Recent intensified erosion and massive sediment deposition in Tibetan Plateau rivers

Rivers from the Tibetan Plateau are key headwaters for major Asian basins, supplying water, sediment, carbon, and nutrients to nearly two billion downstream people. The TP is warming at about 0.32 °C per decade, roughly twice the global average, and is undergoing rapid cryosphere degradation (glacier retreat, permafrost thaw, snowpack reduction). Existing evidence suggests increasing water and sediment fluxes and shifting riverine sediment dynamics, with implications for ecosystems, biogeochemical cycles, hazards, and landscape stability. However, long-term, basin-scale assessments are hindered by sparse and discontinuous in situ monitoring (less than ~30% of TP rivers consistently gauged) and point measurements at outlets that do not capture source-to-sink processes. The study asks how suspended sediment concentration (SSC), suspended sediment flux (SSF), and erosion–deposition patterns have changed across TP headwater basins since the mid-1980s, and what climatic and geomorphic controls drive their spatiotemporal heterogeneity. The purpose is to reconstruct a 36-year, basin-scale history of sediment mobilization, transport, and temporary storage using satellite remote sensing integrated with in situ data, thereby quantifying trends, hotspots, budgets, and the roles of cryosphere and connectivity in shaping sediment dynamics.

Prior work documents elevated sediment yields from cryospheric regions and substantial increases in water and sediment fluxes in TP rivers, with consequences for ecosystems and biogeochemical cycles. TP rivers exported sediment at rates about 1.8 times those of Greenland and 4.5 times those of pan-Arctic rivers in the late 20th to early 21st century. Yet robust model development is limited by scarce, non-public, or short in situ datasets, with most measurements at discrete headwater outlets. Studies highlight cryosphere-driven processes (glacial erosion, thermokarst, permafrost thaw) as primary drivers of increased sediment availability and transport, and emphasize the importance of hydrogeomorphic connectivity and event-driven mobilization (e.g., GLOFs, debris flows). Remote sensing advances have improved sediment mapping globally, but a comprehensive, high-resolution reconstruction for the TP headwaters has been lacking. This study builds on these gaps by integrating extensive Landsat archives with vetted in situ data to quantify SSC/SSF patterns, trends, and sediment budgets along river networks.

Data and study scope: The study focuses on mainstreams and major tributaries of large Asian and inland rivers sourced from the TP (e.g., Yangtze, Yellow, Ganges, Indus, Salween, Mekong, Amu Darya, Tarim), covering average drainage areas >1.9×10^6 km^2. It compiles long-term in situ discharge and sediment data from hydrological agencies in China and Pakistan and literature sources, with strict quality control favoring depth-integrated methods and minimally impacted sites. Ancillary datasets include MERIT DEM/Hydro for hydrography, GRWL for river widths, GLIMS for glaciers, GOODD for dams, and products on soil, land use, vegetation, temperature, precipitation, runoff, permafrost, glacier and snow; ERA5 reanalysis supplements meteorological gaps.

Satellite imagery and preprocessing: Approximately 145,908 Landsat Tier 1 surface reflectance images (L5 TM, L7 ETM+, L8 OLI) were processed in Google Earth Engine. Cross-sensor harmonization (band adjustment and cross-calibration) addressed spectral differences; Landsat 7 SLC gaps were filled via local linear histogram matching per USGS guidelines. Quality masks removed clouds, shadows, snow/ice, and saturated pixels. Water masks were extracted using multi-index methods (NDVI, NDWI, MNDWI, AWEI) with Otsu thresholding tailored to Himalayan versus other cold-region ROIs, limiting retrieval to thaw periods (LST > 0 °C) in permafrost/glacierized areas. Sub-pixel water fraction was estimated to keep RMSE < 7%. Images with cloud cover >20% were excluded.

Clustering and SSC calibration: For each ROI, river water pixels (rivers ≥90 m wide) were buffered to define analysis regions. An unsupervised K-Means clustering (five optimal clusters) classified river spectral properties monthly, accounting for seasonal and spatial variability. After normalizing reflectance across sensors, least-squares multiple regression with Lasso variable selection (cross-validated lambda) was trained separately for each cluster using 32,656 matchup pairs (75% for training) between in situ SSC and Landsat bands/ratios. This produced cluster-specific SSC models that minimize overfitting and multicollinearity while allowing river-specific corrections.

