Plain Language Summary
Hydrological models need information about the soil physical characteristics (soil hydraulic parameters), which are in general not available if the models are applied at larger scales (region to global scale). Therefore, pedotransfer functions (PTFs) are classically used, which relate easily available soil properties such as sand-, silt-, clay-content, soil organic carbon content, and soil bulk density, which are available from soil maps, to the soil hydraulic parameters. Unfortunately, there are many different PTFs available in literature. In the study presented, we analysed the impact of different PTFs on the simulation results of water fluxes and found, that the choice of PTF impacts the simulation results. Further, some PTFs were identified as being less robust compared to others. In general, the study shows that harmonizing PTFs in model-inter-comparisons is needed to avoid artefacts originating from the choice of PTF rather from different model structures.
Keywords: pedotransfer functions, land surface models, LSM, hydrological models, crop models, model inter-comparison, model ensemble mean
Introduction
Water fluxes and soil water content are key variables in the terrestrial system as they control the exchange of water and energy between the land-surface and the atmosphere (e.g., Vereecken et al., 2015). Modelling of the water flow in the unsaturated zone, and the uncertainty in the parameters used to simulate water flow, has been a topic of intense research for many years, both in the soil hydrological and land surface modelling community (Shao & Irannejad, 1999; (Tietje & Tapkenhinrichs, 1993; Vereecken et al., 2008; Iwema et al. (2017)). Moreover, climate modellers have studied the role of soil water content, and strongly related processes such as evapotranspiration, in climate and atmospheric processes (Koster & Suarez 2001; Ek & Holtslag, 2004; van den Hurk et al., 2008; Seneviratne et al., 2010; Groh et al., 2020). In this context, we require reliable estimates of soil hydraulic properties at point to global scale (Cornelis et al., 2001, van Looy et al., 2017). Measuring these properties is tedious, time and cost expensive, and prone to measurement errors. Often, taking measurements is not feasible due to the complexity and/or size of the terrestrial system under investigation. To overcome this problem, pedotransfer functions (PTFs), which estimate these essential soil properties from easily available soil parameters, such as soil texture, soil structure, bulk density, and soil organic matter, have been developed. An extensive overview of existing PTFs was provided by Vereecken et al. (2010) and by van Looy et al. (2017).
Soil properties used as basic input data to estimate the soil hydraulic properties with PTFs are grouped into four categories: (1) soil particle size or soil texture, (2) easily measurable hydraulic properties, (3) morphological properties, and (4) chemical properties (Espino et al., 1995, Vereecken et al. 2010; van Looy et al. 2017; Rahmati et al., 2013; Neyshabouri et al., 2015; Rahmati and Neyshabouri, 2016). In general, two different types of PTF can be distinguished, namely point and parametric PTFs. Point PTFs estimate soil water content (or hydraulic conductivity) values at predefined pressure head values (e.g., field capacity or wilting point), whereas parametric PTFs provide the parameters than can be used in hydraulic functions (water retention curve (WRC) and hydraulic conductivity curve (HCC)) of the Brooks & Corey (1964) or van Genuchten (1980) formulation. The most useful PTFs developed in recent years are the parametric PTFs because they can be used to calculate the WRC and HCC, which are used to simulate the water fluxes in the numerical models. Secondly, PTFs are classified into class and continuous PTFs. Class PTFs are look up tables, where the hydraulic parameters are listed for typical soil textural classes (e.g., 12 USDA soil classes). Continuous PTFs, on the other hand, use mathematical descriptions, e.g., regression functions, to calculate the hydraulic parameters from the entire range of data inputs like e.g. soil texture.
