Just re-emphasizing an important clarification from the organizers that some people might have missed 👇
There seems to be some confusion around what counts as “using latitude/longitude” in this challenge.
The rule is not just about excluding the raw lat/lon columns — it also applies to any features derived from them.
👉 Not allowed:
Even if you don’t explicitly include lat/lon in your model, these features still reintroduce spatial information, which goes against the intent of the competition.
👉 Allowed:
👉 Not allowed:
A simple rule of thumb:
Using lat/lon to extract TerraClimate values = ✅ Allowed Using lat/lon to create model features (including distances) = ❌ Not allowed
The goal here is to ensure models rely on the provided environmental/climate data, not on geographic shortcuts.
Let’s keep things fair and aligned with the spirit of the competition 👍
Discussion that clarified this:
🌍 Challenge Chat: Action Required - Data Rules ... - 666 Views
@Koleshjr Well said 👍. I support this in totality. In fact I needed to reimplement both my extraction and ML pipelines all over again when I saw the same things well pointed out by @meganomaly. Have sent both my old pipeline and newly updated pipeline to Zindi for proper review. So after uploading my new sub file from this new pipeline, my LB score went back to 0.9450 with CV score 0.9412 and another pipeline with LB score 0.9439 and CV score 0.9434 respectively. I hope and wish every participant will embrace this best standard practices and stick to the rules.