Resilience Insights

In the rare cases where there are duplicate parcel IDs, the first parcel is retained and the duplicates are dropped. A verification step further ensures that duplicate parcel IDs do not exist in the dataset.

In some cases, such as parcels for condominiums, multiple geometries that are exactly the same or nearly the same are stacked on top of each other. These geometries are dissolved into a single flat geometry. For numeric data, the maximum across all parcel attributes is retained. For non-numeric data, the attributes of the parcel with the lowest parcel ID are retained.

Figure 1. Representative example of duplicate geometry handling

Nested geometries sometimes occur in master-planned developments or subdivided residential neighborhoods where the outdoor area is owned by one entity, but each unit is privately owned. In these cases, a parent geometry (e.g., a planned unit development parcel) can have several child geometries (e.g., the planned units) as individual parcels completely contained by the larger parent geometry. For scenario development to correctly account for the total land area in these cases, it is essential that no land is double-counted, meaning that the parent geometry does not include the area of the child parcels. To fix this, the child geometries are cut out of the parent geometry. The result is that the parent geometry encompasses the shared common area, while each child parcel contains the information for the particular unit.

Figure 1. Representative example of nest geometry handling

Finally, there are parcels that overlap but are not entirely contained in one or the other. This happens most often with condominiums where the floor plans of each unit differ across floors. For this type of parcelization to work in Analyst, it is important that no geometries overlap, even if they represent different floors in a structure. For these cases, the lower parcel ID has its geometry cut out and the overlapping portion is retained for the parcel with the higher parcel ID. The parcel area is recalculated for each new geometry, but each geometry retains its original attribute data, meaning that the unit number and building area are not impacted by the change in geometry.

Figure 1. Representative example of overlapping geometry handling

Most jurisdictions represent land use at the parcel scale, using codes that reflect specific uses. Analyst “translates” this local information to represent development in terms of Analyst's standardized Building Types, which are foundational to scenario development and analysis. While land use and urban form are the subject of both local/regional land use classification systems and Analyst land use types, they are distinct languages. Land use codes are predominantly use-based and static, whereas Analyst's land use types (including Building Types at the parcel scale or Place Types at the census block scale, and the higher-level generalized categories into which they are classified) are primarily form-based. Analyst land use types are designed to be dynamic and expansive to capture the many variants of built form and land use. To relate Analyst Building and Place Types to the universe of local land use codes (as represented by CoreLogic's set of nearly 300 codes), we developed the Generalized Land Use Classification (GLUC) system.

The GLUC system is comprised of approximately 100 general land use designations. A crosswalk is used to associate each land use code from the CoreLogic parcel data with one of these general land use designations, each of which is associated with one or more Analyst Building Types. As part of Base Canvas creation, a translation algorithm uses this crosswalk to narrow the range of Building Types to which each CoreLogic land use code can be translated. In turn, residential and employment densities, as calculated in later steps, are used to select the closest fitting Building Type.

For example, CoreLogic’s "APARTMENT" land use code (#106) is crosswalked to the general land use designation "residential multifamily - all." A multifamily Building Type will then be selected based on the calculated residential density of the parcel. By contrast, CoreLogic’s "HIGH RISE CONDO" code (#117) is crosswalked to the "residential multifamily - high" designation, which is associated with a more restrictive set of potential Building Types.

The crosswalk between CoreLogic land use codes and Analyst land use are available upon request. (Please contact Analyst Support through the in-app chat, or by email.)

In some cases, CoreLogic commercial-based land use codes are incorrectly assigned to parcels that are not commercial but are held by a commercial entity (for example, a private developer). In such cases, we first check if the parcel contains a building or not. If the parcel does not contain a building, we invalidate the “commercial” land use assigned by CoreLogic and use supplemental datasets to inform the land use code.

One known issue in the parcel data we receive is the presence of right-of-way (ROW) parcels or areas covered by public roads. These records are not readily identifiable in the source data, which historically has led to mistyping of these parcels. To help solve this issue, Analyst manually collected data on a sample of parcels, identifying them as right-of-way or not. We then developed a machine learning model to predict whether each parcel is ROW. Our model relies on a variety of metrics about the parcel geometry, and other, nationally-available data, including:

The Analyst ROW model has been designed to give us higher trust when the model says that a parcel is right-of-way, compared with when the model says that a parcel is not right-of-way. While some ROW parcels may continue to be mistyped, we aren’t likely to incorrectly identify a parcel as ROW when it is truly not ROW. This allows us to place substantial trust in the model, but as a consequence, we may not catch all ROW parcels. After applying the model to all parcels to predict whether they are ROW, we consider this feature as a “supplemental dataset” for the parcel, as described in Supplemental Datasets.

Supplemental datasets provide more specificity where the parcel data providers' land use data may be lacking. Analyst uses a number of additional datasets for locating land uses and points of interest as summarized in the Supplemental Datasets for Built Form Typing and Disaggregation Table.

