Time Series Analysis and Change Detection

Seasonal composites, statistical anomaly detection, and transformer-based change mapping

Time Series Analysis and Change Detection
Intermediate Remote Sensing 4 lessons Samuel Appiah Kubi

About this course

A single satellite image tells you what happened on one day. This course is about what happens over months and years: building cloud-free composites that survive individual bad scenes, detecting statistically meaningful anomalies against a proper multi-year baseline, and applying a modern transformer change-detection model to catch new forest clearing before it spreads — using Ghana's Atewa Range Forest Reserve as the running example.

What you'll learn

  • Build annual and seasonal median composites from cloud-masked Sentinel-2 stacks
  • Build a proper multi-year baseline and detect statistically meaningful NDVI anomalies against it
  • Explain the difference between simple differencing and transformer-based change detection
  • Run bi-temporal change detection with ChangeFormer, and filter its output to a specific failure mode of interest
  • Quantify forest loss area and export a monitoring-ready alert format

Requirements

  • Optical Analysis with Sentinel-2 (or equivalent)
  • Familiarity with the pygeofetch download workflow

Course content

Why Composites Beat Single Scenes Preview 22 min
Detecting Anomalies Against a Real Multi-Year Baseline 26 min

Simple Differencing vs Deep Learning: Knowing Which You Need 28 min
Case Study: A Deforestation Alert System for Atewa Forest Reserve 34 min
Free
Login to Enroll
  • LevelIntermediate
  • Lessons4
  • CertificateYes
  • AccessLifetime
Samuel Appiah Kubi Instructor