| Abstract | Over 70 % of sugarcane industry operates next to the Great Barrier Reef (GBR). Farmers are under pressure to improve practices to minimise off-farm pollution, while at the same time improve fertiliser (e.g. lime) and amelioration (e.g. gypsum) efficiency to minimise yield variation. While the biggest driver of variation is rainfall, differences in soil condition affect yield and farmers need to know its variation.
For example, knowledge about soil cation exchange capacity (CEC – cmol(+)/kg) is important because it is a measure of how many exchangeable (exch.) cations (i.e. calcium [Ca], magnesium [Mg]) can be retained on soil surfaces and because it influences soil stability, nutrient availability, pH and reaction to fertilisers.
If no action is taken to map soil and manage different soil condition, opportunities to sustainably improve application of fertilisers and ameliorants in a cost-effective way will be foregone as well as an opportunity to make a meaningful, economically viable contribution to reducing impacts of sugarcane growing on the GBR.
The six-easy-steps (6ES) nutrient and ameliorant guidelines were developed to minimise in-field variation and reduce losses of inputs to the GBR. However, there was and is no practical way for farmers to apply the 6ES guidelines given there is no in-field data to enable its application. This project aimed to undertake case studies in four sugarcane growing areas to enable precision agriculture via the use of a digital soil map (DSM).
A DSM requires collection of digital data, such as proximally sensed electromagnetic (EM) induction and gamma-ray spectrometry (ϒ-ray) and coupling this to soil data via mathematical models. The study areas, where a DSM approach was taken, include Mossman, Herbert, Burdekin and Proserpine.
The results show that a DSM approach is valid with the potential to implement the 6ES nutrient and ameliorant guidelines to enable precise application of lime, gypsum and other fertilisers demonstrated via various case studies. They are provided here in brief and in summarised form in Section 5. All published papers or submitted manuscripts are provided in the same order and appear in the Appendices.
In the Mossman area (see Section 5.1), a DSM approach was used to characterise soil condition in terms of topsoil (0-0.3 m) soil organic carbon (SOC, %) variation, with the DSM able to be used to apply the 6ES nutrient management guidelines (Schroeder et al., 2010) with varying N application rates for different levels of SOC to achieve a district yield potential of 120 t/ha after a bare fallow (Wang et al., 2021).
In various areas (see Section 5.2), the DSM approach could be used to predict topsoil (0-0.3 m) clay content across any of six study sites in the Mossman, Herbert, Burdekin, and Proserpine districts. The site-specific approach to making DSM of topsoil clay was optimal, however site-independent (universal calibration) and a spiking approach give almost as good prediction agreement and accuracy (Arshad et al., 2021).
In the Herbert (see Section 5.3), a DSM approach was used to characterise soil condition in terms of topsoil (0–0.3 m) and subsoil (0.6–0.9 m) CEC (cmol(+)/kg) variation, with the topsoil DSM able to be used to apply the 6ES nutrient management guidelines (Sugar Research Australia, 2013) with varying lime application rates for different levels of CEC (Li et al., 2018).
In the Herbert (see Section 5.4), a DSM approach was used to identify zones by clustering digital data (i.e. EM and -ray data). The DSM was more accurate in predicting topsoil (0-0.3 m) and subsoil (0.6–0.9 m) chemical (e.g. CEC, exch. Ca and Mg and ESP) properties. The 6ES guidelines of Schroeder et al. (2009) were applicable to ameliorate topsoil ESP; the latter shown to influence yield percentage (Dennerley et al., 2018).
