Votre recherche

<p>Les études portant sur les évènements hydrologiques extrêmes sont d’une grande importance vu leurs nombreux impacts socio-économiques. Ces études reposent dans une large mesure sur la capacité à estimer adéquatement les risques associés à ces évènements. Dans ce cadre, l’analyse fréquentielle (AF) est une des approches les plus utilisées pour la modélisation et l’estimation des risques associés aux évènements extrêmes. Généralement, ces évènements sont caractérisés par plusieurs variables dépendantes, comme le volume, la pointe et la durée pour les crues. Par conséquent, l'analyse de chacune de ces variables séparément ne peut pas fournir une évaluation complète des risques et peut engendrer des pertes de vies humaines ou de biens associés à une sous estimation, ou une augmentation des coûts des ouvrages hydrauliques associés à une surestimation. L’AF multivariée (AFM) permet de pallier ce problème en considérant simultanément ces variables. L’AF classique est basée sur trois hypothèses à savoir l’homogénéité, la stationnarité et l’indépendance. Par ailleurs, différentes conditions non standards telles que la complexité topographique, les perturbations par les aménagements urbains et les changements climatiques peuvent influencer la réponse hydrologique. De telles conditions rendent la prédétermination des crues, par les méthodes classiques d’AFM, un exercice non efficace et mal adapté à de tels contextes. L’objectif de cette thèse consiste à proposer de nouvelles méthodes plus prometteuses pour l'analyse et la modélisation des variables hydrologiques à la fois dans un cadre multivarié et en l’absence de l’hypothèse d’homogénéité. Ces méthodes visent à contourner les limites de celles utilisées dans la littérature. Ces nouvelles méthodes, basées sur les copules, permettent une meilleure estimation des risques des extrêmes hydrologiques en tenant compte des interactions et es dépendances entre les différentes variables. Par conséquent, les gestionnaires des ressources hydriques peuvent prendre des décisions éclairées et mieux adaptées au contexte actuel. Précisément, nous nous intéressons dans une première partie à tester l’homogénéité des séries hydrologiques multivariées. Pour ce faire, nous avons proposé un nouveau test, basé sur les L moments multivariés, capable de détecter la rupture dans la structure de dépendance des variables hydrologiques. Les résultats obtenus montrent la bonne performance du test proposé sous différents scénarios. Par la suite, dans le but de modéliser l’hétérogénéité des séries hydrologiques multivariées, un modèle de mélange de copules a été mis au point. Dans ce cadre, nous avons proposé une nouvelle méthode pour l’estimation des différents paramètres du modèle. La méthode proposée est basée sur l’utilisation des algorithmes génétiques. Sa performance est illustrée sur des séries simulées. Les résultats obtenus montrent que la méthode proposée est un outil efficace capable de fournir de bonnes estimations, en particulier dans un contexte hydrologique. Finalement, deux nouveaux tests d’adéquation pour les copules multiparamètres ont été développés. Le but de ces tests est de vérifier la qualité de l’ajustement de la copule aux données et améliorer ainsi l’estimation des risques associés aux évènements extrêmes. Les résultats révèlent que les tests proposés performent bien dans le contexte hydrologique.<br /><br /> Studies of hydrological extreme events are of great importance given their many socio-economic impacts. These studies rely on the ability to adequately estimate the associated risk. In this context, frequency analysis (FA) is one of the most widely used approaches for modeling and estimating the risks associated with extreme events. Generally, these events are characterized by several correlated random variables. Therefore, analyzing each of these variables separately cannot provide a complete risk assessment and may result in loss of life or property associated with an underestimation, or an increase in the costs of hydraulic structure associated with an overestimation. Multivariate FA (MFA) overcomes this problem by simultaneously considering these variables. Classical FA is based on three assumptions namely homogeneity, stationarity and independence. However, different non-standard conditions such as topographic complexity, disturbances by urban development and climate change can influence the hydrological response. Such conditions make flood predetermination, by conventional MFA methods, an ineffective exercise and unsuitable to such contexts. The main objective of this thesis is to propose new and more promising methods for the analysis and modeling of hydrological variables both in a multivariate framework and in the lack of the homogeneity assumption. These methods aim to overcome the limitations of classical methods used in the literature. These new proposed methods, based on copulas, allow a better estimation of the risks associated to hydrological extremes by taking into account the dependence between the different variables. As a result, water resource managers can make informed decisions that are better suited to the current context. In the first part, we propose a novel statistical test for multivariate heterogeneity detection, based on copula and multivariate L-moments. A simulation study is conducted to evaluate the performance of the proposed test and to compare it with those of existing tests. Results show the ability of the proposed test to discriminate homogeneous and inhomogeneous series. In the second part, we propose a new model based on mixtures of copulas that take into account the heterogeneity of multivariate hydrological series. To estimate the components of this model, we propose a new parameter estimation approach, based on the maximum pseudo-likelihood using genetic algorithms. Results indicate that the proposed method estimates more accurately the parameters even with small sample sizes compared to the existing classical EM method. In the last part, we introduce new goodness of fit (GOF) tests specifically for multiparameter copulas and adapted to hydrometeorological context. More precisely, the proposed GOF tests are based on multivariate L moments. A simulation study is conducted to evaluate and compare the performances of the proposed tests. The results confirm the usefulness of the new GOF tests in comparison with some well-established ones.</p>

