I’m mapping 2023–2024 crop vigor across Iowa in Google Earth Engine using Sentinel-2 SR, but persistent clouds mean the median composite still leaves gaps; has anyone had better luck with S2Cloudless vs QA60 plus temporal interpolation, or do you jump to PlanetScope or Sentinel-1 fusion? I’d love a lightweight approach that doesn’t blow up memory — rolling 10-day medoids got me to 85% coverage, but edges still look smeared.
same pain in Iowa — I’d ditch the “rolling 10‑day medoids” and instead join COPERNICUS/S2_CLOUD_PROBABILITY, mask >45–50%, then sort by cloud_prob and mosaic per 10‑day window; that’s surprisingly light on memory and leaves fewer pinholes than QA60+median. If gaps remain, a single linear fill using the previous/next mosaic beats full temporal interpolation, so I wouldn’t jump to Planet or S1 unless you really need sub‑weekly cadence. Sentinel-2: Cloud Probability | Earth Engine Data Catalog | Google for Developers — does about 50% hold up for your 2023–2024 dates?