Clarifying the spatiotemporal distribution, dominant sources, and ecological risks of soil heavy metals (HMs) is essential for effective pollution prevention and ecological restoration. This study investigated the spatial-temporal variations, source apportionment, and risk assessment of five HMs (Cd, Hg, As, Pb, Cr) in soils surrounding a gold-mining cluster in arid northwestern China. Results indicated pronounced spatial heterogeneity and an overall increasing trend in HM concentrations. Absolute principal component scores-multiple linear regression and positive matrix factorization consistently identified four major sources: gold extraction, mixed source (smelting, fuel combustion, and traffic emissions), geological background with tailings accumulation, and anthropogenic sources. The mixed source contributed the largest share of HM concentrations (42.16 %-44.75 %). By integrating Monte Carlo simulation with receptor models, a source-specific probabilistic risk assessment (PRA) framework was developed to quantify uncertainty and source-specific ecological risk. Compared with deterministic risk assessment, PRA provided a more realistic representation of contamination risk variability. As was the dominant risk-driving element, explaining 92.24 % of total risk variance. Source-specific analysis showed that geological background and mining activities contributed most to integrated ecological risk (65.03 %), followed by the mixed source (23.42 %). The cumulative probability curve of total ecological risk exhibited a constrained synergistic amplification rather than linear superposition under multi-source interactions. Overall, this study constructs a mechanism-based framework linking HMs, pollution sources, and ecological risks, clarifies priority sources and pollutants for regional management, and provides scientific guidance for precise soil pollution control and sustainable restoration in mining-impacted areas.