Extreme rainfall (ER) is intensifying under climate change, providing episodic water resources for arid and semiarid regions while simultaneously enhancing the widespread mobilization of pollutants and associated non-point source pollution risks. However, the regulatory mechanisms by which different hydrological components, particularly quickflow and baseflow, control nitrogen dynamics under ER remain poorly understood. In this study, we investigate the hydrological controls on total nitrogen (TN) concentrations at the daily scale in the Yellow River Basin, a globally representative large arid and semi-arid watershed. A distributed hydrological model (Wflow_hbv) was coupled with random forest-based interpretable machine learning to disentangle the impacts of rainfall-runoff processes on TN responses under ER. Results show that: (1) ER has significantly altered the composition of runoff in the Yellow River Basin. The contribution of quickflow increases markedly from 22.5% during non-extreme rainfall (NER) to 32.8% during extreme events, while baseflow remains a dominant component of total runoff. Concurrently, TN concentrations increase at 61% of monitoring stations. (2) Hydrological and water quality responses exhibit clear temporal asynchrony: quickflow responds in a pulse-like manner, peaking on the second day after rainfall, whereas TN responses are dominated by a first-flush effect with a short lag time of 0-1 day. (3) Under ER, baseflow emerges as the most influential hydrological factor controlling TN concentrations, contributing 14.04% to TN variability. A distinct baseflow threshold is identified (0.92 mm), below which baseflow exerts a dilution effect on TN, whereas exceeding this threshold leads to groundwater table rise and the mobilization of soluble nitrogen stored in the vadose zone into river channels, thereby accelerating nitrogen pollution. Landscape configuration further modulates this process, as highly aggregated grassland patterns (G_AI >= 93.83%) exhibit favorable water retention and pollutant interception capacities, making the relatively clean baseflow in such areas more sensitive to the "source" effect triggered by the rise in groundwater level. This study reveals a cascade mechanism linking landscape configuration, hydrological pathways, and nitrogen pollution under ER, demonstrating how altered hydrological pathways amplify nitrogen pollution risks in large arid and semi-arid river basins.