Application of classification algorithms for the estimation of soil organic carbon in the province of Chimborazo, Ecuador

The objective of the study was to evaluate the decision tree technique using the best supervised classification algorithm, which allows predicting the edaphic carbon content in the province of Chimborazo in native or endemic areas, considering the database of the Ministry of Agriculture and Levestock (MAG). In the estudy, the data set was cleaned and 10 useful variables were determined for the categorization of soil organic carbon, obtaining 4 classes: Very High, High, Medium and Low. The alforithm that provided the best percentage of efficiency and relevant results was Classification and Regression Trees (CART) using the cross-validation method. The refficiency of three algorithms was determined: C5.0, SMV and CART, selecting the CART by means of the cross-validation method for the construction of the tree. The results with the test data set generated a precision of 63.41 percentage points and a prediction error of 36.59 percent, these scopes are presented as a new alternative for SOC quantification, the calibrated model can be extended without the need to sample in situ, very useful in complex areas such as the forest ecosystem. The digital mapping allowed to reveal the existing SOC levels in soils of the Chimborazo province.

• Váldodahkki: Padilla-Sefla, Oscar Roberto (author)
• Eará dahkkit: Haro-Rivera, Silvia Mariana (author)
• Materiálatiipa: article
• Giella: spa
• Almmustuhtton: 2021
• Fáttát: Árboles de clasificación; algoritmos de clasificación supervisada; carbono edáfico; Classification trees; supervised classification algorithms; edaphic carbon
• Liŋkkat: https://revistadigital.uce.edu.ec/index.php/RevFIG/article/view/3518