The Gold Standard of Weather Radar Attenuation Correction | Bounding Intermediate Path Loss from an Endpoint Total and a Local Cap

Imagine a straight ten-kilometre path from a radar to a distant reference. A stipulated one-way propagation loss over that entire path is four dB. We want the accumulated one-way loss before a target seven kilometres away.

If most attenuation occurs near the radar, the intermediate target experiences most of the four dB. If it occurs near the distant endpoint, the target experiences much less. The total alone does not locate the attenuating material.

This is an information question before it is a correction question. Copying the full-path loss into every range bin would assign distant attenuation to nearer targets that do not share the full route. Assuming a uniform coefficient is another possible choice, but it adds a distribution assumption. We will first retain every distribution allowed by the stated constraints and ask what those constraints actually force.

Contents · Next chapter

Let r denote distance along the stipulated path, measured in kilometres from its origin. The full path length is D. Let k(r) be a one-way specific attenuation in dB per kilometre.

Assume k is integrable and satisfies zero ≤ k(r) ≤ K along the path. The supplied constant K is nonnegative. Integrability allows us to add its contributions; the inequalities prohibit negative loss and cap the permitted local loss per distance.

We do not assume a spatially constant coefficient or require a smooth profile. Piecewise-constant profiles are allowed. This matters when proving that an extreme is attained: a sharp transition in a teaching construction is admissible here, although a different physical model might impose continuity, smoothness or additional measurement constraints. Keep the function family attached to the claimed result instead of borrowing a conclusion from a broader family without noticing the difference.

Contents · Previous chapter · Next chapter

Write S(r) = integral from zero to r of k(u) du. This is the one-way accumulated attenuation before range r. Its units are dB because dB per kilometre multiplied by kilometres gives dB in the additive logarithmic loss account.

The supplied total is S(D) = P. For the original example, D = ten kilometres and P = four dB. The value S(0) is zero by construction.

Nonnegative k makes S nondecreasing. Moving farther along this stipulated path cannot remove an earlier accumulated loss. The upper bound on k also gives S(b) − S(a) ≤ K(b − a) for b greater than a. These properties concern accumulated propagation loss. They do not imply that measured reflectivity itself must decrease with range, since scattering strength can vary between targets independently of the path-loss account.

Contents · Previous chapter · Next chapter

The total must satisfy zero ≤ P ≤ KD. A nonnegative coefficient cannot produce a negative integrated loss, and a coefficient never exceeding K cannot produce more than K times the path length.

For D = ten, K = one dB per kilometre and P = four, the condition holds. A supplied P of twelve would be inconsistent with the same cap and path length. A formula that nevertheless returns a pair of numbers would not establish that any allowed profile exists.

An infeasible account prompts review of the observations and assumptions. Perhaps the cap is unjustified, the total includes another loss mechanism, the distance is wrong or one-way and two-way definitions were mixed. Do not label an inconsistent total as an unusually extreme allowed atmosphere. First identify which part of the stated model can no longer be maintained.

Contents · Previous chapter · Next chapter

At an intermediate r, the accumulated loss cannot exceed Kr, because the first r kilometres each contribute no more than K per kilometre. It also cannot exceed P, because the remaining path contributes a nonnegative amount.

Both inequalities must hold, so S(r) ≤ min(P, Kr). This minimum keeps the tighter restriction at the chosen range. Using their maximum would allow values violating one of the original constraints.

For the seven-kilometre target in the original record, Kr is seven dB while P is four. The total is the active upper restriction, giving S(7) ≤ four. The local-cap restriction is still meaningful elsewhere: at a one-kilometre target it gives an upper loss of one dB, even though the full-path total is four. A range-specific bound preserves that distinction.

Contents · Previous chapter · Next chapter

The remaining distance is D − r. It can accommodate at most K(D − r) dB under the cap. Since the full total is P, at least P − K(D − r) must already have accumulated before r.

