There is a need for accurate, quantitative soil information for natural resource planning and management. This information shapes the way decisions are made as to how soil resources are assessed and ...managed. This paper proposes a novel method for whole-soil profile predictions (to 1
m) across user-defined study areas where limited soil information exists. Using the Edgeroi district in north-western NSW as the test site, we combined equal-area spline depth functions with digital soil mapping techniques to predict the vertical and lateral variations of carbon storage and available water capacity (AWC) across the 1500
km
2 area. Neural network models were constructed for both soil attributes to model their relationship with a suite of environmental factors derived from a digital elevation model, radiometric data and Landsat imagery. Subsequent fits of the models resulted in an
R
2 of 44% for both carbon and AWC. For validation at selected model depths,
R
2 values ranged between 20 and 27% for carbon prediction (RMSE: 0.30–0.52 log (kg/m
3)) and between 8 and 29% for AWC prediction (RMSE: 0.01
m/m). Visually, reconstruction of splines at selected validation data points indicated an average fit with raw data values. In order to improve upon our model and validation results there is a need to address some of the structural and metrical uncertainties identified in this study. Nevertheless, the resulting geo-database of quantitative soil information describing its spatial and vertical variations is an example of what can be generated with this proposed methodology. We also demonstrate the functionality of this geo-database in terms of data enquiry for user-defined queries.
Historically, our understanding of the soil and assessment of its quality and function has been gained through routine soil chemical and physical laboratory analysis. There is a global thrust towards ...the development of more time- and cost-efficient methodologies for soil analysis as there is a great demand for larger amounts of good quality, inexpensive soil data to be used in environmental monitoring, modelling and precision agriculture. Diffuse reflectance spectroscopy provides a good alternative that may be used to enhance or replace conventional methods of soil analysis, as it overcomes some of their limitations. Spectroscopy is rapid, timely, less expensive, non-destructive, straightforward and sometimes more accurate than conventional analysis. Furthermore, a single spectrum allows for simultaneous characterisation of various soil properties and the techniques are adaptable for ‘on-the-go’ field use. The aims of this paper are threefold: (i) determine the value of qualitative analysis in the visible (VIS) (400–700 nm), near infrared (NIR) (700–2500 nm) and mid infrared (MIR) (2500–25,000 nm); (ii) compare the simultaneous predictions of a number of different soil properties in each of these regions and the combined VIS–NIR–MIR to determine whether the combined information produces better predictions of soil properties than each of the individual regions; and (iii) deduce which of these regions may be best suited for simultaneous analysis of various soil properties. In this instance we implemented partial least-squares regression (PLSR) to construct calibration models, which were independently validated for the prediction of various soil properties from the soil spectra. The soil properties examined were soil pH
Ca, pH
w, lime requirement (LR), organic carbon (OC), clay, silt, sand, cation exchange capacity (CEC), exchangeable calcium (Ca), exchangeable aluminium (Al), nitrate–nitrogen (NO
3–N), available phosphorus (P
Col), exchangeable potassium (K) and electrical conductivity (EC). Our results demonstrated the value of qualitative soil interpretations using the loading weight vectors from the PLSR decomposition. The MIR was more suitable than the VIS or NIR for this type of analysis due to the higher incidence spectral bands in this region as well as the higher intensity and specificity of the signal. Quantitatively, the accuracy of PLSR predictions in each of the VIS, NIR, MIR and VIS–NIR–MIR spectral regions varied considerably amongst properties. However, more accurate predictions were obtained using the MIR for pH, LR, OC, CEC, clay, silt and sand contents, P and EC. The NIR produced more accurate predictions for exchangeable Al and K than any of the ranges. There were only minor improvements in predictions of clay, silt and sand content using the combined VIS–NIR–MIR range. This work demonstrates the potential of diffuse reflectance spectroscopy using the VIS, NIR and MIR for more efficient soil analysis and the acquisition of soil information.
We use an empirical method where model output uncertainties are expressed as a prediction interval (PI) of the underlying distribution of prediction errors. This method obviates the need to identify ...and determine the contribution of each source of uncertainty to the overall prediction uncertainty. Conceptually, in the context of digital soil mapping, rather than a single point estimate at every prediction location, a PI, characterised by upper and lower prediction limits, encloses the prediction (which lies somewhere on the interval) and ideally the true but unknown value 100(1
−
α)
% of times on average the target variable (typically 95%). The idea is to partition the environmental covariate feature space into clusters which share similar attributes using fuzzy k-means with extragrades. Model error for predicting a target variable is then estimated from which cluster PIs are constructed on the basis of the empirical distribution of errors associated with the observations belonging to each cluster. PIs for each non-calibration observation are then formulated on the basis of the grade of membership each has to each cluster.
