An Improved Triple Collocation Analysis Algorithm for Decomposing Autocorrelated and White Soil Moisture Retrieval Errors

Abstract

If not properly account for, autocorrelated retrieval errors can lead to inaccurate results in soil moisture data analysis and reanalysis. Here we propose a more generalized form of the triple collocation analysis algorithm (GTC) capable of decomposing the total error variance of remotely sensed surface soil moisture retrievals into their autocorrelated and the serially white components. Synthetic tests demonstrate the robustness and accuracy of GTC—even in the presence of significant temporal data gaps. However, the accuracy of GTC error autoregressive parameter estimates is relatively more sensitive to temporal data availability. In addition, land surface model soil moisture predictions collected from phase 2 of the North American Land Data Assimilation System and remotely sensed surface soil moisture retrievals obtained from the European Space Agency Climate Change Initiative (ESA CCI) are applied for a real data demonstration. Despite expectations to the contrary, significant error autocorrelation is found in the remotely sensed‐based ESA CCI soil moisture data sets. In particular, ESA CCI‐Act (i.e., the subset of ESA CCI soil moisture retrievals based on active scattomotter data) demonstrates the largest autoregressive parameters over low biomass areas. Conversely, ESA CCI‐Pas retrievals (based on passive radiometer data) have larger error autoregressive parameters over high biomass areas. As such, results clarify circumstances in which errors in remotely sensed surface soil moisture retrievals cannot be considered serially white.

A Global Assessment of Added Value in the SMAP Level 4 Soil Moisture Product Relative to Its Baseline Land Surface Model

“…However, a key difference is that RR ASC uses an independent soil moisture product (ASCAT in this case) as the instrument, rather than a temporally lagged soil moisture time series. This removes any potential impacts related to autocorrelated soil moisture observation errors (Dong & Crow, ).…”

“…Therefore, RR ASC provides an unbiased estimate of the true value of RR obtained from (5). This removes any potential impacts related to autocorrelated soil moisture observation errors (Dong & Crow, 2017). RR ASC can also be generalized as an instrumental variable based technique shown in Su et al (2014) and Dong et al (2019).…”

“…Provided three independent data products are available, triple collocation (TC; Stoffelen, ) can solve for the error variances of each individual product—even in the absence of high‐quality ground observations. Hence, TC is frequently used for large‐scale soil moisture validation studies (Chen et al, ; Dong & Crow, ; Draper et al, ; Scipal et al, ). However, SMAP L4 ingests both land surface model simulations and radiometer observations and, hence, contains errors that are cross‐correlated with both land surface model estimates and soil moisture retrievals acquired from other space‐borne passive microwave observations (primarily because of shared ancillary data and common approaches to land and radiative transfer modeling).…”

ESA CCI Soil Moisture Assimilation in SWAT for Improved Hydrological Simulation in Upper Huai River Basin

“…RS SM still has considerable uncertainty although its accuracy and reliability have been largely improved in recent years [49,50]. In this study, referencing from previous researches [5,51,52], the error for ESA CCI SM is assumed to be an additive Gaussian distribution with the standard deviation (SD) of σ R . Here, the estimation of σ R is obtained from the equation referring to the study of Lievens et al [30]:…”

Evaluation of Satellite-Derived Soil Moisture in Qinghai Province Based on Triple Collocation

“…Gruber et al reviewed the existing error representations of satellite-based soil moisture data, analyzed the basic assumptions behind the TC method, and proposed a combined investigation of the signal-to-noise ratio (SNR) (expressed in logarithmic units), the unscaled error variances, and the soil moisture sensitivities of the datasets for the evaluation of remotely sensed soil moisture datasets [15]. Dong and Crow discussed the influence of auto-correlated retrieval errors on the TC method and proposed a generalized triple collocation (GTC) analysis algorithm which can decompose the total error variance of remote sensing soil moisture data into its auto-correlated and white-error components [16]. The standard TC method can only provide relative error, and requires a reference dataset to be selected from three parallel data products; this means that the resulting error variance is influenced by the deviations caused by the multiplication and addition operation of the reference dataset.…”