Poznań University of Economics and Business · Real Estate Market Analysis

Data Cleaning & Preparation

Understanding the format behind GML spatial data files
and loading them into Python

Data Cleaning & Preparation · PUEB REMA

Introduction to XML

Vocabulary, then a short RCN GML fragment. Syntax and namespaces follow in later sections.

Introduction to XML

What is XML?

Hierarchical text · elements and attributes

XML is plain-text markup: nested elements, optional attributes, one document tree. You define tags or adopt a standard (e.g. GML); a parser builds the tree.

Why it is used for exchange

Self-describing — tags name the data; you can read structure without a separate data dictionary.
Validation — schemas (e.g. XSD) define allowed elements and types before data enters a pipeline.
Rich fields — attributes carry units, IDs, language, or “nil” flags next to values.
Interoperability — long track record in government, GIS (OGC/ISO), and regulated sectors.

Introduction to XML

Where you see XML

Beyond this course

Geodata and graphics: GML, KML, SVG
Office and tools: OOXML (.docx / .xlsx), RSS, Android layouts, build files
Enterprise: SOAP, B2B payloads where schemas matter

Introduction to XML

RCN export (excerpt)

Declaration (line 1) · comments (2–6, ignored by parsers) · <gml:boundedBy … /> = empty element

Step through the four highlighted sections:

Line 1 — XML declaration (<?xml … ?>): version 1.0, UTF-8. Must be first in the file. standalone="yes" = no external DTD.
Lines 2–6 — comment block (). Export metadata from WEGA; parsers ignore comments entirely.
Lines 7–15 — root element FeatureCollection. Seven xmlns: lines declare namespace prefixes — normal in GML.
Lines 17–24 — first featureMember: RCN_Dokument (notarial deed), with xsi:nil geometry, deed number, date, notary.

Self-closing: <gml:boundedBy xsi:nil="true" /> — no text content; /> closes the tag. Same idea as <boundedBy></boundedBy> when empty.

Ask: "Has anyone opened a file like this before?" — The goal today is that by the end, this looks readable.

The course file Baza_danych_RCN_Poznan_2021-2025.gml is a full multi-year Poznań export (millions of lines). Use streaming (iterparse) or geopandas.read_file(), not a naive full-tree load.

Data Cleaning & Preparation · PUEB REMA

Format overview

One example record (LOK_1) as CSV, JSON, and XML. Later material assumes XML because GML is XML.

Format overview · CSV

CSV

Comma-separated values · flat tables

Comma-Separated Values — one row per record, columns for fields. Typical for spreadsheets and SQL exports.


id,pow,cena
LOK_1,60.25,750000

Works well for quick analysis in pandas or Excel.
No native hierarchy: nested data usually means extra columns, encoding tricks, or another file and a join.
Units and rich geometry have no standard column — often WKT in text or separate layers.

Format overview · JSON

JSON

JavaScript Object Notation · nested objects

JSON — objects { }, arrays [ ], key–value pairs. Common in REST APIs and document stores.


{
  "id": "LOK_1",
  "pow": 60.25,
  "uom": "m2",
  "cena": 750000
}

Nesting maps cleanly to Python dicts and JavaScript objects.
Units or typed values often need extra keys or wrapper objects.
GeoJSON is widely used; cadastre-grade pipelines often still use GML/XML.

Format overview · XML

XML

eXtensible Markup Language · tagged trees

XML — elements in angle brackets, attributes, a strict document tree. Vocabularies such as GML are defined with schemas (e.g. XSD).

Attributes such as uom="m2" can sit on the same element as the value.
Child elements express containment without a second table.
GML, KML, SVG, CityGML, and many government feeds use XML; XSD can validate structure before ingest.

Data Cleaning & Preparation · PUEB REMA

Building blocks

Elements and attributes — toy examples first, then the same ideas in RCN GML.