SSC and SSF reconstruction: Monthly SSC was estimated for observable river sections, then discharge-weighted annual SSC (SSCA) was computed for each cross-section. Annual discharge time series (QA) were generated from in situ monthly records, supplemented by TP-wide runoff products and ERA5; gaps were filled with 3–5-year running averages. Annual SSF for 5-km river segments was calculated as SSFA = SSCA × QA, propagated along the network source-to-outlet for 1986–2021.

Sediment budget and deposition mapping: A basin-wide sediment budget approach quantified erosion versus deposition by comparing upstream and downstream SSF in converging reaches. Proportional deposition P = D/(SSFa + SSFb) was used, where D is the flux difference indicating deposition after confluence. Rivers were stratified into upper, middle, and lower sections to aggregate annual erosion/deposition and to identify locations of persistent or potential massive deposition (e.g., proglacial zones, braided floodplains, valley basins).

Trend analysis and validation: Annual SSC and SSF trends per river section were assessed using the Mann–Kendall nonparametric test (significance via P values). Model performance was evaluated with 19,690 independent satellite–in situ SSC pairs, yielding median relative errors of 0.19–0.32 across clusters, an overall relative error ~0.24, and RMSE ~0.50. Uncertainty sources (sampling mismatches, environmental noise, flood events, classification errors, discharge gaps) were mitigated via stringent QA/QC, thaw-period filtering, clustering, and river-specific calibrations.

• Spatial heterogeneity: SSCs decline from northwest to southeast with an area-weighted difference of 1.78 ± 0.27 kg/m^3. High-SSC hotspots align with glacier- and permafrost-dominated headwaters.
• Basin sediment yields: TP-west (Westerlies) basins show larger yields (e.g., Amu Darya 229.83 ± 56.01 t/km^2/yr; Indus 891.69 ± 125.86 t/km^2/yr) than TP-east (monsoon) basins (e.g., Yellow 30.67 ± 12.98 t/km^2/yr; Salween 137.55 ± 29.14 t/km^2/yr).
• Temporal trends: From 1986–2021, SSC increased at 6.02 ± 2.85 % per decade, outpacing runoff increases of 5.53 ± 0.51 % per decade. In western glacierized basins, upstream SSC increases (4.53 ± 3.26 %/decade) exceeded downstream (1.53 ± 0.48 %/decade), decoupled from runoff. In eastern monsoonal basins, SSC more closely tracked runoff with higher upstream increases (1.50 ± 0.67 %/decade) than downstream (0.63 ± 0.23 %/decade).
• SSF increases in 63% of river sections (P < 0.05). Notable decadal SSF increases include Indus 17.87 ± 7.41 %, Amu Darya 24.57 ± 19.32 %, Ganges 11.27 ± 6.42 %, Brahmaputra 9.46 ± 5.42 %, Salween 18.36 ± 8.23 %, Yangtze 15.21 ± 6.83 %, Mekong 6.89 ± 3.71 %.
• Cryosphere controls: Average SSC correlates positively with glacierization and snowmelt runoff ratio (P < 0.05). Highly glaciated basins showed very high SSC: Indus 6.80 ± 0.35 kg/m^3 (27.61% glacier cover) and Tarim 4.06 ± 1.24 kg/m^3 (20.61% glacier cover), an order of magnitude above many ice-free landscapes.
• Permafrost effects: Thaw-driven thermokarst processes increased sediment availability and upstream transport but created local sinks (thaw subsidence, gullies, lakes) that disconnect signals downstream. In arid-cold upstream basins (e.g., upper Yangtze, Inner, Yellow, Hexi Corridor), widespread thermokarst lakes act as major sediment sinks.
• Deposition magnitude and locations: Approximately 30% of total suspended sediment flux is temporarily deposited within river systems. Structural basins around the TP margin store about -51.86 Mt/yr (~18.21% of headwater outlet export). Specific geomorphic units (braided reaches, alternating wide–narrow valleys) store about -26.34 Mt/yr (~34.33% of downstream transport). Middle reaches of the upper Yangtze accumulated -21.35 Mt (about -21.34% of downstream discharge). Wide valleys of midstream Brahmaputra stored about -45.29% of annual sediment supply.
• Underestimation by outlet stations: Using only outlet stations underestimates SSC by 7.81%–43.52% and SSF by 9.01%–46.85%, due to within-basin deposition and transport limitations.
• Event-driven remobilization: Extreme rainstorms or meltwater floods (e.g., GLOFs) can flush previously stored sediments, abruptly shifting erosion–deposition regimes; a 2016 GLOF in the Bhotekoshi/Sunkoshi triggered a ~30-fold local sediment flux increase.
• Hydrogeomorphic connectivity: Connectivity between hillslopes, proglacial zones, thermokarst features, and channels governs transport efficiency; deposition increases channel shallowing and bar growth, reducing conveyance and flood capacity.

