Effective Prompting Guide
Learn how to craft effective prompts to get the best results from GeoRetina AI (GRAI). Understanding what the platform can and cannot do will help you phrase your requests for optimal outcomes.
The ROI-Based Workflow
GRAI operates on a Region of Interest (ROI) based workflow. This means all spatial analyses are performed on predefined areas that you draw or import, not on text-based location descriptions.
How to Work with ROIs
1. Create Your ROI
Draw a polygon on the map or import a shapefile to define your area of interest. Give it a descriptive name like "downtown_portland" or "amazon_study_area".
2. Use the Mention Menu
When typing @, the mention menu shows all saved ROIs. Select the one you need to avoid typos and keep naming consistent.
3. Reference in Prompts
Use the @ symbol followed by your ROI name in your request: "@downtown_portland" or "@amazon_study_area".
4. Run the Analysis
GRAI evaluates only the geometry you referenced, delivering focused results inside the ROI boundary.
Understanding Platform Capabilities
GRAI specializes in geospatial analysis and querying. Before crafting your prompts, it's important to understand the platform's core strengths and limitations.
What GRAI Excels At
✓ Raster-Based Analysis
Satellite imagery analysis, land cover mapping, change detection, vegetation monitoring, weather forecasting, and environmental monitoring.
✓ Vector Data Querying
Spatial queries, proximity analysis, filtering by attributes, and combining multiple datasets.
✓ Time-Series Analysis
Multi-year trends, seasonal patterns, change detection, and temporal comparisons.
✓ Multi-Modal Integration
Combining raster and vector analysis, custom data integration, and progressive workflows.
What GRAI Cannot Do
GRAI is a specialized geospatial platform. It cannot perform general data analysis, create non-geospatial visualizations, or handle analyses outside its supported capabilities.
✗ Non-Geospatial Analysis
Financial analysis, social media analytics, text processing unrelated to geography.
✗ Unsupported Data Types
Non-spatial databases, proprietary formats not listed in capabilities, real-time streaming data.
✗ Advanced GIS Operations
Complex spatial modeling, advanced geoprocessing workflows, custom algorithm development.
✗ Real-Time Analysis
Live satellite feeds, streaming IoT data, instant change detection.
Crafting Effective Prompts
1. Be Specific About Location
Always work with ROIs (Regions of Interest) that you've drawn or imported. GRAI analyzes the selected ROI, not text-based location descriptions.
Before making analysis requests, you must first draw a polygon on the map or import a shapefile to define your Region of Interest (ROI). Then select your ROI from the mention menu (@ROI_name) when crafting your analysis request.
✓ Good Examples (with ROI workflow)
"Generate a land cover map for @downtown_portland"
"Analyze urban heat island effects in @phoenix_metro"
"Show vegetation changes in @amazon_study_area from 2018 to 2023"
✗ Common Mistakes
"Show me land cover in New York" — Missing ROI reference, no time period
"Analyze heat islands" — No ROI selected, vague request
"Forest changes in Brazil" — Text location instead of ROI, missing timeframe
2. Specify Time Periods Clearly
For time-series analysis, be explicit about your temporal requirements.
✓ Clear Temporal Requests with ROIs
"Compare land cover between summer 2020 and summer 2023 in @study_area"
"Show urban heat trends from 2018 to present in @city_boundary"
"Vegetation monitoring from March to September 2023 in @agricultural_zone"
"10-day weather forecast for @project_area"
3. Use Supported Analysis Types
Frame your requests using the platform's available capabilities.
Raster Analysis Prompting
Be explicit about temporal context for raster requests—call out the season, date range, or acquisition window and any relevant parameters like index type or weather variable.
Land cover change workflows in GRAI are bi-temporal—each change detection request compares exactly two specified dates or seasons.
