Divergent Runoff Responses to 1.5°C, 2.0°C, and 3.0°C Warming Across Headwater and Downstream Regions of the Indus River Basin

Technical Summary: Divergent Runoff Responses to 1.5°C, 2.0°C, and 3.0°C Warming Across Headwater and Downstream Regions of the Indus River Basin

Problem Statement
Mountain river basins, particularly in High Mountain Asia, are highly sensitive to climate warming, which alters runoff magnitude, timing, and phase. While previous studies have established the sensitivity of the Upper Indus Basin (UIB)—dominated by snow and glaciers—to warming, there is insufficient quantification of how these signals propagate to the downstream Lower Indus Basin (LIB). The LIB is a heavily managed, irrigation-dependent system where groundwater dependence and human regulation significantly modify natural flow. A critical knowledge gap exists regarding whether progressive global warming produces a hydrological divergence where upstream runoff gains coincide with downstream deficits. Furthermore, many assessments rely on fixed future time slices rather than equivalent Global Warming Levels (GWLs), which fail to account for varying warming rates across models and emission pathways. This study addresses these gaps by evaluating runoff responses at 1.5°C, 2.0°C, and 3.0°C GWLs across the full upper–lower basin system.

Methodology
The study employs a station-scale, warming-level framework integrating data from 35 hydrological gauging stations across the Indus River Basin (IRB).

• Data Forcing: The analysis utilizes bias-adjusted hydroclimatic forcing data from the Inter-Sectoral Impact Model Intercomparison Project 3b (ISIMIP3b) and Coupled Model Intercomparison Project Phase 6 (CMIP6). Four General Circulation Models (GCMs: CNRM-CM6-1, CNRM-ESM2-1, EC-EARTH3, MIROC6) are used under three Shared Socioeconomic Pathways (SSP1-2.6, SSP3-7.0, SSP5-8.5).
• Modeling Framework: A Temporal Fusion Transformer (TFT), a deep learning architecture designed for multi-horizon forecasting, is trained on observed monthly discharge records and the bias-adjusted climate predictors. The model incorporates 11 dynamic predictors (e.g., precipitation, temperature, radiation, humidity) and static covariates (latitude, longitude) to capture nonlinear hydroclimatic controls.
• GWL Extraction: Instead of fixed years, hydrological responses are extracted based on the attainment of specific GWLs (1.5°C, 2.0°C, 3.0°C) relative to the 1850–1900 baseline. The attainment year is defined as the midpoint of the first 20-year window exceeding the threshold.
• Diagnostics: The study quantifies annual and seasonal runoff anomalies, snowfall fraction, Positive Degree Months (PDM), warming sensitivity (m³/s per °C), and uncertainty (interquartile range, model agreement).

Key Results
The analysis reveals a robust, longitudinally differentiated hydroclimatic divergence between the UIB and LIB under progressive warming:

• Runoff Divergence:
  • Upper Indus Basin (UIB): Median annual runoff increases monotonically with warming, rising from +172.80 m³/s at 1.5°C to +725.49 m³/s at 3.0°C. The warming sensitivity is +289.40 m³/s per °C. This increase is accompanied by a widening uncertainty range (IQR increases from 46.35 to 137.02 m³/s).
  • Lower Indus Basin (LIB): In contrast, median annual runoff deficits deepen, shifting from −78.40 m³/s at 1.5°C to −438.86 m³/s at 3.0°C, with a sensitivity of −187.20 m³/s per °C.
• Seasonal Reorganization:
  • UIB: Gains are concentrated in the warm seasons (JJA and SON), driven by enhanced melt, though the magnitude of anomalies varies by GWL.
  • LIB: Losses persist across MAM–SON, with only modest positive changes in winter (DJF). The seasonal regime in the LIB becomes structurally weakened and increasingly negative.
• Cryospheric Drivers:
  • In the UIB, the runoff increase coincides with a systematic decline in snowfall fraction (from −0.03 at 1.5°C to −0.06 at 3.0°C) and a sharp rise in PDM (from +9.40 to +31.60). This indicates a shift toward rainfall-dominated regimes and enhanced thermal melt energy.
  • The LIB shows similar temperature increases but lacks the cryospheric amplification seen upstream, instead reflecting the compounding effects of evaporation and human regulation.
• Spatial Heterogeneity:
  • Station-scale analysis confirms that basin-scale signals emerge from heterogeneous local responses. UIB stations (e.g., Gilgit, Shigar) show strong positive sensitivity, while LIB stations (e.g., Guddu, Kotri) show strong negative sensitivity. Some intermediate stations show weak or non-significant signals, highlighting local-scale uncertainty.

Significance and Claims
The paper claims that the Indus River Basin does not respond to climate change as a spatially uniform system. Instead, warming drives a pronounced longitudinal reorganization where headwater runoff enhancement does not translate into downstream water availability.

• Hydroclimatic Split: The study demonstrates that the IRB functions as a coupled but functionally heterogeneous system. The UIB acts as a cryosphere-conditioned, energy-driven system where warming temporarily amplifies runoff via melt. Conversely, the LIB acts as a highly regulated system where downstream discharge is decoupled from upstream gains due to irrigation withdrawals, canal diversions, and high evaporative demand.
• Transient Nature of Gains: The authors emphasize that the projected runoff increase in the UIB should be interpreted as a transient "peak water" phase associated with enhanced melt under diminishing cryospheric storage, rather than a permanent water resource gain.
• Policy Relevance: The findings underscore that adaptation planning cannot rely solely on upstream runoff projections. The divergence poses specific risks to irrigation reliability, groundwater dependence, and food security in the LIB. The study provides a physically grounded framework for assessing risks at specific warming thresholds (1.5°C, 2.0°C, 3.0°C), offering decision-relevant insights for water management in one of the world's most stressed river systems.

Limitations
The authors note that the TFT framework captures net runoff response without explicitly separating hydrological processes (e.g., rainfall-runoff partitioning vs. glacier melt). Additionally, while GWLs improve cross-model comparability, uncertainties remain regarding GCM selection and bias correction. The study does not explicitly model glacier mass balance dynamics or reservoir operation schemes, suggesting a need for future integration with physically based models.