Building blocks

Elements

Opening tag · content · closing tag


<city>Poznań</city>

<city> — opening tag
Poznań — text content
</city> — closing tag

Building blocks

Elements in RCN data

Same structure — longer names, namespace prefix

rcn: is a namespace prefix — the element is still opening tag · text · closing tag. Full meaning of prefixes in the namespaces section.

Building blocks

Attributes

Bare value → unit metadata on the opening tag


<pow>60.25</pow>

60.25 has no unit (m², ha, km²…) — for analysis, units must travel with the number, not only in a separate column or comment.

Put metadata on the same element


<pow uom="m2">60.25</pow>

Value unchanged; uom="m2" is unit of measure (real pattern from RCN_Lokal).

Syntax: name="value" inside <…>; quotes; several attributes = space-separated (order does not matter).

Building blocks

Common attributes

Fragments of real geometry and property fields

srsName, gml:id — CRS + stable id
count, srsDimension — how many coordinates, 2D vs 3D
uom — units (check before comparing numbers)
xsi:nil — “no value here” (e.g. missing geometry)

Data Cleaning & Preparation · PUEB REMA

Document structure

One root element, parent–child nesting, and paths to data — then a real RCN_Lokal fragment.

Document structure

XML is a tree

Not a single table — a hierarchy of elements

Each element has at most one parent (except the root, which has none). Children are fully inside the parent — that nesting is how “this address belongs to this flat” is represented.

Root — one outer element wraps the whole document (GML: gml:FeatureCollection).
Records — wrappers like gml:featureMember and typed features such as rcn:RCN_Lokal.
Paths — nested fields are reached by walking the tree (e.g. … → RCN_Adres → miejscowosc).
Well-formed — matching tags, proper nesting, one root; invalid XML fails in the parser (official RCN exports are well-formed).

Document structure

One feature (schematic)

Scalars vs nested block (address)


FeatureCollection
 └─ featureMember
     └─ RCN_Lokal
          ├─ idLokalu · rooms · floor · area · price …
          └─ adresBudynkuZLokalem
               └─ RCN_Adres → miejscowosc · ulica · numer …

Flat fields sit directly under RCN_Lokal; the address is a subtree (container → RCN_Adres → fields).

Document structure

`RCN_Lokal` (trimmed)

Real excerpt — indentation follows depth

1 / 14 — feature open/close + gml:id (stable id in the file).
2–6 — scalar fields; area carries uom="m2".
7–13 — nested address block (adresBudynkuZLokalem → RCN_Adres → fields).

Data Cleaning & Preparation · PUEB REMA

Namespaces

Prefixes (gml:, rcn:, …) point at standard vocabularies; you bind them once, then query with full names in code.

Namespaces

Why prefixes?

Many standards · same local names · different meaning

GML, RCN, XSD, and others all define tags like id or name. A namespace ties each prefix to one vocabulary so gml:id (geometry id) and rcn:id (cadastre id) never clash. prefix:localName is a QName — the colon is only punctuation, not special syntax.

gml: — OGC geometry / exchange shell (FeatureCollection, Polygon, …).
rcn: — Polish RCN types (RCN_Lokal, RCN_Transakcja, …).
xsi: — schema instance: xsi:nil, schemaLocation, …

Namespaces

`xmlns:` on the root

Each line binds a prefix to a URI — a stable vocabulary id

gml / rcn / xsi — main geometry, RCN features, schema hints (see previous slide).
xlink: — xlink:href points at another feature's gml:id.
gmd / gco / gts — ISO metadata / time; often declared; fewer direct paths in exercises.

Declared once on FeatureCollection — like imports for the whole document.