Les inondations présentent une grande menace à la sécurité humaine et matérielle. Les effets associés à ces phénomènes naturels risquent d'augmenter encore plus avec les tendances liées aux changements climatiques. Il est donc important de disposer d'outils de prévision et de prévention des crues fiables afin de mitiger les conséquences dévastatrices reliées. La mise en œuvre de ces outils implique des processus physiques assez complexes et nécessite beaucoup de données avec toute l'incertitude associée. Dans cette thèse, on explore les différentes sources d'incertitudes liée à la détermination des niveaux d'eau en rivières principalement dans un contexte de prévision où l'incertitude liée aux données de forçage est très importante. Les analyses conduites sont appliquées à la rivière Chaudière au Québec. En premier lieu, nous avons exploré les différentes sources paramétriques d'incertitude associées à la modélisation hydraulique dans un contexte de simulation avec un accent sur l'amélioration de la calibration du modèle hydraulique. Par la suite, dans un contexte de prévision opérationnel, on a évalué la propagation des sources d'incertitude de la prévision atmosphérique au modèle de rivière en passant par les prévisions hydrologiques avec des techniques probabilistes d'ensemble. La quantification de l'incertitude a montré que les données de forçage sont celles qui contribuent le plus à la description de l'incertitude dans la détermination des niveaux d'eau. L'incertitude paramétrique, dans un contexte de prévision, est quant à elle négligeable. Le recours à des prévisions d'ensemble a permis de produire une prévision de niveau d'eau assez fiable et a montré que celle-ci est fortement liée à la qualité des données qui proviennent de la chaine de prévision hydrométéorologique à l'amont du système de prévision proposé.

Cette thèse vise à améliorer notre compréhension du modèle hédonique et de son application sur les données des biens immobiliers afin d'étudier l'impact d'un événement / externalité / environnementale liée à la présence d'inondation sur la valeur des propriétés résidentielles. Étant donné que les données immobilières sont réparties dans l'espace et dans le temps, des "corrections" temporelles et spatiales sont nécessaires dans le processus de modélisation économétrique. La recherche prend appui sur l’équation de prix hédonique. L’analyse empirique recours également à l’estimateur de type différence de différences spatio-temporelles (STDID) afin d’étudier l’effet d’une inondation survenue en 1998 sur le prix des résidences dans la ville de Laval au Canada entre 1995-2007. Les résultats suggèrent que l’utilisation des informations sur les zones inondables dans le but d’évaluer l’impact des inondations sur les valeurs résidentielles n’est pas une approche nécessairement appropriée. Les conclusions suggèrent que la grande hétérogénéité des résultats notés dans la littérature n’est probablement pas étrangère à la façon de définir les résidences touchées par les inondations. Cela signifie que les recherches empiriques sur les effets des inondations sur la valeur immobilière mesurent en réalité la valeur liée à la perception du risque d'inondation et non l’effet réel de l'inondation. Les résultats suggèrent que les applications futures dans la littérature devront porter une attention particulière à la manière de définir les zones inondables et d’identifier les résidences réellement touchées.

