Estimating topsoil texture fractions by digital soil mapping - a response to the long outdated soil map in the Philippines

Jeremy P Mondejar,Alejandro F Tongco

doi:10.1186/s42834-019-0032-5

Abstract

Digital soil mapping for soil texture is mostly an understanding of how soil texture fractions vary in space as influenced by environmental variables mainly derived from the digital elevation model (DEM). In this study, topsoil texture models were generated and evaluated by multiple linear regression (MLR), ordinary kriging (OK), simple kriging (SK) and universal kriging (UK) using free and open-source R, System for Automated Geoscientific Analyses, and QGIS software. Comparing these models is the main objective of the study. The study site covers an area of 124 km2 of the Municipality of Barili, Cebu. A total of 177 soil samples were gathered and analyzed from irregular sample points. DEM derivatives and remote sensing data (Landsat 8) were used as environmental variables. Exploratory analyses revealed no outlier in the data. Skewness and kurtosis values of the untransformed data vary greatly between –3.85 to 7.20 and 1.8 to 70.7, respectively; an indication that variables are highly skewed with heavy tails. Thus, Tukey’s ladder of powers transformation was applied that resulted to normal or nearly normal distribution having skewness values close to zero and kurtosis values have lighter tails. All data analysis from MLR modeling, variography, kriging, and cross-validations of models were implemented using the transformed data. Forward selection, backward elimination, and stepwise selection methods were adapted for predictors selection in MLR. The MLR, OK, SK, and UK were applied and cross validated for topsoil texture prediction. Likewise, exponential, Gaussian, and spherical models were fitted for the experimental variograms. Backward elimination method for clay, sand, and silt have the lowest MAE and highest R2 in MLR. The UK fitted with exponential variogram model has the highest R2 of 0.878, 0.821, and 0.893 for clay, sand, and silt, respectively. These models can be adapted as a decision support for agricultural land use planning and crop suitability development in the area.

Highlights

Dealing with global and regional challenges in land degradation, food security, water scarcity, and climate change, an accurate and updated geospatial soil information is imperatively needed [1, 2]
The achieved methodology can lead to valuable outcome in achieving a more comprehensive land use plan since the generated results are useful for watershed management for ecological, hydrological, and crop suitability modeling
simple kriging (SK), and universal kriging (UK) fitted with Gaussian variogram model were less suitable for predicting soil texture fractions whereas, UK coupled with predictors by backward elimination method and fitted with exponential variogram portrayed the most accurate in predicting topsoil texture fractions for clay, sand, and silt compared to any of the spatial and geostatistical modeling applied in this study

Summary

Introduction

Dealing with global and regional challenges in land degradation, food security, water scarcity, and climate change, an accurate and updated geospatial soil information is imperatively needed [1, 2]. Conventional soil survey adapts the manual process of producing a polygon-based soil map, whereby, without the computer-based approach, the map cannot be Mondejar and Tongco Sustainable Environment Research (2019) 29:31 updated rapidly and accurately as the entire production procedure must be repeated [7]. Such method is timeconsuming, requires numerous soil samples, and expensive [8]. The direction of DSM is toward the generation of dynamic and replicable geospatial soil information [10]

Objectives

Methods

Results

Conclusion