Compound hydrogeomorphic cascades and rapid upstream to downstream hazard coupling in the Eastern Himalaya Authors: Kuldeep Dutta, Rupankar Rajkhowa, Mari Riba, Laxmi Linggi, Tana Tage Published: 2026 ABSTRACT AbstractExtreme precipitation in the Eastern Himalaya is increasingly associated with coupled hillslope-floodplain hazards. This study examines the 17th-18th September 2020 rainfall event in Arunachal Pradesh initiating landslides and its downstream impacts in Assam, India, using multi-sensor satellite data and long-term rainfall records. Sentinel-2 imagery was used to map landslides and debris flows, Sentinel-1 SAR data to delineate flood extent, and IMD gridded rainfall (1996–2020) to analyse rainfall spell characteristics. The event triggered widespread slope failures, localized landslide damming, and a subsequent breach, generating sediment-laden flows that inundated ~ 100 km2 of the Dhemaji floodplain. A backscatter-derived Relative Flood Volume Index (RFVI) indicates spatial variability in inundation intensity, although it does not represent absolute flood volume. Rainfall analysis suggests that antecedent wetness from preceding spells preconditioned slopes, while peak daily rainfall (> 170 mm day−1) initiated landsliding. Power-law scaling shows negligible dependence of intensity on duration (R2 ≈ 0.0004), whereas cumulative rainfall exhibits a stronger relationship with duration (R2 ≈ 0.54). These results indicate distinct roles of rainfall intensity and accumulation in controlling landslide initiation and downstream flooding, respectively, highlighting the importance of compound rainfall forcing in rapid hydrogeomorphic cascades.Keywords: Eastern Himalaya, Extreme rainfall, Cascading hazards, Landslides and debris flows, Floodplain processes, Rainfall scalingSubject terms: Climate sciences, Environmental sciences, Hydrology, Natural hazards Introduction The Himalaya and the Brahmaputra Basin are undergoing pronounced hydroclimatic changes, including rising temperatures, altered precipitation regimes, and increasing frequency of extreme events that are reshaping water availability and disaster risk1–4. Northeastern India, particularly Arunachal Pradesh and Assam, exhibits high hydroclimatic sensitivity, with evidence of warming and intensification of precipitation extremes, although long-term mean rainfall trends remain spatially heterogeneous across subregions5–7. This vulnerability is amplified by the geomorphic and atmospheric configuration of the Brahmaputra Valley, where steep surrounding terrain and an approximately 80 km wide orographic corridor funnel southwest monsoon moisture, enhancing rainfall generation. Consistent with these controls, long-term reconstructions of Brahmaputra discharge suggest that instrumental records may underestimate the recurrence of high-flow events, underscoring the basin’s inherent susceptibility to large floods and sediment-driven geomorphic change during extremes8. Multi-decadal analyses indicate increasing trends in extremely wet (R99p) days across parts of Assam and Arunachal Pradesh5, 9, 10 alongside spatially variable changes in consecutive dry days that influence antecedent moisture conditions and, in turn, both landslide initiation and flood magnitude. Attribution and projection studies further suggest a warming-driven intensification of hydrological extremes across the Eastern Himalaya, linked to increased atmospheric moisture availability and circulation variability that enhances the upper tails of rainfall distributions11, 12. The convergence of steep relief, sediment-rich river systems, and climate-sensitive communities renders the Eastern Himalaya especially vulnerable to cascading hydro-geomorphic hazard chains triggered by intense precipitation. Hazard assessments across Arunachal Pradesh consistently identify a multi-hazard environment in which extreme rainfall interacts with tectonic, geomorphic, and anthropogenic factors to produce frequent landslides and floods13, 14. Disturbances initiated in steep headwater regions can propagate downstream as cascading processes, including debris flows, landslide damming, flash floods, sediment pulses, channel aggradation, and floodplain inundation15–17. While such processes are intrinsic to mountain landscape evolution, their magnitude and frequency are increasingly amplified by climate variability and expanding human interventions18. These processes create conditions in which rainfall spells can rapidly escalate into geomorphic disasters, including debris flows, landslide damming, and flash floods as rainfall pulses travel across steep-to-plain transition zones. The Arunachal Pradesh hills and Assam plains within the Brahmaputra River basin (Fig. 1D) exemplify such a transition zone, where intensifying climate extremes and growing human exposure have multiplied the risk of coupled upstream–downstream hazards. Macro to meso-scale landslide susceptibility mapping and national agency assessments (e.g., GSI/NLSM) classify large portions of Arunachal Pradesh as moderately to very highly susceptible to landslides, consistent with observed rainfall-triggered failures along road corridors and near expanding settlements19– 21. Further downstream, prolonged moderate rainfall followed by heavy bursts is known to generate extensive inundation in Assam, highlighting the combined importance of rainfall persistence and intensity in controlling flood magnitude9, 22. Assam’s vast flood-prone zones, combined with monsoon intensification and multiple flood waves within a single season, have resulted in recurrent large-scale flood events. Analysis notes an evolving pattern of flash flood like behaviour superimposed on seasonal riverine flooding, complicating preparedness and risk management under amplified climate signals 23, 24. Overview of the 17th and 18th September 2020 cascading hazard event and study setting, (A) Severe infrastructure damage in the upstream Ego River valley following the breach of the temporary landslide dam formed during the event, (B) Pre and Post outburst google earth imagery of the landslide damming location, (C) Time series of rainfall-spell peak intensities for the year 2020, with the 17–18 September extreme rainfall window highlighted (red box), (D) The study catchment originates at an elevation of approximately 1500 m and descends southward into the Brahmaputra plains, ultimately joining the Brahmaputra River. The identified damming site (marked by a red star) aligns with the outburst flow path (indicated in yellow), along which substantial sediment deposition occurs in the mid-reach, accompanied by intense sediment flushing during the event, (E) Location of the study area within the Eastern Himalaya at the transition between the Arunachal Pradesh hills and the Assam plains, situated within the Brahmaputra River Basin; Maps and spatial analyses were generated using Google Earth Engine (https://earthengine.google.com/) and cartographically refined in QGIS Version 3.34 ‘Prizren’ LTR (https://www.qgis.org/). Within this regional context, the extreme rainfall event of 17th–18th September 2020 (Fig. 1C) in Arunachal Pradesh (Fig. 1E) provides a clear opportunity to investigate upstream to downstream hazard coupling and quantify linked hydrologic—geomorphic transformations across the Arunachal and Assam transition zone. The event triggered widespread landslides, and debris flows in the Leparada and Lower Siang districts. Just upstream of the Egocamp area, a major landslide blocked the Ego River (1B), forming