This study aims to evaluate the effects of the Canadian Regional Climate Model’s (CRCM) spatial resolution on summer-fall floods simulation. Seven different climate simulations issued from the fourth and the fifth version of the CRCM are employed. Four different climate simulations issued from the fourth version of the CRCM (CRCM4) are compared. They are composed of two simulations driven by the Canadian General Circulation Model (CGCM) and two driven by the ERA-40c reanalysis using grid meshes of 15 km and 45 km resolutions for each driver. Three climate simulations issued from the fifth version of the CRCM (CRCM5) driven by the ERA-Interim at 0.44° (≈ 48 km), 0.22° (≈ 24 km) and 0.11° (≈ 12 km) spatial resolutions are used. All comparisons are evaluated on a daily time-step for the 1961-1990 period (for CRCM4) and for the 1981-2010 period (for CRCM5). These seven simulations (four from CRCM4 and three from CRCM5) are used as input for two hydrological models of varying complexity (HSAMI and MOHYSE). Each model is calibrated using three different objective functions based on the Kling-Gupta Efficiency criteria (KGE) to target the summer-fall floods. Three seasonal indices are used to evaluate the CRCM outputs: bias (temperature), relative bias (precipitation) and variances ratio (temperature and precipitation). In an attempt to evaluate the effects of the spatial resolution on the hydrological modelling of summer-fall floods, streamflow simulations are generated using the seven climate datasets. The generated climate-driven streamflow simulations are analysed by two performance statistics: the seasonal values of KGE and the seasonal relative biases. Summer-fall floods are evaluated through the use of four flood indicators, the 2-year, 5-year, 10-year and 20-year return periods. The results revealed an impact of spatial resolution on climate model outputs (temperature and precipitation) and on summer-fall floods simulation by the two hydrological models and the three different calibration approaches, although this can be due to other elements such as domain size and climate model driver. The flood indicators demonstrate an increase on the summer-fall floods return periods with increasing resolution from both hydrological models. On the other hand the hydrological models structure and the calibration approaches did not show significant impacts on the summer-fall floods. The results highlight the need for further research to assess the additional uncertainty due to the impacts of the climate simulations spatial resolution on hydrological studies.

Abstract. Dissolved organic carbon (DOC) trends, predominantly showing long-term increases in concentration, have been observed across many regions of the Northern Hemisphere. Elevated DOC concentrations are a major concern for drinking water treatment plants, owing to the effects of disinfection byproduct formation, the risk of bacterial regrowth in water distribution systems, and treatment cost increases. Using a unique 30-year data set encompassing both extreme wet and dry conditions in a eutrophic drinking water reservoir in the Great Plains of North America, we investigate the effects of changing source-water and in-lake water chemistry on DOC. We employ novel wavelet coherence analyses to explore the coherence of changes in DOC with other environmental variables and apply a generalized additive model to understand predictor–DOC responses. We found that the DOC concentration was significantly coherent with (and lagging behind) flow from a large upstream mesotrophic reservoir at long (> 18-month) timescales. DOC was also coherent with (lagging behind) sulfate and in phase with total phosphorus, ammonium, and chlorophyll a concentrations at short (≤ 18-month) timescales across the 30-year record. These variables accounted for 56 % of the deviance in DOC from 1990 to 2019, suggesting that water-source and in-lake nutrient and solute chemistry are effective predictors of the DOC concentration. Clearly, climate and changes in water and catchment management will influence source-water quality in this already water-scarce region. Our results highlight the importance of flow management to shallow eutrophic reservoirs; wet periods can exacerbate water quality issues, and these effects can be compounded by reducing inflows from systems with lower DOC. These flow management decisions address water level and flood risk concerns but also have important impacts on drinking water treatability.

