Cross-timescale interaction of nonstationary hydrological responses in subtropical mountainous watersheds

Hydrological responses possess watershed systematicity and divergent features at various time scales, but how to quantitatively capture the cross-timescale interaction of nonstationary discharge and sediment concentration is unclear, thus limiting insight into hydrological responses to human alterations and stochastic hydroclimatic drivers. The present study analyzed the regimes of inter-annual, intra-annual, and event-scale hydrological responses using 40-60 years data from 15 subtropical watersheds. Bayesian networks of discharge and sediment concentration were constructed, incorporating watershed characteristics, climate effects, and three timescale hydrological indicators, to implement the nonlinear probability prediction. Broadly, annual discharge and sediment concentration followed a decreasing trend, with a key inflection in 2000. Hydrological nonstationarity was diagnosed in terms of magnitude, frequency, and time based on five temporal indicators and transient types of extreme events. In Bayesian networks, the interactions of hydrological indicators were synergistic, with intraannual hydrological signatures as intermediate nodes between external drivers and annual trends, highly explaining the inconsistent discharge and sediment concentration. The sensitivity of marginal distribution in discharge and sediment concentration to watershed surface alterations was ranked in the terminal position. Nevertheless, the implementation of forest restoration could cause a 90.36% probability of enabling sediment concentration in a decreasing or stationary state under the restrictive scenario. These findings provide refinements in hydrological dynamic causality at different timescales for guiding watershed risk management.