Remote Sensing for Land Cover Mapping Training Course

Participants can select suitable imagery platforms and acquisition characteristics for a mapping brief.

• Satellite, airborne and UAV platforms matched to mapping requirements
• Sentinel-2 MSI bands, revisit cycle and spatial resolution
• Landsat Collection 2 products and surface reflectance applications
• Sentinel-1 SAR polarization, incidence angle and acquisition modes
• PlanetScope and commercial imagery selection criteria for project specifications
• Cloud cover, seasonality and phenology effects on image suitability

Participants can prepare spatially aligned imagery suitable for measurement and multi-date comparison.

• Coordinate reference systems, datums and projections for area measurement
• UTM zone selection and distortion checks for project extents
• Raster georeferencing using ground control points and residual error
• Resampling methods: nearest neighbour, bilinear interpolation and cubic convolution
• Image-to-image registration assessed through root mean square error
• Vector and raster alignment checks before multi-date analysis

Participants can acquire, organise and document source data for a reproducible mapping project.

• Data requirements matrix linking classes, scale, dates and sensors
• Copernicus Data Space Ecosystem queries using location and acquisition filters
• USGS EarthExplorer searches for Landsat scenes and ancillary datasets
• Cloud masks, quality bands and scene metadata screened before download
• STAC catalog fields used to document reproducible image selection
• Project geodatabase structure for imagery, reference data and outputs

Participants can create analysis-ready optical imagery for classification and interpretation.

• Level-1 and Level-2 optical products distinguished by processing state
• Radiometric calibration converts digital numbers to physically meaningful measurements
• Atmospheric correction with Sen2Cor and surface reflectance validation
• Cloud, shadow and cirrus masking from Sentinel-2 quality layers
• Band stacking, clipping and mosaicking prepared for classification
• Normalized Difference Vegetation Index calculated for vegetation separation

Participants can prepare Sentinel-1 radar data and combine radar features with optical imagery.

• Backscatter, speckle and polarization interpreted in Sentinel-1 imagery
• Thermal noise removal and orbit correction applied to SAR scenes
• Radiometric terrain flattening reduces topographic effects in radar backscatter
• Speckle filters balanced against edge preservation and class separability
• VV, VH and ratio features assembled for classification
• Optical-radar feature stacks created for persistent cloud regions

Participants can design reference datasets that support unbiased training and accuracy assessment.

• Reference data sources assessed for date, scale, accuracy and bias
• Probability sampling designs matched to heterogeneous land cover classes
• Stratified random samples allocated using expected class variability
• Field protocols record coordinates, photographs, labels and observation confidence
• Training, validation and test sets separated to prevent optimistic accuracy
• Class imbalance diagnosed before model fitting and accuracy reporting

Participants can create a clear class schema aligned with user needs and image capability.

• Mapping objectives translated into mutually exclusive land cover classes
• FAO Land Cover Classification System concepts applied to class definitions
• Class hierarchy designed for reporting needs and sensor discrimination limits
• Minimum mapping units specified from image resolution and decision use
• Interpretation keys document tone, texture, pattern, context and seasonality
• Crosswalk tables reconcile local categories with national reporting schemes

Participants can interpret imagery systematically and apply object-based classification concepts.

• Visual interpretation combines colour composites, texture and contextual evidence
• False-colour composites selected for vegetation, water and built-up features
• Band ratios and spectral indices discriminate targeted cover types
• Image segmentation parameters tuned for object boundary quality
• Object features include shape, compactness, mean values and neighbourhoods
• Rule-based classification documents thresholds and decision logic explicitly

Participants can train, compare and document supervised land cover classification models.

• Maximum likelihood classification estimates class probability from training signatures
• Random Forest classifiers fitted with bands, indices and ancillary variables
• Support vector machines tuned through kernel and cost parameter selection
• Feature importance ranks predictors and removes redundant variables
• Training polygons reviewed for purity, coverage and spectral consistency
• Classification parameters recorded for repeatable processing and comparison

Participants can use cloud-based image collections and scripts for large-area mapping tasks.

• Google Earth Engine image collections filtered, composited and exported
• Median composites reduce residual cloud and seasonal outliers
• Cloud masking functions inspected before large-area batch processing
• JavaScript scripts parameterise dates, boundaries, bands and classifiers
• Feature stacks combined with terrain derivatives and settlement layers
• Task exports organised with scale, projection and naming conventions

Participants can identify, quantify and interpret land cover change from multi-date imagery.

• Post-classification comparison identifies transitions between independently classified dates
• Image differencing detects index change using defined threshold values
• Change trajectories distinguish seasonal variation from persistent conversion
• Time-series composites establish baseline conditions and disturbance timing
• Transition matrices quantify area gains, losses and stable classes
• Date alignment controls phenology, tide and sensor differences

Participants can assess classification reliability and report statistically defensible area estimates.

• Confusion matrices compare mapped labels against independent reference samples
• Producer's and user's accuracy diagnose omission and commission error
• Overall accuracy and kappa interpreted with stated limitations
• Area-adjusted accuracy estimates account for stratified sampling designs
• Confidence intervals reported for class areas and accuracy metrics
• Error maps locate recurrent confusion and guide class refinement

Participants can refine classified outputs and create usable land cover map products.

• Majority filters remove isolated pixels without erasing narrow features
• Vectorisation converts classified rasters into editable polygon boundaries
• Topology rules detect gaps, overlaps and invalid geometry
• Area calculations use equal-area projections and stated reporting units
• Map layouts present legend, scale, north arrow, sources and date
• Web map tiles and GeoPackage exports prepared for users

Participants can document a mapping workflow for review, reuse and scheduled updates.

• ISO 19115 metadata elements document lineage, quality and distribution
• Processing logs capture software versions, parameters and source scene identifiers
• Quality assurance checklist verifies projections, class codes and attribute domains
• Data governance assigns custodians, access controls and retention periods
• Standard operating procedures define review gates and acceptance criteria
• Update schedules link new imagery to monitored land cover changes

Participants can build, review and present a complete land cover mapping project with documented evidence.

• A defined study area and decision question frame the mapping brief
• Source imagery and reference samples selected against documented requirements
• Preprocessing workflow produces an analysis-ready multi-sensor feature stack
• Class schema, classifier and validation design justified with evidence
• Accuracy report, change analysis and map products assembled for review
• Technical presentation defends methods, limitations and recommended operational actions