Abstract Fluvial biogeomorphology has proven to be efficient in understanding the evolution of rivers in terms of vegetation succession and channel adjustment. The role of floods as the primary disturbance regime factor has been widely studied, and our knowledge of their effects on vegetation and channel adjustment has grown significantly in the last two decades. However, cold rivers experiencing ice dynamics (e.g., ice jams and mechanical breakups) as an additional disturbance regime have not yet been studied within a biogeomorphological scope. This study investigated the long‐term effects of ice dynamics on channel adjustments and vegetation trajectories in two rivers with different geomorphological behaviours, one laterally confined (Matapédia River) and one mobile (Petite‐Cascapédia River), in Quebec, Canada. Using dendrochronological analysis, historical data and aerial photographs from 1963 to 2016, this study reconstructed ice jam chronologies, characterized flood regimes and analysed vegetation and channel changes through a photointerpretation approach. The main findings of this study indicate that geomorphological impacts of mechanical ice breakups are not significant at the decadal and reach scales and that they might not be the primary factors of long‐term geomorphological control. However, results have shown that vegetation was more sensitive to ice dynamics. Reaches presenting frequent ice jams depicted high regression rates and turnovers even during years with very low floods, suggesting that ice dynamics significantly increase shear stress on plant patches. This study also highlights the high resiliency of both rivers to ice jam disturbances, with vegetation communities and channel forms recovering within a decade. With the uncertainties following the reach/corridor and decadal scales, future research should focus on long‐term monitoring and refined spatial scales to better understand the mechanisms behind the complex interactions among ice dynamics, vegetation and hydrogeomorphological processes in cold rivers.

Abstract The flood-prone Saint John River (SJR, Wolastoq), which lies within a drainage basin of 55 110 km 2 , flows a length of 673 km from its source in northern Maine, United States, to its mouth in southern New Brunswick, Canada. Major industries in the basin include forestry, agriculture, and hydroelectric power. During the 1991–2020 reference period, the SJR basin (SJRB) experienced major spring flood events in 2008, 2018, and 2019. As part of the Saint John River Experiment on Cold Season Storms, the objective of this research is to characterize and contrast these three major spring flood events. Given that the floods all occurred during spring, the hypothesis being tested is that rapid snowmelt alone is the dominant driver of flooding in the SJRB. There were commonalities and differences regarding the contributing factors of the three flood years. When averaged across the upper basin, they showed consistency in terms of positive winter and spring total precipitation anomalies, positive snow water equivalent anomalies, and steep increases in April cumulative runoff. Rain-on-snow events were a prominent feature of all three flood years. However, differences between flood years were also evident, including inconsistencies with respect to ice jams and high tides. Certain factors were present in only one or two of the three flood years, including positive total precipitation anomalies in spring, positive heavy liquid precipitation anomalies in spring, positive heavy solid precipitation anomalies in winter, and positive temperature anomalies in spring. The dominant factor contributing to peak water levels was rapid snowmelt.

Abstract Global warming is causing glaciers in the Caucasus Mountains and around the world to lose mass at an accelerated pace. As a result of this rapid retreat, significant parts of the glacierized surface area can be covered with debris deposits, often making them indistinguishable from the surrounding land surface by optical remote-sensing systems. Here, we present the DebCovG-carto toolbox to delineate debris-covered and debris-free glacier surfaces from non-glacierized regions. The algorithm uses synthetic aperture radar-derived coherence images and the normalized difference snow index applied to optical satellite data. Validating the remotely-sensed boundaries of Ushba and Chalaati glaciers using field GPS data demonstrates that the use of pairs of Sentinel-1 images (2019) from identical ascending and descending orbits can substantially improve debris-covered glacier surface detection. The DebCovG-carto toolbox leverages multiple orbits to automate the mapping of debris-covered glacier surfaces. This new automatic method offers the possibility of quickly correcting glacier mapping errors caused by the presence of debris and makes automatic mapping of glacierized surfaces considerably faster than the use of other subjective methods.