There is also the nonnegative lower limit S(r) ≥ zero. Combining them gives S(r) ≥ max[0, P − K(D − r)].

For D = ten, P = four, K = one and r = seven, the final three kilometres can accommodate no more than three dB. At least one dB must therefore occur before the target. This is a capacity argument: the loss that cannot fit after the target must fit before it. It uses the supplied cap and total, without assuming where individual rain cells or other attenuating regions are located.

Contents · Previous chapter · Next chapter

The resulting limits are max[0, P − K(D − r)] ≤ S(r) ≤ min(P, Kr). For the original seven-kilometre target they give one ≤ S(7) ≤ four dB.

This interval is conditional on exact P, the specified cap, the path definition and the allowed coefficient family. It is not a confidence interval. No probability distribution over profiles has been supplied, so the middle of the range is not automatically the expected or most likely loss.

The calculation answers a narrower question than identifying a physical attenuation profile. It states how much loss can lie before one range while remaining consistent with the entire account. To call the limits sharp, we still need to exhibit allowed profiles attaining each end. Without those constructions, we would have only necessary inequalities that might conceivably leave unattainable extreme values.

Contents · Previous chapter · Next chapter

Place coefficient one dB per kilometre over the final four kilometres, from range six to ten, and zero before range six. Its full integral is four dB, matching P, and its coefficient never exceeds K.

At range seven, only the one-kilometre segment from six to seven has contributed. Thus S(7) is exactly one dB, attaining the lower bound.

The placement is a mathematical witness, not a reconstruction of an actual storm. We did not observe a rain region beginning at six or a uniform coefficient thereafter. The construction proves that the supplied information cannot exclude an intermediate loss as small as one. Any stronger lower claim would need additional evidence or a stricter profile family. Keeping that logical role explicit prevents a convenient extreme profile from becoming an invented meteorological narrative.

Contents · Previous chapter · Next chapter

Place coefficient one over the first four kilometres and zero over the remaining six. Again the total is four and the cap is respected. All four dB have accumulated by range four, so S(7) is exactly four.

The near-loaded and far-loaded profiles share the same full-path loss while giving different losses at the intermediate target. Their contrast directly demonstrates why one endpoint total does not identify a range-resolved correction.

Both constructions can be checked by rectangular areas: coefficient times occupied length. Drawing them helps a learner see that the uncertainty concerns location of the permitted loss, not an arithmetic uncertainty in four divided by ten. The same endpoint average coefficient can describe very different distributions, and the path seen by an intermediate target ends before some of those differences have contributed.

Contents · Previous chapter · Next chapter

For an interior range r, choose any candidate s inside the derived bounds. Set k equal to s/r on the first segment and to (P − s)/(D − r) on the second.

The upper inequalities ensure that the first coefficient is between zero and K. The lower inequalities ensure the same for the second. The two rectangular integrals are s and P − s, so their sum is P and S(r) equals the chosen s.

Consequently every value in the interval is feasible for this integrable bounded-coefficient model. This conclusion depends on the construction and does not follow merely because we reported an enclosing interval. Other inverse problems can have disconnected feasible answer sets. Here the simple two-segment witness supplies the missing attainability argument and shows precisely which regularity assumptions the proof uses.

Contents · Previous chapter · Next chapter

The construction using s/r requires an interior target with r greater than zero. At the origin, the integral is zero directly; no division is needed. At the full endpoint, S(D) equals the supplied P directly, and the second segment has zero length. These boundary cases are already consistent with the bound formula.

If P is zero, nonnegative k must have zero integral, so every prefix loss is zero. Values at isolated points do not affect an ordinary integral. If P equals KD, the cap must be saturated almost everywhere and S(r) equals Kr. In both limiting totals, the bounds collapse throughout the path rather than only at its endpoints.

