Zoning of Qanat Systems Extension using Machine Learning Models (Case Study: East and Northeast of Iran)

Abstract

Aims: As a valuable human heritage, the Qanat is of great significance to groundwater systems. This research aims to evaluate the effectiveness of environmental variables in the construction of Qanat systems in the east and northeast of Iran and to present the best machine learning model for modeling. Materials & Methods: Using GIS and R-biomod2, 40 environmental parameters were selected as predictive variables, and GLM, GBM, CTA, SRE, FDA, MARS, RF, and ESMs models were used to determine the relationship between Qanat potential areas and environmental factors. Their Accuracy was evaluated using Kappa, Accuracy, TSS, and ROC. Findings: Results revealed that random forest (RF) and ensemble (ESMs) models achieved the highest Accuracy in determining Qanat potential reas. SRE performed worse than the other eight models. The results also indicated that climatic factors (BIO4), physiographic factors (DEM& Topographic wetness index & slope), soil factors (Organic 60-100 cm, Cations 60-100 cm, Land Surface Temperature) and Geology has a considerable significance in geographical distribution of areas prone to Qanat existence but terrain roughness index showed the least contribution in determining the groundwater potential areas. Conclusion: According to the results, the areas of regions with good to outstanding potential for the existence of Qanats were estimated at 13.15% and 13.31% of the total area using ESMs and RF models, respectively. In general, the use of RS in combination with DEM can reveal numerous significant correlations in groundwater research.