EO Summer School 2025 Hackathon - NDVI time-series

Satellite imagery is often affected by clouds and other artifacts, which can obscure or distort the data. Although such contaminants are filtered out, many applications, particularly machine learning models, require complete datasets to be effective. The task of reconstructing missing values, commonly referred to in the literature as gap-filling or time-series reconstruction, is the focus of this challenge.You will work with real NDVI data derived from satellite imagery at a 30 m resolution. The dataset covers three tiles (each approximately 120 km × 120 km): one central tile, along with one adjacent tile to the east and one to the west. For each tile, cloud-free observations are available over a 10-year period, with a temporal resolution of 16 days. In addition, 10,000 time series have been sampled from pixels within the central tile. To these time series, we have artificially introduced an additional 25% of missing values.Your goal is to reconstruct the missing values as accurately as possible, using both the remaining values within each time series and information from the neighboring tiles. The quality of your solution will be evaluated based on the Root Mean Square Error (RMSE)—the lower, the better

https://kaggle.com/competitions/eo-summer-school-2025-hackathon-ndvi-reconstruct