The study demonstrates that TP river sediment dynamics are strongly heterogeneous across space and time, primarily governed by climate-driven cryosphere processes and hydrogeomorphic connectivity rather than hydraulics alone, especially in western glacierized basins. By reconstructing SSC and SSF along river networks, the work reveals that increased warming-driven erosion coexists with substantial in-channel and valley deposition, which decouples upstream signals from downstream export. This explains why outlet-based measurements underestimate true erosion and miss critical source-to-sink adjustments. The west–east gradient reflects differing climatic regimes and landscape states: western basins exhibit high SSC and strong upstream increases driven by glacier erosion and debris production; eastern monsoonal basins show rainfall-driven mobilization and trends more closely tied to runoff. Permafrost degradation in eastern and northern headwaters expands erodible areas and new flow paths, enhancing hillslope–channel coupling but also creating numerous local sinks (thermokarst lakes and thaw features) that interrupt transport. The findings are significant for predicting hazard cascades (e.g., GLOFs, debris flows), safeguarding hydropower and infrastructure in narrowing–widening valley nodes and braided systems, and managing downstream water quality and ecosystem health. The documented temporary storage (~30%) implies latent sediment that could be rapidly remobilized during extreme events, raising flood stages and bed aggradation risk. The remote sensing approach resolves basin-internal variability and offers a template for monitoring sediment responses to continued warming in other high mountains and polar regions.

This work provides a four-decade, basin-scale reconstruction of suspended sediment concentrations, fluxes, and erosion–deposition budgets across major TP headwaters using harmonized Landsat imagery integrated with vetted in situ data and river-specific machine-learning calibration. It reveals (1) a pronounced west–east heterogeneity in SSC and SSF linked to glacierization, permafrost, precipitation, and connectivity; (2) significant SSF increases in most rivers since 1986; and (3) widespread temporary in-channel and valley deposition, totaling roughly 30% of suspended flux, which causes major underestimation by outlet-only measurements and poses latent hazard potential. The study advances understanding of cryosphere–sediment–hydrology coupling and identifies key geomorphic units that regulate transport capacity. Future directions include: expanding in situ networks for discharge and sediment in headwaters; integrating higher-cadence sensors and SAR to reduce cloud/shadow gaps; explicitly modeling connectivity and thresholds for remobilization; quantifying vegetation and land-use contributions disentangled from climate signals; and developing early-warning systems for event-driven sediment pulses affecting water quality, ecosystems, and hydropower safety.

• Monitoring gaps: In situ discharge and sediment records are sparse, temporally discontinuous, and concentrated at outlets; some discharge gaps were filled with running averages, adding uncertainty.
• Remote sensing constraints: Landsat 30 m resolution excludes sub-30 m streams and narrow channels, potentially missing small tributary contributions and causing spatial discontinuities. Cloud, shadow, snow/ice, and mountain terrain introduce noise despite QA/QC.
• Calibration transferability: Although clustering and river-specific calibrations reduce bias, residual errors arise from timing mismatches between sampling and image acquisition, spectral variability, and environmental anomalies.
• Ancillary data inconsistencies: Supplementary datasets (soil, vegetation, climate, permafrost, glacier) have varying spatiotemporal resolutions and uncertainties, complicating attribution analyses.
• Process attribution: The study cannot directly isolate vegetation restoration or human activities from climate-driven effects because of strong coupling of drivers; contributions of greening, overgrazing, damming, and deforestation remain uncertain locally.
• Hydrological connectivity simplifications: Sediment budget computations rely on flux differences and proportional deposition parameters that simplify complex storage–remobilization dynamics and do not fully resolve short-term event-scale processes.