Gap-filled, speckle-reduced, monthly Sentinel-1 dual-polarization (VV, VH) radar backscatter with a per-estimate uncertainty layer, organized on the MGRS tile grid and distributed as cloud-native GeoZarr inside Icechunk repositories.
This dataset takes the NASA/JPL OPERA Radiometric Terrain Corrected Sentinel-1 (RTC-S1) archive, and estimates, for every pixel, a smooth and regularly sampled monthly backscatter series together with its uncertainty. The result is ready for time-series analysis without the per-scene gaps, speckle, and irregular revisit of the raw archive.
Tiles follow the MGRS 100 km grid:
T11SLB.[30.0, 0.0, x_origin, 0.0, -30.0, y_origin].To find the tile(s) covering an area of interest, intersect your AOI with the
MGRS grid (widely available, e.g. from ESA/Copernicus) and select
the tiles published here. All tiles live
in one Icechunk repository, each under {ASC | DESC}/tiles/{MGRS tile}.
The dataset is one Icechunk repository. The repository follows the GeoZarr conventions for georeferencing.
The primary consumer-facing layer is estimate, the time-series model's
smoothed backscatter estimate, published beside its uncertainty:
Each layer ships with a multiscales pyramid for fast visualization: 7 levels downsampled ×1, ×2, ×4, …, ×64. The overviews are convenience layers derived from the full resolution level 0.
One tile and one month of estimate, written as a COG with
rioxarray, continuing from tree above:
A defining feature of this product is that every smoothed estimate carries a quantified uncertainty. The estimator tracks not just the most likely backscatter value but how confident it is, given how many observations were available, how noisy they were, and how far the estimate is from the nearest acquisition.
Use the uncertainty layer to weight estimates, flag low-confidence months (e.g. long gaps, heavily masked terrain), or propagate error into downstream analysis.