ArcGIS Pro: Makes Complex Land Mapping Easy!

I've been using ArcGIS Pro to map different types of land use based on LCMAP Level 1 Land Cover Classes (Class 1: Developed, Class 2: Cropland, Class 3: Grass/Shrub, 4. Tree Cover, Class 5: water, Class 6: Wetland, Class 7: Ice/Snow and Class 8: Barren) , and I'm excited to share a tool in ArcGIS Pro that makes this process easier: AutoDL.

Imagine you have a massive amount of satellite images, and you need to figure out what's in each pixel. Doing this manually would take forever! That's where deep learning and AutoDL come in.

Deep learning is like teaching a computer to see patterns. AutoDL is a tool in ArcGIS Pro that automatically trains and tests different "computer vision" models on your data. It figures out the best model for the job.

Here's how it works:

• Data Preparation, you can use tools like Export Training Data For Deep Learning in ArcGIS Pro to do so, this step is critical and usually time-consuming.
• You give AutoDL your satellite images and tell it what each type of land looks like (e.g., "this is a water, this is a tree and.....").
• AutoDL tries out different models, like HRNet, PSPNetClassifier, and UnetClassifier.
• It compares their performance and picks the winner based on how well they classify the land.
• The chosen model is then used to automatically classify all your images, saving you tons of time and effort.

I used this workflow with radar images from satellites, which can see through clouds. I trained the model on images from 2018 and then used it to classify images from 2024. This allowed me to see how land use has changed over time.

Tips and Resources

• It is recommended using an NVIDIA GPU with at least 8GB of dedicated memory for optimal performance.
• ArcGIS Living Atlas offers pretrained deep learning models that can be used directly or fine-tuned.

Key Takeaways

• ArcGIS Pro provides a comprehensive workflow for LULC classification using deep learning.
• AutoDL simplifies the process of training and selecting the best model.
• SAR imagery offers advantages for monitoring land cover change.

By leveraging these tools and resources, you can effectively apply deep learning for land-use land-cover classification, contributing to environmental monitoring, resource management, and other critical applications. AutoDL is a powerful tool that makes deep learning accessible to everyone. If you're working with land cover data, I highly recommend giving it a try!