Mathias Bellat

Past human spatial dynamics are strongly influenced by water availability. Since the development of agrarian sedentary societies, water is seen as an essential ressource. Yet quantifying its influence on past populations across different environmental and chronological contexts remains methodologically challenging. This study investigates the relative importance of thirteen water-related variables alongside five additional environmental covariates in shaping archaeological site distributions. Using machine learning model of Maximum Entropy (MaxEnt), a total of 461 Bronze Age and Iron Age (3,500 - 50 BCE) sites from four regions with contrasting climatic conditions - the Rhine Valley, Lazio, the Shephelah, and the Kurdistan region of Iraq - were analysed. Results show that distance to water flow and stable topographic variables consistently emerge as the strongest predictors of settlement location. Palaeoclimatic variables, particularly precipitation seasonality and precipitation of the wettest quarter, outperform modern remote sensing indices. Despite contrasting environmental conditions, the four regions share broadly comparable settlement trajectories, with a consistent shift toward greater site density and spatial dispersion from the Bronze Age to the Iron Age. These findings highlight the value of temporally robust environmental proxies for modelling past human-landscape interactions and offer a replicable framework for water-focused archaeological predictive modelling across diverse climatic settings.

We present the first detailed soil property maps at multiple depths for the northwestern autonomous Kurdistan region of Iraq (Dohuk). A total of 532 soil samples from 122 sites were collected at five depth increments (0–10, 10–30, 30–50, 50–70, and 70–100 cm), and their mid-infrared (MIR) spectra were measured. A subset of 108 samples, selected via Kennard–Stone sampling, was analysed in a laboratory on ten soil properties. A Cubist model was trained and used from these measured values to predict all samples’ soil properties from their MIR spectra. Digital soil mapping was conducted using various machine learning regression techniques (ensemble learning, linear classifier, nearest neighbour classifier, decision trees), trained on the predicted soil properties and using a total of 85 covariates at 25 m pixel resolution, resulting in 50 prediction maps in total. Results were compared with the SoilGrids 2.0 product and a regional texture model. Soil depth was also mapped using a quantile random forest with 26 covariates. Our regional model outperformed global SoilGrids 2.0 predictions in resolution and accuracy, with texture RMSEs (sand: ∑RMSE = 9.35; silt: ∑RMSE = 6.8; clay: ∑RMSE = 10.28) comparable to local models. Quantile random forest achieved the best performance in 51 % of the models, and key predictors included Sentinel 2 SWIR, EVI, NDVI, and SAVI. Spatial patterns reflected the contrast between the flat areas of the Simele and Zakho plains, as opposed to the shallower and steeper Little Khabur Valley and anticline formations. Furthermore, the soil depth prediction model (R² = 0.57; RMSE = 2.59 cm^-0.5) showed strong correlation with slope and a similar pattern distribution with deeper soils in the flat areas of the Simele and Zakho plains, while shallow soils are visible in the anticline and strongly erodible areas. Our comprehensive dataset (Bellat et al., 2024a, b, c, d, 2025) offers substantial insights for soil knowledge in the region, as well as for aridic and semi-aridic areas.