To Average or Not to Average? Using ground reference networks to validate satellite soil moisture products

The validation of remotely sensed Soil Moisture (SM) products against ground reference data is strongly affected by the spatial scale mismatch between the large satellite footprints and point-scale in situ measurements, exacerbated by differences in station density across in situ monitoring networks. A common but inconsistently applied practice is to average multiple in situ sensors falling within a single satellite grid cell prior to validation. However, there is currently no clear consensus on whether spatial averaging provides a more reliable reference for satellite validation.

In this study, we systematically assess the impact of spatially averaging in situ soil moisture measurements from different sensors of the International Soil Moisture Network (ISMN) on the validation of soil moisture products from the ESA Climate Change Initiative (CCI). The ESA CCI SM product is provided on a 0.25° grid (approximately 25 × 25 km). More than 20% of the CCI grid cells, with in situ stations on them, contain two or more in situ sensors, with an average of 9 sensors per such grid cell. Averaging these sensors within single grid cells may provide a more reliable proxy for grid cell average soil moisture dynamics, but only if their measurements are mutually consistent. Alternatively, satellite products may be compared against each sensor individually, but this causes grid cells that contain multiple sensors to be disproportionately represented in validation summary statistics, potentially biasing validation metrics.
We therefore examine the implications of different spatial averaging choices to answer the question of when and how in situ measurements from dense networks should be averaged for the validation of satellite products.

Our results suggest that, in most cases, averaging measurements from multiple, spatially distributed sensors yields more reliable reference time series that improve in situ-satellite comparison metrics. However, we also see a considerable number of cases where averaging sensors reduces the reliability of the time series, most commonly when averaging measurements from different sensor types. Our findings highlight the importance of methodological consistency and provide guidance for the validation of current and future satellite soil moisture products.