We demonstrate how we can apply this method for mapping continuous soil depth functions. First, using soil depth functions and digital soil mapping (DSM) methods, we map the continuous vertical and lateral distribution of organic carbon (OC) and available water capacity (AWC) across the Edgeroi district in north-western NSW, Australia. From those predictions we define a continuous PI for each prediction node, generating upper and lower prediction limits of both attributes. From an external validation dataset, preliminary results are encouraging where 91% and 93% of the OC and AWC observations respectively fall within the bounds of their 95% PIs. Ideally, 95% of instances should fall within these bounds.
► The empirical uncertainty model presented complements the continuous prediction of soil attributes. ► The empirical uncertainty model explicitly accounts for all sources of uncertainty without the requirement to separate out the contribution of each error source. ► We have demonstrated that this method performed well for both OC and AWC. ► This methodology represents a pragmatic approach to estimating uncertainties both spatially and laterally in a digital soil mapping framework.
Many soil science laboratories are now equipped with technology platforms in portable visible near infrared (vis-NIR) and X-ray fluorescence (XRF) spectrometers. These technologies have complementary ...capabilities, where XRF is known to accurately measure the soil's inorganic element concentration, and vis-NIR has the ability to estimate the soil's organic component and mineralogy suites. In this study data mining was used to estimate soil properties from the vis-NIR spectra, and in a novel way from the XRF spectral data. The prediction outcomes were combined into a single prediction outcome, using formal methods called model averaging procedures. Combining model outcomes derived from spectra using model averaging techniques improves or maintains the prediction status (as ratio of performance to inter-quartile distance) of vis-NIR and XRF models for a wide range of soil properties of agronomic importance. Overall, the relative improvement in %RMSE ranged from 4 to 44%. Weight preference in model averaging was related to the inference of soil chemical and physical properties from vis-NIR and XRF spectra. For example, the weights preference the vis-NIR predictions for soil pH, soil carbon (C), clay, and XRF predictions for most of the elemental soil properties. Based on both the relative improvement in RMSE and RPIQ status, model averaging was found to be suitable for soil pH, soil C (soil organic C (SOC) and total C (TC)), soil texture (sand and clay), CEC and total elements K, Mg, Co, Cr and Mn. Optimum prediction performance for total elements Cu and Zn is achieved by XRF alone. The unreliable RPIQ status for total elements P, S, Mo, Se and exchangeable Ca, Mg, K, and Na derived from vis-NIR and XRF predictions in this study did not improve with model averaging. Overall, Granger-Ramanathan averaging produced similar or better outcomes compared to variance weighted averaging. This model averaging approach is more simple to compute requiring only to fit a simple multiple linear regression model, unlike the VWA approach in which the weighting is estimated for each soil property, and thus in the interests of parsimony is recommended as the model averaging technique to be adopted as protocol. More conventional use of portable XRF in soil analysis is to employ the elemental concentrations measured by the XRF device to predict other soil properties, generally applying multiple linear regression models. When XRF is used in a conventional way to determine elemental concentrations it was demonstrated to be highly reliable for elemental concentrations present in high concentrations, but predictions of elemental content derived from XRF spectra was more effective for elements present in low concentrations. This in turn reduces the capacity of XRF elemental concentrations to be used in the prediction of other soil properties by inference.
•Model averaging proposed to increase predictive power of portable spectral devices.•An improvement of between 4 and 44% in RMSE was achieved by model averaging.•Model averaging was found to be suitable for a range of soil properties.•The Granger-Ramanathan model averaging technique is recommended as protocol.•Prediction of other soil properties from XRF elemental concentrations is limited.
