Towards soil hydraulic parameter retrieval from land surface models using near-surface soil moisture data

Abstract

The water and energy fluxes at the land-atmosphere interface depend heavily on the soil moisture content, which imposes a significant control on evaporation, infiltration and runoff. Nonetheless, temporal soil moisture evolution is not easy to measure or monitor at large scales due to its spatial variability, which is largely driven by the local variation in soil properties and the vegetation cover. As a consequence, soil moisture dynamics are generally estimated using land surface models, with model physics based on low- resolution soil property maps, which may include significant errors due to their underpinning information and spatial scale. Consequently, there is a need for more accurate and detailed soil parameter data sets than are currently available in order for the model to perform reliably. There are a large number of land surface models with a range of characteristics and model physics. This paper presents an assessment of the soil hydraulic parameter retrieval capability from calibration to surface soil moisture observations using two prominent land surface models, the Community Atmosphere Biosphere Land Exchange (CABLE) model and the Joint UK Land Exchange Simulator (JULES). A synthetic twin experiment was used to achieve the objective of identifying the most suitable of these land surface models to be used in soil hydraulic parameter retrieval using near-surface soil moisture data. This study was conducted in three steps: (a) identifying the soil parameters to which the soil moisture shows the greater sensitivity, (b) assessing the capability of the land surface model to allow the retrieval of identified parameters, and (c) assessing the ability of the model to simulate real conditions. A range of sensitivity studies were performed for both the CABLE and JULES land surface models using the parameter sensitivity index (S) to identify the model parameters with the highest impact on soil moisture prediction. Having identified the key parameters, a set of assumed 'true' parameter values were prescribed for calculating 'true' soil moisture dynamics. Those parameters found most sensitive to surface soil moisture observations were then changed to 'best guess' values, and subsequently retrieved by optimizing the surface soil moisture predictions against those from the 'truth run' using the Parameter ESTimation (PEST) software. Parameter retrieval capability was assessed for both land surface models based on comparison with the 'true' soil parameters and deeper layer soil moisture. Compared to field observation data, both models captured the wetting and drying trends of the real-world scenario, but over-estimated the soil moisture after wet-up periods and under-estimated the moisture for deeper layers. However, JULES showed greater sensitivity to changes in the soil hydraulic parameters when compared to CABLE. Moreover, an unrealistic value for the volume of water at wilting point, in the form of the observed lowest soil moisture, had to be used as an input parameter for CABLE to simulate sensible soil moisture. It is also important to note that in contrast to CABLE, JULES has been formulated to allow depth varying soil parameter data to be assigned to different soil layers with the added flexibility of allowing the user to vary the number of soil layers and their depths. It was therefore concluded that the JULES model is better suited for the long-term work of retrieving soil hydraulic parameters from surface soil moisture observations.