Predicting and mapping the soil available water capacity of Australian wheatbelt
Introduction
Soil available water capacity (AWC) is defined as the amount of water soil can store between field capacity or drainage upper limit (DUL) and wilting point or crop lower limit (CLL). It is the main source of water for vegetation development and is related to the potential amount of water a soil could make available for the atmosphere through evapotranspiration (Dunne and Willmott, 1996). Information about its distribution in space is crucial for planning and management in agriculture, and for ecological modelling.
To model the spatial distribution of AWC, digital soil mapping has been proposed (McBratney et al., 2003). The scorpan model describes that soil properties can be predicted from its predicting factors in the form of empirical regression equations. The general steps in the modelling process involve: collection of a dataset of soil observations over the chosen area of interest; compilation of relevant covariates for the area; calibration or training of a spatial prediction function based on the observed dataset; interpolation and/or extrapolation of the prediction function over the whole area of interest; calculation of uncertainty; and finally validation using existing or independent datasets.
Despite the importance of AWC, not many studies present a mapping methodology at national scale. Hong et al. (2013) successfully predicted AWC for Korea based on detailed soil series maps and modal profiles, also recognising the shortcomings due to variability within mapping units. Poggio et al. (2010) used morphological features as covariates, obtaining an optimal model selecting covariates using generalised additive mixed models, to map AWC in Scotland. Ugbaje and Reuter (2013) used two different covariate combinations (remote sensing data; terrain, climate, and vegetation attributes) and pedotransfer functions (PTFs) to map AWC in Nigeria, not finding a clear effect of number of covariates on model accuracy. Most of these studies used PTFs to predict the AWC. Thus the uncertainty of the map depends also on the accuracy of the PTFs.
In digital soil mapping, the visual representation of the product (map) depends on the covariates and the models used. Several studies that looked at the selection and parsimony of the covariates, and also studies have compared different data mining predictions. However no work has looked at the effect of both covariates and models on the visual representation of the map.
A good digital soil map should have a balance of model parsimony (number of covariates), accuracy (numerical performance) and realism of the visual representations (maps). The aim of this work is to obtain a continuous spatial prediction of AWC over Australia, based on field measured data that reconcile these three aspects, exploring the use of different covariate combinations and modelling techniques, and visually inspecting the generated maps.
Section snippets
Data sets and study area
The data set used correspond to a CSIRO Ecosystem Sciences (APSRU) compilation of 806 soil profiles that includes field measurements of DUL and CLL for the most commonly grown crops of Australia (Dalgliesh et al., 2012). Procedures for determination of these properties are described in the accessory publication of the article by Dalgliesh et al. (2009), “Procedures for determination of soil properties and states relevant to crop simulation and farmer crop management decision making”. The method
Covariate selection
We generally observed that adding groups of covariates decreased the magnitude of the cross-validated error (Fig. 2).
As we mentioned in Section 2.3.1 in some studies the addition of extra covariates does not yield better results. In other cases, a smaller number of covariates are preferred, following parsimony and Occam's razor principle (Blumer et al., 1987). In this study we decided to select all the covariates for two reasons: there is not loss of accuracy when using the maximum number of
Conclusions
We explored the use of digital soil mapping approach to model AWC in Australia, balancing three important aspects of it, which are not discussed in previous studies: model parsimony, accuracy and realism of the visual representations. We also explored the use of ensemble methods (i.e.: model averaging) as an alternative to single model selection.
We used different combinations of environmental covariates to represent the various processes involved in soil formation. In many studies the use of
Acknowledgments
The first author was supported through a scholarship by the Chilean Government (“Becas Chile”, CONICYT).
References (44)
- et al.
Modelling soil attribute depth functions with equal-area quadratic smoothing splines
Geoderma
(1999) - et al.
Occam's razor
Inf. Process. Lett.
(1987) Combining forecasts: a review and annotated bibliography
Int. J. Forecast.
(1989)Geostatistical modelling of uncertainty in soil science
Geoderma
(2001)Optimal linear combinations of neural networks
Neural Netw.
(1997)- et al.
Australia-wide predictions of soil properties using decision trees
Geoderma
(2005) - et al.
Irrelevant features and the subset selection problem
ICML
(1994) - et al.
Using expert knowledge with control of false discovery rate to select regressors for prediction of soil properties
Geoderma
(2007) - et al.
Empirical estimates of uncertainty for mapping continuous depth functions of soil attributes
Geoderma
(2011) - et al.
An overview of pedometric techniques for use in soil survey
Geoderma
(2000)
On digital soil mapping
Geoderma
Soil available water capacity interpolation and spatial uncertainty modelling at multiple geographical extents
Geoderma
A weathering intensity index for the Australian continent using airborne gamma-ray spectrometry and digital terrain analysis
Geoderma
Determining the optimum number of predictors for a linear prediction equation
Mon. Weather Rev.
Support-vector networks
Mach. Learn.
Re-inventing model-based decision support with Australian dryland farmers. 2. Pragmatic provision of soil information for paddock-specific simulation and farmer decision making
Crop Pasture Sci.
APSoil-providing soils information to consultants, farmers and researchers
Comparison of point forecast accuracy of model averaging methods in hydrologic applications
Stoch. Env. Res. Risk A.
Assessment and propagation of model uncertainty
J. R. Stat. Soc. Ser. B Methodol.
Global distribution of plant-extractable water capacity of soil
Int. J. Climatol.
Improved methods of combining forecasts
J. Forecast.
Contributions of soil and crop factors to plant available soil water capacity of annual crops on Black and Grey Vertosols
Crop Pasture Sci.
Cited by (31)
Global mapping of volumetric water retention at 100, 330 and 15 000 cm suction using the WoSIS database
2023, International Soil and Water Conservation ResearchCitation Excerpt :According to Dai, Xin, et al. (2019), application of a single-PTF to globally predict soil hydraulic properties leads to biases, underestimation of uncertainties, and overconfidence in model performances of predictive applications. Present global maps of water retention generally show the parameters of a soil water retention (SWR) function that were obtained by PTFs, implying that the uncertainty of the resulting maps depends on the accuracy of the PTFs (Padarian et al., 2014) as well as the underpinning soil data. Table 1 shows examples of studies that present global estimates of soil water retention based on PTFs.
Diagnosis of winter-wheat water stress based on UAV-borne multispectral image texture and vegetation indices
2021, Agricultural Water ManagementCitation Excerpt :Therefore, this study took the Gs of winter wheat as an indicator of water stress, and used image texture, VIs and their combinations to estimate the Gs and assess crop water stress. In particular, Gs estimation models of Cubist (Padarian et al., 2014), BPNN (Zeng et al., 2015; Zhang and Subbarayan, 2002) and ELM (Liang Huang et al., 2006) were built based on sensitive vegetation indices, image texture, as well as their combination. Based on the optimal Gs estimation model, the water stress inversion map of winter-wheat was drawn so as to evaluate whether the combination of image texture and VIs could effectively improve the inversion accuracy of water stress.
Scope to map available water content using proximal sensed electromagnetic induction and gamma-ray spectrometry data
2021, Agricultural Water Management