GeoAI Spatial Data Science Training Course

Participants can diagnose coordinate, geometry and attribute problems before spatial analysis.

• Geographic and projected coordinate reference systems compared accurately
• EPSG codes selected for distance, area and directional calculations
• Datum transformations checked when integrating datasets from different sources
• Geometry validity tested using self-intersection, gaps and duplicate features
• Positional accuracy, temporal currency and attribute completeness profiled systematically
• Metadata captured using ISO 19115 principles and source documentation

Participants can create analysis-ready layers and clear operational maps in QGIS.

• QGIS project structure organised with relative paths and layer groups
• Attribute joins, spatial joins and field calculations applied correctly
• Buffers, overlays and proximity analysis answer location-based questions
• Geoprocessing results inspected for slivers, missing records and topology errors
• Classification methods selected: equal interval, quantile, natural breaks
• Map layouts communicate scale, source, uncertainty and analytical purpose

Participants can build a reproducible Python workflow for vector-based geospatial analysis.

• Conda environments created with pinned geospatial package versions
• Jupyter notebooks structured for inputs, transformations, outputs and commentary
• GeoPandas reads, filters and reprojects vector feature datasets
• Shapely geometry operations calculate buffers, intersections and centroids
• Pandas merges spatial attributes with operational and demographic tables
• Python functions package repeated geoprocessing tasks into reusable components

Participants can use R to analyse, visualise and report spatial datasets.

• sf objects store feature geometry and attribute data efficiently
• terra workflows manage raster stacks, masks and zonal summaries
• dplyr pipelines transform spatial attributes with traceable operations
• ggplot2 and tmap produce exploratory spatial visualisations and maps
• Quarto reports combine R code, findings, figures and interpretation
• R project folders separate raw data, scripts, outputs and documentation

Participants can design spatial queries that retrieve and summarise operational geographic data.

• PostGIS geometry types, spatial indexes and SRID values explained
• SQL filters reduce records before expensive spatial operations begin
• ST_Intersects and ST_Within support boundary and service-area queries
• ST_Distance and nearest-neighbour queries identify accessible facilities efficiently
• Spatial aggregation calculates counts, rates and totals by zone
• Query plans inspected to diagnose slow spatial database performance

Participants can process raster datasets and extract spatially meaningful numerical features.

• Rasterio reads georeferenced bands, profiles and nodata values
• Raster alignment checks extent, resolution, transform and coordinate system
• Clipping and masking isolate analysis areas without altering source pixels
• Map algebra calculates thresholds, weighted surfaces and categorical outputs
• Zonal statistics summarise raster values within administrative or asset polygons
• Cloud-Optimized GeoTIFFs support efficient distribution and remote access

Participants can prepare satellite imagery for reliable environmental and land-use analysis.

• Electromagnetic spectrum bands linked to surface-material reflectance characteristics
• Spatial, spectral, temporal and radiometric resolution compared for applications
• Sentinel-2 and Landsat imagery selected using task requirements
• Atmospheric correction and surface reflectance products distinguished carefully
• Cloud masks, shadows and invalid pixels removed before analysis
• Image composites built using median, percentile and seasonal criteria

Participants can derive and interpret spectral features for land and environmental monitoring.

• NDVI measures vegetation greenness from near-infrared and red bands
• NDWI identifies water-related signals using green and infrared reflectance
• NDBI highlights built-up surfaces using shortwave infrared relationships
• Band ratios and textures generate candidate predictive image features
• Feature distributions inspected for outliers, saturation and seasonal effects
• Reference samples linked to imagery dates and spatial locations

Participants can classify imagery and measure defensible land-cover change.

• Training labels sampled across classes, regions and acquisition conditions
• Random Forest classification predicts land-cover categories from image features
• Confusion matrices report producer, user and overall classification accuracy
• Class imbalance addressed through sampling strategy and performance metrics
• Post-classification comparison identifies transitions between land-cover categories
• Change maps distinguish observed change from registration and classification error

Participants can use Google Earth Engine to process satellite image collections at scale.

• Earth Engine image collections filtered by date, area and metadata
• Server-side processing distinguished from client-side JavaScript execution
• Cloud masking functions applied to Sentinel-2 surface reflectance imagery
• Reducers calculate regional summaries, composites and grouped statistics
• Exports create analysis-ready raster assets and tabular summaries
• Earth Engine scripts documented with parameters, sources and assumptions

Participants can identify spatial patterns, clusters and outliers before model development.