Analyst tags parcels with a land use from the parcel data providers and land uses from these supplemental datasets where they apply. The land use for a parcel is then “resolved” by picking one land use from this set of options. The land use that is most in agreement from the set of options is given priority, unless there is an exception-based rule. For example, between a choice of one Industrial - All, two Commercial - Alls and vs. one Commercial - Office, the Commercial - Office Land Use is chosen since the parcel is most likely a commercial parcel with one land use specifically calling it a “Commercial - Office”.

Supplemental datasets can contain either polygon or point features. With polygon datasets, at least 25% of a parcel must be covered by the dataset to be tagged. With point datasets, a parcel is tagged directly if a point intersects it. If a parcel touches a 50-meter buffer around a point, it receives a buffer tag, which is only used to resolve land uses if a parcel’s original land use code is vacant or null.

The process of typing parcels using a polygon dataset is exemplified in Polygon-based typing using the Census Landmarks Dataset. The campus geometry defined by the Census TIGER dataset is shown with the orange boundary, while parcels typed as Campus College - Low are shaded blue. All parcels where more than 25% of the polygon intersect with the landmark geometry are retyped accordingly. The large parcel in the upper right side of the image is not typed as campus, as less than 25% of its area intersects with the landmark polygon.

Figure 1. Polygon-based typing using the Census Landmarks Dataset

Polygon-based typing using the Census Landmarks Dataset shows examples of direct and buffered tagging using point data, in this case the SABINS schools dataset (which provides point locations of K-12 schools in the National Center for Education Statistics' Common Core of Data). The blue parcels are all typed correctly as schools, either through direct tagging, or buffer tagging and subsequent typing because the parcels had null land use codes.

Figure 2. Point-based Typing of Schools using SABINS Dataset

Lastly, there are some supplemental sources that are used to assign land use types directly. For example, golf courses are assigned the Golf Course Building Type. The Analyst generalized land use designation plays a role only for disaggregating data as part of a later step (Disaggregating Block-Level Data to Parcels).

CoreLogic provides two attribute columns that help identify the number of dwelling units present on a parcel: Building Units and Units Number. The definition of each is shown in CoreLogic Dwelling Unit Attributes.

The Building Units attribute is useful for identifying single-family homes. However, it is insufficient for calculating numbers of multifamily units as it would only count the building that houses all of the units. For that reason, we use the Units Number attribute to count the numbers of multifamily dwelling units.

As Units Number contains information on business units as well as residential units, the first step is to differentiate between the two. We do this by using the CoreLogic land use codes to classify parcels as residential, commercial/employment, or mixed use. Residential and mixed use parcels are assigned their Units Number value as the count of dwelling units, while parcels classified as commercial/employment are ignored.

Even after all of these transformations, there are still instances where the raw parcel data simply does not provide enough information to impute dwelling units. In these cases, data from the 2010 US Decennial Census is used to identify and fill in gaps. Parcel-level data is first aggregated to the block scale to allow for a direct comparison of dwelling unit totals against the Census block totals. Blocks that satisfy the following conditions are flagged as cases where raw parcel data should be substituted with census dwelling unit totals:

For parcels that match these cases, block-level totals are disaggregated down to parcels using land use codes to identify parcels that can accommodate residential data (residential or mixed use parcels). This process is detailed further in Disaggregating Block-Level Data to Parcels.

Beyond the process described above, some modifications need to be made in cases where unit counts or land use codes originating with the original parcel source data are not consistent with the CoreLogic attribute definitions, or where the values do not follow the pattern of the rest of the dataset. In these cases, modifications are made. There are several types of data inconsistencies: unreliable building units data; overtyping with the apartment land use code; overtyping with the single family residential land use code; and the representation of attached single-family units.

Each modification is described in more detail in the sections that follow.

Where data is not available at the parcel level, census block-level counts for dwelling units and employment is disaggregated down to the parcels. Disaggregation is guided by the parcels' general land use designation, each of which has rules as to the types of dwelling units and/or employment sectors it can include. For example, a parcel with a Single Family Detached land use designation can take on dwelling units, but not employment, from its parent block. Similarly, a parcel with a Retail land use designation can take on retail employees, but not industrial employees.

From there, dwelling units and employment are distributed among parcels in proportion to their land area, such that larger parcels receive more jobs while smaller parcels receive less.

By default, parcels that are classified as Vacant, Open Space, or Other are excluded from disaggregation. That said, there are some exceptions. If there are dwelling unit or employment counts at the block level, yet all parcels within the block are classified as one of Vacant, Open Space, or Other, the dwelling units or employment get assigned to only the Open Space or Other parcels. If all the parcels are Vacant, the disaggregation logic distributes the dwelling units and employment by land area.