a temporary landslide dam that subsequently breached and generated an outburst flood. The resulting surge devastated downstream settlements by destroying houses, bridges, and critical infrastructure (Fig. 1A) and, within 3 to 4 h, propagated into Dhemaji and parts of Lakhimpur district in Assam (Fig. 1D), causing extensive flooding, agricultural losses, and observable shifts in land-use patterns (Fig. 1C). Despite local narratives and post-event documentation of these impacts, significant gaps remain in integrating multi-sensor remote sensing, geomorphic response mapping and rainfall spell statistics across the entire disaster continuum. In particular, coupled hydrologic-geomorphic evaluations that translate multi-day rainfall persistence into slope failures, sediment pulses, channel responses, and floodplain transformations remain underdeveloped for the Arunachal and Assam corridor. Strengthening these linkages are critical for early-warning preparedness, especially in regions where short travel time allow upstream disturbances to rapidly impact downstream communities. This study addresses these gaps by combining multi-sensor satellite observations with long-term rainfall analysis to examine the hydrometeorological and geomorphic controls of cascading hazard events. Specifically, the objectives are to: (i) map co-event landslides and debris flows in upstream catchments, (ii) delineate flood extent and characterize inundation patterns in downstream areas, and (iii) analyse rainfall spell characteristics and scaling relationships to identify conditions associated with compound hazard occurrence. The methodology integrates Sentinel-1 SAR, Sentinel-2 MSI, Shuttle Radar Topography Mission (SRTM) DEM, and Indian Meteorological Department (IMD) gridded rainfall data (1996–2023), supported by statistical scaling analysis to link rainfall dynamics with observed geomorphic responses. Study area, geology and tectonics The study area (Fig. 1E) captures a tightly coupled upstream and downstream hydrogeomorphic system in the eastern Himalaya, forming an ideal natural laboratory for examining compound cascade processes and rapid hazard transmission. It extends from the steep, tectonically active hill ranges of Leparada and Lower Siang districts in Arunachal Pradesh (~ 2600 km2) to the low-gradient alluvial plains of Dhemaji and Lakhimpur districts in Assam (~ 5500 km2). The cascade pathway spans ~ 50 km, with a sharp altitudinal drop from > 1400 m in the headwaters to ~ 100 m in the plains. This pronounced physiographic gradient, combined with intense monsoon forcing, enables highly efficient sediment and water transfer, preconditioning the system for rapid upstream-to-downstream hazard coupling. During extreme events, this connectivity facilitates the transformation of localized slope failures and landslide dam outburst floods into downstream flooding, overbank sedimentation, and channel instability. Severely impacted locations during the 2020 event include Daring, Egocamp, Tapo, and Loglu in the upstream reaches, and Likabali, Silapathar, and adjoining floodplain corridors downstream. Geologically, the region is situated within the active Himalayan fold-and-thrust belt, where tectonic processes exert a first-order control on landscape evolution and hazard generation (Fig. 2). The regional structural framework is dominated by major crustal discontinuities such as the Main Central Thrust (MCT), which marks the boundary between the Higher and Lesser Himalayan sequences. In the Arunachal Himalaya, the Siyomi structural window represents a key tectonic feature, exposing the Lesser Himalayan Sequence beneath the MCT due to tectonic exhumation25–27. This zone is characterized by intense deformation, fractured rock masses, and reduced mechanical strength, making it highly susceptible to landsliding and slope failures. The landslide-damming site in Leparada district is located near the southwestern extremity of this structural window, highlighting the critical role of inherited tectonic weaknesses in controlling slope instability and initiating cascade processes. Geological framework of the study area illustrating the tectonically active Himalayan thrust belt in the upstream reaches, showing the structural complexity of the landslide damming site. The upstream terrain is predominantly composed of Lesser Himalayan Sequences (LHS), while the downstream sector transitions into the alluvium-dominated floodplains of the Assam Valley; Geological data were sourced from the Geological Survey of India, GSI, Bhukosh portal in 2025, which is now integrated to the National Geoscience Data Repository, NGDR in 2026 (https://geodataindia.gov.in/). Southward from the LLOF source region, the geological framework transitions into the Neogene foreland basin deposits of the Siwalik Group. These sediments were deposited within a foredeep formed during the Early Miocene as a result of flexural loading and thrust sheet advancement28. Continued compressional tectonics led to the development of imbricate thrust systems, duplex structures, and associated ramp-flat geometries, facilitating crustal shortening and structural thickening within the foreland basin29–31. The southward propagation of deformation, expressed through branching splays of major structures such as the MCT and the Main Boundary Thrust (MBT), has resulted in pervasive shearing, folding, and lithological heterogeneity28. These structural conditions enhance sediment production, slope instability, and connectivity between hillslopes and channel systems. Further downstream, as the river exits the Siwalik foothills and enters the Brahmaputra foreland basin, the geological regime shifts to Quaternary alluvial deposits that dominate the Dhemaji and Lakhimpur districts (Fig. 2). These deposits comprise unconsolidated sand, silt, clay, and gravel derived from high sediment flux Himalayan rivers32, 33. Their loose, poorly consolidated nature results in a highly dynamic fluvial system characterized by rapid channel migration, lateral erosion, avulsions, and recurrent flooding. The study area represents a tectonically integrated system, where active orogenesis, structural discontinuities, and sediment-rich river systems interact across spatial scales. Data and methods This study integrates multi-sensor satellite observations, gridded rainfall datasets, and statistical modelling to assess the hydrometeorological drivers and geomorphic impacts of the 17th-18th September 2020 extreme rainfall event across the Arunachal Pradesh-Assam corridor (Fig. 3). Data processing was performed in Google Earth Engine (GEE) (https://earthengine.google.com/), with supplementary R and Python workflows implemented in Visual Studio Code (VS Code) to ensure reproducibility. Landslide detection, flood mapping, rainfall spell construction, and scaling analyses followed established remote sensing and hydrometeorological approaches. Integrated methodological workflow illustrating the end-to-end analytical framework adopted in this study. The workflow combines multi-sensor satellite observations, topographic data, and long-term rainfall records, progressing from data acquisition and pre-processing to (i) landslide mapping using vegetation change (RdNDVI) and terrain constraints, (ii) SAR-based flood inundation detection and relative inundation intensity proxy estimation, and (iii) rainfall spell construction followed by power-law scaling analysis of intensity-duration and volume-duration relationships. The framework integrate remote