N. Robinson

Modelling and prediction of soil water contents at field capacity and permanent wilting point of dryland cropping soils

Field capacity (FC) and permanent wilting point (PWP) are two critical input parameters required in various biophysical models. There are limited published data on FC and PWP of dryland cropping soils across north-western Victoria. Direct measurements of FC and PWP are time-consuming and expensive. Reliable prediction of FC and PWP from their functional relationships with routinely measured soil properties can help to circumvent these constraints. This study provided measured data on FC using undisturbed samples and PWP as functions of geomorphological unit, soil type, and soil texture class for dryland cropping soils of north-western Victoria. We used a balanced, nested sampling strategy and developed functional relationships of FC and PWP with routinely measured soil properties using residual maximum likelihood based mixed-effects regression modelling. Using the data, we also tested the adequacy of nine published pedotransfer functions (PTFs) in predicting FC and PWP. Significant differences were observed among the three soil types and nine texture classes for most soil properties. FC and PWP were higher for Grey Vertosols (FC 43.7% vol, PWP 29.1% vol) than Hypercalcic Calcarosols (38.4%, 23.5%) and Red Sodosols (20.2%, 9.2%). Of the several functional relationships developed for prediction of FC and PWP, a quadratic single-predictor model based on dg (geometric mean particle size diameter) performed better than other models for both FC and PWP. It was nearly bias-free, with a root mean square error (RMSE) of 3.18% vol and an R 2 of 93% for FC, and RMSE 3.47% vol and R 2 89% for PWP. Another useful model for FC was a slightly biased, two-predictor quadratic model based on clay and silt, with RMSE 3.14% vol and R 2 94%. For PWP, two other possibly useful, though slightly biased, models included a single-predictor quadratic model based on clay (RMSE 3.45% vol, R 2 89%) and a three- predictor model based on clay, silt, and sg (geometric standard deviation of particle size diameter) (RMSE 3.27% vol, R 2 90%). We observed a strong quadratic relationship of FC with PWP (RMSE 1.61% vol, R 2 98%). This suggests the possibility to further improve the prediction of FC indirectly through PWP. These predictive models for FC and PWP, though developed for the dryland cropping soils of north-western Victoria, may be applicable to other regions with similar soil and climatic conditions. Some validation is desirable before these models are confidently applied in a new situation. Of the nine published PTFs, the multiple linear regression and artificial neural network based NTh5 for FC and NTh3 and CAM for PWP performed better on our data for the prediction of FC and PWP. The root mean square deviation of these PTFs, for both FC and PWP, was higher than the RMSE of our models. Our models are therefore likely to perform better under the dryland cropping soils of north-western Victoria than these PTFs. As a safeguard against arriving at optimistic inferences, we suggest that the modelling of functional relationships needs to account for the hierarchical structure of the sampling design using appropriate mixed effects regression models.

Advances in precision agriculture in south-eastern Australia. IV. Spatial variability in plant-available water capacity of soil and its relationship with yield in site-specific management zones

Spatial variability in grain yield can arise from variation in many different soil and terrain properties. Identification of important sources of variation that bear significant relationship with grain yield can help achieve more effective site-specific management. This study had three aims: (i) a geostatistical description/modelling of the paddock-level spatial structure in variability of plant-available water capacity (PAWC) and related soil properties, (ii) to determine optimal number of management zones in the paddock, and (iii) to assess if the variability in PAWC and related soil properties is significantly associated with the variability in grain yield across the management zones. Particle size distribution, bulk density (BD), field capacity (FC), permanent wilting point (PWP), and soil water content (SWC) at sowing were measured at 4 soil depths (to 0.60 m) at 50 representative spatial sampling locations across a paddock near Birchip (Victoria). PAWC and plant-available water at sowing (PAWs) were derived from these data. Moderate to strong spatial dependence across the paddock was observed. The magnitude of the structural variation and of range varied widely across different soil properties and depths. The south-east edge and the central areas of the paddock had higher clay content, FC, PWP, PAWC, and lower PAWs. The paddock was divided into 6 potential management zones using combined header yield and normalised difference vegetation index (NDVI). The adequacy of zoning was evaluated using relative variability (RV) of header yield and soil properties. The mean RV for 3 zones differed little from that of 6 management zones for header yield and for each measured soil property, indicating division of the paddock into 3 zones to be adequate. The results from residual maximum likelihood (ReML) analysis showed that low yield zones had significantly higher clay content, FC, PWP, SWC, and PAWC and significantly lower PAWs than both medium and high yield zones. The mean FC, PWP, and PAWC in the low yield zones were, respectively, 25%, 26%, and 28% higher, and PAWs 36% lower than their corresponding values in the high yield zones. Linear regression analysis indicated that 59–96% of the observed variation in grain yield across management zones could be explained by variation in PWP. The practical implications of these results are discussed.

