Advanced Remote Sensing Data Analysis Training Course

Participants can select and assess imagery sources against technical and operational requirements.

• Landsat, Sentinel-2, Sentinel-1 and commercial imagery acquisition characteristics
• STAC catalogues queried using location, date, cloud cover and collection filters
• Metadata inspection for acquisition geometry, processing level and quality flags
• Scene suitability assessed against seasonal timing and target-feature visibility
• Licensing conditions reviewed for publication, sharing and derivative product distribution
• Source-data register created with identifiers, dates, coverage and limitations

Participants can prepare spatially aligned imagery for multi-source and multi-date analysis.

• Coordinate reference systems selected using extent, distortion and reporting requirements
• Georeferencing errors distinguished from registration errors and orthorectification limitations
• Ground control points assessed for distribution, stability and positional confidence
• Image-to-image registration checked using stable features and residual displacement
• Resampling methods compared: nearest neighbour, bilinear interpolation and cubic convolution
• Raster grids aligned by pixel origin, resolution, extent and projection

Participants can prepare optical imagery with corrections appropriate to quantitative comparison.

• Digital numbers converted to radiance, reflectance and surface reflectance products
• Atmospheric correction assumptions compared for scene-specific and multi-date analysis
• Dark object subtraction applied where higher-order correction inputs are unavailable
• Cloud, cirrus, shadow and snow masks generated from quality bands
• Topographic illumination effects identified on steep terrain and forested slopes
• Radiometric normalization performed using pseudo-invariant features between image dates

Participants can create analytical raster features that distinguish target land-surface conditions.

• Histogram stretching selected for visual interpretation without altering source measurements
• Band combinations designed for vegetation, water, burn scars and built surfaces
• NDVI, NDWI, NBR and SAVI calculated with documented band definitions
• Texture measures derived using grey-level co-occurrence matrix neighbourhoods
• Principal component analysis used to reduce correlated multispectral band information
• Feature stacks assembled from spectral, texture, elevation and contextual variables

Participants can derive and evaluate terrain variables for environmental and infrastructure analysis.

• DEM sources compared for resolution, vertical datum and terrain artefacts
• Sink filling and hydrological conditioning applied before drainage analysis
• Slope, aspect, curvature and hillshade derived using documented neighbourhood settings
• Flow direction and accumulation calculated for catchment and flood-screening models
• Terrain derivatives incorporated into land-cover and susceptibility classification features
• Vertical accuracy checked against surveyed points and independent elevation references

Participants can interpret and preprocess SAR data for monitoring where optical imagery is limited.

• SAR backscatter explained through wavelength, polarization, geometry and surface roughness
• Sentinel-1 GRD processing steps reviewed from orbit files through terrain correction
• Speckle filtering balanced against edge preservation and spatial-detail requirements
• VV, VH and polarization ratios interpreted for vegetation, water and urban surfaces
• Incidence-angle effects identified before comparing backscatter across acquisition geometries
• Coherence products assessed for surface disturbance and infrastructure monitoring

Participants can create representative reference data for classification and validation.

• Class schema defined using operational categories, hierarchy and minimum mapping units
• Reference samples collected from field records, imagery interpretation and authoritative datasets
• Stratified random sampling designed for rare classes and heterogeneous landscapes
• Spatial autocorrelation reduced through sample spacing and independent validation partitions
• Training labels reviewed for ambiguity, temporal mismatch and interpreter disagreement
• Sample inventories documented with source, date, confidence and class rationale

Participants can build and compare pixel-based land-cover classification models.

• Maximum likelihood classification configured from class signatures and covariance estimates
• Random forest models trained using balanced samples and predictor variables
• Support vector machines tuned through kernel selection and parameter testing
• Feature importance examined to remove redundant or misleading predictor layers
• Classification probabilities retained for uncertainty screening and threshold decisions
• Post-classification filters applied without removing meaningful small-area features

Participants can use image objects and contextual rules to classify complex features.

• Multiresolution segmentation configured using scale, shape and compactness parameters
• Image objects characterized by spectral, geometric, texture and neighbourhood features
• Rule sets constructed for buildings, fields, roads and water bodies
• Object-based and pixel-based approaches compared against target-feature spatial structure
• Classification hierarchies applied to separate broad classes before detailed subclasses
• Segmentation quality reviewed against boundaries, over-segmentation and under-segmentation

Participants can quantify classification performance and report limitations clearly.

• Independent validation samples separated from model training and parameter tuning
• Confusion matrices calculated with producer, user and overall accuracy measures
• F1 scores and class-specific errors reported for imbalanced class distributions
• Area-adjusted accuracy estimates calculated using stratified sampling weights
• Confidence intervals produced for thematic area estimates and accuracy metrics
• Uncertainty statements linked to reference quality, class confusion and scale

Participants can detect, quantify and validate change using multi-date imagery.

• Image dates selected to control seasonality, phenology and acquisition differences
• Post-classification comparison used to produce transition matrices and change maps
• Image differencing applied to reflectance, indices and radar backscatter variables
• Change thresholds determined from reference samples, distributions and decision risk
• Time-series trajectories analysed for vegetation recovery, disturbance and persistent change
• Detected changes validated against independent imagery, field evidence and records

Participants can combine remote sensing outputs with spatial models and statistical tests.

• Zonal statistics summarized by administrative units, parcels and management zones
• Spatial joins connect classified areas with assets, populations and field observations
• Moran's I applied to test spatial clustering in mapped conditions
• Hot spot analysis identifies statistically significant clusters of change indicators
• Regression models assessed for multicollinearity, residual patterns and spatial dependence
• Modifiable areal unit effects considered when aggregating pixel-based evidence

Participants can automate repeatable raster processing while preserving traceability.

• GDAL commands used for format conversion, reprojection, clipping and mosaicking
• QGIS Processing Modeler chains configured with parameters and intermediate outputs
• Python scripts structured for raster input, processing, logging and output validation
• Batch processing designed for multiple scenes, dates and geographic areas
• Version-controlled project folders separate raw data, code, outputs and documentation
• Processing logs capture software versions, parameters, warnings and execution dates

Participants can communicate remote sensing findings with suitable maps, evidence and caveats.

• Map scale, symbology and generalization selected for analytical and decision audiences
• Classification legends designed to distinguish categories, uncertainty and no-data areas
• Change maps paired with baseline context, dates and quantified area statistics
• Methods statements record sensor inputs, corrections, models and validation results
• Limitations communicated using positional, thematic, temporal and scale-specific evidence
• Technical review packs assembled with metadata, outputs, assumptions and approvals

Participants can build, review and present a complete documented remote sensing analysis addressing a defined operational question.

• Project scope specifies decision question, study area, classes and acceptance criteria
• Source imagery screened for temporal suitability, quality flags and spatial coverage
• Processing workflow applies corrections, feature engineering and documented parameter choices
• Analytical model produces classification, index or change-detection outputs with uncertainty
• Independent validation quantifies accuracy, error sources and confidence limitations
• Peer technical review tests reproducibility, evidence quality and decision relevance
• Final presentation communicates methods, findings, maps and recommended operational actions