Developing a SmartSoil Classifier Tool for Classifying Soils According to Soil Taxonomy at Subgroup level for Some Soils in Iraqi Alluvial Plain Using Python

Abstract

A software tool was developed in Python to automate soil classification according to Soil Taxonomy (1999) and Keys to Soil Taxonomy (2022) for the Entisols and Aridisols orders, down to the most common Subgroups in Iraqi Alluvial Plain. The reference database was used of 309 Pedons from four previously studied projects Sheikh Saad, Najmi, Awadiyat, and Shataniyah. These Pedons were classified by experts to evaluate the performance of the developed software. Three areas were selected for field soil surveys: Sheikh Saad, Al-Warkaa, and Al-Islah, encompassing 18 Pedons and 47 boreholes, with a total area of 12,317 hectares. The program's performance evaluation results, when compared with expert classifications, showed an overall accuracy of 93.5% at the order level. At the sub-order level, the accuracy was 89.32%, while the overall accuracy at the Great group level was 78.96%. At the sub-group level, the overall accuracy was 70.55%. Analysis of variance revealed 14 categories of discrepancies, most of which stemmed from errors in reference data and human estimation. After correcting these errors, the classification accuracy theoretically reached 100%. When the field results for the described 18 Pedon’s, distributed across three governorates within the Iraqi alluvial plain, were presented, the program's performance evaluation results showed 100% accuracy across all classification levels. This reflects the program's ability to process field and laboratory data. However, this accuracy remains dependent on the quality of the morphological description, as results may be affected if the person describing the pedon changes, particularly in criteria that rely on estimation, such as waterlogging conditions.