Prediction of soil Olsen P through mixed pasture leaf tissue biochemical and biophysical properties, topography and farm management in New Zealand hill country : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Agricultural Science at Massey University, Turitea, New Zealand

Thumbnail Image
Open Access Location
Journal Title
Journal ISSN
Volume Title
Massey University
The Author
In New Zealand hill country, soil Olsen Phosphorus (P) is a key piece of information used to decide rates, products and areas to which aerial applications of phosphate fertilisers are made. Laboratory based soil Olsen P measurements are made on bulked soil cores collected along transects, laid out across slope faces. On most hill country farms, aerial fertiliser applications are applied uniformly over large blocks or the whole property. Accurate, detailed soil maps are scarce, but essential for site specific nutrient management. Current soil sampling techniques provide spatially sparse information and attempts to interpolate point measurements of soil properties in hill country, have not been successful. The potential improvement in nutrient use efficiency would lead to increases in pasture production and quality, and an increase in production of meat and wool produced off the same land area. As sheep and beef production is forced from more productive land into more marginal areas by other land uses, managing hill country landscapes efficiently will become critical for the sheep and beef industry. Increases in global food demand, a growing interest in product origins and production practices by the consumer, and tightening of environmental regulations will further put pressure on these systems. Appropriate soil and fertiliser management has suffered from a lack of information to make sound decisions. Maps of soil Olsen P are a first step, with much potential in the applications of hyperspectral imaging yet to be discovered. The objective of this thesis was to develop a model that could be applied to readily available data layers, to make continuous predictions of phosphate availability in the soil (Olsen P) across New Zealand hill country farms. This research was one part of a larger project that firstly aimed to derive estimates of pasture parameters from hyperspectral imagery. This information could then be used in conjunction with ancillary data to determine soil nutrient status. Finally, this information would be used to inform variable rate fertiliser applications through a prescription map loaded into a computer controlled aerial top dressing system. A multi-site, multi-seasonal database from eight commercial hill country farms incorporating a range of leaf tissue nutrient concentrations, pasture biophysical properties, and topographic, soil and farm management information was built up alongside soil chemical properties. Model development was based on in-situ measurements and laboratory analysis of leaf tissue and soil samples collected on 0.5m x 0.5m plots. A total of 3,030 plots were sampled in the autumn and spring. Simple plant P indices are usually used on single species samples of actively growing tissue. Here they were used on mixed pasture samples at various stages of maturity with mixed success. Although overall correlations were weak, leaf tissue P concentration and PNI were more strongly correlated to soil Olsen P than the P:N ratio or PNIc. Soil Olsen P was more strongly correlated to leaf tissue P concentration and PNI in spring (R² = 0.21 and 0.24 respectively) than in autumn (R² = 0.12 and 0.12) and both seasons combined (R² = 0.13 and 0.13). For individual sampling events, all P indices were generally more strongly correlated to soil Olsen P in spring than in autumn. Of the individual sampling events, the strongest correlation was at the hill country farm Cleardale in the spring (R² = 0.56) using leaf tissue P concentration. The database was then used in exploratory analysis to identify important input variables across farms and seasons through stepwise multiple linear regression. Leaf tissue P and copper concentration, slope, fertiliser history, the proportion of green tissue in the sample and seasonal information were consistently selected. For all seasons and farms combined, 13 variables (slope, 30 day rainfall, soil moisture deficit, time since the last fertiliser application, rate of the last fertiliser application, leaf tissue P, Cu, Na, Mn and Zn concentrations, DM%, the dead vegetation fraction and legume content) were selected as predictors for soil Olsen P with an R² of 0.42 for the mean of the 5 plots at each site. For season specific models, different sets of predictors were selected and achieved higher levels of explanation, R² of 0.45 in autumn and R² of 0.50 in spring for the mean of the 5 plots at each site. The approach taken to predict soil Olsen P was to develop Bayesian hierarchical multiple linear regression models. Models were developed using different parameters and hierarchical structures. The pasture biochemical and biophysical parameters along with topographic and farm management factors all contributed significantly to the model. The model fit was significantly increased and the residual error greatly reduced in the combined model compared to models containing only plant biochemical and biophysical or only physical inputs. The posterior predictive distribution from a Bayesian hierarchical multiple linear regression model provided an estimate of soil Olsen P and the uncertainty of the prediction made. A leave-one-out-cross-validation showed an improvement compared to an average value used to inform the most basic and risk averse uniform fertiliser application as a benchmark. The Bayesian hierarchical model can be used for predictive soil mapping, which was demonstrated, however still needs to be validated. Predictive soil mapping exhibited the potential of using input data layers derived from hyperspectral imagery and digital elevation models, to provide continuous predictions of soil Olsen P across a hill country farm. Maps of soil Olsen P were produced, where methods attempting to use interpolation techniques in hill country have been unsuccessful. These maps provide vast amounts of data compared to traditional spatially sparse soil sampling. Soil phosphate availability has a significant effect on pasture productivity, and species composition which affects pasture quality. As considerable research and programmes have focused on genetics and breeding for animal performance, it is now thought that pasture productivity and management is restraining the potential performance of sheep and beef systems. Of the factors driving pasture productivity, fertiliser applications are one factor farmers have control over. The levels of P available in soils observed in this study suggest that much potential exists to increase pasture production and quality through increasing P availability in the soil. In summary, a Bayesian hierarchical linear regression model significantly improved the predictions of soil Olsen P made across hill country farms compared to a benchmark traditional uniform approach. From this Bayesian hierarchical model, estimates of soil Olsen P can be made on locations and time points outside of the dataset with a known level of uncertainty. This model can be used in predictive soil mapping to produce maps of soil Olsen P across hill country farms.
Soils, Phosphorus content, New Zealand, Mathematical models, Composition