In the tropical environment such as Brazil, the frequency of rainfall-induced landslides is particularly high because of the rugged terrain, heavy rainfall, increasing urbanization, and the orographic effect of mountain ranges. Since such landslides repeatedly interfere with human activities and infrastructures, improved knowledge related to spatial and temporal prediction of the phenomenon is of interest for risk management. This study is an analysis of empirical rainfall thresholds, which aims to establish local and regional scale correlations between rainfall and the triggering of landslides in Angra dos Reis in the State of Rio de Janeiro. A statistical analysis combining quantile regression and binary logistic regression was performed on 1640 and 526 landslides triggered by daily rainfall over a 6-year period in the municipality and the urban center of Angra dos Reis, in order to establish probabilistic rainfall duration thresholds and assess the role of antecedent rainfall. The results show that the frequency of landslides is highly correlated with rainfall events, and surprisingly the thresholds in dry season are lower than those in wet season. The aspect of the slopes also seems to play an important role as demonstrated by the different thresholds between the southern and northern regions. Finally, the results presented in this study provide new insight into the spatial and temporal dynamics of landslides and rainfall conditions leading to their activation in this tropical and mountainous environment.

Abstract Resilience has become a cornerstone for risk management and disaster reduction. However, it has evolved extensively both etymologically and conceptually in time and across scientific disciplines. The concept has been (re)shaped by the evolution of research and practice efforts. Considered the opposite of vulnerability for a long time, resilience was first defined as the ability to resist, bounce back, cope with, and recover quickly from the impacts of hazards. To avoid the possible return to conditions of vulnerability and exposure to hazards, the notions of post-disaster development, transformation, and adaptation (build back better) and anticipation, innovation, and proactivity (bounce forward) were then integrated. Today, resilience is characterized by a multitude of components and several classifications. We present a selection of 25 components used to define resilience, and an interesting linkage emerges between these components and the dimensions of risk management (prevention, preparedness, response, and recovery), offering a perspective to strengthen resilience through the development of capacities. Despite its potential, resilience is subject to challenges regarding its operationalization, effectiveness, measurement, credibility, equity, and even its nature. Nevertheless, it offers applicability and opportunities for local communities as well as an interdisciplinary look at global challenges.

Abstract The highly fissile lithology of the rockwalls and the diversity of mass‐wasting processes provide a specific character to the active talus slopes of the northern Gaspé Peninsula since deglaciation. At a regional scale, the geology of the rockwalls, the patterns and modalities of deglaciation and the evolution towards a cold temperate morphoclimatic regime in a maritime context still influence the geomorphological dynamics of scree slopes today. At a local scale, the south–north orientation of the main coastal valleys influences insolation and exposure to prevailing winds, which in turn influence the snow cover regime and the occurrence of freeze–thaw cycles. The statistical analyses carried out from the mapping of 43 talus slopes and their geometric variables allowed the identification of significant environmental factors for the characterization of the dominant geomorphic processes: snow avalanches, frost‐coasted clast flows, debris flows and rockfalls. Slope aspect appears to be a key parameter in the nature of the processes acting on the talus slopes. East‐ and north‐facing talus slopes are generally covered by a significant snowpack in winter and the dominant processes are snow avalanches and debris flows. West‐ and south‐facing talus slopes face prevailing winds and insolation and are subject to frost‐coated clast flows, the main driver for forest regression, and rockfalls. However, the evolution of scree slopes in forested environments remains extremely complex due to the multiscale components that affect their evolution in the short, medium and long term.

Abstract Overcooled talus slopes are generally described as islands of sporadic permafrost below the lower alpine limit of permafrost. The negative thermal anomaly of the ground is mainly consecutive to the internal ventilation of the deposit, but it is also conditioned by multiple factors as topography, slope aspect and incline, openwork structure and coarseness of the deposit, air temperature, solar radiation and wind regime. Therefore, the study of the spatiotemporal dynamics of ventilation processes allows a better understanding of the phenomenon. At Cannon Cliff, New Hampshire (USA), several field visits and environmental monitoring allowed us to describe the varying nature and significance of the ventilation mechanisms that can be observed at the ground surface and associated with both the intensity and direction of the airflows in a talus debris accumulation/protalus rampart system. The thermal negative anomalies are strong enough to lower the ground temperature to the point of preserving ice during the late spring and summer seasons. The monitoring of the gradient between external (air) and internal (talus) temperatures coupled with several dendroecological and geomorphological analyses provided a complete environmental picture of the impacts, feedback and extent of the phenomenon.

