Landcover Classification of Amsterdam

Landcover Classification using Unsupervised Pixel-Based Classification

Unsupervised pixel-based classification in ArcGIS Pro involves analyzing the spectral signatures of individual pixels in satellite images or aerial photos. Pixels with similar spectral characteristics are grouped together into distinct classes, effectively creating a detailed map of various landcover types across Amsterdam.

Landcover Classification Map Legend

Strengths & Weaknesses

This method excels in identifying specific land or soil types with high precision, offering a more granular view compared to object-based classification. However, it lacks context recognition, making it less effective for distinguishing between similar classes in different contexts. Despite this, it can reveal unexpected classes, providing unique insights.

Unique Insight

One surprising finding from this project was the identification of sandy areas within Amsterdam—something I hadn't anticipated. The precision of this method made it easier to name and categorize landcover classes accurately, enhancing the overall usefulness of the map.

Steps to Create the Map

The map was created in ArcGIS Pro. The steps I followed are:

  1. Downloaded satellite imagery of Amsterdam from Sentinel 2 from Google Earth Engine.
  2. Imported the imagery into ArcGIS Pro and adjusted the band combination for better visualization.
  3. Measured the length and width of each pixel to understand the spatial resolution.
  4. Used the spectral profile tool to create spectral profiles for different land cover types.
  5. Ran an unsupervised classification using the Classification Wizard in ArcGIS Pro.
  6. Renamed the classes from the unsupervised classification and adjusted the symbology to reflect the identified land cover types.
  7. Compared the results with supervised and object-based classifications.
  8. Generated NDVI to analyze vegetation health.

Best Classification Scheme

The unsupervised pixel-based classification method worked best for my chosen area, which was an urban area. This method excelled because boundaries are hard to recognize on a 10x10m pixel scale in the city, where multiple boundaries can exist within a single pixel. Object-based classification may have worked better in rural areas where objects are larger and more distinct. Supervised pixel classification was less effective as it was challenging to manually identify distinct areas for every category in a highly pixelated urban environment. The computer's ability to detect subtle differences in land properties that I might overlook made unsupervised classification the most effective approach.

The map was created in ArcGIS Pro and provides valuable insights into the distribution and types of landcover within Amsterdam, which can be particularly useful for urban planning, environmental studies, and resource management.