Land Cover Change Analysis Training Course

Participants can select imagery and supporting data appropriate to a defined mapping objective.

• Multispectral reflectance principles for visible, near-infrared and shortwave infrared bands
• Sentinel-2 MSI spatial, spectral and temporal resolution assessed
• Landsat Collection 2 surface reflectance products selected appropriately
• Planetary data, aerial imagery and field observations matched to requirements
• Cloud cover, phenology and acquisition dates screened systematically
• Metadata records capture source, processing level and positional limitations

Participants can organise multi-source spatial data in a consistent and traceable project structure.

• Coordinate reference systems selected for regional extent and measurement purpose
• Raster cell size, extent and alignment controlled before analysis
• Geodatabases organise imagery, reference layers, outputs and documentation
• File naming conventions preserve dates, sensors, versions and processing status
• Vector boundaries checked against imagery and authoritative reference sources
• Data lineage recorded from acquisition through final map publication

Participants can prepare analysis-ready image stacks while documenting preprocessing decisions.

• Surface reflectance products distinguished from top-of-atmosphere reflectance products
• Cloud, shadow and cirrus masks applied using quality assurance bands
• Multi-date scenes co-registered and visually checked for displacement
• Image mosaics created with acquisition-date and seamline controls
• Seasonal imagery matched to vegetation cycles and target classes
• Histogram inspection identifies haze, saturation and radiometric inconsistencies

Participants can interpret land cover features and design reliable reference datasets.

• Interpretation keys define tone, texture, pattern, shape and context
• Reference samples sourced from field surveys, imagery and authoritative datasets
• Sample locations distributed across classes, gradients and geographic areas
• Training and validation samples separated to prevent optimistic accuracy estimates
• Mixed pixels, edge effects and temporal mismatch documented explicitly
• Field data forms capture class evidence, photographs and observation dates

Participants can derive spectral features that improve separation of land cover classes.

• Normalized Difference Vegetation Index calculated for vegetation condition assessment
• Normalized Difference Water Index separates open water and moisture
• Normalized Burn Ratio identifies burned areas and fire severity
• Tasseled Cap transformation summarises brightness, greenness and wetness
• Band ratios highlight built surfaces, bare soil and vegetation
• Principal component analysis reduces correlated spectral information

Participants can create representative training samples and run supervised land cover classifications.

• Training polygons digitised within homogeneous and representative land-cover areas
• Class sample sizes balanced against within-class spectral variability
• Spectral signatures inspected using band statistics and feature-space plots
• Maximum likelihood classification applied with documented class assumptions
• Random Forest classification configured using labelled training samples
• Training datasets revised after reviewing confusion and class overlap

Participants can refine raw classification outputs into usable land cover map products.

• Majority filtering reduces isolated pixels without obscuring small features
• Sieve operations remove patches below the minimum mapping unit
• Ancillary elevation, slope and road layers constrain implausible classes
• Rule-based recoding resolves known classification ambiguities consistently
• Object-based segmentation groups neighbouring pixels into meaningful units
• Manual editing procedures preserve an auditable record of corrections

Participants can quantify classification quality and report area estimates with appropriate caution.

• Confusion matrices compare mapped classes against independent reference samples
• Producer's and user's accuracy interpreted for class-specific reliability
• Overall accuracy calculated without concealing minority-class performance
• Stratified random sampling allocates validation points across mapped classes
• Area-adjusted estimates correct mapped area using reference sample evidence
• Confidence intervals communicate uncertainty around class area estimates

Participants can identify and quantify transitions between classified land cover maps.

• Post-classification comparison identifies transitions between two land-cover maps
• Cross-tabulation produces class-to-class transition matrices and change totals
• Persistence, gains, losses and swaps calculated for each class
• Change maps coded to distinguish conversion pathways clearly
• Transition hotspots located using zonal statistics and spatial aggregation
• Class-definition consistency checked before comparing historical map series

Participants can apply image-based methods for detecting change before final classification.

• Image differencing measures spectral change between comparable acquisition dates
• Change Vector Analysis quantifies magnitude and direction of spectral change
• NDVI differencing identifies vegetation loss, gain and seasonal anomalies
• Multitemporal composites reduce cloud effects and short-term noise
• Threshold selection uses reference samples, histograms and visual review
• Direct-detection outputs validated against imagery and independent observations

Participants can use repeated observations to distinguish persistent trends from temporary change.

• Monthly and seasonal composites create consistent observation time series
• Time-series charts reveal disturbances, recovery and recurring phenological patterns
• Theil-Sen slope estimates long-term vegetation index trends
• Mann-Kendall testing assesses monotonic trend significance in pixel series
• Breakpoint detection identifies abrupt shifts in land surface condition
• Temporal segmentation separates temporary disturbance from permanent conversion

Participants can structure repeatable large-area analyses using cloud geospatial processing methods.

• Google Earth Engine collections filtered by date, location and quality
• Server-side processing distinguished from local desktop geoprocessing tasks
• Cloud masking functions applied consistently across Sentinel-2 image collections
• Image composites generated using median, percentile and quality mosaics
• Classification scripts parameterised for repeatable regional monitoring workflows
• Export settings preserve projection, scale, extent and output metadata

Participants can assess when advanced classification approaches are justified and how to evaluate them.

• Object-based image analysis combines segmentation, shape and spectral attributes
• Random Forest variable importance identifies informative predictor layers
• Support Vector Machine classification handles complex spectral boundaries
• Hyperparameter tuning uses validation results without contaminating test data
• Feature selection reduces redundancy among spectral and contextual variables
• Model transferability tested across locations, dates and sensor conditions

Participants can prepare maps and technical documentation that communicate findings and limitations clearly.

• Map layouts include scale, north arrow, legend, sources and projection
• Colour schemes distinguish stable classes, gains, losses and transitions
• Transition matrices summarised for technical and management audiences
• Methods sections document imagery, training data, algorithms and validation
• Limitations describe uncertainty, temporal constraints and interpretation risks
• ISO 19115 metadata elements applied to published geospatial datasets

Participants can establish quality controls and an operational plan for recurring land cover monitoring.

• Quality assurance plan defines checks, acceptance criteria and review roles
• Version control records changes to data, scripts and classification rules
• Peer review examines class definitions, samples, outputs and conclusions
• Data governance assigns custodianship, access controls and retention periods
• Monitoring schedules align imagery acquisition with reporting deadlines
• Procurement specifications define deliverables, accuracy thresholds and metadata requirements

Participants can build, review and present a complete documented land cover change analysis for a defined study area.

• Project brief confirms decision question, study area and reporting audience
• Multi-date imagery prepared using documented masking and alignment procedures
• Classification schema, training samples and algorithms selected with justification
• Change outputs include transition matrix, hotspot map and area estimates
• Accuracy findings and limitations reviewed against independent reference evidence
• Technical presentation communicates methods, findings, uncertainty and recommended actions