Comparing the ensemble and extended Kalman filters for in situ soil moisture assimilation with contrasting conditions

Abstract

Two data assimilation (DA) methods are compared for their ability to produce an accurate soil moisture analysis using the Météo-France land surface model: (i) SEKF, a simplified extended Kalman filter, which uses a climatological background-error covariance, and (ii) EnSRF, the ensemble square root filter, which uses an ensemble background-error covariance and approximates random rainfall errors stochastically. In situ soil moisture observations at 5 cm depth are assimilated into the surface layer and 30 cm deep observations are used to evaluate the root-zone analysis on 12 sites in south-western France (SMOSMANIA network). These sites differ in terms of climate and soil texture. The two methods perform similarly and improve on the open loop. Both methods suffer from incorrect linear assumptions which are particularly degrading to the analysis during water-stressed conditions: the EnSRF by a dry bias and the SEKF by an oversensitivity of the model Jacobian between the surface and the root-zone layers. These problems are less severe for the sites with wetter climates. A simple bias correction technique is tested on the EnSRF. Although this reduces the bias, it modifies the soil moisture fluxes and suppresses the ensemble spread, which degrades the analysis performance. However, the EnSRF flow-dependent background-error covariance evidently captures seasonal variability in the soil moisture errors and should exploit planned improvements in the model physics.Synthetic twin experiments demonstrate that when there is only a random component in the precipitation forcing errors, the correct stochastic representation of these errors enables the EnSRF to perform better than the SEKF. It might therefore be possible for the EnSRF to perform better than the SEKF with real data, if the rainfall uncertainty was ac-curately captured. However, the simple rainfall error model is not advantageous in our real experiments. More realistic rainfall error models are suggested.Published by Copernicus Publications on behalf of the European Geosciences Union.

The effect of satellite-derived surface soil moisture and leaf area index land data assimilation on streamflow simulations over France

“…Previous work by Muñoz Sabater et al (2007) and Fairbairn et al (2015) clearly demonstrated that the assimilation of SSM observations with an SEKF can improve WG2 with the three-layer ISBA-A-gs model. Barbu et al (2014) also demonstrated that the assimilation of LAI reduces phase errors in the modelled LAI evolution.…”

“…Draper et al (2009) found the flowdependence from a 24 h assimilation window was sufficient to enable the SEKF to perform similarly to an EKF (which cycles the background-error covariance). Likewise, Muñoz Sabater et al (2007) and Fairbairn et al (2015) found that the SEKF and EnKF performed similarly, in spite of different linear assumptions.…”

“…The linear assumptions in deriving the Jacobians are generally acceptable for these perturbation sizes. However, occasionally the linear assumptions can break down, especially during dry periods in summer Fairbairn et al, 2015). For this reason we set an upper bound on the soil moisture Jacobians of 1.0.…”

“…During dry conditions in summer, the SEKF ∂WG1 ∂WG2 Jacobian can be excessive. This is linked to a rapid increase in transpiration when water is added to WG2 following dry conditions Fairbairn et al, 2015). The origin of this nonlinearity is partly related to an unrealistic feature of the surface energy balance.…”

Estimating Root Zone Soil Moisture at Continental Scale Using Neural Networks

“…To estimate RZSM over a large geographical area, many computational techniques have been proposed to retrieve the soil moisture profile using surface information, based on the fact that RZSM is related to surface soil moisture and meteorological forcing through various hydrological processes. Among various alternatives, soil-vegetation-atmosphere transfer (SVAT) models and land surface models (LSMs) are more widely used for large-scale estimation, particularly in combination with data assimilation techniques (Kumar et al, 2009;Xia et al, 2014;Fairbairn et al, 2015). These models use physically based equations to represent the hydrological processes governing RZSM dynamics.…”