Geo-Information

E1: Selecting and Editinfg Features

Warm Up

Firstly, let’s take a moment to talk about a very important (and maybe slightly mysterious) word: 👉 Feature.

What Does “Feature” Mean in Data Science?

In data science and machine learning, a feature is just a fancy word for: A column in your table that describes something about your data.

For example, if you’re building a model to predict house prices, features might include:

• Number of bedrooms 🛏️
• Square footage 📐
• Distance to the nearest train station 🚉
• Year built

These features feed into your models, visualizations, and decisions. This process — turning messy, real-world data into usable, structured information — is called data engineering(machine learning engineer also covers this).🛠️Typical Workflow of DS or MLE is like:

Data collection -> Raw data → Preprocessing → Feature extraction → Model / Analysis

```{admonition} Features of Special Format :class: dropdown In other fields, “features” can look very different:

• 🖼️ In computer vision:
  Each pixel can be considered a feature — or features might be higher-level elements like color patterns, shapes, or edges.
• 🧠 In NLP (Natural Language Processing):
  Features could include word counts, sentence length, or embeddings from a language model.

  For example, in Transformers, each token is represented as a 512-dimensional vector.

• 🧊 In 3D data:
  Features might be points in a point cloud, meshes, or implicit surfaces — used in fields like LiDAR, AR/VR, or 3D modeling & generation.
• 🎥 In 4D data (e.g., video):
  Each frame can be treated as a feature, and you might also extract features within a frame (like objects), or track features across time (motion, trajectory, etc.).
• 🗺️ In spatial data (our focus today!):
  A feature is a real-world geographic object — something that has both location and descriptive attributes.

  Think of roads, buildings, rivers, or tram stops — all stored with geometry and fields in shapefiles or geodatabases.

And just like in other fields, we still care about their properties — but now they’re stored in attribute tables.

> Take it easy — imagine you're the computer. For every item in a dataset, the things you store or know about it are called features.
Features can be:
- Built-in (directly collected from raw data)
- Extracted, filtered, or converted from original information (e.g., converting time to day of week, [PCA](https://scikit-learn.org/stable/modules/generated/sklearn.decomposition.PCA.html))
- Learned through machine learning or deep learning models — though these learned features can sometimes be a bit mysterious 🤖✨

---

### What Are Feature and what are Fields in a Shapefile?
In this exercise, a **feature** specifically refers to a **geometry object** — such as a point, line, or polygon — representing a real-world location. Each feature typically consists of two parts:
- **Geometry**:
Describes where it is, including spatial shape and location.
Also includes auxiliary information like the Coordinate Reference System (CRS).
- **Attribute Fields**:
Describe what it is — these are the additional properties or metadata stored in a table (e.g., name, type, status, ID, etc.).
> If you read or import shapefiles (or other spatial formats) in your program, you’ll notice that **all the information** — including geometry-related data — is stored as attributes. So in a way, everything becomes a **feature describing the data** — whether it’s where something is (geometry) or what it is (attributes). From the program’s perspective, it’s just a structured collection of information.
```{admonition} ⚠️  Reminders
Always check the **CRS (coordinate reference system)** when working with multiple layers or datasets — mismatched CRS is a common source of confusion!

What makes geoinformation special?

In the lecture, we mentioned three key points that make geospatial data different from other types of data:

1. Spatial and Temporal Dependence Example: Traffic congestion in one district affects neighboring areas; deforestation in one region impacts climate patterns elsewhere.

2. Spatial Heterogeneity Example: Rainfall patterns or land use can differ dramatically from one region to another.

3. Place-Sensitive User Interest Example: Local air quality reports are much more relevant to nearby residents.

```{admonition} 🔬 Dive more

• In probability theory(e.g. Bayesian Estimation), the assumption of I.I.D. (Independent and Identically Distributed) is fundamental.
  But due to spatial and temporal dependence, this assumption often fails in geospatial contexts.
  👉 How can we address this? → Use models that account for spatial autocorrelation rather than assuming independence.

• We can only collect samples at specific locations — e.g., we have limited and static observations about weather conditions.
  👉 How can we estimate values in unsampled areas? → Use spatial interpolation techniques such as Kriging or IDW.

• How can we quantify spatial heterogeneity?
  👉 Use variograms, local statistics (e.g., Local Moran’s I), or model non-stationarity with techniques like GWR (Geographically Weighted Regression). ```

  In today’s exercise, you’ll get hands-on experience with how to interact with these features and fields — the fundamental building blocks of geospatial data.

---

Task

Descriptions

Detailed instructions in {download}Lesson 1 <../doc/Lesson 1.docx>

& You can Click here to look

Data

• Data Students.gdb
• orthophoto.tif

Overview

• Create a new project, import the provided dataset as a layer, and add a base map or image layer to your map view.
  • Connect a floder to your project
  • Add a database and then add the buildings and roads to the map
  • Add the .tiff file to your map as a layer
• Learn how to select features using different methods:
  • Select by click
  • Select by location
  • Select by attributes
• Process attribute fields:
  • Rename fields
  • Convert field formats (e.g., from string to number)
• Create a new feature class named Lawn for your own data.
• Edit new features:
  • Add now the new Feature Class Lawn to the Map according to the orthophoto
  • Add features such as tram stops and underground stations
  • Create a new feature template
• Edit attribute data:
  • Add new fields
  • Edit existing values
  • Calculate field values
  • Combine editing techniques for efficient workflows

Advance Task

• Explore: ESRI Shapefile, TIFF.
• Write a program that loads a shapefile, and try to complete the following tasks:
  • Display the shape (rows and columns) and print an overview of the data (e.g., first 5 rows or metadata).
  • Create a new geometry feature manually (e.g., add a point or a polygon).
  • Process the attribute fields, including:
    • Rename a field
    • Filter features based on specific conditions
    • Add a new field and calculate its value
    • Perform a CRS (Coordinate Reference System) transformation to reproject the data
  • Return features that match a specific rule or condition, such as:
    • All areas larger than a threshold
    • All points within a bounding box
    • All features of a certain type or category
• If using raster data: perform a basic raster operation, such as clipping, masking, or resampling.

```{admonition} 🔬 Coding Hints

• For Python: we suggest using GeoPandas, Shapely, Rasterio (for TIFF), and Pyproj.
• For C++ or C#: use GDAL or OGR for both vector and raster data handling. ```

---

Materials

• What is a Feature in Machine Learning and Data Science?(by Domino)
• Introduction to GeoPandas
• What is a shapefile?
• What is a Shapefile? by AMDGS

• [Edit Features Using Editor Tools in ArcGIS Pro

  Beginners Guide](https://www.youtube.com/watch?v=i-HDJaZw6dU&ab_channel=TerraSpatial)

• Editing the Attribute Table in ArcGIS Pro
• GTiff – GeoTIFF File Format
• ESRI Shapefile / DBF, GDAL