sensing and statistical approaches for evaluating coupled hydrometeorological and geomorphic processes. Sentinel-2 Level-2A (COPERNICUS/S2_SR_HARMONIZED) and Sentinel-1 SAR GRD (COPERNICUS/S1_GRD) datasets were used for landslides, sedimentation and flood mapping. Terrain attributes were derived from the SRTM DEM (USGS/SRTMGL1_003). Surface water dynamics and hydrological structure were analysed using the JRC Global Surface Water dataset (JRC/GSW1_3/GlobalSurfaceWater) and WWF HydroSHEDS flow accumulation data (WWF/HydroSHEDS/15ACC), while vegetation dynamics were assessed using the Global Forest Change dataset (UMD/hansen/global_forest_change_2018_v1_6). Additional validation was carried out using Cloud to Street permanent water data (projects/cloudtostreet/Hiring/Remote_Sensing_Scientist/RSSTestPermWater) and high-resolution imagery from Google Earth Pro. India Meteorological Department (IMD) daily gridded rainfall data (1996–2023) (https://www.imdpune.gov.in/cmpg/Griddata/Rainfall_25_Bin.html) were used for rainfall analysis, and power-law scaling was applied to intensity-duration and volume-duration relationships. Landslide mapping Landslides were mapped using a multitemporal change-detection approach34–36 that integrated Sentinel-2 imagery with terrain derivatives from the SRTM DEM. The workflow is executed within the Google Earth Engine (GEE) Application Programming Interface (API), in conjunction with ArcMap and QGIS software platforms, to integrate and analyse the data. The process commences by overlaying the pre-prepared historical landslide inventory onto imagery sourced from Google Earth (GE) and Sentinel-2 (S2) to meticulously delineate landslide polygons. Subsequently, a co-event landslide polygon dataset (in KML/SHP format) is generated for the study. This dataset is later utilized to determine the threshold of spectral raster in semi-automatic detection. Pre-event (2019 October to 2020 September 15th) and post-event (2020 September 25th to 2021 April) Sentinel-2 scenes were filtered for low cloud cover and corrected using the cloud and cirrus mask. Normalized Difference Vegetation Index (NDVI) was computed for all cloud-free images, and a Relative Difference NDVI (Rd-NDVI) was derived between the maximum pre-event and post-event NDVI. The composites are then exploited to map slope failures using the DEM derivatives and vegetation change as proxies. To suppress non-landslide disturbances, a set of terrain constraints was applied. Slope, elevation, and aspect masks derived from the DEM were used to retain only steep, geomorphologically susceptible slopes within relevant altitude and orientation ranges. Rd-NDVI values below the defined threshold were then extracted as candidate landslide pixels. A connected pixel filter was applied to remove isolated noise and retain only spatially coherent landslide features. The final binary landslide raster was polygonised to generate discrete landslide boundaries, which were exported in GIS ready format for further spatial analysis. Flood depth, area, and volume estimation Flood inundation was mapped from Google Earth Engine using Sentinel-1 Ground Range Detected (GRD) VV-polarized imagery following a SAR terrain-fusion approach37, 38. Monthly mosaics for the September 2020 flood period were generated to ensure spatial continuity, and speckle noise was reduced using a median filter. Permanent water bodies were masked using the Hansen Global Forest Change “data mask”, retaining only dynamically inundated areas for classification. Flooded pixels were identified using an adaptive threshold derived from Otsu’s method applied to the backscatter histogram of the mosaicked VV image. To minimize terrain-induced misclassification, the initial flood mask was refined using slope and elevation constraints derived from SRTM DEM and further restricted to hydraulically plausible zones by enforcing connectivity to a river network derived from HydroSHEDS. Flood extent was computed by summing the pixel areas of all classified inundated cells. Within the GEE environment, each pixel has an associated surface area, enabling direct calculation of total inundated area, which was subsequently converted to square kilometres. Flood depth was not estimated as an absolute hydraulic quantity; instead, a relative inundation intensity proxy was derived from SAR backscatter. This proxy was defined as the difference between the Otsu threshold and the VV backscatter value for each flooded pixel (ΔVV). Given that radar backscatter generally decreases with increasing surface smoothness and open-water conditions, ΔVV provides a relative measure of inundation intensity, rather than a direct estimate of water depth. Accordingly, flood volume was not treated as a physically absolute quantity. Instead, a Relative Flood Volume Index (RFVI) was computed by integrating the depth proxy with pixel area. For each inundated pixel, the product of ΔVV and pixel area yields a relative volume contribution, and aggregation across the inundated domain provides a spatially integrated index of flood magnitude. A first-order uncertainty assessment was performed using threshold perturbation (± 1 dB), which quantifies the sensitivity of flood extent and RFVI to classification threshold selection. This represents the dominant controllable parameter in SAR-based flood mapping. However, additional sources of uncertainty including non-linear SAR backscatter response, vegetation-induced scattering effects and the absence of calibration with observed hydraulic data were not explicitly quantified and remain important limitations of the present approach. Spell construction and trend analysis A quantitative framework was developed to analyse rainfall spell characteristics, with emphasis on the statistical relationships and temporal evolution of key parameters, including spell duration, maximum daily intensity, and cumulative rainfall volume39–45. Prior to analysis, the dataset was quality-controlled by removing missing or invalid entries, and all numerical fields were standardized to ensure internal consistency. Rainfall spells were defined using a threshold-based exceedance approach designed to capture moderate-to-high intensity events relevant for hydro-geomorphic responses. Specifically, a spell was initiated when daily rainfall exceeded 30 mm day⁻1 at any grid cell within the study domain and was terminated when no grid cell satisfied this threshold. Consecutive exceedance days were grouped into a single event, such that each spell represents a temporally continuous period of elevated rainfall. The analysis is based on grid-cell-level exceedance rather than basin-averaged rainfall, thereby preserving spatial heterogeneity in rainfall occurrence. No explicit minimum dry-day separation criterion was imposed; spells terminate naturally when the exceedance condition is no longer met. To characterize scaling behaviour, power-law relationships were evaluated between spell duration (D) and (i) maximum daily intensity (I), and (ii) cumulative rainfall volume (V). A functional form of was fitted in log–log space using least-squares regression, allowing estimation of scaling coefficients and exponents. Only strictly positive values were retained to ensure mathematical validity and physical interpretability of the fitted relationships. The identified extreme event (17th-18th September 2020) was explicitly highlighted within the dataset to contextualize its position relative to the broader population of rainfall spells. Scaling relationships