images → rainfall intensity without a gauge; potential for vast ground-truth dataset creationEarly research — Yin et al. (2023)Yin et al. (2023)
Real-time gauge-mounted undercatch sensorsSmall sensors on TBRG measure local wind speed, drop size, turbulence → real-time undercatch quantification and correctionEmerging — Dutton & Balsamo (2024)Dutton & Balsamo (2024)
Rainfall simulator for CF developmentKnown controlled rainfall applied to gauge in field — eliminates ground truth ambiguity entirely; true catch efficiency measurable directlyProposed — not yet demonstrated for undercatch research specificallyDunn et al. (2025) recommendation

Dunn et al.'s Three Formal Recommendations

Recommendations (Verbatim Content, Paraphrased)
  1. All future studies incorporating rain gauge data should acknowledge undercatch — either by explicitly recognising the uncertainty it introduces, or by applying an appropriate CF. Silence on undercatch is no longer acceptable given the breadth of evidence.
  2. Further research using rainfall simulators — to ensure the 'ground truth' is known and controlled. Existing pit gauges and shielded gauges still underestimate; a truly known-input experiment is the next frontier.
  3. Development of a UK-specific CF — versatile enough to be broadly adopted (possibly a categorical approach for liquid/mixed/solid × exposure class) but not so universal it leads to misrepresentation and over-correction. Explicitly: the Legates & Willmott global CF is insufficient for UK precision needs.

Connection to the PAPRISM500 and Pennsylvania Work

Dunn et al. IssuePennsylvania AnalogPractical Implication
No single CF widely adoptedAdam & Lettenmaier (2003) still primary reference despite 22-year ageFor PAPRISM500, A&L corrections are the best available but acknowledge their 1979–1998 basis
Heterogeneous network — mixing gauge types/heightsPA COOP network: mix of NWS 8" (1.1m), FP-5 ASOS (heated, ~1m), CoCoRaHS manual (0.3m)Each gauge type needs its own CF; mixing without correction introduces systematic spatial bias in PRISM training data
Metadata absence prevents CF applicationHistorical PA COOP records: limited shield metadata; unknown exposure classUse ERA5 wind climatology as surrogate; apply exposure class assumption based on land cover / station type
Seasonality of undercatch in temperate climatesPA DJF corrections 2× JJA correctionsSeasonal weighting essential for accurate PAPRISM500 monthly fields — do not apply a single annual CF
Over-correction risk from universal CFsApplying A&L gridded CRs directly to valley stations in Ridge & ValleyValley stations have lower effective wind than the surrounding ½° cell average implies; consider downscaling CF by exposure class
ML emerging for bias removalPAPRISM500 GROS framework (Claude + ChatGPT + Gemini)Future pipeline: train a random forest on known-exposure ASOS stations to predict exposure-corrected CRs for data-sparse COOP sites