Quantifying the delayed contributions to runoff change is crucial for water resource management and ecological conservation in water-deficient regions. This study combined the advantages of sample entropy and DFI to propose a new DFI-SE method that automatically optimizes the Nmax parameter based on sample entropy convergence to identify the spatial and temporal delayed contributions of runoff in the Wuding River basin (1977-2020) of the Chinese Loess Plateau. The spatial and temporal changes in four runoff components: short (DS), intermediate (DI), long (DL), and baseline (DB) were calculated in wind-erosion sandy areas (WESAs) and loess areas (LAs). The results revealed significant variations in runoff components between WESAs and LAs. In WESAs, mean runoff contributions were DS (12 %), DI (41 %), DL (7 %), and DB (40 %), whereas in LAs, they were 43 %, 25 %, 19 %, and 13 %, respectively. The higher DS in LAs indicated a greater proportion of surface runoff and potentially contributing to increased flood risks. DI and DL, associated with interflow or preferential flow, exhibited higher volume but longer transit time in WESAs but generated more rapidly in LAs due to denser preferential pathways. Groundwater discharge (piston flow DB) played a crucial role in sustaining dry-season river flow, particularly in WESAs, where it was three times higher than in LAs. Seasonally, WESAs provide 95 % of dry-season baseflow (DI: 48 %, DB: 39 %), while LAs exhibit stronger wet-season runoff surges (DS: 68 %). The long-term trends (1977-2020) revealed a decline in total runoff, DS, and DB, but an increase in DL, coinciding with vegetation restoration after 1997 under the "Grain for Green" project. Although the baseflow increased by 6.25 % in the basin, ensuring sustainable groundwater management remains a critical challenge in this arid region. This study provides new insights into the hydrological dynamics of the Loess Plateau and informs strategies for improving water management in the future.