Lastly, data is not disaggregated to parcels under 100 square feet.

The total living square feet for a parcel is allocated to the canvas attributes for building area by housing type and employment by subsector according to the dwelling units and employment present on the parcel. The logic used to distribute the square footage is summarized in the Building Area Distribution Logic Table.

For cases where Living Square Feet data is missing for the parcel, building area is imputed using default assumptions for building area per unit by housing type, and per employee by subsector. The assumptions vary according to broader land development category as identified by the intersection density of the census block in which a parcel is located. Areas with intersection densities above 150 per square mile are considered Urban or Compact, while those with lower densities can be Suburban or Rural. The assumptions are summarized in the Default Building Area Assumptions Table.

Within the top-level residential and employment categories, there are subcategories by dwelling type and employment sector. These parcel area subcategories are not mutually exclusive—each receives the total parcel area if the associated uses are present on the parcel. For example, if a parcel has both retail employment and office employment, both the retail parcel area and office parcel area will be populated with the same value—that for the total area of the parcel. The All Parcel Area Attributes Table includes a full list of the parcel area columns.

As described earlier, parcels are first assigned an Analyst general land use designation based on their CoreLogic land use codes (see Crosswalking to General Land Use Designations). Each general land use designation is associated with one or more Building Types, effectively narrowing down the potential candidates. In this step, the Building Type that "most closely" matches the density of each parcel is identified.

“Closeness” is measured as the lowest standardized absolute difference between the Building Type and parcel densities. To do this, the densities of the parcels and Building Types are first standardized into scores by subtracting the mean and dividing by the standard deviation of the corresponding building types set. Then, each parcel’s standardized density score is compared to the standardized scores of its candidate Building Types. The differences are then squared and summed. The Building Type that corresponds to the least sum-of-squares is selected and assigned to the parcel.

If the dwelling unit density and employment density were the axes of a two-dimensional graph, this sum-of- squares would represent the distance between the parcel’s data point and the Building Type data point. Therefore, the least sum-of-squares would represent the building type that is closest to the parcel’s data point. Other attributes could be represented similarly on an n-dimensional graph.‌

Currently, the process uses dwelling unit density for residential land use designations, employment density for employment land use designations, and both for mixed land use designations.

Relying purely on the density-based approach will not capture Building Types representative of special land uses such as parks, open space, water, schools, or cemeteries. These types are not density-based, so their identification is based on the use of supplemental datasets (see the Supplemental Datasets section for more details). A few exceptions are detailed below.

Parcels are assigned the Water land use if census data indicates that the area of the parent block geography is covered entirely by water¹, unless they are classified with residential land uses. This exception accounts for residential parcels at the edge of water bodies.

Institutional types, such as courthouses, libraries, or city halls, cannot be identified using the density-based approach. In the CoreLogic data, these parcels are sporadically categorized as public, tax exempt, or state property, all of which are hard to parse into specific Building Types. If these parcels have not already been typed using supplemental datasets, they are categorized as Open Space if they are rural (often they are state or regional parks), Non-Urban Civic if they are in developed areas with intersection densities under 150 per square mile, or Urban Civic if they are in developed areas with intersection densities over 150 per square mile.

Intersection density is recognized as a proxy for walkability. An intersection is defined as the intersection of any two walk or drive network segments, as derived from OpenStreetMap (OSM) data. Intersections within 15 meters (about 50 feet) of each other are consolidated into a single intersection. This resolves counts for boulevard intersections and other cases where the network configuration can lead to overcounting (for example, by capturing two sides of the same intersection as separate intersections). Intersection densities are calculated over a buffered area of 400 meters around each block to smooth out local variations and normalize densities for all locations with respect to their surroundings.

Analyst Land Development Category is a classification that reflects broad development patterns. They include Urban Infill (Urban), Compact Walkable (Compact), and Suburban (Standard). The Urban category represents areas (typically within moderate and high density urban centers) that have the highest intensity and mix of uses. Compact areas are less intensely developed than Urban areas but very walkable in part because of their mix of residential, commercial, and civic uses. Standard represents auto-oriented, separate-use suburban development patterns. (For custom canvases, a Rural category can be used to represent rural development.)

Land Development Category is assigned to census blocks, and in turn parcels, according to two criteria: intersection density per square mile and activity density (i.e., dwelling unit and employment densities). The categories are used in Base Canvas land use typing, as well some analysis modules (namely the Fiscal Impacts module). The categories also serve to communicate scenario concepts and results.

The criteria for the categories are summarized in Land Development Category Criteria.

Lastly, the Base Canvas includes estimates of residential and commercial irrigated area. The values in the parcel canvas are modeled based on general assumptions for the percentage of parcel area that is irrigated. Assumptions are associated with the Building Type assigned to each parcel.

Updated parcel data. All of our input parcel data has been updated to the most recently available.

