盗掘古墓葬犯罪的空间分异特征及防控脆弱区识别OA

Tomb robbery poses a severe threat to cultural heritage preservation,necessitating precise prevention strategies adaptable to grassroots grid-based governance.This study examines the spatial differentiation characteristics and influencing factors of tomb robbery in City B.Using 150 crime records(2011-2019)from the China Heritage Crime Information Center,it aims to identify vulnerable localities for targeted control.The"township"served as the basic analytical unit,aligning with grassroots local administrative and policing units in China.At this scale,we integrated multi-source data including demographic/economic statistics,points of interest,and transportation networks.Methodologically,spatial patterns were first identified using kernel density estimation(KDE)and Standard Deviational Ellipse(SDE)analysis.Subsequently,to build a vulnerability assessment model,we tested several machine learning classifiers(Random Forest,XGBoost,CatBoost)predicting crime occurrence(binary:0=no crime,l=crime occurred)based on theoretically derived environmental,guardianship,and population indicators.XGBoost demonstrated superior performance(Accuracy ≈ 75.83％,AUC≈ 80.02％)and informed the selection of eight key factors.Critically,we improved the traditional Vulnerable Localities Index(VLI)method by employing Shapley Additive exPlanations(SHAP)analysis on the trained XGBoost model to objectively derive data-driven weights(contributions)for these factors,replacing subjective expert scoring.The results highlight distinct spatial patterns and dynamics:(1)Tomb robbery crimes display a"broad coverage,local concentration"pattern.While 41.8％of the 122 townships recorded incidents,high-frequency townships(≥2 incidents)constituted nearly 20％,concentrated in relic-rich central/eastem regions.SDE analysis confirmed a strong spatial association between the overall crime distribution and the concentration of both national and provincial Key Protected Heritage Sites(KPSs),particularly aligning with provincial KPSs.(2)A multi-scale target selection strategy emerged:Macro-level KDE hotspots are spatially adjacent to dense clusters of KPSs.However,micro-level SHAP interpretation reveals criminals tend to bypass the well-protected core areas of these KPSs,shifting instead towards selecting more vulnerable,less-monitored targets situated in surrounding fields,reflecting rational risk-reward assessment.(3)SHAP quantified key factor impacts,identifying significant inhibitors and facilitators of crime:low population density,geographical remoteness(evidenced by negative contributions from total road length and railway presence),and low economic activity(negative from per capita industrial output)are associated with higher vulnerability,aligning with reduced guardianship.Water bodies significantly inhibit crime,likely by restricting accessibility.Conversely,farmland/forest influence was indistinct.Notably,the geographical distribution of public security authorities and cultural heritage administrations showed negligible impact on location selection at the township scale.Building upon these SHAP-derived weights,the study generated a township-level graded Prevention and Control Vulnerability Map,classified into five distinct levels using the Jenks natural breaks method.This map provides actionable intelligence directly serving grid-based governance.It offers scientific support for implementing tiered responses and dynamic adjustments based on vulnerability levels,facilitating differentiated resource allocation:prioritizing enhanced monitoring in high-vulnerability zones while maintaining standard protocols elsewhere.This data-driven framework aims to enhance the overall efficiency of regional cultural heritage protection,extending crime geography applications to rural heritage crime and offering empirical insights for optimizing policing and heritage management strategies.