Introduction
The Qazvin Plain, located on the northern margin of Iran’s Central Plateau, is considered one of the most significant prehistoric settlement centers in the region. Over thousands of years, this landscape has been continuously reshaped by complex geomorphological processes such as erosion, sedimentation, and fluvial dynamics. These processes have not only modified the physical environment but have also influenced the preservation and visibility of archaeological sites. Understanding the relationship between sedimentation and site distribution is therefore critical for reconstructing past environments, interpreting patterns of human settlement, and refining predictive models for future archaeological research. Beyond its theoretical implications, this line of inquiry helps shed light on the decision-making logic of prehistoric communities, their perception of environmental stability, and their long-term strategies for sustaining habitation within dynamic landscapes.
Objectives
This study aims to analyze the relationship between sedimentation intensity and the spatial distribution of prehistoric sites (Neolithic and Chalcolithic) within three sub-basins of the Qazvin Plain—Abharrood, Kharrood, and Hajiarab. The research further seeks to evaluate the relative importance of multiple geomorphological and environmental factors in shaping settlement suitability and to develop an integrated framework that combines machine learning and multi-criteria decision-making techniques for environmental archaeological modeling.
Materials and Methods
The study was conducted in two main phases.
Phase 1 – Sediment Yield Modeling:
Sediment yield maps were reconstructed using the Fournier Index, expressed in tons per square kilometer, to quantify the spatial variability of sedimentation. These initial outputs were subsequently refined using three machine learning algorithms—Multiple Linear Regression (MLR), Random Forest (RF), and Artificial Neural Networks (ANN). For this section ten environmental predictor variables were incorporated, including elevation, slope, clay content, sand content, Topographic Position Index (TPI), Normalized Difference Vegetation Index (NDVI), channel network distance, profile curvature, plan curvature, and lithology. Model performance was assessed using Root Mean Square Error (RMSE) to identify the most accurate algorithm.
Phase 2 – Settlement Suitability Modeling:
A settlement suitability model was developed to identify areas with high potential for prehistoric occupation. Seven key environmental and geomorphological criteria were selected: elevation, slope, aspect, flow accumulation, distance to water sources, distance from faults, and NDVI. These criteria were standardized and weighed using a hybrid approach that combined Analytic Hierarchy Process (AHP) with Principal Component Analysis (PCA). The integration of AHP and PCA ensured that statistical variance structure contributed to the final weighting scheme, thereby reducing subjective bias and improving model robustness. Raster maps were generated in the Python programming environment and subsequently analyzed in ArcGIS Pro, where overlay operations, reclassification, and weighted linear combination were applied to produce the final settlement suitability map.
Results
The machine learning comparison revealed that ANN and RF significantly outperformed MLR, achieving lower RMSE values and higher spatial accuracy. ANN demonstrated superior capability in capturing the non-linear and complex relationships between environmental variables and sediment yield, thereby producing more realistic sedimentation maps.
The suitability analysis showed that the majority of known archaeological sites are located in areas characterized by lower sedimentation rates. Statistical testing confirmed a significant negative correlation between sedimentation intensity and site presence probability. This finding indicates that regions with high sedimentation are less likely to preserve visible archaeological sites, either because such areas were less frequently chosen for habitation or because existing sites have been buried beneath thick sediment layers.
Discussion
The results suggest that prehistoric communities in the Qazvin Plain tended to occupy geomorphologically stable zones with lower sedimentation rates. This pattern likely reflects an experiential understanding of landscape dynamics, even if not formally articulated scientific knowledge. The preference for stable locations may have been shaped by the need for long-term settlement sustainability, reduced risk of flood damage, and better preservation of arable land. It should be noted that the lower number of identified sites in areas with high sediment production may result from their burial beneath sediments rather than deliberate avoidance. Therefore, site burial should be considered when interpreting spatial distribution and settlement patterns.
Furthermore, the use of machine learning techniques, particularly ANN and RF, highlight the potential of artificial intelligence along with Analytic Hierarchy Process to improve environmental reconstruction and predictive modeling in archaeology. These approaches allow researchers to capture complex, non-linear relationships that traditional statistical methods may fail to represent.
Conclusions
This study demonstrates that combining machine learning models with multi-criteria decision-making methods offers a powerful framework for understanding the interplay between environmental processes and human settlement patterns. The integrated approach not only enhances the accuracy of sedimentation modeling but also improves the reliability of archaeological predictive models.
From a practical standpoint, the findings can assist archaeologists in identifying high-probability areas for future excavations, prioritizing regions for survey, and allocating resources more efficiently. Moreover, this research underscores the importance of considering geomorphological stability as a key factor in cultural heritage management. By understanding where sites are most likely to be buried or preserved, heritage managers can design more effective conservation strategies and anticipate potential threats posed by ongoing erosion and sedimentation processes.
On a broader level, the study bridges the gap between archaeology, geomorphology, and data science, offering a replicable methodological template for other regions and time periods. Ultimately, the synergy between AI-driven modeling, PCA, AHP, GIS-based spatial analysis, and multi-criteria evaluation represents a forward-looking approach to environmental archaeology—one that not only reconstructs the past but also informs sustainable management of cultural landscapes for the future.