Approach for extending the Deering coastline time series

[Vrittigyl]

Hi @fwitmer, @rawann31 and @Ritika-K7,
I’m Vritti Goyal, a third-year B.Tech student in Information Technology. I have experience with Python, PyTorch, and machine learning, and I’m interested in contributing to Project Idea Automated coastline extraction for erosion modelling.

I’ve set up and run the existing pipeline to understand the current data flow and constraints. As a first step, I plan to focus on extending the Deering dataset beyond 2019, while keeping the rest of the pipeline unchanged initially. In parallel, I’ll also spend time studying and understanding the later phases of the pipeline to better inform future improvements.

Along with manually restricting date ranges (e.g., focusing on summer months), I’d like to consider a small set of metadata-based factors that directly affect NDWI label reliability:

• snow_ice_percent — exclude from processing
• sun_elevation — reject low values to avoid long-shadow effects
• cloud_cover — already handled in the current pipeline
• shadow_percent — optional secondary signal

NDWI assumes that water absorbs NIR strongly, land reflects more NIR, and that this contrast remains relatively stable. Snow, ice, and shadows violate these assumptions: snow and ice alter reflectance behavior, while shadows reduce NIR return and can be misclassified as water, particularly near cliffs and steep terrain.
I’d appreciate your thoughts on whether this approach aligns with your expectations for extending the Deering time series, and if there are any additional considerations you’d recommend at the data ingestion stage.

[Vrittigyl]

@fwitmer Please let me know if this looks good so I can start working on it.

[Ritika-K7]

Thank you for your interest.
Yes, this makes sense. Shadows are an important factor to handle, particularly near cliffs and steep terrain, where NDWI can be misleading.

[tanu-shree06]

Hello @Ritika-K7, can we combine NDWI and DEM data to improve the accuracy of coastline?

[Ritika-K7]

Yes, it is possible to combine NDWI with DEM data, and it can help improve the accuracy in some areas especially near cliff regions .