There are tens of millions of contaminated soil sites in the world, and with an increasing population and associated risk there is a growing pressure to remediate them. A barrier to remediation is ...the lack of cost-effective approaches to assessment. Soil contaminants include a wide range of natural and synthetic metallic and organic compounds and minerals thus making analytical costs potentially very large. Further, soil contaminants show a large degree of spatial variation which increases the burden on sampling costs. This paper reviews potentially cost-effective methods for measurement, sampling design, and assessment. Current tiered investigation approaches and sampling strategies can be improved by using new technologies such as proximal sensing. Design of sampling can be aided by on-the-go proximal soil sensing; and expedited by subsequent adaptive spatially optimal sampling and prediction procedures enabled by field spectroscopic methods and advanced geostatistics. Field deployment of portable Visible & Near Infrared wavelength 400–2500nm (Vis-NIR) and X-ray fluorescence (PXRF) spectroscopies will require special calibration approaches but show huge potential for synergistic use. The use of mid-infrared spectroscopy wavelength 2500–25,000nm, wavenumber 4000–400cm−1 (MIR) for field implementation requires further adaptive research. We propose an integrated field-deployable methodology as a basis for further developments.
•There is still a large number of contaminated soil sites worldwide.•Lab analysis of soil contaminants increases investigation and remediation costs.•Field-deployable portable instruments for Vis-NIR and PXRF are now widely available.•A new approach includes adaptive sampling supported by on-the-go Vis-NIR and PXRF.
On digital soil mapping McBratney, A.B; Mendonça Santos, M.L; Minasny, B
Geoderma,
11/2003, Letnik:
117, Številka:
1
Journal Article
Recenzirano
We review various recent approaches to making digital soil maps based on geographic information systems (GIS) data layers, note some commonalities and propose a generic framework for the future. We ...discuss the various methods that have been, or could be, used for fitting quantitative relationships between soil properties or classes and their ‘environment’. These include generalised linear models, classification and regression trees, neural networks, fuzzy systems and geostatistics. We also review the data layers that have been, or could be, used to describe the ‘environment’. Terrain attributes derived from digital elevation models, and spectral reflectance bands from satellite imagery, have been the most commonly used, but there is a large potential for new data layers. The generic framework, which we call the scorpan-SSPFe (soil spatial prediction function with spatially autocorrelated errors) method, is particularly relevant for those places where soil resource information is limited. It is based on the seven predictive scorpan factors, a generalisation of Jenny's five factors, namely: (1)
s: soil, other or previously measured attributes of the soil at a point; (2)
c: climate, climatic properties of the environment at a point; (3)
o: organisms, including land cover and natural vegetation; (4)
r: topography, including terrain attributes and classes; (5)
p: parent material, including lithology; (6)
a: age, the time factor; (7)
n: space, spatial or geographic position. Interactions (*) between these factors are also considered. The scorpan-SSPFe method essentially involves the following steps:
(i) Define soil attribute(s) of interest and decide resolution
ρ and block size
β.
(ii) Assemble data layers to represent
Q.
(iii) Spatial decomposition or lagging of data layers.
(iv) Sampling of assembled data (
Q) to obtain sampling sites.
(v) GPS field sampling and laboratory analysis to obtain soil class or property data.
(vi) Fit quantitative relationships (observing Ockham's razor) with autocorrelated errors.
(vii) Predict digital map.
(viii) Field sampling and laboratory analysis for corroboration and quality testing.
(ix) If necessary, simplify legend or decrease resolution by returning to (i) or improve map by returning to (v).
Finally, possible applications, problems and improvements are discussed.
The rapid development in NIR and information technologies saw the development of various initiatives that have generated large scale databases of soil spectroscopy globally. Models generated within a ...specific spectral or geographical domain should be carefully used in other contexts since they may lose their validity. This includes the application of a global, continental or national spectral libraries to local areas or regions. Both, global and local models are valuable and, ideally, we would like to transfer some of the rules learnt by the more general global models to a local domain. In machine learning, the process of sharing intra-domain information is known as transfer learning. This paper aims to describe and evaluate the effectiveness of transfer learning to “localise” a general soil spectral model. The transfer process consists in, first, training a model with a big volume of data covering a diverse group of cases. Second, some layers of the trained neural network are used to build a local model, which is fine-tuned by using a smaller amount of local data. We demonstrated this method using the LUCAS database, an European dataset, comprising spectral data from 21 countries. For each country, we generated three models: a) Global, with data from all except the country of interest; b) Local, with data from the country; and c) Transfer, pre-trained as the Global model and fine-tuned with data from the country. The results showed that the Transfer model can lower the error (expressed as RMSE) 91% of the cases, with a mean reduction of RMSE: 10.5, 11.8, 12.0 and 11.5% for organic carbon, cation exchange capacity, clay content and pH, respectively. This paper demonstrates the usefulness of transfer learning for soil spectroscopy, which will enhance the use of global spectral libraries for local application.
