Improving the SMAP Daily Soil Moisture Time Series with Land Surface Model Datasets Using Power Spectrum-Adjustment Techniques
AbstractLand-atmosphere feedbacks act through process chains that link variables in the land-atmosphere system. For the global energy and water cycles, the first link in the chain is soil moisture. Flux tower sites provide in-situ observations, including land surface states, surface fluxes, and nearsurface atmospheric states, to validate these links; however, they are unevenly distributed over the globe. Therefore, to obtain a global view of observationally based land-atmosphere coupling metrics, satellite data are useful. Among satellite products, the Soil Moisture Active Passive (SMAP) satellite provides the closest match to in-situ observations. However, SMAP exhibits stochastic random noise that can deflate coupling estimates. Since soil moisture variability closely follows a first-order Markov process, it typically has a distinct red noise spectrum. Satellite data with random noise has a whiter spectrum at high frequencies that can be compared to the expected red spectrum. Also, missing data in SMAP are not entirely random; its 8-day repeating polar orbit creates a cadence of missing data for both ascending and descending overpasses, depending on the location. This creates additional artifacts in the power spectrum, calculated through lagged autocovariance in the time series, with harmonic spikes at 8, 4 (8/2), 2 2/3 (8/3), and 2 (8/4) days that broaden due to the satellite's orbital variations. To be optimally useful for quantifying land-atmosphere feedbacks, the effects of random noise and periodic missing data must be minimized. A power spectrum adjustment technique has been designed to remove the orbital harmonic spikes from Level 3 (L3) SMAP data. This is achieved by fitting and removing a catenary function to the power spectrum between harmonic spikes. This adjusted spectrum is then scaled to match surface layer soil moisture observations at sites of the AmeriFlux network (in-situ data), which exhibit relatively low noise and have spectra that are very similar to those produced by offline land surface models (LSMs). Utilizing validated spectral data from gridded LSM-based datasets, a global L3 SMAP product with removed noise and harmonic effects is being produced. We will present results quantifying the extent to which this technique improves SMAP data and its temporal correlation with observations.