To ensure their long-term viability and stability, complex infrastructure assets such as rail bridges require meticulous attention to critical uncertainties arising during their design process. Such uncertainties come from material properties and behaviors, variable operating conditions, and changing environmental conditions. Climate change causes severe weather conditions such as extreme temperatures, excess precipitation, convective storms, and sustained gales. This leads to various extreme actions on infrastructures like thermal expansions, differential settlements, landslides, and scour some of which are identified as leading causes of bridge failure worldwide. The main objective of the research work introduced below is to develop and implement efficient numerical approaches for designing reinforced concrete rail bridges in the multi-hazard context. Designing bridges using finite element simulations can be challenging in the presence of uncertainties due to their high associated computational costs. To overcome this, surrogate models have proved to be an effective solution, as they offer the ability to explore a large parameter design space with minimal computational resources. These proxies offer the ability to generate prompt and precise predictions of bridge performance under different design scenarios having enough available accurate bridge responses. Our work explores the performance of different surrogate modeling algorithms for the design of railroad bridges. The case study we are considering consists of a continuous reinforced concrete beam. For this beam, three different material models (elastic, plastic, and damage) are considered in combination with two loading types (static and dynamic), which means six surrogate models are being generated combining each of the three mechanical models with each of both loading types considered. Each surrogate model is developed to quantify the effect of variable material properties and bridge span lengths on different quantities of interest such as bending moments, stresses, deflections, and damage. Sampling plans are created using the LHS algorithm, and necessary data is collected through finite element simulations. To compare and analyze their performance, artificial neural networks, kernel functions algorithms (Kriging, RBF, and SVR), and least squares methods such as polynomial regression are used to generate each of the six surrogates. Accuracy measurement and comparison of these algorithms is done based on MSE calculation (mean squared error), R2 (Coefficient of determination) value, and comparing the number of needed simulations for each algorithm to reach the same accuracy level (same MSE and R2 values).