Advances in precision agriculture in south-eastern Australia. II. Spatio-temporal prediction of crop yield using terrain derivatives and proximally sensed data

The effects of seasonal as well as spatial variability in yield maps for precision farming are poorly understood, and as a consequence may lead to low predictability of future crop yield. The potential to utilise terrain derivatives and proximally sensed datasets to improve this situation was explored. Yield data for four seasons between 1996 and 2005, proximal datasets including EM38, EM31, and γ-ray spectra for 2003–06, were collected from a site near Birchip. Elevation data were obtained from a Differential Global Positioning System and terrain derivatives were formulated. Yield zones developed from grain yield data and yield biomass estimations were included in this analysis. Statistical analysis methods, including spatial regression modelling, discriminant analysis via canonical variates analysis, and Bayesian spatial modelling, were undertaken to examine predictive capabilities of these datasets. Modelling of proximal data in association with crop yield found that EM38h, EM38v, and γ-ray total count were significantly correlated with yield for all seasons, while the terrain derivatives, relative elevation, slope, and elevation, were associated with yield for one season (1996, 1998, or 2005) only. Terrain derivatives, aspect, and profile and planimetric curvature were not associated with yield. Modest predictions of crop yield were established using these variables for the 1996 yield, while poor predictions were established in modelling yield zones.

Identification and interpretation of sources of uncertainty in soils change in a global systems-based modelling process

In the past, uncertainty analysis in soil research was often reduced to consideration of statistical variation in numerical data relating to model parameters, model inputs or field measurements. The simplified conceptual approach used by modellers in calibration studies can be misleading, because it relates mainly to error minimisation in regression analysis and is reductionist in nature. In this study, a large number of added uncertainties are identified in a more comprehensive attention to the problem. Uncertainties in soil analysis include errors in geometry, position and polygon attributes. The impacts of multiple error sources are described, including covariate error, model error and laboratory analytical error. In particular, the distinction is made between statistical variability (aleatory uncertainty) and lack of information (epistemic uncertainty). Examples of experimental uncertainty analysis are provided and discussed, including reference to error disaggregation and geostatistics, and a systems-based analytic framework is proposed. It is concluded that a more comprehensive and global approach to uncertainty analysis is needed, especially in the context of developing a future soils modelling process for incorporation of all known sources of uncertainty.

100 Years of superphosphate addition to pasture in an acid soil - Current nutrient status and future management

Pasture-based animal production systems, which occupy a significant proportion of the landscape in Victoria, Australia, have historically been nutrient-limited, with phosphorus (P) often the most limiting nutrient. The Permanent Top-Dressed (PTD) pasture experiment was established in 1914 at the Rutherglen Research Station, Victoria, to investigate the management of this deficiency. The main objective of the PTD experiment was to demonstrate the value of adding P fertiliser at two rates to increase pasture productivity for lamb and wool production. We report on the status of the PTD soils after 100 years, investigating the long-term implications of continuous grazing and fertiliser management (0, 125 and 250 kg/ha of superphosphate every second year) of non-disturbed pasture. We investigated the long-term effects of P fertiliser on the forms and distribution of P and other relevant soil parameters. In the fertilised treatments, P has accumulated in the surface soils (0–10 cm) as both orthophosphate and organic P, with an Olsen P of 16–21 mg P/kg, which is non-limiting for pasture production. In the treatment with 250 kg superphosphate, there has also been movement of P down through the soil profile, probably due to the high sand content of the surface soil and the transfer through the profile of small quantities of water-soluble P and P bound to organic ligands. Over time, the site has continued to acidify (surface 0–10 cm); the soil acidity combined with aluminium (Al) concentrations in the fertilised treatments approach a level that should impact on production and where broadcast lime would be recommended. After 100 years of non-disturbed pasture, the surface soils of these systems would be in a state of quasi-equilibrium, in which the fertilised systems have high levels of carbon (C), nitrogen, P and exchangeable Al. The continued stability of this system is likely dependent upon maintaining the high C status, which is important to nutrient cycling and the prevention of Al phytotoxicity. There are two risks to this system: (i) the declining pH; and (ii) soil disturbance, which may disrupt the equilibrium of these soils and the bio-chemical processes that maintain it.

