Geocoding and Location-Level Hazard Assignment

Why Geocoding Matters

A catastrophe model can only relate an insured exposure to a geographic hazard if it has a usable location. Geocoding is the process of converting a location description—often an address—into coordinates. Those coordinates are then used to assign a location-level hazard value according to a specified map or lookup rule. A plausible-looking coordinate is not necessarily an accurate one: a street address may match to a nearby intersection, a postal-area centroid, or the wrong parcel.

The practical concern is not simply whether an address has been converted into latitude and longitude. It is whether the result identifies the insured location at a resolution appropriate for the hazard map, and whether the assignment can be reproduced and reviewed. A location error can place an exposure in a different mapped area even when the address text appears nearly correct.

Definition: Geocoding converts a location description into one or more geographic coordinates. Location resolution describes how specifically those coordinates identify the insured property—for example, a building point, parcel centroid, street segment, or broad-area centroid. Hazard assignment applies a specified geographic lookup rule to the selected coordinates to attach a hazard value to the exposure.

From Address to Assigned Hazard Value

A geocoding workflow begins with the address and the identity of the insured location. Before looking up coordinates, the address should be parsed into components such as street number, street name, unit, municipality, and postal code. Standardizing abbreviations and correcting a known formatting error can improve matching, but a standardized address is not proof that the underlying property is identified correctly. Unit or building information is especially important when a single street address contains multiple insured locations.

A geocoder may return one candidate or several candidates, together with a match type, a quality indicator, or both. A rooftop or building-point match is generally more specific than a parcel centroid; a parcel centroid is generally more specific than a street-segment point; and a postal-area centroid is usually a broad proxy. These descriptions are not guarantees of positional accuracy. A rooftop point may still be misplaced, and a parcel centroid may be unsuitable for a large parcel whose buildings span different mapped areas.

A match score is a measure generated by the geocoding process. Unless the score is explicitly calibrated as a probability for the relevant population of addresses, it must not be treated as the probability that the returned coordinates are correct. Keep the score, its meaning, the match type, the source, and any manual correction as separate pieces of information.

Once a coordinate is selected, a hazard assignment requires a defined lookup. For example, a location may be assigned the value attached to the polygon containing its point, or the value of the raster cell containing it. The lookup rule should specify the coordinate reference system, map version, treatment of points on boundaries, and handling of points outside mapped coverage. This tutorial treats the map values as given inputs; it does not estimate hazard values between map locations.

Identify the insured location.
Confirm which building, parcel, or unit the exposure record represents. Do not assume that a mailing address uniquely identifies the insured property.

Generate and review coordinate candidates.
Record candidate coordinates, match types, sources, and quality indicators. Investigate conflicting or incomplete matches rather than selecting one solely because it is the first result.

Select a resolution appropriate to the exposure.
Prefer a verified building or parcel location when available. If only a broader proxy is usable, label it as such and retain the reason for using it.

Apply a consistent hazard lookup.
Use the selected coordinate and documented map rules to assign the location-level value. Preserve enough information to reproduce that assignment.

Resolution, Uncertainty, and Exposure Weighting

A coordinate is not meaningful without its resolution and provenance. For example, a point derived from a verified building outline has a different interpretation from a postal centroid, even if both are stored with several decimal places. Decimal places describe how a coordinate is written; they do not establish how accurately the underlying location is known.

When location is uncertain, a useful diagnostic is to retain a set of plausible coordinate scenarios. Let $H_j$ be the hazard index assigned under scenario $j$, and let $w_j$ be that scenario's weight. Assume that the scenarios are mutually exclusive and collectively exhaustive for the uncertainty being represented, and that $w_j\geq 0$ with $\sum_j w_j=1$. Then the weighted index is:

\bar{H}=\sum_j w_j H_j

If the insured value $V$ is the same exposure under every scenario, the corresponding index-weighted exposure is $V\bar{H}$. This quantity is a useful comparison measure; it is not an insured loss or a monetary estimate of damage. If multiple exposures have different insured values $V_k$, their portfolio exposure-weighted index is $\sum_k V_kH_k/\sum_k V_k$, provided $\sum_kV_k>0$. A simple average of the location indices would give every location equal weight instead.

The weighted average alone does not describe the full uncertainty. Also report the plausible range of assigned values, and consider a dispersion measure where the scenario weights have a defensible interpretation. A wide range signals that the location decision may matter materially. Weights should come from a documented basis, such as verified address alternatives or an explicit judgment process; they should not be invented merely to make an average available.