Due to limitations in traditional concrete gravity dam (CGD) design, a new approach is necessary. In this study, the lean analysis as a novel approach for CGD design, considering the interaction between dam and reservoir was considered. Maximum and minimum stresses at the heel and displacement of the crest were obtained as crucial input values of bubble sorting based on seismic analysis using Finite element analysis (FEA), and the Fuzzy Analytic Hierarchy Process (FAHP). The fuzzy bubble sorting analytic process, aimed at developing a novel method for selecting the best CGD configuration, was developed. Required Criteria, Sub-Criteria and developed models were applied to optimize the body of CGD. The weight of each sub-criterion and models were calculated based on pairwise comparison matrices. The novel approach was designed in MATLAB with the OPT-CGD code to select the best CGD model. The best weight of the Criteria, for selecting the best CGD model, based on the lean construction principles was selected from 60 developed models under implicit dynamic analysis. Statistical analysis reveals a 20% reduction in the concrete mass of the case study’s optimal body compared to the traditionally designed dam.

Droughts are increasingly recognized as a significant global challenge, with severe impacts observed in Canada's Prairie provinces. While less frequent in Eastern Canada, prolonged precipitation deficits, particularly during summer, can lead to severe drought conditions. This study investigates the causes and consequences of droughts in New Brunswick (NB) by employing two drought indices: the Palmer Drought Severity Index (PDSI) and Standardized Evapotranspiration Deficit Index (SEDI)– at ten weather stations across NB from 1971 to 2020. Additionally, the Canadian Gridded Temperature and Precipitation Anomalies (CANGRD) dataset (1979–2014) was utilized to examine spatial and temporal drought variability and its alignment with station-based observations. Statistical analyses, including the Mann–Kendall test and Sen's slope estimator, were applied to assess trends in drought indices on annual and seasonal timescales using both station and gridded data. The results identified the most drought-vulnerable regions in NB and revealed significant spatial and temporal variability in drought severity over the 1971–2020 period. Trend analyses further highlighted the intensification of extreme drought events during specific years. Coastal areas in southern NB were found to be particularly susceptible to severe drought conditions compared to inland regions, consistent with observed declines in both the frequency of rainy days and daily precipitation amounts in these areas. These findings underscore the need for targeted drought mitigation strategies particularly in NB’s coastal zones, to address the region’s increasing vulnerability to extreme drought events.

Seasonal snowpack deeply influences the distribution of meltwater among watercourses and groundwater. During rain-on-snow (ROS) events, the structure and properties of the different snow and ice layers dictate the quantity and timing of water flowing out of the snowpack, increasing the risk of flooding and ice jams. With ongoing climate change, a better understanding of the processes and internal properties influencing snowpack outflows is needed to predict the hydrological consequences of winter melting episodes and increases in the frequency of ROS events. This study develops a multi-method approach to monitor the key snowpack properties in a non-mountainous environment in a repeated and non-destructive way. Snowpack evolution during the winter of 2020–2021 was evaluated using a drone-based, ground-penetrating radar (GPR) coupled with photogrammetry surveys conducted at the Ste-Marthe experimental watershed in Quebec, Canada. Drone-based surveys were performed over a 200 m2 area with a flat and a sloped section. In addition, time domain reflectometry (TDR) measurements were used to follow water flow through the snowpack and identify drivers of the changes in snowpack conditions, as observed in the drone-based surveys. The experimental watershed is equipped with state-of-the-art automatic weather stations that, together with weekly snow pit measurements over the ablation period, served as a reference for the multi-method monitoring approach. Drone surveys conducted on a weekly basis were used to generate georeferenced snow depth, density, snow water equivalent and bulk liquid water content maps. Despite some limitations, the results show that the combination of drone-based GPR, photogrammetric surveys and TDR is very promising for assessing the spatiotemporal evolution of the key hydrological characteristics of the snowpack. For instance, the tested method allowed for measuring marked differences in snow pack behaviour between the first and second weeks of the ablation period. A ROS event that occurred during the first week did not generate significant changes in snow pack density, liquid water content and water equivalent, while another one that happened in the second week of ablation generated changes in all three variables. After the second week of ablation, differences in density, liquid water content (LWC) and snow water equivalent (SWE) between the flat and the sloped sections of the study area were detected by the drone-based GPR measurements. Comparison between different events was made possible by the contact-free nature of the drone-based measurements.

