GeoRetinaDocs

Effective Prompting Guide

3 min read · Last updated Jun 30, 2026

GeoRetina AI works best when your prompt names the site or asset context, the risk or conservation question, the time period, and the output you want. You do not need technical map syntax, but you should be precise about the decision you are trying to support.

Prompt Anatomy

Use this pattern for most analysis requests:

Recommended structure

Run [analysis type] for [@region_or_layer] during [time period], and return [map/table/report/export].

Examples:

Land-cover trend

Run a 2018-2025 land-cover trend analysis for @study_area and show the class area chart.

Vegetation stress

Analyze NDVI trends for @farm_blocks from 2018 to 2025 and identify the highest-priority stress zones.

Flood exposure

Where physics-based coverage exists, map the 100-year flood depth and requested OSM feature exposure for @ottawa_study_area.

Wildfire

Map current wildfire susceptibility for @wildfire_audit and summarize high-risk zones.

Reference Assets Clearly

Regions, layers, and reports created in a chat become assets. Open the Assets control or use the mention menu to select the exact item you want to reference.

Good asset references:

  • "Use @detroit_roi for the analysis."
  • "Compare this against the previous vegetation stress layer."
  • "Open the flood exposure table and summarize the highest-risk OSM features."

Raster Prompting Tips

Be clear about dates and seasons. For vegetation and land-cover work, use consistent growing-season or seasonal windows unless you have a reason to compare different seasons.

Common raster patterns

Land cover: "Generate a land-cover map for @roi for summer 2025."

Land-cover trend: "Show land-cover composition from 2018 to 2025 for @roi."

Vegetation: "Analyze NDVI trends for @roi from 2018 to 2025 and identify stress hotspots."

Remote-sensing flood: "Run remote-sensing flood risk assessment for @roi."

Physics-based flood: "Run flood depth and OSM feature exposure for @roi using the 100-year return period where coverage exists."

Wildfire: "Map current wildfire susceptibility for @roi and summarize high and very high areas."

Physical risk: "In Super mode, screen @site_boundary for physical climate risk and identify dominant hazard drivers."

Forest loss: "Analyze forest loss and likely drivers for @roi since 2018."

Asset and Exposure Prompting Tips

Asset and exposure prompts should specify the feature type, risk layer, and any filtering criteria.

Public assets

Find buildings and critical facilities within 500 meters of @floodplain.

Uploaded parcels

Show parcels in @uploaded_parcels with conservation priority greater than 3.

Follow-up

Of these buildings, keep only those in the very high flood-risk class.

Export

Export the selected features as GeoJSON and include the main attributes.

Use Follow-ups Deliberately

Follow-up prompts are strongest when they point to a specific previous output:

  • "Of these results..."
  • "Using the flood exposure layer..."
  • "Based on the report you just created..."
  • "Compare this chart with the 2018-2025 land-cover trend..."

Common Mistakes

No site or asset context

"Analyze vegetation in Ontario" is too broad. Use a saved ROI, imported layer, or explicit boundary.

Ambiguous dates

"Recent change" can be unclear. Use exact years, a date range, or a season such as summer 2025.

Too many tasks

Large multi-part requests are harder to inspect. Run the key layer first, then ask follow-ups.

Missing output format

Say whether you need a map, attribute table, report, export, or concise summary.

Quick Templates

Map: "Map [analysis] for [@roi] in [time window]."

Trend: "Show [metric] trends for [@roi] from [start] to [end]."

Exposure: "Intersect [hazard layer] with [asset layer] and summarize exposed features."

Report: "Create a report that explains [result] for [audience]."

Knowledge base: "Use our uploaded documents to interpret [analysis result]."