Multi-sensor active-passive blending

Multi-sensor active-passive blending answers the question: what is the best estimate of a geophysical variable when multiple satellite sensors each observe it with different error characteristics and sampling patterns? The method merges measurements from active sensors (scatterometers) and passive sensors (microwave radiometers) that observe the same variable, exploiting the complementary noise structures of the two instrument types to produce a harmonised record more accurate and more complete than either alone.

The canonical application is the ESA CCI Soil Moisture climate data record. The blending pipeline produces three output streams: an ACTIVE-only product from scatterometers (ERS-1/2, MetOp ASCAT), a PASSIVE-only product from radiometers (SMMR, SSM/I, TMI, AMSR-E, AMSR2, WindSat, SMOS), and a COMBINED product blending both.^[1] All inputs are first rescaled to a common dynamic range using CDF-matching against a GLDAS land surface model reference. Error weights are then estimated using triple collocation analysis (TCA), which infers per-dataset error variance from the statistical agreement among three independent data streams without requiring a ground truth.^[2] Inverse-error-variance weighting combines the rescaled inputs into the blended estimate. In regions where TCA cannot be applied (deserts, dense vegetation, high latitudes), a VOD-based polynomial regression fills the error-weight gap.

The merging approach has evolved through successive algorithm versions: version 2 applied decision-tree selection, version 3 introduced least-squares fusion with triple-collocation weights, and version 4 added the VOD polynomial regression for gap filling.^[1] The combined product covers 1978 to present at 0.25 degree global daily resolution. The method fails where soil moisture is physically undefined (frozen ground, dense tropical forest) and over open water.