Claude Does Permanent Scatterer Interferometric Synthetic Aperture Radar: A tutorial for the use of large language models in replication studies

In this tutorial, the application of large language models (LLMs) in replicating complex remote sensing algorithms is explained, using permanent scatterer interferometric synthetic aperture radar (PSInSAR) as a case study. The article highlights the transformative potential of LLMs in scientific code development. In the tutorial, we show the problems in coding with LLMs, in terms of obvious as well as hidden errors in the code they provide. The PSInSAR workflow is implemented iteratively using Claude Sonnet 3.5. The process involves structured prompting, contextual guidance, testing, and debugging. This tutorial, therefore, not only demonstrates LLM usage but also provides a very detailed step-by-step implementation guide to PSInSAR. The LLMs generate functional Python code for PSInSAR, reducing the development time significantly. However, the code is not error-free and requires debugging. Domain knowledge in SAR interferometry (InSAR) is essential for guiding the LLM via detailed prompts and also for validating outputs. LLMs can serve as powerful collaborators in scientific replication, shifting focus from manual coding to high-level problem solving and debugging. This tutorial demonstrates how to use LLMs to assist scientific code development and also underscores the changes artificial intelligence (AI)-assisted research will bring in geospatial sciences.