Relationships between soil water retention and soil composition

Relationships between soil water retention and physical composition have been examined in this study, using a database of two independent sets of soils from two regions in New Zealand: Canterbury in the South Island (Set 1) and Waikato in the North Island (Set 2). Five soils (Wakanui, Templeton, Temuka, Timpendean and Cookson) were selected from the Canterbury region, the first three being representative of the dominant texture types (i.e. silt and sandy loams) in the Canterbury Plains, and the Cookson and Timpendean soils being chosen to widen the range of textures to include soils higher in clay. Also, to investigate the possible effects of cultivation history, which affects both structure and organic matter level, four Wakanui soil sites representing four different management treatments, ranging from continuous arable to permanent grassland, were included. The second set, from the Waikato region, was selected from results of a sampling programme completed by staff of the NZ Soil Bureau (Joe and Watt, 1986). The soils from this set (Horotiu, Te Kowhai, Hamilton, Otorohanga, and Netherton) cover a wide range of textures, developed from different parent materials, and in different physiographic positions and climates. Water content measurements were made on undisturbed samples at suctions ranging from 0.98 kPa to 1500 kPa, using tension tables and pressure plate apparatus. For the Set 1 (Canterbury) soils, for which all measurements were completed by the author, particular attention was directed to the accurate measurements of composition data. Detailed particle-size analysis was carried out using a combination of sieving and sedimentation techniques. For the latter, both Sedigraph and pipette measurements were made and compared. The Sedigraph method was found to systematically overestimate the mass percentage at all equivalent diameters in the sedimentation range, compared to the standard pipette method. The results of this comparison were found to be in remarkably close agreement with results of a similar independent comparison undertaken previously by the New Zealand Soil Bureau. Measurements on soils in the Canterbury set showed, for all soils except one, a highly significant correlation between the total iron content of the sample and the difference between the Sedigraph and pipette methods. This strongly supported the hypothesis that the greater concentration of iron (a strong X-ray absorber) in the smaller size fractions (particularly clay) is the main factor causing the difference. Regression equations are also developed for converting Sedigraph data to their pipette equivalents. Organic carbon, a second key compositional parameter, was determined using three different methods: (i) Walkley and Black (1934); (ii) Dumas Combustion (the reference method); and (iii) Loss on ignition. The results of this study revealed a novel result for converting loss on ignition (L) to total organic carbon (T). In contrast to the traditional assumption a ≏ 0 and b ≏ 1.72 in the linear regression L = a + b(T), the results showed a significant intercept effect (i.e. a > 0), related to residual effects of adsorbed moisture and other components driven off on ignition. Inclusion of the clay term in the bivariate regression, i.e. L = a + b(T) + c(C), is shown to significantly improve the regression and accounts for the non-zero intercept effect. This study also confirmed that the Walkley and Black method can be used for reliable determination of organic carbon of these soils. Particular emphasis is placed in this thesis on the improved functional representation of particle-size distribution and the soil moisture characteristic. The modelling of particle-size distribution is a relatively neglected area of soil science research, so five different parametric models are evaluated. Three of these have not previously been introduced in the soil science literature: a simple one-parameter model borrowed from the geotechnics literature; and two modified lognormal models. One of the latter two models, the 'offset-renormalized lognormal' (ORL) model, is found to provide the optimum description of PSD for the majority of the soils. The general statistical approach to selection of an 'optimum' model is discussed. Both these topics, i.e. exploration of functional models for PSD, and a systematic approach to model selection, have been given little attention in the soil science literature. For representation of the soil moisture characteristic, the applicability of a simple two-parameter power-function model was tested. This model was found to provide a good fit to data over suction ranges important for key unsaturated processes in the field. A single-parameter version of this model, proposed by Gregson et al. (1987), based on an alleged correlation between the two parameters, is shown to be a result of a mathematical artefact (Buchan and Grewal, 1990). This study concludes that a minimum of two parameters are required to model the unsaturated portion of the soil moisture characteristic, while a third, θs, is required to define its saturation limit. When a proper scaling basis is selected for θv and ψ, the two parameters are found to be essentially independent. The real reason enabling an approximate one-parameter approach is also revealed. The relationship between power-function model parameters and composition parameters is then explored. The results of the study indicate that clay content of the soil is the most important composition parameter and is highly correlated with the exponent b (the 'pore-size distribution index') of the power-function model. This variable alone accounts for 66% of the variation in the b parameter. Correlation between the a-parameter and composition parameters is weak, presumably reflecting the relatively narrow range of values of the a parameter. Thus it is concluded that 'a' (=ln ψe, where ψe is a notional air-entry potential) cannot be reliably estimated from composition data. The best practical estimation for a is to assume that it is a constant, given by its mean value for a soil set (a = -1.0 in this study). However this approximation results in large (>±3%) differences between observed and estimated θv values for more than half of the soils. The possibility of rapid estimation of field capacity (FC) and wilting point (WP) water contents from the readily-measured saturation percentage (SP) of the soil, was also assessed in this study. A strong linear relationship between the mass percentage water contents at FC and WP, and the SP, indicates that SP can be used to provide a rapid estimate of these limits to plant-available water. However, the lack of correlation found between SP and available water capacity (AWC) indicates the inability of the SP method to provide good estimates of AWC in these soils