Namespaces

URIs and ElementTree

Identifiers — not necessarily pages to open in a browser

http://… and urn:… strings are globally unique vocabulary names. Parsing does not require downloading them.


ns = {
    'gml':   'http://www.opengis.net/gml/3.2',
    'rcn':   'urn:gugik:specyfikacje:gmlas:rejestrcennieruchomosci:1.0',
    'xlink': 'http://www.w3.org/1999/xlink',
}
tree.findall('gml:featureMember/rcn:RCN_Lokal', ns)

Elements may appear as {http://www.opengis.net/gml/3.2}Polygon — Clark notation — in errors and repr.

Data Cleaning & Preparation · PUEB REMA

RCN in the file

What the Poznań .gml represents in business terms: sales, properties, parcels, buildings, flats, and deeds — as linked records, not one nested story.

RCN in the file

What is in the export?

Rejestr Cen Nieruchomości — price and transaction facts

The file is a long sequence of gml:featureMember blocks. Each block is one business object (one transaction, one parcel, one apartment, …). The same real-world deal is split across several such objects, connected by identifiers — not by putting everything inside one big XML subtree.

You will filter by type (RCN_Transakcja, RCN_Lokal, …) and follow links to assemble a full case.
Geometry and address detail live on parcel / building / unit features; the transaction row carries price and pointers.

RCN in the file

Six feature types (domain)

Polish tag names — plain-language role

RCN_Transakcja — one reported deal: transaction price, type (e.g. sale vs other), links to the legal deed and to the property aggregate.
RCN_Nieruchomosc — the economic property sold as a bundle (may combine plot, building, unit; totals and roll-up attributes).
RCN_Dzialka — a cadastral plot (land parcel): identifiers, land use, polygon geometry.
RCN_Budynek — a building on a plot: footprint geometry, type/classification, address hooks.
RCN_Lokal — a dwelling unit (flat): rooms, floor, usable area, unit price — what you use for typical apartment analytics.
RCN_Dokument — the notarial deed (number, date, notary office) that is the legal basis of the transaction.

RCN in the file

How records tie together

A graph of IDs — not one nested XML document per deal

Each feature has a gml:id. Elsewhere, xlink:href attributes store references to another feature's id (deed, property bundle, parcel, …). So the business view is: many rows + foreign-key-style links, same idea as joins in a database.

Typical path to “full picture”: Transakcja → nieruchomosc href → RCN_Nieruchomosc → hrefs to Dzialka / Budynek / Lokal; podstawaPrawna → Dokument.

RCN in the file

One transaction in XML

Price on the row · rest via xlink:href

Wrapper + id · scalars (incl. gross price) · link to deed · link to property aggregate — then follow those ids elsewhere in the file.

Up next

Now let's parse it

Open gml_xml_tasks.ipynb and work with
Baza_danych_RCN_Poznan_2021-2025.gml

Brief summary of what was covered:

XML structure: elements, attributes, nesting — the three building blocks
Declaration, comments, empty tags — on the RCN excerpt; well-formedness — with document structure
Namespaces: 7 prefixes in the RCN file, the xmlns: declaration pattern, URIs as unique labels
RCN business structure: six feature types (transaction, property, plot, building, unit, deed) linked by gml:id / xlink:href

Hand off to next section: loading and querying this file in Python.

Remind students: Baza_danych_RCN_Poznan_2021-2025.gml is very large — use iterparse() instead of ElementTree.parse() to avoid loading everything into RAM. geopandas.read_file() can also read GML directly and is the fastest path to a GeoDataFrame.

Data Cleaning & Preparation

Introduction to XML

What is XML?

Why it is used for exchange

Where you see XML

RCN export (excerpt)

Format overview

CSV

JSON

XML

Building blocks

Elements

Elements in RCN data

Attributes

Common attributes

Document structure

XML is a tree

One feature (schematic)

RCN_Lokal (trimmed)

Namespaces

Why prefixes?

xmlns: on the root

URIs and ElementTree

RCN in the file

What is in the export?

Six feature types (domain)

How records tie together

One transaction in XML

Now let's parse it

`RCN_Lokal` (trimmed)

`xmlns:` on the root