Improved Dwelling Unit numbers. We have improved our typing for multifamily units, leading to more accurate dwelling unit numbers.

Improved data processing algorithm. We updated our data handling process for the parcel data from one of our providers, resulting in improved estimates to many fields.

Updated parcel data. All of our input parcel data has been updated to the most recently available.

Updated contextual data. An update to the underlying context data has improved our land use typing to more accurately reflect current conditions.

Improved commercial land use typing. An update to the underlying point-of-interest data has improved the typing of commercial land use.

More accurate land use and employment information. An update to the underlying OpenStreetMap point and polygon data yields more accurate land use typing and employment information.

Incorporated an additional parcel provider. We integrated an additional data provider with nationwide coverage.

Improved Dwelling Unit assignment. We leveraged additional parcel data to better align our dwelling unit estimation with the Census.

Improved land use typing for Right of Way parcels. We built a machine-learning model to identify parcels that are actually public Right of Way (such as roads). Incorporating this model into the Base Canvas improved our overall parcel typing, and increased the share of parcels typed as Right of Way from 0.002% to 0.48% nationwide.

More accurate land use and employment information. An update to the underlying OpenStreetMap point and polygon data yields more accurate land use typing and employment information.

Improved commercial land use typing. An update to the underlying point-of-interest data has improved the typing of commercial land use.

More accurate land use and employment information. An update to the underlying OpenStreetMap point and polygon data yields more accurate land use typing and employment information.

Improved commercial land use typing. An update to the underlying point-of-interest data has improved the typing of commercial land use.

Expanded coverage. 60 new counties have been added, pushing our nationwide cover up to 99% of the U.S.

Improved commercial land use typing. An update to the underlying point-of-interest data has improved the typing of commercial land use.

More accurate land use and employment information. An update to the underlying OpenStreetMap point and polygon data yields more accurate land use typing and employment information.

Near-nationwide coverage. Nearly 2,000 new counties have been added, tripling the number supported and resulting in coverage of 97% of the US.

Improved commercial land use typing. An update to the underlying point-of-interest data has further improved the typing of commercial land uses.

Updated building footprints data. An update to the underlying building footprints data (Microsoft Building Footprints) has increased the coverage of our dwelling units data and irrigated land area calculations, and in some cases has improved our commercial land use typing.

Improved commercial land use typing. An update to the underlying point-of-interest data has improved the typing of commercial land uses.

More accurate land use and employment information. An update to the underlying OpenStreetMap point and polygon data yields more accurate land use typing and employment information.

Bug Fix. Previously, calculated columns in the parcel reference data layer such as Floor Area Ratio, Improvement to Land Value Ratio, Market Improvement to Land Value Ratio, and Assessed Improvement to Land Value Ratio erroneously reported a zero when inputs to the calculation were missing or not applicable. With this release, the calculated value will also be reported as not available in those cases.

Addition of Oconee County, GA

Addition of download capabilities for Imperial County, CA and Cochise County, AZ Updated parcel geometries and attribute data for hundreds of U.S. counties Improved typing of commercial land use

Expanded coverage. The Base Canvas for California has been expanded to cover an additional 15 counties for complete parcel coverage of the entire state. Parcel geometries and attribute coverage have also been updated for hundreds of counties across the U.S.

Improved commercial land use typing. An update to the underlying point-of-interest data improves the typing of commercial land uses.

More accurate land use and employment information. An update to the underlying OpenStreetMap point and polygon data yields more accurate land use typing and employment information.

Updated point-of-interest data. Updated point-of-interest (POI) data from SafeGraph and OpenStreetMap, now current as of February 2021, improve the accuracy of commercial typing and employment information assigned to parcels.

Updated American Community Survey (ACS) data. The Base Canvas now uses ACS 2019 5-year estimates of population and households.

Updated parcel geometries and attributes. Parcel geometries have been updated for 183 counties. Parcel attributes have been updated for 1,023 counties.

Updated employment data. By updating the underlying data source from the U.S. Census Bureau’s LODES 2017 to the newly released LODES 2018, the UrbanFootprint Base Canvas now provides you with the latest available census-based employment data, adding nearly 3.5 million jobs nationwide.

Updated parcel geometries and attributes. Parcel geometries have been updated for 132 counties. Parcel attributes have been updated for 1,022 counties.

More accurate typing of parks and multifamily parcels. The assignment of current land use to parcels with parks and multi-family residential buildings has been improved to result in fewer incorrectly assigned park or residential parcels.

In addition to our regular update, this quarter’s release includes some exciting improvements:

More accurate typing of commercial parcels based on newly integrated SafeGraph point-of-interest (POI) data, improving how commercial use is identified (including offices, retail sites, medical services) and how job categories are separated at the parcel level

Updated census metrics, using the latest Census ACS 2018 dataset, to improve how missing population or household rates are handled at the census tract level