Justification: Treat the scenario index as a discrete random variable $H$ that takes value $H_j$ with probability $w_j$. By the definition of expectation, $E[H]=\sum_jP(H=H_j)H_j=\sum_jw_jH_j$. If the exposure value $V$ is fixed across scenarios, linearity of expectation gives $E[VH]=V E[H]=V\sum_jw_jH_j$. This explains the weighted-index calculation. It does not imply that an expected index is an expected loss; that would require a specified loss relationship as well.

Worked Location Assessments

Worked Example: Choosing Between Two Address Matches

Method and assumptions. A hypothetical insurer reports a single building at 24 Basin Road with insured value $V=\$850{,}000$. A geocoder returns a rooftop candidate with match score 0.92 and a street-segment candidate with score 0.78. The scores are not calibrated probabilities. A property record independently confirms that the rooftop candidate is the insured building. Under the stipulated map lookup, the rooftop point receives index 1.4 and the street point receives index 0.9. Find the selected assignment and quantify the effect of using the street point instead.

Step 1: Select the location. The property record confirms the rooftop candidate. The selection is based on that corroboration, not on interpreting 0.92 as a 92% probability.

Step 2: Assign the index. The documented lookup gives $H=1.4$ at the confirmed rooftop coordinate. Thus the selected index-weighted exposure is:

\$850{,}000(1.4)=\$1{,}190{,}000

Step 3: Compare the alternative. The street-segment point would give $\$850{,}000(0.9)=\$765{,}000$. The difference is $\$1{,}190{,}000-\$765{,}000=\$425{,}000$ in index-weighted exposure units. The index difference is $1.4-0.9=0.5$.

Check and trap avoided. The selected building point is supported by independent location evidence, and the alternative is retained as a sensitivity result. Treating the score 0.92 as a probability and averaging the two indices would be unjustified because the score has not been calibrated as a probability. Neither index-weighted amount is an estimate of a claim payment.

Worked Example: Combining Three Plausible Coordinate Scenarios

Method and assumptions. A hypothetical insured location has value $V=\$3{,}000{,}000$. Address review leaves three plausible coordinate scenarios. For this exercise, documented scenario weights are 0.50, 0.30, and 0.20, and the corresponding mapped indices are 0.8, 1.2, and 2.0. Assume the scenarios cover the uncertainty under review and the weights sum to one. Calculate the weighted index and index-weighted exposure.

Step 1: Check the weights. $0.50+0.30+0.20=1.00$, so they form a complete set of scenario weights.

Step 2: Multiply each index by its weight. The three contributions are $0.50(0.8)=0.40$, $0.30(1.2)=0.36$, and $0.20(2.0)=0.40$.

Step 3: Add the contributions.

\bar{H}=0.40+0.36+0.40=1.16

Step 4: Apply the fixed insured value.

V\bar{H}=\$3{,}000{,}000(1.16)=\$3{,}480{,}000.00

Check and trap avoided. The weighted index of 1.16 lies between the smallest index, 0.8, and the largest, 2.0, as a weighted average with nonnegative weights must. A tempting but incorrect total is $0.40+0.36+0.40=1.12$; adding those three displayed contributions actually gives 1.16. The corrected index-weighted exposure is $\$3{,}480{,}000.00$, not $\$3{,}360{,}000.00$. These are index-weighted exposure units, not loss dollars.

Worked Example: Diagnosing a Missing Building Coordinate

Method and assumptions. A hypothetical insured property has value $V=\$2{,}400{,}000$, but its building coordinate is missing. The available address identifies a broad area that overlaps two mapped zones. Review supports two possible locations: scenario A has weight 0.40 and index 0.7; scenario B has weight 0.60 and index 1.5. Assume the scenario weights are valid for this location uncertainty. Calculate the weighted index, the range of index-weighted exposure, and the weighted standard deviation of the index.

Step 1: Calculate the weighted index. The contributions are $0.40(0.7)=0.28$ and $0.60(1.5)=0.90$. Therefore $\bar{H}=0.28+0.90=1.18$.

Step 2: Calculate the index-weighted exposure.

\$2{,}400{,}000(1.18)=\$2{,}832{,}000.00

Step 3: Show the scenario range. Scenario A gives $\$2{,}400{,}000(0.7)=\$1{,}680{,}000.00$; scenario B gives $\$2{,}400{,}000(1.5)=\$3{,}600{,}000.00$. The range is therefore $\$1{,}680{,}000.00$ to $\$3{,}600{,}000.00$.

