Abstract: Addressing the inherent challenges in determining design parameters,the complexity of simulation modeling,and the over-reliance on craftsmanship experience in the design and construction of timber arch bridges,this study proposes a data-knowledge hybrid-driven two-stage prediction model for parameter estimation.The proposed framework integrates the SSA-XGBoost model with an expert knowledge-enhanced cascade forward BP neural network.Field-measured data from more than 130 timber arch bridges in the Fujian and Zhejiang provinces were collected and subjected to statistical analysis to identify distribution characteristics and key influencing factors of design parameters.In the first stage,an SSA-XGBoost model is developed to predict the rise-span ratio,where the Sparrow Search Algorithm (SSA) is employed to optimize hyperparameters,thereby enhancing prediction performance in complex regression contexts.In the second stage,a cascade forward BP neural network incorporating domain expert knowledge is constructed,with a modified loss function embedding empirical rules to improve prediction accuracy for root diameter ranges.Experimental evaluations on real-measured datasets reveal that the proposed approach delivers superior predictive performance,with predicted values exhibiting high agreement with actual design parameters (error rate<8.2%).The findings demonstrate that this methodology not only facilitates the assessment of structural parameter rationality by intangible cultural heritage inheritors but also supports the preservation of traditional craftsmanship.Furthermore,it offers robust theoretical and technical guidance for the parameterized design and material selection of Chinese timber arch bridges,underscoring its significant engineering applicability.