Fitting and comparing water retention curves for soils under contrasting experimental treatment: Examples from conservation agriculture trials in southern Africa

Conservation Agriculture (CA) is proposed as a ‘climate-smart’ intervention for resilient crop production in dryland areas affected by climate change. Evidence is needed for how these practices affect fundamental properties of the soil. The soil water retention curve (SWRC) is a physical attribute of the soil which provides information on its porous structure and physical quality. It is also critical for modelling processes in the soil such as water movement, water availability for plants and infiltration into the soil during rainfall events. In this paper we estimate parameters of the van Genuchten model of the SWRC from experiments on CA interventions in southern Africa, using a linear mixed modelling framework. The method we use, stochastic approximation maximization, allows for maximum likelihood estimation of the parameters without use of linearizing approximation. We show how sequential fitting of model parameters, with marginal false discovery rate control, allows us to make robust inferences about differences in the SWRC between soils under contrasting experimental management. We also show how the method allows us to draw samples from distribution of SWRC parameters, reflecting the uncertainty which arises from variation within the management treatments. Indices of soil physical quality may be computed from the parameter estimates to compare treatments, and by computing them from the samples, the uncertainty in these indices can also be assessed. We use the estimated model parameters to simulate infiltration of water into the soils under different management during a rainfall event. Again, by using the samples from the joint distributions of the parameters the effects of uncertainty in these parameters as propagated through the model can be computed. We applied these methods to soils collected from experimental plots under CA and conventional tillage (CV) at sites in Zimbabwe, Zambia and Malawi. We observed differences in the SWRC for the CA and CV plots at the Zambian site where a physically vulnerable soil showed greater macroporosity under CA than CV. In contrast, a sandy and organic-poor soil at the site in Zimbabwe showed somewhat greater macroporosity under cultivation rather than CA management. There was no detectable treatment effect of the management system on the SWRC for the soils at the site in Malawi.