Step 4: Calculate weighted dispersion. The weighted variance is $\sum_jw_j(H_j-\bar{H})^2$. The deviations from 1.18 are $0.7-1.18=-0.48$ and $1.5-1.18=0.32$. Their weighted squared contributions are $0.40(0.48^2)=0.40(0.2304)=0.09216$ and $0.60(0.32^2)=0.60(0.1024)=0.06144$. The variance is $0.09216+0.06144=0.15360$, so the standard deviation is $\sqrt{0.15360}=0.391918$, or 0.392 to three decimals.

Check and trap avoided. The mean 1.18 lies between 0.7 and 1.5, and the range shows that the missing coordinate has a material effect on the assigned index. Reporting only the average would conceal that spread. The standard deviation summarizes dispersion under the stated weights; it does not repair the missing coordinate or establish that the weights are objectively correct.

Controls and Common Traps

A useful location record allows another analyst to understand not only the selected coordinate, but also how it was selected and what uncertainty remains. At a minimum, retain the source address, standardized address, coordinates, coordinate reference system, match type, geocoder or reference-data source, map version, lookup result, and any review or override. Where a coordinate is a proxy, label the proxy explicitly. Where multiple candidates remain plausible, preserve the alternatives and the basis for any weights.

Do not equate coordinate precision with accuracy. Many decimal places can represent an approximate centroid just as easily as a verified building point.
Do not treat a match score as a probability by default. Use it according to its documented meaning.
Do not hide a broad proxy. A postal-area centroid should not be recorded as though it were a verified property location.
Do not silently choose among materially different candidates. If their assignments differ, document the decision and assess the sensitivity.
Do not confuse index-weighted exposure with loss. It is a weighted comparison measure unless a separate, explicit loss model gives it a loss interpretation.

For a location near a mapped boundary, the lookup rule matters: small coordinate changes may switch the assigned area. Apply the stated boundary convention consistently and flag cases where plausible positions produce different assignments. If no credible coordinate or scenario set can be established, report the location as unresolved rather than implying that a broad proxy is precise.

Key takeaway: A defensible location-level hazard assignment connects a clearly identified insured location to a documented coordinate, a stated location resolution, and a reproducible map lookup. When coordinates are uncertain, retain plausible alternatives and report both the weighted result and the sensitivity that the average alone can hide.

Check Your Understanding

For each hypothetical case, state the requested assignment or diagnostic using the information given.

A building worth $\$1{,}200{,}000$ has two plausible coordinate scenarios: index 0.8 with weight 0.25 and index 1.6 with weight 0.75. Find the weighted index and index-weighted exposure. (Answer: Weighted index $=1.40$; index-weighted exposure $=\$1{,}680{,}000.00$.)
A location worth $\$4{,}000{,}000$ has three plausible assignments: index 0.9 with weight 0.40, index 1.1 with weight 0.35, and index 1.8 with weight 0.25. Find the weighted index and index-weighted exposure. (Answer: Weighted index $=1.195$; index-weighted exposure $=\$4{,}780{,}000.00$.)
A geocoder returns a street-segment point and a postal-area centroid for the same address, with similar-looking coordinate precision. What should be checked before using either point for assignment? (Answer: Verify which insured location each candidate represents, compare match type and provenance, establish the appropriate location resolution, and apply the documented lookup; decimal places alone do not establish accuracy.)
Two exposures have values $\$2{,}000{,}000$ and $\$1{,}000{,}000$, with assigned indices 0.8 and 1.4. Find the portfolio exposure-weighted index and compare it with the simple average of the indices. (Answer: Exposure-weighted index $=(\$2{,}000{,}000(0.8)+\$1{,}000{,}000(1.4))/\$3{,}000{,}000=1.0$; simple average $=(0.8+1.4)/2=1.1$.)
A $\$5{,}000{,}000$ exposure has two supported coordinate scenarios: index 0.5 with weight 0.60 and index 2.0 with weight 0.40. Find the weighted index, index-weighted exposure, and scenario range of index-weighted exposure. (Answer: Weighted index $=1.10$; index-weighted exposure $=\$5{,}500{,}000.00$; range $=\$2{,}500{,}000.00$ to $\$10{,}000{,}000.00$.)

What You'll Learn