were visualized using log–log scatter plots with fitted power-law curves, enabling assessment of overall structure, dispersion, and deviations associated with extreme events. Uncertainty estimation and limitations Uncertainty in this study arises from data resolution, methodological assumptions, and the use of proxy-based estimations inherent in remote sensing and statistical analyses. The following approaches were adopted to identify and partially quantify these uncertainties within the analytical framework. The landslide mapping uncertainty is associated with the semi-empirical nature of the detection workflow. Thresholds applied to rdNDVI and terrain filters were calibrated using visually interpreted landslide polygons derived from satellite imagery. While this ensures internal consistency, the absence of an independent validation dataset and the dependence on spatial resolution (10 m Sentinel-2) introduce uncertainty in the delineation and representation of landslide extent. For flood mapping, uncertainty is primarily associated with the use of Sentinel-1 SAR backscatter as a proxy for inundation. The backscatter-water relationship is inherently non-linear and influenced by surface roughness, incidence angle, and vegetation structure. In densely vegetated floodplains, volume scattering from canopy elements can partially obscure the open-water signal, introducing ambiguity in flood detection and depth inference. To account for classification uncertainty, sensitivity to threshold selection was evaluated by perturbing the Otsu-derived threshold (± 10%). The resulting variability in mapped inundation extent and the Relative Flood Volume Index (RFVI) was used as a first-order estimate of classification sensitivity. Flood depth is represented using a relative backscatter difference metric; therefore, the RFVI is treated as a proxy for spatial variability in inundation intensity rather than an absolute hydraulic parameter. Rainfall-related uncertainty arises from the use of IMD gridded daily rainfall data at 0.25° spatial resolution. This dataset may smooth localized convective extremes and underrepresent peak rainfall intensities in complex terrain. Additionally, the daily temporal resolution limits the representation of sub-daily rainfall bursts that are critical for landslide initiation processes. For rainfall scaling analysis, uncertainty in power-law parameters was assessed using bootstrap resampling. This approach provides confidence intervals for the estimated scaling exponents and allows evaluation of the statistical robustness of the derived relationships. Results Hydroclimate extreme event of 17th–18th september 2020 and long-term rainfall The 17th-18th September 2020 rainfall event in Leparada and Lower Siang represents one of the intense short-duration episodes within the 1996–2020 IMD rainfall record for the region (Fig. 4). Daily rainfall exceeded ~ 170 mm day⁻1, placing the event within the upper percentiles of Himalayan orographic rainfall46. Although brief, the event occurred during a monsoon season characterized by multiple short- to moderate-duration wet spells. These closely spaced spells likely maintained elevated soil moisture and baseflow conditions, effectively preconditioning hillslopes prior to the extreme rainfall burst47. The peak intensity during the event window exceeded 170 mm day⁻1, consistent with convective rainfall associated with late-monsoon synoptic disturbances along the Eastern Himalayan front48. Long-term rainfall spell characteristics derived from IMD daily gridded rainfall data for the period 1996–2020. Variations in spell duration (days), maximum daily intensity (mm day⁻1), and cumulative rainfall volume (m3) are shown to illustrate multi-decadal variability and evolving rainfall behaviour. The 17th-18th September 2020 event (star symbol) stands out as one of the most extreme spells in the record, characterized by exceptionally high cumulative rainfall volume, above-average peak intensity, and a comparatively long duration within the 25-year dataset. Long-term rainfall spell analysis (1996–2020) indicates an increase in the frequency of rainfall spells, longer cumulative seasonal durations, and rising accumulated rainfall volumes, with more pronounced changes after ~ 2014 (Fig. 4). This period exhibits clustering of high-volume spells, suggesting a tendency toward more frequent extreme precipitation conditions. Trend analysis further indicates gradually increasing peak intensities per spell, modest increases in annual spell counts, and near-steady to slightly declining average spell duration. Collectively, these patterns point toward a regime characterized by more frequent and temporally concentrated rainfall pulses. These variations, accompanied by extended monsoon persistence into September, place the 2020 event within a broader pattern of evolving rainfall behaviour in the Eastern Himalayas49, 50. Short-term rainfall clustering prior to the September 2020 event played a critical hydro-geomorphic role. A sequence of closely spaced short-duration spells (Fig. 1B) sustained elevated discharge in upstream catchments and likely facilitated downstream transfer of sediment mobilized from earlier slope instabilities. These antecedent conditions may have reduced effective infiltration capacity at the onset of the extreme rainfall burst, resulting in shorter time-to-peak, sharper hydrographs, and more rapid hillslope responses. Late-monsoon synoptic disturbances, reinforced by orographic moisture convergence along the Brahmaputra Valley, further contributed to rainfall concentration and the observed late-season intensity peak. The analysis suggests a potential early-warning indicator, whereby geomorphic hazard likelihood increases when (a) daily rainfall exceeds local R99p thresholds and (b) two or more short-duration spells occur within a 3–7-day window, a pattern consistent with the 2020 cascade event. Within the 1996–2020 record, this event corresponds to one of the highest cumulative rainfall volumes and among the longer-duration high-intensity spells (Fig. 4), with intensity exceeding the long-term mean. Notably, periods of elevated spell intensity broadly align with major flood years (1996–1998) in the Brahmaputra Valley51, supporting an association between clustered extreme rainfall and large-scale flood responses. Upstream slope failures, debris flows, and sediment pulses Post-event observations revealed widespread slope instability across the Leparada and Lower Siang districts, with approximately 115 landslide scars mapped (Fig. 5A). Failures were concentrated along steep hillslopes, road benches, slope toes, and recently modified terrain, highlighting strong hillslope-channel coupling and the amplifying role of anthropogenic disturbances under extreme rainfall conditions. Remote sensing-based assessment of upstream slope failures and downstream floodplain impacts associated with the 17th–18th September 2020 event. (A) Semi-automatic mapping of landslides and debris flows using Sentinel-2 imagery, identifying > 150 initiation zones across the Leparada and Lower Siang districts, with pronounced clustering near the landslide-damming site. (B) Upstream flood extent delineated from Sentinel-1 SAR VV backscatter using an Otsu threshold-based classification, highlighting the location of landslide-induced channel blockage. (C) Relative Difference NDVI (rdNDVI) map of the downstream Lakhimpur and Dhemaji floodplains, indicating zones of channel adjustment and floodplain