If K is zero, feasibility forces P to zero. Formulas locating a filled segment through P/K would then involve zero divided by zero, even though the underlying account is simple. Use the integral definition and feasibility conditions to resolve such cases before applying a witness construction derived for a positive cap. A failed division need not signal a physically mysterious loss profile; it may mean that a simpler boundary case was overlooked.

Contents · Previous chapter · Next chapter

A uniform profile would use k = P/D, giving S(r) = Pr/D. In the original record k is 0.4 dB per kilometre and S(7) is 2.8 dB.

That value lies inside the sharp range [1, 4], as it should, because the uniform profile satisfies the cap. It is one allowed answer, not a consequence uniquely forced by the total.

A useful report distinguishes the uniform-profile estimate from the wider model-family bounds. If independent information supports approximate uniformity, that can motivate a narrower model. If no such evidence exists, placing a smooth line through the accumulated-loss endpoints does not create it. The visual neatness of a correction curve is not evidence that the unknown spatial distribution followed that curve.

Contents · Previous chapter · Next chapter

The coefficient and P in this guide describe one-way loss. In the stipulated ideal monostatic radar model, a return travels outward and back through the same attenuating path. Its two-way propagation loss before r is therefore 2S(r).

For the seven-kilometre target, the conditional two-way range is two to eight dB. At the full endpoint it is exactly eight because P is exactly four one-way.

This doubling follows the specified symmetry and quantity definition. If an input already represents two-way path-integrated attenuation, doubling it again would be a mistake. Before using a coefficient or published formula, identify whether it describes a one-way quantity, a two-way quantity or a differential channel quantity. Similar names and the common unit dB do not guarantee that two records use the same convention.

Contents · Previous chapter · Next chapter

Suppose a perfectly calibrated, attenuated target return is reported as thirty-five dBZ, and stipulate that the only missing correction is the two-way propagation term in our model. The restored value would be 35 + 2S(7).

The sharp loss range therefore gives restored values from thirty-seven to forty-three dBZ. The uniform-profile assumption would give 35 + 5.6 = 40.6 dBZ, one compatible value inside that interval.

This does not estimate surface rainfall or identify particle type. Nor does it certify that the endpoint total was correctly measured. It propagates an explicitly declared attenuation account into one simplified corrected quantity. The Weather Radar Rainfall Estimation guide explains the additional chain from radar observables to a precipitation estimate.

Contents · Previous chapter · Next chapter

Because the displayed power-related quantity is logarithmic, a two-way correction C dB corresponds to a factor 10^(C/10) under the stipulated power convention. Adding two dB is not adding two percent.

The lower and upper corrections two and eight dB correspond to factors about 1.58 and 6.31. Their ratio is 10^(6/10), about 3.98. Thus uncertainty in the spatial allocation of a fixed total can imply a substantial range of multiplicative correction factors at one intermediate target.

These factors belong to the simplified loss restoration, not directly to rain-rate multipliers. Converting reflectivity into a rain rate introduces another model and may be nonlinear. Keep each transformation attached to its input quantity and assumptions rather than carrying the same percentage through several different observables.

Contents · Previous chapter · Next chapter

For D = ten, P = four and K = one, evaluate the same formulas at different ranges. The endpoint total remains unchanged throughout this comparison.

Near the origin, the cap limits how much loss could already have occurred. Near the endpoint, the remaining distance cannot hold much of the fixed total, so most must already have accumulated. At the endpoint the bounds meet because its total was supplied exactly. This pattern follows the geometry of the account, not a claim that uncertainty must always behave this way in an operational product.

Contents · Previous chapter · Next chapter

Separate bounds at two ranges are marginal statements. Values selected independently from their intervals might not arise from one common coefficient profile.

For example, the original bounds permit S(3) as large as three and S(7) as small as one when each is considered separately. Selecting both together would require accumulated loss to fall from three to one while travelling farther, contradicting nonnegative k.