Abstract Accelerating mountain glacier recession in a warming climate threatens the sustainability of mountain water resources. The extent to which groundwater will provide resilience to these water resources is unknown, in part due to a lack of data and poorly understood interactions between groundwater and surface water. Here we address this knowledge gap by linking climate, glaciers, surface water, and groundwater into an integrated model of the Shullcas Watershed, Peru, in the tropical Andes, the region experiencing the most rapid mountain‐glacier retreat on Earth. For a range of climate scenarios, our model projects that glaciers will disappear by 2100. The loss of glacial meltwater will be buffered by relatively consistent groundwater discharge, which only receives minor recharge (~2%) from glacier melt. However, increasing temperature and associated evapotranspiration, alongside potential decreases in precipitation, will decrease groundwater recharge and streamflow, particularly for the RCP 8.5 emission scenario. , Plain Language Summary Mountain regions play an important role in water supply, because meltwater from snow and ice feeds rivers during dry periods. Groundwater (water stored in the pore spaces of soils and rock), which flows into rivers, is also an important store of water in mountain areas and may help to protect water resources against the negative impacts of shrinking mountain glaciers. We used extensive field measurements and computer modeling of the Shullcas Watershed in the Peruvian Andes to determine the current and future role of groundwater in the face of climate change. Our model projects that glaciers in our study area will disappear by 2100. The loss of glacier meltwater is buffered in the short term (~30 years) by consistent groundwater flow to rivers. However, in the long term (>60 years), precipitation is expected to decrease and rising temperatures lead to increased evaporation and water use by plants. These factors reduce groundwater recharge and storage, causing dry season streamflow to drop. , Key Points Groundwater accounts for a large fraction of streamflow and only receives minor (~2%) recharge from glaciers in the study catchment in Peru As meltwater decreases, groundwater provides consistent discharge in the near term (~30 years), becoming a larger fraction of streamflow In the long term (>60 years), groundwater storage and discharge decrease in response to higher evapotranspiration and lower precipitation

Abstract. Climate models predict amplified warming at high elevations in low latitudes, making tropical glacierized regions some of the most vulnerable hydrological systems in the world. Observations reveal decreasing streamflow due to retreating glaciers in the Andes, which hold 99 % of all tropical glaciers. However, the timescales over which meltwater contributes to streamflow and the pathways it takes – surface and subsurface – remain uncertain, hindering our ability to predict how shrinking glaciers will impact water resources. Two major contributors to this uncertainty are the sparsity of hydrologic measurements in tropical glacierized watersheds and the complication of hydrograph separation where there is year-round glacier melt. We address these challenges using a multi-method approach that employs repeat hydrochemical mixing model analysis, hydroclimatic time series analysis, and integrated watershed modeling. Each of these approaches interrogates distinct timescale relationships among meltwater, groundwater, and stream discharge. Our results challenge the commonly held conceptual model that glaciers buffer discharge variability. Instead, in a subhumid watershed on Volcán Chimborazo, Ecuador, glacier melt drives nearly all the variability in discharge (Pearson correlation coefficient of 0.89 in simulations), with glaciers contributing a broad range of 20 %–60 % or wider of discharge, mostly (86 %) through surface runoff on hourly timescales, but also through infiltration that increases annual groundwater contributions by nearly 20 %. We further found that rainfall may enhance glacier melt contributions to discharge at timescales that complement glacier melt production, possibly explaining why minimum discharge occurred at the study site during warm but dry El Niño conditions, which typically heighten melt in the Andes. Our findings caution against extrapolations from isolated measurements: stream discharge and glacier melt contributions in tropical glacierized systems can change substantially at hourly to interannual timescales, due to climatic variability and surface to subsurface flow processes.