transformation following the event. (D) Downstream flood extent showing ~ 100 km2 of inundation across Dhemaji and Lakhimpur districts. (E) Identification of mid-channel and overbank siltation zones derived from rdNDVI, representing areas of significant sediment deposition. (F) Landslide damming and temporary lake formation site, illustrating the spatial relationship between slope failures and channel obstruction; Maps and spatial analyses were generated using Google Earth Engine (https://earthengine.google.com/) and cartographically refined in QGIS Version 3.34 ‘Prizren’ LTR (https://www.qgis.org/). The 17th–18th September 2020 rainfall event represents one of the intense short-duration episodes within the 1996–2020 IMD record for the region, with daily rainfall exceeding ~ 170 mm day⁻1. Although short-lived, this event occurred within a monsoon phase characterized by multiple closely spaced rainfall spells. These antecedent spells likely maintained elevated soil moisture and baseflow conditions, progressively increasing subsurface saturation and shear stress. As a result, hillslopes were effectively preconditioned prior to the extreme rainfall. Short-term rainfall clustering immediately preceding the event played a critical hydro-geomorphic role. Successive spells within a 3–7- day window sustained high antecedent wetness and reduced effective infiltration capacity, leading to rapid runoff generation and elevated pore-water pressures during the peak rainfall phase. This preconditioning significantly lowered the threshold for slope failure. Consequently, the high-intensity rainfall pulse acted as the immediate trigger, initiating shallow landslides and debris flows within hours of the intensity peak. Although the rainfall duration was limited, geomorphic impacts were substantial because landslide initiation in steep, orographic terrain is governed primarily by peak rainfall intensity acting on already saturated slopes, rather than storm duration alone52, 53. In this context, antecedent moisture buildup from prior spells and the subsequent extreme intensity acted in combination to drive widespread slope failure. Newly constructed road cuts further enhanced susceptibility by providing preferential pathways for runoff concentration and debris mobilization, while toe erosion along valley margins removed basal support and promoted additional instability. The resulting slope failures delivered large volumes of coarse and fine sediment into tributaries and main channels, generating sediment pulses that propagated rapidly downstream (Fig. 5C–E). Near the Ego Camp area, a major landslide temporarily blocked the Ego River, forming a short-lived landslide-dammed lake (Fig. 5B,F). The subsequent breach of this dam produced a secondary outburst flood, which merged with the ongoing sediment-laden flow. This sequence significantly amplified both discharge and sediment load, illustrating a characteristic cascade in high-relief Himalayan basins, where intense rainfall, slope failure, temporary damming, and outburst processes interact over short timescales54. Downstream floodplain inundation, sedimentation, and geomorphic adjustment Sediment-laden flows propagated from steep headwater reaches into the low-gradient floodplains of Dhemaji within approximately 3–4 h following the rainfall peak and the subsequent landslide-dam outburst (Fig. 5C,D). Sentinel-1 SAR-based flood mapping indicates that roughly 100 km2 of agricultural land, settlements, transport corridors, and riverine areas were inundated. Instead of directly estimating hydraulic volume, a backscatter-derived metric was employed to characterize inundation intensity. The resulting Relative Flood Volume Index (RFVI) highlights the substantial spatial extent and relative magnitude of floodplain inundation associated with the water–sediment mixture (Fig. 5D). It should be emphasized that RFVI represents relative variability rather than absolute volumetric discharge. Floodplain geomorphic changes were pronounced (Fig. 5C). Upstream sediment pulses promoted channel bed aggradation, reducing conveyance capacity and elevating water levels under low-gradient conditions. Downstream, this resulted in extensive overbank flooding, widespread siltation, development and enlargement of mid-channel bars, localized channel avulsions, and intensified bank erosion. Field-consistent impacts included deposition of coarse overbank sediments that buried croplands, disrupted transport infrastructure, and degraded soil quality across the Likabali-Silapathar-Laimekuri belt. Process synthesis and cascading hazard dynamics The extreme rainfall event of 17th-18th September 2020 triggered a tightly coupled hazard cascade that propagated rapidly from the Eastern Himalayan headwaters to the Assam floodplains. Short, intense rainfall bursts exceeding 170 mm day⁻1 occurred within a cluster of closely spaced spells, generating elevated antecedent wetness and rapid pore-pressure rise in the steep terrain of Leparada and Lower Siang districts. This hydrometeorological forcing almost immediately translated into geomorphic failures, with shallow landslides and debris flows initiating across steep hillslopes, recent road cuts, slope toes, and other pre-weakened zones. These slope failures delivered substantial volumes of unconsolidated material into channel heads and tributaries, producing sediment pulses that were efficiently transmitted downstream through strong hillslope–channel coupling (Fig. 6A). Cascading hazard processes and evolving hydroclimatic stress in the Arunachal Pradesh-Assam corridor. (A, B) Conceptual representation of the compound cascade triggered by the 17th-18th September 2020 extreme rainfall event, illustrating the progression from intense rainfall to hillslope erosion, widespread landslides, debris flows, landslide dam formation, and subsequent breach. (C) Pre- and post-event imagery of the landslide-damming site, showing channel blockage and temporary lake formation associated with slope failure. The resulting sediment-laden surge propagated downstream into the floodplains, driving channel aggradation, river course adjustments, extensive siltation, bank erosion, infrastructure damage, agricultural losses, and ecosystem disruption, as illustrated in panels (D–F). These geomorphic responses characterize the 2020 event and reflect recurring patterns of annual landscape reorganization that are progressively increasing regional hazard risk under intensifying hydroclimatic conditions; Maps and spatial analyses were generated using Google Earth Engine (https://earthengine.google.com/) and cartographically refined in QGIS Version 3.34 ‘Prizren’ LTR (https://www.qgis.org/). Within hours, these sediment-laden flows propagated into trunk channels as turbidity waves and aggradation surges. Near the Ego-camp area, a major landslide temporarily dammed the Ego River (Fig. 6C), impounding water and forming a short-lived landslide-dammed lake. The subsequent dam breach generated a secondary outburst flood, which merged with the ongoing sediment–water pulse. As these combined pulses moved from confined mountain reaches into lower-gradient floodplain sections toward Assam, channel aggradation and constricted conveyance amplified stage levels, accelerating the transition from upstream geomorphic failures to downstream floodplain inundation (6D, 6E & 6F). The cascade reached portions of the Dhemaji and Lakhimpur floodplains within approximately 3–4 h of the rainfall peak and initial hillslope failures. Sentinel-1 