Operational sampling challenges to digital soil mapping in Tasmania, Australia

Abstract Digital soil mapping (DSM) was used to generate soil property surfaces at 30 m resolution for Tasmanian Government Land Suitability Modelling in Tasmania, Australia. Soil predictions were required for pH, EC, clay percentage, stone content, drainage, and depth to sodic and impeding layer. Empirical modelling using a suite of environmental covariates and the relevant soil attribute data from field-collected soil cores was used to generate the digital maps. Environmental covariates included: SRTM DEM and derivatives, gamma radiometry, legacy soil maps, surface geology, and multi-spectral satellite imagery. An integral component of any DSM process is a sound sampling design that represents the full range of environmental variables used. However, in cases where there are operational constraints, the approach needs to remain flexible, efficient, and compatible with project area land use and terrain. In two separate study areas, a combined 700 training and 230 validation sites were sampled over 70,000 ha. A conditioned Latin hypercube (cLHS) sampling design was used for the initial sampling for DSM training sites, with ‘contingency sites’ created for alternative sampling if access was constrained. The pre-defined (‘strict’) sample locations proved difficult to implement in the field, with a variety of access issues making sampling slow and arduous. In an attempt to increase sampling progress rates to meet tight project milestones an alternative ‘relaxed’ sample design based on random sampling of fuzzy k-means covariate clusters (strata) was used for the second study area. A map of clusters provided to soil sampling staff allowed difficult sites to be relocated within the same cluster type, maintaining stratification. The relaxed approach still adequately represented the covariate distribution while providing greater flexibility to site placement. This paper provides background to the Tasmanian DSM project, some discussion of sampling designs for DSM, and the pros and cons of their implementation in the field with due consideration of operational constraints in a Tasmanian case study, highlighting the need for sampling flexibility within ‘real-world’ conditions.

[1]  J. MacQueen.  Some methods for classification and analysis of multivariate observations , 1967 .

[2]  Peter A. Burrough,et al.  High-resolution landform classification using fuzzy k-means , 2000, Fuzzy Sets Syst..

[3]  J. Gallant,et al.  A multiresolution index of valley bottom flatness for mapping depositional areas , 2003 .

[4]  Gerard Grealish,et al.  Soil survey design for management-scale digital soil mapping in a mountainous southern Philippine catchment , 2012 .

[5]  H. L. Le Roy,et al.  Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability; Vol. IV , 1969 .

[6]  N. McKenzie,et al.  Spatial prediction of soil properties using environmental correlation , 1999 .

[7]  Budiman Minasny,et al.  Digital soil assessment: Guiding irrigation expansion in Tasmania, Australia , 2012 .

[8]  Budiman Minasny,et al.  On digital soil mapping , 2003 .

[9]  J. J. de Gruijter,et al.  An R package for spatial coverage sampling and random sampling from compact geographical strata by k-means , 2010, Comput. Geosci..

[10]  Alex B. McBratney,et al.  Soil pattern recognition with fuzzy-c-means : application to classification and soil-landform interrelationships , 1992 .

[11]  Gerard B. M. Heuvelink,et al.  Sampling design optimization for multivariate soil mapping , 2010 .

[12]  J. L. Boettinger,et al.  Conditioned Latin Hypercube Sampling: Optimal Sample Size for Digital Soil Mapping of Arid Rangelands in Utah, USA , 2010 .

[13]  J. Beek,et al.  Developments in Soil Science , 2019, Global Change and Forest Soils.

[14]  J. Bezdek,et al.  FCM: The fuzzy c-means clustering algorithm , 1984 .

[15]  G. Laslett.  Discussion of the paper by D.J. Brus and J.J. de Gruijter , 1997 .

[16]  James E. Payne,et al.  Pragmatic soil survey design using flexible Latin hypercube sampling , 2014, Comput. Geosci..

[17]  Lotfi A. Zadeh,et al.  Fuzzy Algorithms , 1968, Inf. Control..

[18]  Pierre Roudier,et al.  A conditioned Latin hypercube sampling algorithm incorporating operational constraints , 2012 .

[19]  J. Walker,et al.  Australian Soil and Land Survey Field Handbook , 1984 .

[20]  D. J. Brus,et al.  Random sampling or geostatistical modelling? Choosing between design-based and model-based sampling strategies for soil (with discussion) , 1997 .

[21]  Budiman Minasny,et al.  A conditioned Latin hypercube method for sampling in the presence of ancillary information , 2006, Comput. Geosci..

[23]  B. Hanquet,et al.  Analysis of Spatial Soil, Crop and Yield Data in a Winter Wheat Field , 2003 .

[24]  J. W. van Groenigen,et al.  Chapter 14 Designing Spatial Coverage Samples Using the k-means Clustering Algorithm , 2006 .

[25]  J. Deckers,et al.  World Reference Base for Soil Resources , 1998 .

[26]  D. Kidd.  Land Degradation and Salinity Risk Investigations, Tunbridge, Tasmania , 2015 .

[27]  Luca Montanarella,et al.  Digital soil assessments: Beyond DSM , 2007 .

[28]  J. Hartigan.  Statistical theory in clustering , 1985 .

[29]  Alfred E. Hartemink,et al.  Digital soil mapping: bridging research, environmental application, and operation , 2010 .