It has been shown that PTFs are highly accurate for the area (or the input data range) they were developed for, but have limited accuracy if applied outside these regions (Vereecken et al., 2010). Several reviews about the accuracy and reliability of PTFs for the van Genuchten model (VGM) have already been published (e.g., Wösten et al., 2001; Schaap, 2004; Donatelli et al., 2004). Hereby, the predicted hydraulic function from the PTFs were compared to the measured data and the goodness of fit of the prediction was evaluated. The authors used two metrics to determine the performance of the PTF: 1) the term accuracy was related to the comparison between predicted and measured values of water content or hydraulic conductivity that were used to develop the PTF; 2) reliability was related to the evaluation of PTFs on measured values that were different from those that were used to develop the PTFs (Wösten et al., 2001, Zhang et al., 2020). Reliability studies are typically validation studies such as those performed by Tietje & Tapkenhinrichs (1993), Wösten et al. (2001), and Wagner et al. (2001). Despite much progress in developing PTFs and in identifying appropriate PTF predictor candidates, some unresolved or unexplained variability still exists at the level of the soil sample (Schaap & Leij, 1998), which plays and important role when functional aspects of soils are being studied and analysed using numerical models (e.g., Christiaens & Feyen, 2001). Functional aspects already studied are the impact of PTFs on water supply capacity (Vereecken et al., 1992), ground water recharge (Vereecken et al., 1992), and aeration (Wösten et al., 2001). In the study of Vereecken et al. (1992), the authors showed that 90% of the variation in the predicted soil water supply was attributed to estimation errors in hydraulic properties using the PTFs developed by Vereecken et al. (1989, 1990). Chirico et al. (2010), on the other hand, evaluated the effect of PTF prediction uncertainty on the components of the soil water balance at the hillslope scale. One major result was that the simulated evaporation was much more affected by the PTF model error than by errors resulting from uncertainty in the input data (e.g., soil texture).
Land surface models (LSMs), when embedded in numerical weather prediction or climate models, generally operate at large scales (regional, continental to global scales) and rely on PTFs to predict the hydraulic functions needed to solve the Richards equation for the water flow. Different LSMs use different PTFs for this purpose (Vereecken et al., 2019). As Vereecken et al. (2019) showed, not only different PTFs but also different hydraulic models (Campbell, Brooks and Corey, or Mualem-van Genuchten) are in use. Knowing that different PTFs and/or the choice of the hydraulic model will impact the outcome of the water flow simulations (e.g., Gruber et al., 2006; Yakirevich et al., 2013), a key question is how recently launched LSM inter-comparison activities of the Land Surface Schemes (LSSs) embedded in LSMs, such as those under the Global Energy and Water Exchanges (GEWEX) GLASS project (https://www.gewex.org/panels/global-landatmosphere-system-study-panel/) or model inter-comparisons initiated by the World Climate Research Programme (https://www.wcrp-climate.org/wgcm-cmip/wgcm-cmip6, of which GEWEX is part), such as CMIP6 and its predecessors, will be impacted by the choice of PTF.
Therefore, the aim of this study is to systematically analyse the functional sensitivity to the choice of different PTFs using a physically-based numerical model. As the ‘truth’ of this model exercise is unknown, the performance of the model runs with its hydraulic parameters derived from a set of individual PTFs will be evaluated against the ensemble mean as best predictor, as well as against the 70 and 90 tolerance intervals of the ensemble range. The numerical exercise is structured in the following way: 1) model runs for homogeneous soil profiles without vegetation, 2) homogeneous soil profile covered with grass and wheat, 3) layered bare soil, 4) layered vegetated soil (grass and wheat), and 5) influence of a fluctuating water table in a layered grass vegetated soil. Finally, additional soil physical properties were calculated based on the estimated soil hydraulic parameters obtain from the PTFs, which were used to explain the differences observed in simulated water fluxes. As some LSMs also use class PTFs (van Looy et al., 2017, Vereecken et al., 2019), we will also analyse the use of this type of PTF, and the associated errors when simulating water fluxes. We formulate three hypotheses 1) the use of different PTFs will lead to systematically different hydrological states and fluxes (e.g., net infiltration, evapotranspiration, root zone water availability, drainage), 2) some PTFs can be identified which perform distinctively differently from the ensemble spread in terms of 90 % tolerance interval outliers, and 3) the differences in predicted states and fluxes simulated with inputs from different PTFs will be reduced with increasing model setup complexity.
Materials and Methods