The Tibetan Plateau, the source of major Asian rivers, faces increasing risks of hydrological extremes due to climate change. Seasonal hydrological forecasts are crucial for water resource management and food security. Seasonal catchment memory extends the lead times and predictability of streamflow forecasts. Understanding catchment response to precipitation is vital for improving forecast skill. Incoming precipitation is retained in various forms (soil water, snow, ice, groundwater) and released as streamflow and evapotranspiration (ET). This water storage impacts the catchment's response to subsequent precipitation, exhibiting memory. Catchment memory is categorized into multiyear (long-term) and seasonal (short-term) memory. Lagged correlation methods describe memory but can't establish functional relationships. Alternative methods calculate parameters like baseflow index and recession coefficient, focusing on catchment state for forecasting but neglecting memory processes and durations. Hysteresis loops provide qualitative analysis but lack quantitative assessment of memory duration. Current methods lack a well-defined approach for characterizing seasonal catchment memory's process and duration. This study addresses the linkage among precipitation, streamflow, and terrestrial water storage (TWS) to understand how catchments memorize precipitation and quantifies seasonal catchment memory's process and duration using a precipitation memory curve, influence time, and domination time.

Previous research has explored catchment memory using various methods, including lagged correlations between hydrological variables and the analysis of hysteresis loops. Studies have utilized hydrological models to simulate water storage and have increasingly incorporated Gravity Recovery and Climate Experiment (GRACE) satellite data for more accurate TWS measurements. GRACE data has shown to be applicable to the Tibetan Plateau, but there is still a lack of a standardized method for characterizing the process and duration of seasonal catchment memory, particularly in the complex hydrological systems of the Tibetan Plateau. The existing methods either focus on correlations or specific parameters without fully addressing the memory process and its duration. This research builds on the prior work by offering a more comprehensive model that addresses these shortcomings.

This study analyzes eight upstream rivers in the Tibetan Plateau: upper Brahmaputra River (UBR), Salween River (SWR), Lancang River (LCR), upper Yangtze River (YTR), upper Yellow River (YLR), upper Tarim River (TRM), upper Indus River (UIR), and upper Amu Darya (AMU). The basins were classified into precipitation-dominated and non-precipitation-dominated groups based on their water supply sources. A modified Variable Infiltration Capacity (VIC) Macroscale Hydrological Model was used to simulate runoff and evapotranspiration (ET). The model was calibrated using a multi-objective optimization algorithm (ε-NSGA-II) and validated by comparing simulated streamflow and ET against observed data. GRACE terrestrial water storage anomaly (TWSA) data (2003-2018) was used to evaluate the VIC model's reproduction of TWSC. The VIC-simulated streamflow, GRACE-derived TWSA, and precipitation data were used to analyze hysteresis loops to illustrate the relationships among these variables. A new precipitation-to-TWSC model was developed to describe the catchment memory process, using a precipitation memory curve to quantify the impact of precipitation on TWSC. Influence time and domination time metrics were introduced to quantify the duration of seasonal catchment memory, considering both the memory process and intra-annual precipitation distribution. For non-precipitation-dominated basins, a revised model incorporating a temperature index function was used due to the significant influence of temperature on these systems. Multiple precipitation datasets were employed to analyze the model's uncertainty, and the Mann-Kendall method was used to assess temporal trends in GRACE-derived TWSA, streamflow, precipitation, and ET. The glacier melting water was calculated using a simple temperature-index model. The VIC model's water budget approach was used to estimate TWSC, and uncertainty was assessed based on the uncertainty in GRACE TWSA.

In precipitation-dominated basins, a positive correlation between streamflow and precipitation was observed, with increasing TWSA accompanying increased precipitation. Annual hysteresis loops (TWSA-streamflow, precipitation-TWSA, precipitation-streamflow) showed clockwise (S-Q), anticlockwise (P-S), and anticlockwise (P-Q) patterns, respectively. These indicate that precipitation is temporarily stored within the catchment before being gradually released into the streamflow. The narrow variability range of the P-Q loop, compared to the P-S loop suggests that streamflow responds more rapidly to precipitation changes. Exceptions in annual hysteresis loops were observed in some years, mainly attributed to precipitation anomalies, particularly during July. In non-precipitation-dominated basins, increasing precipitation didn't correspond to increasing streamflow, as a substantial portion of precipitation is stored as snow and ice. In these basins, temperature strongly influenced TWSA and streamflow. The TRM basin showed some seasonal precipitation memory during warmer months. The precipitation-to-TWSC model accurately reproduced basin water storage variability, and it performed better than the VIC model in most basins. In precipitation-dominated basins, the influence and domination times were around 6 and 4 months during the monsoon season, and 3 and 1 month during winter. Longer influence time corresponded to a longer domination time. In non-precipitation-dominated basins, the UIR and AMU basins exhibited weak seasonal catchment memory due to meltwater dominance, while the TRM basin showed shorter influence and domination times, consistent with its steep precipitation memory curve.

The findings demonstrate distinct hydrological behavior between precipitation-dominated and non-precipitation-dominated basins on the Tibetan Plateau. The proposed precipitation-to-TWSC model, enhanced with a temperature index function for non-precipitation-dominated basins, successfully captures the seasonal catchment memory in the region. The influence and domination time metrics provide valuable insights into the duration and magnitude of the memory effect. These results highlight the importance of considering both precipitation and temperature when assessing seasonal catchment memory in high mountain river basins. The study's results have important implications for improving seasonal streamflow forecasts and water resource management in the region, providing more accurate lead times for predictions and better informed water management strategies.

This study provides a novel framework for quantifying seasonal catchment memory in high-mountain river basins. The precipitation-to-TWSC model, coupled with influence and domination time metrics, offers a more comprehensive approach than previous methods. Future research could explore the impacts of climate change and human activities on catchment memory dynamics and the model's applicability to other high-altitude regions. Further investigation into the long-term catchment memory observed in some basins is also warranted.

The study's reliance on GRACE data introduces limitations related to its spatial resolution and uncertainty, potentially affecting the accuracy of TWSC estimations, particularly in smaller sub-basins. The modified VIC model used, which does not explicitly model glaciers, might introduce some limitations for the accurate assessment of the role of glacier meltwater. The assumption of time-invariant parameters in the model for certain basins simplifies reality and might affect the model's representation of the complex hydrological processes.