•The method is capable of transferring knowledge from a continental calibration model to generate a localised model.•The method performs better than a global or local model individually.•There was a reduction of the RMSE in 91% of the cases.•The mean reduction was 10.5, 11.8, 12.0 and 11.5% for organic carbon, CEC, clay content and pH.•With this method, collaboration can be beneficial for everyone, including the data-rich countries or organisations.
A digital soil mapping exercise over a large extent and at a high resolution is a computationally expensive procedure. It may take days or weeks to obtain the final maps and to visually evaluate the ...prediction models when using a desktop workstation. To increase the speed of time-consuming procedures, the use of supercomputers is a common practice. GoogleTM has developed a product specifically designed for mapping purposes (Earth Engine), allowing users to take advantage of its computing power and the mobility of a cloud-based solution. In this work, we explore the feasibility of using this platform for digital soil mapping by presenting two soil mapping examples over the contiguous United States. We also discuss the advantages and limitations of this platform at its current development stage, and potential improvements towards a fully functional cloud-based soil mapping platform.
•It is possible to include the platform as part of a digital soil mapping workflow.•Map generation is 40–100 times faster compared with traditional digital soil mapping.•To be a complete solution, implementation of geostatistical methodologies is needed.•We encourage researches to participate during the beta-testing period to improve it.
Diffuse reflectance infrared spectroscopy allows the rapid acquisition of soil information in the field or the laboratory. The vis-NIR spectroscopy research enthusiasm around the world has created ...regional to global soil spectral libraries. While machine learning methods have been utilised in processing spectral data, such large regional datasets are better dealt with big data analytics. Deep learning is an exciting discipline that has already transformed the way data are analysed in many fields and could also change the way we model soil spectral data. This study developed and evaluated convolutional neural networks (CNNs), a type of deep learning algorithm, as a new way to predict soil properties from raw soil spectra. We demonstrated the effectiveness of CNNs on the LUCAS soil database, which consists of around 20,000 topsoil observations with physicochemical and biological properties from Europe. To fully utilise the capacity of the CNN model, we represented the soil spectral data as a 2-dimensional spectrogram, showing the reflectance as a function of wavelength and frequency. We showed the capacity of a CNN to be trained in a multi-task setting to simultaneously predict six soil properties in one model (OC, CEC, clay, sand, pH, total N). Compared with conventional methods such as PLS regression and Cubist regression tree, the CNN performed significantly better, especially the multi-tasking model. In the case of soil organic carbon prediction, the multi-task CNN decreased the error by 87% compared to PLS and 62% compared with Cubist. This approach proved to be effective when trained on a relatively large dataset. The high accuracy of CNN makes it an ideal tool for modelling soil spectral data.
•CNN model is able to predict soil properties from raw spectral data.•CNN is able to predict multi properties simultaneously and synergically.•The multi-task CNN reduced the error compared to the Cubist model by 62%.•This approach works better with large spectral datasets.
Parametric pedotransfer functions (PTFs), which predict parameters of a model from basic soil properties are useful in deriving continuous functions of soil properties, such as water retention ...curves. The common method for deriving parametric water retention PTFs involves estimating the parameters of a soil hydraulic model by fitting the model to the data, and then forming empirical relationships between basic soil properties and parameters. The latter step usually utilizes multiple linear regression or artificial neural networks. Neural network analysis is a powerful tool and has been shown to perform better than multiple linear regression. However neural‐network PTFs are usually trained with an objective function that fits the estimated parameters of a soil hydraulic model. We called this the neuro‐p method. The estimated parameters may carry errors and since the aim is to be able to estimate water retention, it is sensible to train the network to fit the measured water content. We propose a new objective function for neural network training, which predicts the parameters of the soil hydraulic model and optimizes the PTF to match the measured and observed water content, we called this neuro‐m method. This method was used to predict the parameters of the van Genuchten model. Using Australian soil hydraulic data as a training set, neuro‐m predicted the water retention from bulk density and particle‐size distribution with a mean accuracy of 0.04 m3 m−3 The relative improvement of neuro‐m over neural networks that was optimized to fit the parameters (neuro‐p) is 13%. Compared with a published neural network PTF, the new method is 30% more accurate and less biased.