Common Formats for Geospatial AI Models

GeoWGS84
Jul 29
1 min read

The way we study, interpret, and work with spatial data has undergone a complete transformation in recent years due to Geospatial Artificial Intelligence (GeoAI). Advanced solutions for urban planning, environmental monitoring, agriculture, disaster management, and other fields are made possible by GeoAI, which combines artificial intelligence (AI) and machine learning with geographic information systems (GIS). The data formats used for model training, storage, deployment, and inference, however, have a significant impact on the performance of these models.

GeoAI Model File Extensions

These include general deployment extensions and Esri-specific formats for the AI/ML model files and metadata used in GeoAI processes.

AI/ML Model File Extensions

Extension	Description	Framework
.onnx	Open Neural Network Exchange format	Cross-platform (ONNX)
.pb	Protocol Buffer for model storage	TensorFlow
.h5	HDF5 format for model and weights	Keras / TensorFlow
.pt	PyTorch model weights	PyTorch
.pth	PyTorch model checkpoint	PyTorch
.pkl	Pickled model (traditional ML)	Scikit-learn / Python
.joblib	Optimized serialization for scikit-learn models	Scikit-learn
.tflite	Lightweight model for edge/mobile deployment	TensorFlow Lite
.mlmodel	CoreML model for Apple devices	Apple CoreML

Esri-Specific GeoAI Model Extensions

Extension	Description	Used In
.dlpk	Deep Learning Package – includes model + metadata.	ArcGIS Pro/Online
.emd	Esri Model Definition – metadata describing input/output	ArcGIS
.rft.xml	Raster Function Template – defines raster model operations.	ArcGIS Pro
.sd	Service Definition – for publishing models as services	ArcGIS Server