A range-resolved profile must preserve shared-path constraints. For r₂ greater than r₁, its values satisfy zero ≤ S(r₂) − S(r₁) ≤ K(r₂ − r₁), as well as compatibility with the final total. Retain these links when presenting multiple uncertainty ranges. Drawing an arbitrary line through independently chosen endpoint bounds can produce an impossible accumulated-loss curve even when each plotted point passed its individual interval check.

Contents · Previous chapter · Next chapter

For proposed values s₁ at r₁ and s₂ at r₂, examine three segment averages: s₁/r₁, (s₂ − s₁)/(r₂ − r₁) and (P − s₂)/(D − r₂).

When both target ranges are strictly inside the path, a piecewise-constant construction is feasible exactly when all three averages lie between zero and K. Each segment then supplies the required share of the total without exceeding its capacity.

For r₁ = three, r₂ = seven and P = four, choose s₁ = one and s₂ = three. The coefficients are one third, one half and one third, all within a cap of one. This gives an explicit jointly compatible profile. The check is stronger than separately confirming s₁ and s₂ lie in their marginal ranges, because it preserves the relation between them along the common path.

Contents · Previous chapter · Next chapter

Suppose the original account is supplemented with an exact one-way accumulated loss S(5) = two dB. This divides the path into two subpaths, each with a known total and the same cap.

To bound S(7), start from the known two at range five. Across the five-to-ten segment, another two dB must occur. Of that segment, two kilometres lie before seven and three after it. Its loss before seven can range from zero to two.

Thus S(7) lies between two and four, narrower than the original [1, 4]. The new information removes profiles that would place too little loss before range five. It still does not determine a unique correction at seven. An added observation can narrow a family without eliminating every compatible distribution, which is why the revised bounds should be recalculated rather than replaced by an unjustified point estimate.

Contents · Previous chapter · Next chapter

Repeating the exact supplied statement P = four adds no new spatial constraint in this deterministic model. Every admissible profile already satisfies it. By contrast, the intermediate value at five restricts where the loss can lie.

Actual repeated measurements may improve an uncertainty estimate, reveal inconsistency or provide independent information about a noisy total. That is a different observation model from treating the original total as exact.

The useful question is what the additional record distinguishes. Another decimal place in an endpoint total may not separate near-loaded and far-loaded profiles. An appropriate interior observation can. This habit transfers beyond radar: before requesting more data, identify the alternatives still allowed and ask which measured quantity would make those alternatives predict different results.

Contents · Previous chapter · Next chapter

Suppose P is known only to belong to [A, B]. First intersect this interval with the feasible total range [0, KD]. If the intersection is empty, no profile satisfies the stated account.

For feasible endpoints A* and B*, the smallest possible S(r) is max[0, A* − K(D − r)] and the largest is min(B*, Kr). These are extrema over both the allowed total and the allowed spatial distribution.

For D = ten, K = one and P between three and five, the seven-kilometre loss can range from zero to five. The lower witness can place all three dB after seven. The upper witness can place five before it. This is a deterministic admissible range, not a statistical confidence statement. Its wider extent reflects two distinct unresolved features: the total and its allocation.

Contents · Previous chapter · Next chapter

The interval calculation assumes every total in the stated range can be paired with a permitted coefficient profile. Additional evidence may link the total to another quantity, such as a shared calibration parameter or a locally estimated coefficient.

Combining independent-looking ranges without retaining such links can admit impossible combinations. A small endpoint total and a large interior contribution, for example, cannot coexist with a nonnegative remaining loss even if each value appears in a separate summary interval.

A rigorous extension defines the joint admissible set first, then optimises S(r) over that set. The simple formula remains useful for the simpler model it actually solves. More complicated uncertainty should prompt a more explicit constraint account rather than a silent claim that all interval endpoints can be combined freely.

Contents · Previous chapter · Next chapter

A constant K is not essential to the basic argument. Suppose a supplied nonnegative upper function K(u) varies along range, and k(u) must lie between zero and K(u).

