Services on Demand
Journal
Article
Indicators
- Cited by SciELO
Related links
- Similars in SciELO
Share
Ciencia del suelo
On-line version ISSN 1850-2067
Abstract
PACCIORETT, Pablo Ariel; KURINA, Franca Giannini and BALZARINI, Monica Graciela. Site sampling at regional scale for digital mapping based on soil properties. Cienc. suelo [online]. 2020, vol.38, n.2, pp.310-320. ISSN 1850-2067.
The objective of this study was to evaluate the performance of the sampling method called conditioned Latin Hypercube (cLHS) to identify convenient sites for obtaining data on edaphic properties that are used in the construction of models for digital mapping of a spatially distributed variable as it is soil organic carbon (SOC). Given N sites with information on p explanatory variables (X), cLHS selects a sample of n sites in such a way that the multivariate distribution of X is fully characterized. In this work, data from a regional soil study of the Province of Córdoba were used to compare the performance of the cLHS sampling method with simple random sampling (RS). To evaluate the sampling method, the population of sites with data was repeatedly sampled and, in each sample, the relationship between SOC and the edapho-climatic properties of the site was adjusted, using both linear regression models and random forest as machine learning algorithm. The prediction errors of each sampling method were evaluated with each statistical method used for the prediction of SOC in sites where this variable was not measured. The sampling method impacted the overall reliability of the predictions derived from both regression models and site-specific prediction errors. The cLHS method was more efficient than RS to identify sites with sufficient variability to estimate the model of the relationship between SOC and edapho-climatic properties, used to predict the SOC value in other sites of the territory. The estimated model can be used for digital mapping of SOC.
Keywords : Conditioned Latin Hypercube Sampling; Random Sampling; Multiple Linear Regression; Random Forest.