Common Raster Workflows
Land Cover Mapping
"Generate a [land cover/LULC] map for @[roi_name] for [time period]"
Example: "Generate a land cover map for @manhattan_area for summer 2024"
Change Detection
"Compare [analysis type] between [start date] and [end date] in @[roi_name]"
Bi-temporal only: "Detect forest loss by comparing 2020 and 2024 scenes in @amazon_reserve"
Urban Heat Islands
"Analyze [heat metric] in @[roi_name] for [time window]"
Example: "Analyze urban heat island effects in @downtown_la since 2018"
Vegetation Monitoring
"Monitor [vegetation index] in @[roi_name] from [start] to [end]"
Example: "Monitor NDVI changes in @agricultural_fields from March to October 2023"
Weather Forecasting
"Display [weather parameters] for @[roi_name] for [forecast window]"
Example: "Show me the 10-day weather forecast for @study_region"
Air Quality & Pollution
"Analyze [pollutant] in @[roi_name] for [time frame]"
Example: "Analyze NO₂ concentrations in @beijing_area for the past month"
Vector Analysis Prompting
Vector prompts work best when you combine spatial filters with precise attribute constraints and follow-up refinements.
Common Vector Workflows
Proximity Searches
"Show all [feature type] within [distance + unit] of @[roi_name]"
Example: "Show all schools within 2 km of @residential_district"
Attribute Filtering
"Find [feature type] with [attribute condition] in @[roi_name]"
Example: "Find all restaurants with outdoor seating in @downtown_area"
Refine Previous Results
"Of the previous results, show only those that [additional criteria]"
Example: "Of these buildings, show only those built after 2010"
Advanced Integration Prompting
Knowledge Base Integration
RAG-Enhanced Queries
"Using our environmental impact reports, analyze land cover for @project_area"
"Based on our planning documents, show relevant infrastructure near @development_site"
"Reference our ecological studies and analyze current vegetation patterns in @study_zones"
Esri Atlas Integration
Connect your ArcGIS Online account in GRAI to unlock Esri Atlas raster layers. Only Esri-published datasets are currently available.
Select an ROI and request your analysis just like native raster workflows. You may specify a date, but availability depends on the source dataset and is not guaranteed.
Example Prompts
"Need a slope analysis in @mountain_roi from Esri Atlas"
"Pull the official Esri Atlas soil layer for @farm_blocks"
"Use Esri Atlas imagery to estimate snow cover in @winter_area for January 2024"
Common Prompting Mistakes
Avoid These Patterns
❌ Common Mistakes
Requesting Unsupported Analysis Types:
"Calculate population density from satellite imagery" (not supported)
→ The platform cannot derive demographic data from imagery alone
Missing ROI References:
"Show recent changes in New York" (missing ROI)
→ Use ROIs: "Show recent changes in @nyc_study_area from 2020 to 2023"
Non-Geospatial Requests:
"Analyze stock market trends" (not geospatial)
→ Focus on geographic and spatial analysis only
Overly Complex Single Requests:
"Show land cover, calculate population, find schools, and predict weather all at once" (too complex)
→ Break into focused, sequential queries with proper ROI references
Pro Tips for Better Results
💡 Start Simple, Build Complex
Begin with basic analysis, then use follow-up questions to dive deeper. The platform maintains context between related queries.
🎯 Be Goal-Oriented
State your objective clearly: "I need to assess deforestation for a conservation report" helps the platform understand context.
🔄 Use Iterative Refinement
Refine your analysis progressively. Start broad, then narrow down based on initial results.
📊 Leverage Interactive Features
After getting results, use interactive querying tools to extract specific values and generate detailed charts.
⚠️ Split Analyses When Possible
GRAI can process up to two analyses in a single message, but you'll get faster, clearer responses by sending each analysis request as its own message.
Quick Reference: Prompt Templates
Raster Analysis Templates
Land Cover: "Generate a [land cover/LULC] map for @[roi_name] for [time period]"
Change Detection: "Compare [analysis type] between [start date] and [end date] in @[roi_name]" Bi-temporal: only the two specified dates are used.
Time Series: "Show [parameter] trends in @[roi_name] from [start] to [end]"
Weather: "Display [weather parameters] forecast for @[roi_name] for [time period]"
Environmental: "Analyze [pollutant/vegetation index] in @[roi_name] for [time period]"
Vector Analysis Templates
Proximity: "Show all [feature type] within [distance] of @[roi_name]"
Filtering: "Find [feature type] with [attribute condition] in @[roi_name]"
Counting: "Count the number of [features] in @[roi_name] that meet [criteria]"
Multi-step: "Of the previous results, show only those that [additional criteria]"
Remember: GRAI is most effective when you work within its specialized geospatial capabilities. Clear, specific prompts that align with the platform's strengths will yield the best results for your analysis needs.