Define the capacity before r as Cbefore = integral from zero to r of K(u) du, and after r as Cafter = integral from r to D of K(u) du. The accumulated loss satisfies max(0, P − Cafter) ≤ S(r) ≤ min(P, Cbefore).

For a fictional seven-kilometre first segment with cap one half, Cbefore is 3.5 dB. If the final three kilometres have cap one, Cafter is three. With P = four the range is [1, 3.5]. The upper restriction is now tighter than before. These cap values remain supplied teaching assumptions; the extension does not derive a meteorologically valid capacity map from a radar display.

Contents · Previous chapter · Next chapter

If independent evidence supplies k(u) ≥ m(u) as well as an upper bound, there is unavoidable loss on each segment before any remaining total is allocated. That changes both the feasible total and the range-specific limits.

One way to analyse it is to write k = m + q. The residual coefficient q is nonnegative and no greater than K − m. Subtract the integrated lower profile from the total, apply the capacity argument to q, then add the compulsory prefix contribution back.

The transformation is useful because it separates what is forced locally from what remains free to move along the path. It is valid only when the lower and upper functions themselves are compatible and the residual total is feasible. A lower bound should come from justified information, not be chosen to make a preferred correction appear inevitable.

Contents · Previous chapter · Next chapter

Our sharp witnesses include abrupt transitions between zero and the cap. They are allowed by the integrable bounded-coefficient family. If the scientific question also imposes a bound on how quickly k can vary, some of those profiles are excluded.

The previous inequalities remain necessary because they used only constraints still present. Their endpoints may cease to be attainable in the smaller family, however. A new optimisation problem must determine whether the sharp limits move or become limiting values approached by smoother profiles.

Do not add a smoothing procedure to the plotted witness and assume it preserves both its total and its extreme prefix value. Smoothing can redistribute loss across the target range. State the revised regularity requirement, preserve the integral constraint and verify the new witnesses. The meaning of sharpness always belongs to a named family of allowed profiles.

Contents · Previous chapter · Next chapter

A supplied endpoint loss may include losses near the instrument, target-dependent effects or components that do not belong to the same bounded distributed coefficient. Treating all of it as rainfall along the path changes the physical interpretation.

The Ultrasonics echo-ratio guide develops a different ambiguity between boundary reflection and bulk propagation loss. It is useful background on why a combined observed reduction does not automatically identify one mechanism.

For weather radar, the subject root discusses wet-radome and differential-channel issues. Our bounds assume the supplied P already belongs to the stated one-way distributed account. They do not separate those mechanisms or validate that assumption. A mathematically sharp answer can remain physically inappropriate if the input loss was assigned to the wrong process.

Contents · Previous chapter · Next chapter

Restoring a modelled propagation loss is different from recovering an observation the instrument never made. Blockage, inadequate coverage, nonweather scatterers and low-quality measurements can require other treatments.

The Weather Radar Dual-Polarization Hydrometeor Classification guide provides a related route through target mixtures and classification limits. A plausible attenuation allocation does not prove that a particular echo represents liquid rain or that a rain-specific conversion is appropriate.

Keep unavailable observations marked as unavailable under the relevant product definition. Assigning them a bright corrected value because the path-loss upper bound is large would confuse an allowance with evidence of an actual target. The bound constrains one missing propagation term within a stipulated model; it supplies neither a target’s original scattering strength nor a complete quality assessment.

Contents · Previous chapter · Next chapter

A correction that reproduces the total used to constrain it has satisfied a fitting or accounting condition. That agreement is not an independent validation of its distribution across range.

Near-loaded and far-loaded witnesses both match P exactly. An observation distinguishing their predicted interior effects is needed to decide between them. In an actual study, validation also needs compatible spatial footprints, times, radar bands, precipitation regimes and measurement quality.

The KNMI institutional publication record linked below describes a 2021 study comparing attenuation corrections for two operational Dutch C-band radars using gauge references. That study supplies research context, not the cap or total in our examples. Its existence does not make an invented bounded-profile model operationally validated. Preserve the distinction between deriving what an assumed account allows and testing whether that account improves a real measurement product.

