GIS Chat for ESA SNAP

AI-powered chat assistant for ESA SNAP Desktop that executes remote sensing operations via natural language.

Features

• Chat panel docked inside SNAP Desktop (Tools menu)
• Natural language to GPT (Graph Processing Tool) command translation
• Python/snappy code generation and execution
• Google Earth Engine integration -- query, process, and download GEE data directly from the chat
• Automatic context awareness (open products, bands, CRS, dimensions)
• Smart code type detection (GPT command vs. Python) from content
• Markdown rendering in chat (tables, headers, code blocks, bold)
• Automatic error recovery with "Analyzing error and working on a fix..." feedback
• Adaptive tiled downloading for GEE (automatically splits large downloads to stay under the 50 MB API limit)
• Confirmation dialog before executing operations

Supported AI Providers

Provider	Cost	Notes
Google Gemini	Free tier available	Recommended to start
Ollama	Free (local)	Runs on your machine, fully offline
Anthropic (Claude)	Paid	claude-sonnet-4-6, claude-opus-4-6
OpenAI (GPT)	Paid	gpt-4o, gpt-4o-mini
OpenAI-compatible	Varies	LM Studio, vLLM, any compatible endpoint

Installation

Manual Install

1. Download org-gischat-snap.jar from Releases

2. Copy the JAR to your SNAP user modules directory:

   • Windows: %APPDATA%\SNAP\modules\
   • Linux/Mac: ~/.snap/snap-desktop/modules/

3. Create the module config file at:

   • Windows: %APPDATA%\SNAP\config\Modules\org-gischat-snap.xml
   • Linux/Mac: ~/.snap/snap-desktop/config/Modules/org-gischat-snap.xml

   With this content:

   <?xml version="1.0" encoding="UTF-8"?>
   <!DOCTYPE module PUBLIC "-//NetBeans//DTD Module Status 1.0//EN"
                           "http://www.netbeans.org/dtds/module-status-1_0.dtd">
   <module name="org.gischat.snap">
       <param name="autoload">false</param>
       <param name="eager">false</param>
       <param name="enabled">true</param>
       <param name="jar">modules/org-gischat-snap.jar</param>
       <param name="reloadable">false</param>
   </module>

4. Restart SNAP Desktop

5. Open via Tools > GIS Chat

Build from Source

Requirements: Java 21+, Maven 3.9+

git clone https://github.com/Nagyhoho1234/SNAPChat.git
cd SNAPChat
mvn package -DskipTests

The JAR is produced at target/org-gischat-snap.jar. Copy it to the modules directory and create the config XML as described above.

Windows gotcha: SNAP's user modules directory is %APPDATA%\SNAP\modules\, not %USERPROFILE%\.snap\snap-desktop\modules\ (the latter is the Linux/Mac path). Deploying to the wrong location means SNAP loads the old JAR silently.

Google Earth Engine Setup (Optional)

GIS Chat can query, process, and download data from Google Earth Engine directly. To enable:

1. Install the GEE Python package in your system Python:

   pip install earthengine-api
   

2. Authenticate (one-time):

   import ee
   ee.Authenticate()

3. Configure in GIS Chat -- go to Settings and enter your GEE project ID (e.g. my-gee-project).

If you don't have a GEE project, create one at code.earthengine.google.com.

After setup, you can ask things like "Calculate a soil moisture map for Debrecen using Sentinel-1 from GEE" and the AI will handle the full pipeline -- querying GEE, tiled downloading at native resolution, merging tiles, and telling you the output file path.

Usage

1. Open SNAP and load a satellite product
2. Open Tools > GIS Chat
3. Click Settings to choose your AI provider and enter an API key (if needed)
4. Type a natural language request, for example:

You say	GIS Chat does
"Apply radiometric calibration to this product"	Runs GPT Calibration operator
"Calculate NDVI from bands B8 and B4"	Runs GPT BandMaths operator
"Subset this product to a bounding box"	Runs GPT Subset operator
"What bands does this product have?"	Reads context and answers directly
"Download Sentinel-2 NDVI for my area from GEE"	Queries GEE, downloads tiles, merges
"Calculate soil moisture from S1 for Debrecen"	Builds S1 composite on GEE, downloads at 10m

The assistant reads your current SNAP state (open products, bands, CRS) and generates the appropriate GPT command or Python/snappy code.

Known Limitations

• Cannot open products into SNAP GUI from code. Due to Java security and classloader restrictions, the esa_snappy Python bridge runs in a separate JVM and cannot access SNAP Desktop's GUI classes (SnapApp, ProductManager, OpenProductAction). Generated files must be opened manually via File > Open Product (Ctrl+O). The chat will always show the exact output file path.
• esa_snappy multi-band GeoTIFF reading can fail with "Cannot construct DataBuffer". The AI is instructed to use GDAL for reading and snappy only for writing when this occurs.
• GPT operator names are case-sensitive and not all operators support --version. Use gpt -h to list available operators.

Project Structure

SNAPChat/
  pom.xml                         # Maven build (Java 21, SNAP 13.0)
  src/main/java/org/gischat/snap/
    ChatTopComponent.java          # NetBeans TopComponent (UI panel)
    SettingsDialog.java            # Settings dialog (provider, key, model, GEE project)
    LlmService.java                # Multi-provider LLM client (no dependencies)
    LlmProvider.java               # Provider enum (endpoints, models)
    LlmResponse.java               # Response model with tool call support
    ChatSettings.java              # Persistent settings (java.util.prefs)
    SnapContextService.java        # Reads SNAP state (products, bands, CRS)
    CommandExecutor.java           # Runs GPT commands and Python scripts
  src/main/resources/org/gischat/snap/
    layer.xml                      # NetBeans filesystem registration
    icon24.png                     # Toolbar icon

Security

• API keys are stored locally in Java Preferences (java.util.prefs)
• Keys are never transmitted anywhere except to the configured AI provider endpoint
• All LLM communication uses HTTPS (except local Ollama)

Citation

If you use GIS Chat in your research, please cite the preprint:

Fehér, Zs. Z. (2026). GIS Chat: Bridging Natural Language and Desktop GIS Automation with LLM-Powered GIS Plugins. EarthArXiv preprint, submitted to SoftwareX. DOI: 10.31223/X54Z09

@article{feher2026gischat,
  title={GIS Chat: Bridging Natural Language and Desktop GIS Automation with LLM-Powered GIS Plugins},
  author={Feh{\'e}r, Zsolt Zolt{\'a}n},
  year={2026},
  doi={10.31223/X54Z09},
  note={EarthArXiv preprint, submitted to SoftwareX}
}

License

MIT - Zsolt Zoltan Feher