Globally estimating root mean square errors in remotely sensed soil moisture

To make the best use of remotely sensed near surface soil moisture observations requires a confident understanding of their accuracy across the globe. In this study Root Mean Square Errors (RMSE) are estimated over North America for the soil moisture data sets retrieved from the Advanced Microwave Scanning Radiometer (AMSR-E; using the Land Parameter Retrieval Model) and the Advanced Scatterometer (ASCAT), using two methods: triple colocation (RMSE_TC) and error propagation through the soil moisture retrieval models (RMSE_EP). Due to the systematic differences between different global scale soil moisture data sets, RMSEs estimated by comparing two or more soil moisture data sets rely on rescaling the data sets into a common "reference" climatology, for example by matching to the mean and variance of the reference. The resulting RMSE is a measure of the random (i.e., the non-biased) component of the error only, and is a direct function of the time series standard deviation of the reference data set. To remove the signal of the reference standard deviation it is recommended that RMSE for remotely sensed soil moisture be presented as a fraction of the local time series standard deviation (fRMSE). Visually, maps of the fRMSE_TC and fRMSE_EP have similar large-scale spatial patterns, and both are consistent with the soil moisture errors expected from the land cover type. The absolute values of both fRMSE estimates are benchmarked against errors estimated from comparison to in situ soil moisture observations from the SCAN/SNOTEL network in the US (RMSE_IS). After correcting the RMSE_IS to account for the errors contributed by the in situ data, the agreement between fRMSE_TC and fRMSE_IS is reasonable for both ASCAT and AMSR-E. However, for fRMSE_EP, only the ASCAT errors agree well with fRMSE_IS, since the AMSR-E fRMSE_EP are unrealistically high. These results suggest that triple colocation could be useful for estimating the absolute values and spatial variability of the (random component of) remotely sensed soil moisture errors, while error propagation could be useful for estimating the spatial variability.