Contents · Previous chapter · Next chapter

Record D, the coordinate origin, path geometry, coefficient units, one-way or two-way convention, P and the evidence supporting the local cap. Check feasibility before entering a target range.

For each target, record the capacity before it, the capacity after it and the resulting lower and upper losses. Keep the profiles or segment coefficients used as witnesses. If multiple targets are reported together, test whether the selected values share one feasible cumulative curve.

Finally identify any subsequent transformation, such as doubling a one-way loss or adding a two-way correction to a calibrated logarithmic reflectivity. Record that operation separately from a rainfall conversion or ground-impact interpretation. A later reader should be able to reconstruct every numerical step while seeing which assumptions concern mathematics, which concern the instrument and which concern atmospheric or hydrological processes.

Contents · Previous chapter · Next chapter

A learner may divide the total by distance, obtain 0.4 and declare the intermediate loss known. The division correctly calculates the full-path average coefficient. The unstable step is treating that average as the coefficient at every location.

Repair it by drawing the two rectangular witness profiles. Both have area four, but their areas before seven are one and four. Once that contrast is secure, derive the capacity bounds and place the uniform answer 2.8 inside them.

Another common mistake is applying the final eight-dB two-way loss to every target. Ask which parts of the path an intermediate return actually traverses in the stipulated geometry. That question returns the explanation to the physical path definition before the arithmetic. Correct units and a correct factor of two cannot repair a correction assigned to the wrong segment.

Contents · Previous chapter · Next chapter

Use D = twelve kilometres, P = six dB one-way, K = 0.75 dB per kilometre and target range r = eight kilometres. Confirm that the total is feasible, then calculate the two capacities and the sharp prefix range.

Construct one allowed profile attaining the lower value and one attaining the upper value. Then calculate the uniform-profile prefix loss and its corresponding two-way correction. Explain why the uniform answer is compatible without being uniquely determined.

Finally propose values of accumulated loss at two different ranges and test their joint compatibility through segment averages. Include at least one pair that passes the separate interval checks but fails the shared-path check. The purpose is to connect the numerical bounds with the family of distributions they describe, rather than memorise a formula detached from the path account.

Contents · Previous chapter · Next chapter

The total capacity is nine dB, so P = six is feasible. Before eight kilometres the capacity is six; after it the capacity is three. The sharp one-way prefix interval is therefore [3, 6] dB and the two-way interval is [6, 12].

The uniform coefficient is one half dB per kilometre, giving a prefix loss of four and a two-way correction of eight. A lower witness places the cap over the last eight kilometres, starting at four, so three dB have accumulated by eight. An upper witness places it over the first eight, giving six by the target.

For extension, use the same two-segment construction to attain a prefix value of 4.5. The first coefficient is 4.5/8 = 0.5625 and the remaining coefficient is 1.5/4 = 0.375, both within the cap. This independent witness checks the claim that intermediate values are feasible.

Contents · Previous chapter · Next chapter

Does a known full-path loss identify its distribution? No. Does a uniform average supply a unique correction at each range? Only after adding that distribution assumption. Does sharp mean probable? It means that an allowed profile attains the bound. Can separate range intervals be combined independently? Their shared-path constraints must still hold.

For meteorological context, NWS guidance on radar rainfall estimation and limitations distinguishes attenuation and other observation difficulties. The KNMI institutional record for the 2021 C-band attenuation-correction study provides a research route. Both were checked on 24 October 2026. The numerical caps, totals and sharp-bound constructions here are original teaching models, not settings or performance results taken from those sources.

Return to Weather Radar Attenuation Correction for phase-based methods and quality limits. Continue through How X Works and The Gold Standard of Education: Learning Transfers into Life with a reusable question: what does the total constrain, and what evidence would locate the part that matters?

Contents · Previous chapter · Return to Weather Radar Attenuation Correction