Use of NIRS in Soil Properties Evaluation Related to Soil Salinity and Sodicity in Colombian Caribbean Coast

Abstract

The banana sector contributes to 3% of the Colombia total exports, becoming the most important crop in the north of the country, benefiting more than 2.5 million families. Banana production is concentrated on the Colombian Caribbean coast where most of 90% of the soils are affected to some degree of salt affection in soils. This study was carried out in the municipality of Zona Bananera, Department of Magdalena (Colombia) elsewhere eleven geomorphological units were delimited through geomorphological surveying with geopedological methods. Given the high costs of implementing salt monitoring and management programs in the field, the implementation of Near Infrared spectroscopy (NIR) and image analysis are proposed as an alternative for mapping soils salt affectation. Geostatistical methods, traditional soil laboratory methods and multispectral analysis of 697 soil samples were analyzing using machine learning and spectral models. The Orthogonal Partial Linear Square- Discriminant Analyses (OPLS-DA), Principal Component Regression (PCR), Partial Linear Square (PLS–PLSR), Least Absolute Shrinkage and Selection Operator (LASSO) were implemented. Soil cartographies for SAS were designed in areas under banana cultivation, determining the affectation degree. The results obtained showed that 45.1% of the soils are affected by salts, with R² 0.76 and RMS 0.15 for the applied of supervised models. OPLS-DA had a better performance being the high above sea level was the principal covariable to improve the model accuracy. LASSO and PLS were useful to Mg⁺² and K⁺ with RMSE 0.92 and 0.34 and R² of 0.37 and 0.44, while Saitsky&Golay filter improved the predictions model for pH and Ca⁺². The use of combined techniques of geopedology, geostatistics and spectroscopy were efficient, practical and cheap methodologies for evaluate soil properties associate to SAS in the stablished banana crops.