Unravelling the effects of soil properties on water infiltration: segmented quantile regression on a large data set from arid south-west Africa

Relationships were sought between infiltrability and the properties of hundreds of surface soils (pedoderms) sampled across Namibia and western South Africa. Infiltrability was determined using a laboratory method, calibrated against a rainfall simulator, which measures the passage of a suspension of soil in distilled water through a small column packed with the same soil. Other properties determined were EC, pH, water-soluble cations and anions, ammonium acetate-extractable cations, organic C, total N, a 7-fraction particle size distribution, water-dispersible silt and clay, and clay mineral composition. Our objective was to ascertain whether general principles pertaining to infiltrability can be deduced from an analysis of a wide diversity of soils. To achieve this we compared correlation analysis, generalised linear models (GLMs), and generalised additive models (GAMs) with a segmented quantile regression approach, in which parametric regression lines were fitted to the 0.9 and 0.1 quantile values of equal subpopulations based on the x variable. Quantile regression demarcated relational envelopes enclosing four-fifths of the data points. The envelopes revealed ranges for soil properties over which infiltrability is potentially maximal ( spread over a wide range) or predictably minimal (confined to small values). The r(2) value of the 0.9 quantile regression line was taken as an index of reliability in being able to predict limiting effects on infiltrability associated with a variety of soil properties. Prediction of infiltration was most certain from textural properties, especially the content of water-dispersible silt (r(2) = 0.96, n = 581), water-dispersible clay (0.88, n = 581), very fine sand (0.86, n= 174), and medium sand ( 0.84, n = 174). Chemical properties such as EC, sodium status, organic C content, and clay mineralogy were less clearly related to infiltrability than was texture. The role of fine-particle dispersion in blocking pores was highlighted by the stronger prediction in all statistical analyses provided by the water-dispersible as opposed to total content of silt and clay. All the statistical analyses revealed a probable skeletal role of medium and. ne sand fractions in shaping pores and a plasmic ( mobile) role of finer fractions in blocking pores. A noteworthy discovery was an apparent switch in role from skeletal to plasmic at a particle diameter of about 0.1 mm (i.e. between. ne and very. ne sand).