In the Herbert (see Section 5.5), a wavelet transform of the digital data (i.e. EM and -ray data) was used to enable prediction of topsoil (0-0.3 m) ESP. The DSM, using all the wavelet transformed digital data (i.e. elevation, EM and -ray data) gave the most accurate predictions. The 6ES guidelines of Schroeder et al. (2006) to manage ESP through variable rates of gypsum was also demonstrated (Li et al., 2021a). Sugar Research Australia Final Report 2017/014
4
In the Herbert (see Section 5.6), a DSM approach was again used to identify zones by clustering digital data (i.e. EM and -ray data). The DSM was more accurate in predicting topsoil (0-0.3 m) and subsoil (0.6–0.9 m) chemical (e.g. CEC, exch. Ca and Mg) properties than a traditional texture map or field delineations. The 6ES guidelines of Schroeder et al. (2009) were applicable for these properties (Arshad et al., 2019).
In the Burdekin (see Section 5.7), a DSM approach was used to predict topsoil (0-0.3 m) exch. Ca and Mg. The DSM was more accurate than a traditional map (Li et al., 2019a) and useful for applying lime and magnesium, respectively, using 6ES guidelines (Schroeder et al., 2009). In terms of calibration, 30 samples were enough to predict exch. Ca with 40 for exch. Mg (Li et al., 2019b).
In Proserpine (see Section 5.8), a DSM was developed to predict topsoil (0-0.3 m) ESP. A map generated using ordinary kriging of 120 soil samples was satisfactory, but, a minimum of 100 samples was required. When digital data was used to value add to soil data, Cubist-RK outperformed OK with only 60 samples required. The 6ES guidelines of Schroeder et al. (2009) were applicable to ameliorate topsoil ESP (Li et al., 2021b).
In Proserpine (see Section 5.9), a DSM was developed to predict topsoil (0-0.3 m) and subsoil (0.9-1.2 m) CEC. Topsoil prediction required 80 calibration samples whereas for subsoil only 30 were needed. Using both digital gave best results although -ray used alone slightly better than EM. Small transect spacing (i.e. 5 m) was recommended for topsoil, but larger spacing OK for subsoil (i.e. 5 – 60 m). The 6ES guidelines of Proserpine (Calcino et al., 2010) were applicable to ameliorate topsoil CEC (Zhao et al., 2020).
Given the results presented in this Final Report and the published research, it can be concluded that the DSM approach can be applied to map various topsoil and subsoil physical (e.g. clay, silt and sand) and chemical (i.e. CEC, Exch. Ca, Exch. Mg and ESP) properties at the field and multi-field scale in different sugarcane growing districts. The final DSM can be used to apply the 6ES nutrient and ameliorant guidelines in the four sugarcane growing areas investigated and including Mossman, Herbert, Burdekin, and Proserpine.
In terms of operational aspects, the following key conclusions can be made;
i) Various soil physical (e.g. clay, silt and sand) and chemical (i.e. CEC, Exch. Ca, Exch. Mg and ESP) properties can be mapped using a DSM approach, but regardless of modelling technique, the number of soil samples required to make a calibration was approximately the same (i.e. 1 sample per hectare) regardless of the soil property (i.e. topsoil Exch. Ca and Mg and ESP) or study area.
ii) Mathematical methods such as LMM are useful when digital data are correlated with soil data, with hybrid methods of machine learning (i.e. Cubist) and regression kriging (Cubist-RK) useful when correlations were statistically significant but not as strong and if residuals were spatially auto-correlated. Alternatively, wavelet analysis can also be useful to predict soil properties (i.e. topsoil ESP) where there was no direct relationship with digital data but a relationship with scale specific variation in digital data (i.e. ϒ-ray, EM and DEM). Moreover, fuzzy k-means or k-means clustering can be used to make management zones from -ray and EM data when the digital data is not directly correlated to the soil data of interest and produce superior predictions than traditional soil texture maps and or using field delineations to predict soil properties.
iii) Digital data of elevation, ϒ-ray and EM were best used in combination rather than alone, regardless of which modelling technique was considered (e.g. LMM, Cubist-RK and wavelet analysis). In terms of the density of digital data transect spacing, the smaller the spacing the better (i.e. transect every 7.5 m) with a maximum transect spacing of 30 m allowing large areas to be measured in a day (~ 400 ha). |