flood mapping indicated an inundation extent of roughly 100 km2, with an estimated RFVI of ~ 1.8 × 10⁸. The flood pulse caused rapid overbank flow, mid-channel bar growth, and severe bank erosion, reshaping floodplain morphology and depositing fresh silt and coarse sediment across agricultural lands and settlements. Transport corridors, embankments, and low-lying villages in the Likabali, Silapathar, and Laimekuri belts experienced severe impacts, consistent with a sediment-charged flood wave that prolonged inundation and delayed recession. Scaling behaviour of rainfall controls and implications for hazard thresholds Analysis of 25 years of rainfall spell statistics reveals two distinct power-law scaling relationships that differentiate the controls exerted by peak rainfall intensity and cumulative rainfall volume on hazard initiation across the Arunachal Pradesh and Assam corridor (Fig. 7). Power-law regressions were fitted following established hydrometeorological scaling approaches44, 55–57, enabling evaluation of the relative importance of rainfall metrics for landslide initiation, debris-flow generation, and downstream flooding. Empirical rainfall-scaling thresholds derived from 1996 to 2020 rainfall spell statistics for the study area. Power-law regressions illustrate the contrasting controls of spell duration on peak daily intensity and cumulative rainfall volume. The 17th and 18th September 2020 event (red box) plots within the lower intensity regime but ranks among the highest volume spells in the 25-year record. The near-zero duration dependence in the intensity—duration relationship indicates that short spells can generate extreme daily rainfall capable of triggering landslides and debris flows, whereas the strong duration dependence of the volume—duration relationship highlights the role of sustained rainfall in driving basin scale flooding and siltation. Together, these relationships provide a basis for scenario development and early warning thresholds in the Eastern Himalayan foothill to floodplain system. The first relationship links peak daily intensity (I) to spell duration (D) and follows the empirical form: The near-zero exponent (~ 0.01) indicates negligible duration dependence. This contrasts with most reported landslide intensity-duration thresholds, which typically exhibit a negative exponent, implying that longer-duration rainfall events require lower mean intensities to trigger failures58, 59. In the present study, the weak dependence of I on D reflects a multi-day wet-spell dominated climatology, where rainfall intensity varies only minimally with event duration, and variations are primarily expressed through frequency and cumulative spell length. Such behaviour is characteristic of prolonged, multi-day rainfall regimes in humid basins60–64, including the eastern Himalaya. For the 2020 event, intensities exceeded 170 mm day⁻1 even during short spells, consistent with thresholds known to trigger shallow landslides and debris flows in Himalayan settings55, 56. The second relationship links cumulative rainfall volume (V) to spell duration (D) (Fig. 7), following the scaling: The exponent (~ 1.5) reflects strong duration control on accumulated rainfall. A doubling of spell duration produces an approximately threefold increase in cumulative rainfall volume. This scaling underscores the importance of multi-day persistence in generating basin-scale flooding, downstream aggradation, and floodplain transformation61, 63–65. While landslides are primarily sensitive to peak intensities, flood magnitude and sediment-wave propagation respond more strongly to multi-day accumulation66, 67. These patterns are consistent with studies showing that multi-day rainfall intensity varies weakly with duration, whereas cumulative volume increases super-linearly with spell length. Taken together, these scaling relationships distinguish routine monsoon spells from those associated with cascading hazard responses. Peak intensity followed by persistent rainfall duration drives landslide and debris-flow initiation, whereas cumulative rainfall volume influences channel response, siltation, and floodplain inundation. Based on these patterns, indicative thresholds can be proposed for early-warning considerations. Daily rainfall in the range of ~ 100–120 mm day⁻1 aligns with previously reported landslide-triggering conditions in global and Himalayan contexts39, 41, 42, 44, 45, 68, suggesting elevated hazard potential even for short-duration events. Hazard levels may increase when such intensities occur consecutively, reflecting cumulative pore-pressure effects. Similarly, rolling 3–7-day accumulations within the upper decile of the historical distribution indicate higher likelihood of channel aggradation, overbank flow, and reduced conveyance capacity. The most hazardous scenarios arise when both conditions co-occur with one or more days exceeding high-intensity thresholds while the event also satisfies elevated multi-day accumulation. These compound conditions correspond to the hydro-geomorphic cascade observed during 17th-18th September 2020 and align with emerging frameworks for compound hazards in mountainous regions62, 69, 70. This integrated intensity-accumulation framework provides a physically consistent basis for assessing landslide and flood hazards in the Eastern Himalaya, although operational application for early warning requires further multi-event validation. Uncertainty estimation and limitations Uncertainty in this study arises from data resolution, methodological assumptions, and the use of proxy-based estimations inherent in remote sensing and statistical analyses. These uncertainties are explicitly considered within both the analytical framework and the interpretation of results. For landslide mapping, uncertainty is associated with the semi-empirical detection workflow. Thresholds applied to rdNDVI and terrain-based filters were calibrated using visually interpreted landslide polygons derived from Sentinel-2 and high-resolution imagery. While this ensures internal consistency with observed co-event failures, the absence of an independent validation dataset and formal accuracy assessment (e.g., confusion matrix, precision/recall) introduces uncertainty in landslide detection performance. The minimum mapping unit is constrained by the 10 m spatial resolution of Sentinel-2 data and subsequent post-processing filters. Although long-runout landslides are preserved using a connected-pixel approach and cross-checked against pre- and post-event imagery, the threshold-dependent and semi-automated nature of the method may influence landslide counts and mapped extents. Flood mapping uncertainty primarily stems from the use of Sentinel-1 SAR backscatter as a proxy for inundation. The backscatter-water relationship is non-linear and influenced by surface roughness, incidence angle, and vegetation structure. In densely vegetated floodplains, volume scattering from canopy elements can obscure the open-water signal, introducing ambiguity in flood detection and depth estimation. Sensitivity to classification threshold selection was evaluated by perturbing the Otsu-derived threshold (± 10%), providing a first-order estimate of variability in mapped inundation extent and the Relative Flood Volume Index (RFVI). However, flood magnitude estimation is based on a relative backscatter difference metric and does not represent absolute hydraulic depth or volume. Consequently, RFVI is interpreted as a proxy for spatial variability in