The main objective of the study at hand is to improve the parameter estimation for meso-scale hydrological models in order to realistically reproduce the soil water interaction. Soil water movement and storage can be particularly sensitive to many hydrological variables, and therefore affect various simulation outputs, like runoff, infiltration, groundwater recharge, and actual evapotranspiration. Especially, the water content in the soil near the surface has a decisive influence on runoff generation [3
]. Soil water simulations on the catchment scale require parameters, which describe the physical characteristics of the soil. Hence, the soil hydraulic parameterisation is an essential aspect in any hydrological model application [6
]. At the hydrological meso- or macroscale, information regarding soil properties is mostly only available on very coarse spatial resolutions, e.g., soil maps with scale of 1:50000. Commonly, such maps provide simplified soil characterisations that are based on their texture [7
]. Soil texture is defined as the particle size composition expressed as relative fractions of sand, silt, and clay. The advantage of this classification is that the measurability of the particle sizes is easy and affordable. Furthermore, a regionalisation of these (point) measurements is feasible. The most established classification is the system of the United States Department of Agriculture (USDA), which initially was elaborated by [8
]. The general assumption of any soil classification is that the “soils” within each class are as similar as possible to each other. The similarity has to be objectively assessed, e.g., as it is done by texture in the USDA system. This classification is reasonable regarding particle size. The pioneering studies of [9
] and [10
] questioned the physical reliability of this system concerning soil hydraulic parameters. [9
] investigated, whether a classification that was based on hydraulic characteristics—and not texture as single criterion—led to different patterns within the texture triangle. They selected properties, such as field capacity, wilting point, porosity, saturated hydraulic conductivity, drainability, and capillary pressure at field capacity. The classification results were different to the USDA classification for soils with low sand content. [9
] argued that, from a soil hydraulic point of view, a classification that is based on texture does not work well for soils with a considerable high impact from capillary forces. However, the question of parameterisation of large scale hydrological soil models in a way that considers hydraulic behaviour still remains. [10
] applied a physically based soil-atmosphere-transfer model [11
] with more than 5000 realisations of synthetic soil hydraulic parameterisations, according to the Brooks and Corey function [12
]. The simulation results (water balance components), as well as the associated soil hydraulic parameters, were used for cluster analyses. The results showed similarities between clusters that were based on soil hydraulic parameters and texture based classification, but different patterns for classification based on simulated water balance. [10
] argued that soil texture classification with regard to hydrological aspects could improve the applicability of classified soil data on large scales. [13
] applied the HYDRUS one-dimensional (1D) model [14
], with 1326 realisations of synthetic parameterisations according to the van Genuchten function [15
]. Different kinds of cluster analyses were conducted and various variants of the model were used. One cluster analyse was based on saturated hydraulic conductivity. The others were based on the comparison of water storage prior to and after the simulation. The first model variant only accounted for infiltration, the second variant for drainage, and the third one for both. The cluster patterns differed from USDA texture based pattern, in addition, they also differed from each other.
The above-mentioned studies indicate that a texture based classification of soils may not be the optimal solution with respect to hydrological modelling. The alternative, which could be a classification that is based on soil hydraulic parameters or soil water model results, on the other hand, has not been sufficiently addressed so far and the water balance or the change of water storage is a very general input variable for the clustering.
In the study at hand, simulated time series of soil water flux in high temporal resolutions are utilised as the input in the creation of classes, based on soil hydraulic behaviour, since the flux controls the subsurface discharge components as well as the groundwater recharge. Furthermore, the considered number of samples for the hydraulic parameterisation is much larger than in previous research. We applied more than one-million simulations that are based on Richards equation [16
] for two different atmospheric boundary conditions. The k-means clustering [17
] was applied as a classification algorithm with the water flux as the classification variable. The time series of water flux representative of the new classes were compared to time series representative for USDA texture classification.