Flooding, an increasing risk in Rwanda, tends to isolate and restrict the mobility of rural communities. In this work, we developed a streamflow model to determine whether floods and rainfall anomalies explain variations in rural trail bridge use, as directly measured by in-situ motion-activated digital cameras. Flooding data and river flows upon which our investigation relies are not readily available because most of the rivers that are the focus of this study are ungauged. We developed a streamflow model for these rivers by exploring the performance of process-based and machine learning models. We then selected the best model to estimate streamflow at each bridge site to enable an investigation of the associations between weather events and pedestrian volumes collected from motion-activated cameras. The Gradient Boosting Machine model (GBM) had the highest skill with a Kling-Gupta Efficiency (KGE) score of 0.79 followed by the Random Forest model (RFM) and the Generalized Linear Model (GLM) with KGE scores of 0.73 and 0.66, respectively. The physically-based Variable Infiltration Capacity model (VIC) had a KGE score of 0.07. At the 50% flow exceedance threshold, the GBM model predicted 90% of flood events reported between 2013 and 2022. We found moderate to strong positive correlations between total monthly crossings and the total number of flood events at four of the seven bridge sites (r = 0.36–0.84), and moderate negative correlations at the remaining bridge sites (r = -0.33– -0.53). Correlation with monthly rainfall was generally moderate to high with one bridge site showing no correlation and the rest having correlations ranging between 0.15–0.76. These results reveal an association between weather events and mobility and support the scaling up of the trail bridge program to mitigate flood risks. The paper concludes with recommendations for the improvement of streamflow and flood prediction in Rwanda in support of community-based flood early warning systems connected to trail bridges.