inundation intensity rather than a quantitative volumetric measure. The absence of observed stage-discharge data or physically based hydraulic modelling further limits quantitative interpretation. Rainfall-related uncertainty arises from the use of IMD gridded daily rainfall data at 0.25° spatial resolution. This dataset may smooth localized convective extremes and underestimate peak rainfall intensity in complex terrain. The daily temporal resolution also limits the ability to resolve sub-daily rainfall bursts that are critical for landslide initiation. In addition, rainfall scaling relationships are conditioned by methodological choices, particularly the threshold-based spell definition (≥ 30 mm day⁻1 exceedance), which biases the dataset toward moderate-to-high intensity events and excludes lower-intensity rainfall. These factors contribute to the observed near-zero intensity-duration exponent, which should therefore be interpreted as a methodological artefact rather than a fundamental physical relationship. In contrast, the volume-duration scaling is more robust, reflecting cumulative rainfall processes that are less sensitive to temporal aggregation. Uncertainty in scaling parameters was quantified using bootstrap resampling, confirming that the intensity-duration relationship is not statistically significant, whereas the volume-duration relationship remains consistently positive. The proposed rainfall thresholds (~ 100–120 mm day⁻1 and 3–7-day accumulation windows) are empirically derived from a single event and have not been validated through hindcasting or performance-based evaluation (e.g., hit rate, false alarm ratio). Accordingly, these thresholds are interpreted as indicative rather than operational. Robust early-warning applications would require multi-event calibration and validation across diverse hydro-climatic and geomorphic settings. The analyses represent basin-scale integrated behaviour and do not explicitly resolve local heterogeneity in lithology, land use, slope modification, or anthropogenic disturbances. These factors can significantly influence landslide initiation and flood response at finer spatial scales and are not fully captured within the current framework. Discussion The results collectively indicate an evolving hydroclimatic-geomorphic regime in the Eastern Himalaya, where short-duration high-intensity rainfall and multi-day rainfall persistence interact through complementary but distinct pathways. Long-term analysis (1996–2020) suggests an increase in the frequency of high-intensity spells, clustering of high-volume rainfall events, and modest changes in spell duration, particularly after ~ 2014. These patterns point toward a monsoon regime characterized by more frequent and temporally concentrated rainfall pulses. The 17th-18th September 2020 event exemplifies this behaviour, demonstrating how extreme rainfall can rapidly translate into coupled hillslope-channel-floodplain responses61, 63–65, 71, 72. A key control on landslide initiation during this event was the combination of antecedent wetness and peak rainfall intensity. Successive short- to moderate-duration rainfall spells prior to the event likely maintained elevated soil moisture and baseflow conditions, effectively preconditioning hillslopes. This antecedent saturation reduced matric suction and infiltration capacity, enabling rapid pore-pressure buildup during the subsequent high-intensity rainfall burst (> 170 mm day⁻1). In this context, slope failure was not driven by intensity alone, but by the interaction between cumulative wetness and short-duration extreme forcing. Anthropogenic modifications, particularly road cuts and slope toe disturbances, further amplified susceptibility by concentrating runoff and reducing slope stability. The rainfall scaling analysis provides additional insight into these processes but must be interpreted cautiously. The near-zero exponent in the intensity-duration relationship indicates no meaningful statistical dependence of maximum daily intensity on event duration (R2 ≈ 0.0004; confidence interval includes zero). However, this behaviour is primarily a consequence of methodological factors, including daily temporal aggregation, spatial smoothing in gridded datasets (0.25° IMD), and the threshold-based definition of rainfall spells. As such, it should not be interpreted as a physical property of rainfall systems. In contrast, the volume-duration relationship exhibits a robust positive scaling (R2 ≈ 0.54), confirming that cumulative rainfall is strongly controlled by event duration. This distinction reinforces that while peak intensity governs rapid slope failure initiation, rainfall persistence governs basin-scale hydrological response and flood magnitude39, 41, 42, 44, 45, 68. These hydrometeorological controls are consistent with the observed geomorphic cascade. Approximately 115 landslides were mapped across Leparada and Lower Siang, with strong clustering along steep slopes, road benches, and slope toes, reflecting tight hillslope-channel coupling. Rapid pore-pressure rise during the rainfall peak triggered shallow landslides and debris flows, which delivered substantial sediment volumes into tributaries and trunk channels. These sediment inputs generated downstream-propagating pulses that were further amplified by localized geomorphic feedback. In particular, temporary channel blockage along the Ego River led to the formation of a short-lived landslide-dammed lake, followed by breach and release of an outburst surge. Although small in scale compared to large transboundary events, this local dam-breach sequence enhanced discharge and sediment load, illustrating the importance of intermediate process amplification within rainfall-driven cascades62, 69, 70. Downstream impacts in the Dhemaji floodplain highlight the sensitivity of low-gradient alluvial systems to such sediment-laden flows. Sentinel-1 SAR analysis indicates that approximately 100 km2 of floodplain were inundated. The use of a backscatter-derived RFVI provides a spatial proxy for inundation intensity73, although it does not represent absolute hydraulic volume. The observed floodplain response including overbank flooding, bar accretion, channel adjustment, and widespread silt deposition reflects the combined effects of elevated discharge and reduced channel conveyance due to rapid aggradation. The rapid propagation of the cascade from upstream rainfall peak and slope failure to downstream inundation (within ~ 3–4 h) underscores the extremely short response times characteristic of mountain-plain coupled systems. The study further distinguishes between rainfall characteristics associated with different components of the cascade. Peak daily rainfall intensity is associated with landslide and debris-flow initiation, whereas multi-day rainfall accumulation is linked to downstream flooding, sedimentation, and channel adjustment. The co-occurrence of these conditions high-intensity rainfall within a period of elevated cumulative rainfall appears to define the most hazardous scenarios. These relationships are derived from event-based analysis and should be interpreted as indicative rather than predictive thresholds. No formal hindcasting, validation, or performance evaluation (e.g., hit rate or false alarm ratio) has been conducted, and therefore these thresholds are not suitable for operational early warning. The implications for hazard monitoring are nevertheless clear. Effective risk assessment in this region requires simultaneous