Improving the information content in soil pH maps: a case study

Summary Uncertainties associated with legacy data contribute to the spatial uncertainty of predictions for soil properties such as pH. Examples of potential sources of error in maps of soil pH include temporal variation and changes in land use over time. Prediction of soil pH can be improved with a linear mixed model (LMM) to analyse factors that contribute to uncertainty. Probabilities from conditional simulations in combination with agronomic critical thresholds for acid-sensitive species can be used to identify areas that are likely, or very likely, to be below these critical thresholds for plant production. Because of rapid changes in farming systems and management practices, there is a need to be vigilant in monitoring changes in soil acidification. This is because soil acidification is an important factor in primary production and soil sustainability. In this research, legacy data from south-western Victoria (Australia) were used with model-based geostatistics to produce a map of soil pH that accommodates a variety of error sources, such as the time of sampling, seasonal variation, differences in analytical method, effects of changes in land use and variable soil sample depth in legacy data. Spatial covariates that are representative of soil-forming factors were used to improve predictions. To transform spatial prediction and estimates of error in soil pH into more informative and usable maps with more information content, simulations from the conditional distribution were used to compute the probability of a soils pH being less than critical agronomic production thresholds at each of the prediction locations. These probabilities were mapped to reveal areas of potential risk. Highlights Can maps of soil pH be improved by accounting for temporal variation and change in land use? First example of taking account of temporal variability in sampling for pH in spatial models. Key factors for uncertainty in spatial prediction include time of sampling and sample depth. Accuracy improved by accounting for additional sources of error combined with conditional simulations.

Quantification of Uncertainty in Mathematical Models: The Statistical Relationship between Field and Laboratory pH Measurements

The measurement of soil pH using a field portable test kit represents a fast and inexpensive method to assess pH. Field based pH methods have been used extensively for agricultural advisory services and soil survey and now for citizen soil science projects. In the absence of laboratory measurements, there is a practical need to model the laboratory pH as a function of the field pH to increase the density of data for soil research studies and Digital Soil Mapping. The accuracy and uncertainty in pH field measurements were investigated for soil samples from regional Victoria in Australia using both linear and sigmoidal models. For samples in water and CaCl2 at 1 : 5 dilutions, sigmoidal models provided improved accuracy over the full range of field pH values in comparison to linear models (i.e., pH 9). The uncertainty in the field results was quantified by the 95% confidence interval (CI) and 95% prediction interval (PI) for the models, with 95% CI < 0.25 pH units and 95% PI = pH units, respectively. It was found that the Pearson criterion for robust regression analysis can be considered as an alternative to the orthodox least-squares modelling approach because it is more effective in addressing outliers in legacy data.

Error propagation in computer models: analytic approaches, advantages, disadvantages and constraints

Uncertainty and its propagation in computer models has relevance in many disciplines, including hydrology, environmental engineering, ecology and climate change. Error propagation in a model results in uncertainty in prediction due to uncertainties in model inputs and parameters. Common methods for quantifying error propagation are reviewed, namely Differential Error Analysis and Monte Carlo Simulation, including underlying principles, together with a discussion on their differences, advantages and disadvantages. The separate case of uncertainty in the model calibration process is different to error propagation in a fixed model in that it is associated with a dynamic process of iterative parameter adjustment, and is compared in the context of non-linear regression and Bayesian approaches, such as Markov Chain Monte Carlo Simulation. Error propagation is investigated for a soil model representing the organic carbon depth profile and also a streamflow model using probabilistic simulation. Different sources of error are compared, including uncertainty in inputs, parameters and geometry. The results provided insights into error propagation and its computation in systems and models in general.

Assessment of Error Sources in Measurements of Field pH: Effect of Operator Experience, Test Kit Differences, and Time-Of-Day

ABSTRACT Various methods exist to measure soil pH, and while there is general agreement between the existing published laboratory and field-based methods, the latter are subject to uncertainties including test kit reliability, accuracy, precision, and environmental factors. The contribution of this study is to quantify three uncertainties that affect the conversion between field pH and laboratory pH measurements, namely operator experience, choice of test kit, and the time-of-day for measurement. Soil samples from western Victoria, representing the pH range 4.5–10.0, were used in a randomized complete block design with 10 assessors split into two groups representing experienced and inexperienced users. Statistical analysis of laboratory and field pH was based on using the Maximum Likelihood Functional Relationship (MLFR) to determine if there was any bias between the two methods. Significant differences were found between experienced and inexperienced users, and between test kits.

Land use change: understanding and managing soil dynamics.

Lessons of the past show that care for the soil is fundamental to the rise and sustainability of agrarian civilisations. The responsibility for this care devolves to individual farmers, land managers and investors in agricultural production, all of whom are, by default, soil custodians. Soil condition is affected by land use practices; understanding the dynamics of soil and land use interaction is therefore critical in achieving sustainable soil management and the maintenance of soil health. Understanding and managing this dynamic requires good data and sound knowledge of farming systems and their interaction with soil properties and processes. The North Central Catchment Management Authority (NCCMA) region has soil and land use data at a range of scales that can assist in understanding the region’s soil assets with respect to current and future farming systems. Programs that fully engage farmers in planning for soil health, and provide appropriate tools and information, will be a cornerstone for managing soil dynamics under pressure from climate and land use change. This paper explains the contexts in which soil spatial information and land use data are collected, provides examples for the NCCMA region, and briefly describes the challenge of providing soil information at the farm scale.