Combining two soil property rasters using an adaptive gating approach
David Clifford A C and Yi Guo B DA CSIRO, GPO Box 2583, Brisbane, Qld 4001, Australia.
B CSIRO, PO Box 184, North Ryde, NSW 1670, Australia.
C National Institute of Applied Statistics Research Australia, University of Wollongong, NSW, Australia.
D Corresponding author. Email: Yi.Guo@csiro.au
Soil Research 53(8) 907-912 https://doi.org/10.1071/SR14275
Submitted: 1 October 2014 Accepted: 6 May 2015 Published: 18 September 2015
Abstract
Given the wide variety of ways one can measure and record soil properties, it is not uncommon to have multiple overlapping predictive maps for a particular soil property. One is then faced with the challenge of choosing the best prediction at a particular point, either by selecting one of the maps, or by combining them together in some optimal manner. This question was recently examined in detail when Malone et al. (2014) compared four different methods for combining a digital soil mapping product with a disaggregation product based on legacy data. These authors also examined the issue of how to compute confidence intervals for the resulting map based on confidence intervals associated with the original input products. In this paper, we propose a new method to combine models called adaptive gating, which is inspired by the use of gating functions in mixture of experts, a machine learning approach to forming hierarchical classifiers. We compare it here with two standard approaches – inverse-variance weights and a regression based approach. One of the benefits of the adaptive gating approach is that it allows weights to vary based on covariate information or across geographic space. As such, this presents a method that explicitly takes full advantage of the spatial nature of the maps we are trying to blend. We also suggest a conservative method for combining confidence intervals. We show that the root mean-squared error of predictions from the adaptive gating approach is similar to that of other standard approaches under cross-validation. However under independent validation the adaptive gating approach works better than the alternatives and as such it warrants further study in other areas of application and further development to reduce its computational complexity.
References
Bates JM, Granger CW (1969) The combination of forecasts. OR, pp. 451–468.Bivand RS, Pebesma E, Gomez-Rubio V (2013) ‘Applied spatial data analysis with R.’ 2nd edn. (Springer: New York)
Bui EN, Moran CJ (2001) Disaggregation of polygons of surficial geology and soil maps using spatial modelling and legacy data. Geoderma 103, 79–94.
| Disaggregation of polygons of surficial geology and soil maps using spatial modelling and legacy data.Crossref | GoogleScholarGoogle Scholar |
Byrd RH, Lu P, Nocedal J, Zhu C (1995) A limited memory algorithm for bound constrained optimization. SIAM Journal on Scientific Computing 16, 1190–1208.
Cressie N, Johannesson G (2008) Fixed rank kriging for very large spatial data sets. Journal of the Royal Statistical Society: Series B (Statistical Methodology) 70, 209–226.
| Fixed rank kriging for very large spatial data sets.Crossref | GoogleScholarGoogle Scholar |
de Bruin S, Wielemaker WG, Molenaar M (1999) Formalisation of soil-landscape knowledge through interactive hierarchical disaggregation. Geoderma 91, 151–172.
| Formalisation of soil-landscape knowledge through interactive hierarchical disaggregation.Crossref | GoogleScholarGoogle Scholar |
Diks CG, Vrugt JA (2010) Comparison of point forecast accuracy of model averaging methods in hydrologic applications. Stochastic Environmental Research and Risk Assessment 24, 809–820.
Eberhart RC, Kennedy J (1995) A new optimizer using particle swarm theory. In ‘Proceedings of the Sixth International Symposium on Micro Machine and Human Science’. Vol. 1, pp. 39–43. (Institute of Electrical and Electronics Engineers: New York)
Fletcher R, Reeves CM (1964) Function minimization by conjugate gradients. The Computer Journal 7, 149–154.
| Function minimization by conjugate gradients.Crossref | GoogleScholarGoogle Scholar |
Granger CW, Ramanathan R (1984) Improved methods of combining forecasts. Journal of Forecasting 3, 197–204.
Grundy MJ, Viscarra Rossel RA, Searle RD, Wilson PL, Chen C, Gregory LJ (2015) Soil and Landscape Grid of Australia. Soil Research 53, 835–844.
Hoerl AE, Kennard RW (1970) Ridge regression: Biased estimation for nonorthogonal problems. Technometrics 12, 55–67.
Hoeting JA, Madigan D, Raftery AE, Volinsky CT (1999) Bayesian model averaging: A tutorial. Statistical Science 14, 382–401.
Holmes KW, Griffin EA, Odgers NP (2015) Large-area spatial disaggregation of a mosaic of conventional soil maps: evaluation over Western Australia. Soil Research 53, 865–880.
Kidd D, Webb M, Malone B, Minasny B, McBratney A (2015) Eighty-metre resolution 3D soil-attribute maps for Tasmania, Australia. Soil Research 53, 932–955.
Liddicoat C, Maschmedt D, Clifford D, Searle R, Herrmann T, Macdonald LM, Baldock J (2015) Predictive mapping of soil organic carbon stocks in South Australia’s agricultural zone. Soil Research 53, 956–973.
Malone BP, Minasny B, Odgers NP, McBratney AB (2014) Using model averaging to combine soil property rasters from legacy soil maps and from point data. Geoderma 232–234, 34–44.
Masoudnia S, Ebrahimpour R (2014) Mixture of experts: a literature survey. Artificial Intelligence Review 42, 275–293.
McBratney A, Mendonça Santos M, Minasny B (2003) On digital soil mapping. Geoderma 117, 3–52.
| On digital soil mapping.Crossref | GoogleScholarGoogle Scholar |
Odgers N, McBratney A, Minasny B (2014a) Digital soil property mapping and uncertainty estimation using soil class probability rasters. GlobalSoilMap: Basis of the Global Spatial Soil Information System. p.341.
Odgers NP, Sun W, McBratney AB, Minasny B, Clifford D (2014b) Disaggregating and harmonising soil map units through resampled classification trees. Geoderma 214–215, 91–100.
| Disaggregating and harmonising soil map units through resampled classification trees.Crossref | GoogleScholarGoogle Scholar |
Odgers NP, Holmes KW, Griffin T, Liddicoat C (2015) Derivation of soil-attribute estimations from legacy soil maps. Soil Research 53, 881–894.
Pebesma EJ, Bivand RS (2005) Classes and methods for spatial data in R. R News 5, 9–13. http://CRAN.R-project.org/doc/Rnews/
Viscarra Rossel R, Chen C (2011) Digitally mapping the information content of visible-near infrared spectra of surficial Australian soils. Remote Sensing of Environment 115, 1443–1455.
| Digitally mapping the information content of visible-near infrared spectra of surficial Australian soils.Crossref | GoogleScholarGoogle Scholar |
Viscarra Rossel RA, Chen C, Grundy MJ, Searle R, Clifford D, Campbell PH (2015) The Australian three-dimensional soil grid: Australia’s contribution to the GlobalSoilMap project. Soil Research 53, 845–864.
Wilford J, Thomas M (2013) Predicting regolith thickness in the complex weathering setting of the central Mt Lofty Ranges, South Australia. Geoderma 206, 1–13.