consideration of (i) short-duration rainfall extremes capable of triggering slope failures, and (ii) multi-day accumulation that governs basin scale hydrological response. This dual-control framework highlights the need for integrated monitoring systems spanning upstream catchments and downstream floodplains. However, the current study should be viewed as a precursor analysis toward such systems, rather than a finalized early-warning framework. Uncertainty remains an important consideration. Rainfall estimates derived from IMD 0.25° gridded data may under-represent localized convective extremes, particularly in complex orographic terrain74, 75. SAR-based flood mapping is subject to uncertainties arising from vegetation-induced scattering, surface roughness, and non-linear backscatter response76, 77. The RFVI metric, while useful for relative comparisons, cannot be interpreted as an absolute volumetric measure without hydraulic calibration. Additionally, the rainfall scaling relationships are influenced by methodological choices, including event definition and temporal resolution, which constrain their physical interpretation. Landslide mapping is based on a semi-empirical approach calibrated against visually interpreted polygons, without independent validation or formal accuracy assessment, and is limited by the spatial resolution of Sentinel-2 imagery. While large transboundary LLOFs in the Eastern Himalaya occur infrequently, smaller rainfall-triggered cascades involving localized slope failures and temporary damming appear to be increasingly significant at sub-basin to watershed scales. Concurrently, emerging evidence points toward increasing frequency of extreme precipitation events at local scales, which may enhance the likelihood of such compound cascades. This evolving hazard regime suggests that future risk will be shaped by more frequent, moderate-scale cascades that propagate efficiently through coupled mountain-floodplain systems. Future work should focus on integrating higher-resolution rainfall observations (e.g., AWS data), validating landslide inventories, and coupling hydrological and hydraulic models to better constrain flood magnitude and sediment flux. Multi-event analysis and hindcasting will be essential to evaluate the robustness of proposed rainfall indicators and to advance toward operational forecasting frameworks. Conclusions The 17th-18th September 2020 Leparada-Dhemaji event demonstrates a rapid and tightly coupled upstream–downstream cascade, in which rainfall-triggered slope failures, localized channel blockage, and sediment pulses collectively amplified downstream inundation and geomorphic impacts across the mountain-floodplain transition. The 17th-18th September 2020 Leparada-Dhemaji event demonstrates a rapid and tightly coupled upstream–downstream cascade, in which rainfall-triggered slope failures, localized channel blockage, and sediment pulses collectively amplified downstream inundation and geomorphic impacts across the mountain-floodplain transition. Landslide initiation during the event was governed by the interaction between antecedent wetness and peak rainfall intensity. While high daily rainfall (> 170 mm day⁻1) triggered rapid pore-pressure rise and slope failure, the near-zero intensity-duration scaling exponent (R2 ≈ 0.0004) reflects methodological constraints rather than a physically meaningful relationship. Landslide initiation during the event was governed by the interaction between antecedent wetness and peak rainfall intensity. While high daily rainfall (> 170 mm day⁻1) triggered rapid pore-pressure rise and slope failure, the near-zero intensity-duration scaling exponent (R2 ≈ 0.0004) reflects methodological constraints rather than a physically meaningful relationship. Floodplain response is strongly controlled by rainfall persistence. The robust volume-duration scaling (R2 ≈ 0.54) highlights the dominant role of multi-day accumulation in governing basin-scale hydrological response, sediment transfer, channel aggradation, and overbank flooding. Floodplain response is strongly controlled by rainfall persistence. The robust volume-duration scaling (R2 ≈ 0.54) highlights the dominant role of multi-day accumulation in governing basin-scale hydrological response, sediment transfer, channel aggradation, and overbank flooding. The results support a dual-control hydrogeomorphic framework, wherein (i) short-duration high-intensity rainfall drives landslides and debris flows, and (ii) multi-day cumulative rainfall governs basin scale response. The co-occurrence of these conditions defines the most hazardous scenarios; however, the identified thresholds (e.g., ~ 100–120 mm day⁻1 and elevated 3–7-day accumulation) are event-specific and indicative, and have not been validated for predictive or operational use. The results support a dual-control hydrogeomorphic framework, wherein (i) short-duration high-intensity rainfall drives landslides and debris flows, and (ii) multi-day cumulative rainfall governs basin scale response. The co-occurrence of these conditions defines the most hazardous scenarios; however, the identified thresholds (e.g., ~ 100–120 mm day⁻1 and elevated 3–7-day accumulation) are event-specific and indicative, and have not been validated for predictive or operational use. Long-term rainfall analysis (1996–2020) indicates increasing frequency of high-intensity events and clustering of high-volume rainfall spells, suggesting a shift toward more temporally concentrated precipitation regimes in the Eastern Himalaya, although trends in individual spell characteristics remain modest. Long-term rainfall analysis (1996–2020) indicates increasing frequency of high-intensity events and clustering of high-volume rainfall spells, suggesting a shift toward more temporally concentrated precipitation regimes in the Eastern Himalaya, although trends in individual spell characteristics remain modest. Findings indicate a growing likelihood of compound rainfall-driven cascade events, where short-duration extremes and antecedent accumulation jointly amplify basin scale hydrogeomorphic responses. Such cascades, even when locally generated, can propagate rapidly across connected mountain-plain systems with limited response time. Findings indicate a growing likelihood of compound rainfall-driven cascade events, where short-duration extremes and antecedent accumulation jointly amplify basin scale hydrogeomorphic responses. Such cascades, even when locally generated, can propagate rapidly across connected mountain-plain systems with limited response time. Acknowledgements Authors would like to thank Director, Centre for Earth Sciences & Himalayan Studies (CESHS), Department of Science and Technology, Government of Arunachal Pradesh for providing the laboratory and computational facility to carry out the work. Author contributions KD: Conceptualization, Data Processing, Analysis, Figures and Plots, Interpretation & Writing; RR: Data Processing, Analysis, Figures and Plots, Interpretation & Writing; MR: Field verification and information; Analysis & Writing; LL: Analysis & Writing; TT: Supervision, Writing. Data availability Data used in the study are mostly open-sourced and the processing GEE/Python/R codes, processed data will be made available upon request for further use. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement Data used in the study are mostly open-sourced and the processing GEE/Python/R codes, processed data will be made available upon request for further use.