• Spatial weights matrices define neighbour relationships and interaction structures
• Global Moran's I tests overall spatial autocorrelation in values
• Local Indicators of Spatial Association identify clusters and spatial outliers
• Getis-Ord Gi* detects statistically significant hot and cold spots
• Kernel density estimation maps concentrated incidents or service demand
• Rate smoothing stabilises small-area indicators with sparse event counts

Participants can select and interpret regression approaches that account for spatial structure.

• Ordinary least squares assumptions checked using residual diagnostics
• Residual Moran's I identifies unmodelled spatial dependence patterns
• Spatial lag and spatial error models address autocorrelated residuals
• Geographically weighted regression examines relationships varying across locations
• Multicollinearity assessed using correlation matrices and variance inflation factors
• Model coefficients interpreted with uncertainty, scale and causal limitations

Participants can create geospatial machine-learning pipelines for classification and prediction tasks.

• Prediction targets, predictor features and observation units specified clearly
• scikit-learn pipelines combine preprocessing, modelling and evaluation steps
• Random Forest, XGBoost and logistic regression compared for suitability
• Feature engineering incorporates proximity, density, terrain and spectral variables
• Hyperparameters tuned using documented search spaces and evaluation criteria
• Feature importance reviewed for plausibility, bias and operational interpretation

Participants can evaluate GeoAI models without overstating predictive performance.

• Random train-test splits compared with blocked spatial cross-validation
• Spatial leakage identified when nearby records enter training and testing
• Spatial blocks designed using grid, cluster and administrative boundaries
• Classification performance measured through precision, recall, F1 and ROC-AUC
• Regression error measured using MAE, RMSE, bias and residual maps
• Thresholds selected using error costs and operational decision consequences

Participants can assess when deep-learning methods are appropriate for imagery-based spatial tasks.

• Convolutional neural networks extract spatial patterns from image patches
• Semantic segmentation assigns classes to individual image pixels
• Transfer learning adapts pretrained models to limited labelled imagery
• Training datasets checked for class representation and geographic diversity
• Overfitting detected through learning curves and independent validation areas
• Compute, annotation and maintenance costs assessed before model selection

Participants can analyse geographic change over time and identify temporal modelling considerations.

• Time-enabled datasets structured with consistent dates, locations and identifiers
• Seasonal decomposition separates trend, periodicity and irregular variation
• Space-time cubes organise repeated observations across grids and periods
• Temporal aggregation selected to match decision cycles and data reliability
• Trajectory analysis compares changes in land cover or asset condition
• Forecast uncertainty communicated using intervals, scenarios and update schedules

Participants can document, govern and communicate GeoAI work for review and reuse.

• Git version control records changes to scripts and analytical decisions
• README files specify data sources, dependencies, assumptions and execution steps
• STAC metadata describes imagery assets, bands and acquisition properties
• Data protection risks assessed for location traces and sensitive facilities
• Bias audits test performance differences across places and population groups
• Model cards document intended use, limitations and monitoring requirements

Participants can translate spatial model results into usable products, actions and monitoring arrangements.

• Decision briefs connect spatial findings to defined operational choices
• Web-map layers designed with appropriate symbology, filters and user tasks
• Uncertainty maps distinguish confidence, data gaps and prediction reliability
• Alert thresholds link model outputs to review and intervention actions
• Monitoring indicators track drift, accuracy and data-quality deterioration
• Deployment options compared: notebooks, scheduled jobs, APIs and dashboards

Participants can design a GeoAI project around a realistic organisational use case.

• Flood exposure screening combines terrain, rainfall, assets and settlement layers
• Agricultural monitoring identifies crop vigour anomalies from seasonal imagery
• Branch catchment analysis estimates accessibility, competition and demand patterns
• Infrastructure inspections prioritise assets using condition, location and risk
• Project charters define users, decisions, data, methods and acceptance criteria
• Analysis plans identify dependencies, controls, milestones and review points

Participants can build, review and present a complete GeoAI analysis addressing a defined operational decision.

• Problem statement defines decision, users, geography and measurable analytical objective
• Data inventory documents sources, quality checks, permissions and transformations
• Python or R pipeline produces reproducible spatial features and outputs
• Selected model or spatial method addresses the stated analytical question
• Validation evidence demonstrates performance, uncertainty and geographic limitations
• Maps, technical documentation and recommendations support stakeholder review