Forests are crucial for the production of high-quality freshwater resources. Complex interactions between climate change and forest processes can result in uncertainty in the availability of freshwater to downstream communities and the environment. Previous studies reported consistent increasing trends in global river discharge during the last century, which has been explained by either climate factors (usually called “hydrological intensification”) or suppressed transpiration due to CO₂-induced stomatal closure. In this study, we study long-term changes in hydrological partitioning of precipitation between evapotranspiration and runoff generation (as mm per year) along a gradient of forested watershed along the eastern temperate forest biome. The precipitation is increasing at faster rates than runoff at most of these study catchments, which suggests long-term increases in evapotranspiration. These divergent trends in precipitation versus runoff rates are significantly correlated to long-term trends in NDVI and growing season length at the watershed scale, while climate variables cannot provided significant explanation. These findings suggest that the combined effect of increased temperatures and CO₂ fertilization have led to increased leaf area and lengthened growing season, which may counteract the effect of the CO₂-induced stomatal closure across the eastern US. This study emphasizes the importance of understanding vegetation responses to climate change to predict future flow regimes in forested watersheds.

How to cite: Hwang, T., Band, L., Creed, I., and Green, M.: Greening mediates climate and CO2 induced water use efficiency effects on freshwater yield, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-3597, https://doi.org/10.5194/egusphere-egu24-3597, 2024.

How to cite: Avanzi, F., Munerol, F., Milelli, M., Gabellani, S., Massari, C., Girotto, M., Cremonese, E., Galvagno, M., Bruno, G., Morra di Cella, U., Rossi, L., Altamura, M., and Ferraris, L.: From snow to socio-hydrology: mechanisms behind the 2022 drought in the Alps, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-5634, https://doi.org/10.5194/egusphere-egu24-5634, 2024.

The hydroelectrical potential is derived from the hydraulic head and the available water discharge. In certain hydroclimatic regions, the water discharge is only a small part of the regional precipitation and the water cycle. This is particularly true when evaporation accounts for a large proportion of the regional water balance. Such conditions prevail, for example, in the catchment area of Lake Malawi and the Shire River in Malawi in south-east Africa. The country produces over 95% of its electricity from hydropower plants in the Shire River. This river is the outlet of Lake Malawi. Our study examines the sensitivity of regional water resources and hydropower generation to climate change, covering various aspects:

• Processing of rainfall, runoff and evaporation data for this tropical region, with particular attention to data scarcity.
• Hydrological simulation of the water balance and water level of Lake Malawi as well as runoff.
• Calculation of possible hydropower generation in the Shire River.
• Performing scenario calculations for climate change conditions and associated sensitivity analyses.

The most important results of these analyses are

• Between 1970 and 2013, meteorological droughts have increased in intensity and duration. This can be attributed to a decrease in precipitation and an increase in temperatures and evaporation rates.
• The hydrological system of Lake Malawi reacts to meteorological droughts with a time lag (up to 24 months), so that hydrological droughts can be predicted up to 10 months in advance by meteorological drought parameters.
• Despite the uncertainties in the regional climate projections, it is clear that the water level of Lake Malawi, as a residual of the catchment water balance, is very sensitive to changes in precipitation and evaporation.
• The discharge from the lake is a direct function of the lake's water level, and the combination of the projected decrease in precipitation and increase in temperature leads to a significantly lower flow in the Shire River.
• This suggests a future decline in annual hydropower production of between 1 % and 2.5 % (2021-2050) and 5 % and 24 % (2071-2100)
• Some projections even result that the outflow of Lake Malawi would temporarily dry up and the country's electricity supply would be interrupted.
• It is shown that regional evaporation and its changes are the key variable for assessing future water availability. This process is characterized by a particularly high degree of uncertainty.

The example of Lake Malawi basin shows that a careful hydro-climatic analysis is required to assess such sensitive hydro-systems. Global-scale analyses do not have sufficient predictive power and explanatory potential.

How to cite: Bronstert, A., Mtilatila, L., and Vormoor, K.: Hydro-electrical potential at risk und climate change: the case of Lake Malawi basin and the Shire River in Malawi / Southeast Africa, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-11698, https://doi.org/10.5194/egusphere-egu24-11698, 2024.

Persistent drought conditions may alter catchment response to precipitation, both during and after the drought period. Drought impacts on vegetation and hydrological dynamics may persist beyond the drought event, challenging accurate streamflow forecasting especially under flooding conditions. Yet, the influence of drought characteristics on the catchment response to precipitation remains unclear. In this study, we use a comprehensive dataset consisting of observations and remote sensing data of streamflow, precipitation, soil moisture, and total water storage for 3957 catchments worldwide. By employing multivariate statistical analysis, we identify significant abrupt shifts in the precipitation-streamflow relationship and examine the role of drought in driving these shifts. Our analysis shows that drought events may generally lead to significantly lower streamflow than expected from the historical norm during and after drought conditions. While warm-temperate and equatorial regions generally experience a slight decrease in streamflow during drought compared to expected values, arid regions predominantly exhibit an unexpected increase during soil moisture drought. In snow-influenced regions both increases and decreases of streamflow compared to expected were found. Notably, soil moisture drought events emerge as main drivers of hydrological regime shifts, particularly in snow-influenced and arid regions.  This study sheds light on the importance of considering regional characteristics in predicting dynamic catchment response to precipitation under and after persistent drought conditions.

How to cite: Matano, A., Barendrecht, M., Brunner, M., Hamed, R., and Van Loon, A. F.: Drought influences on hydrological regimes, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-14704, https://doi.org/10.5194/egusphere-egu24-14704, 2024.

Prolonged droughts have led to impacts on streamflow by strongly reducing flows, and in some cases gives rise to permanent shifts in streamflow regimes, despite rainfall recovery. Understanding changes from perennial to non-perennial regimes is crucial for water management, such as the allocation of available water for crop irrigation to environmental flows. Furthermore, given the prospect of a drier and warmer future, to combat increasing water scarcity and environmental degradation, understanding regime changes will help decision makers develop water-related mitigation and adaptation strategies.

In this study we investigate streamflow regime change from perennial to non-perennial flow, combining well established metrics using both climatic and hydrologic indices.  A case study of the Victorian region in south-eastern Australia will be presented applying our analysis to pre-, during and post Millennium drought conditions. We analysed streamflow from 116 Hydrological Reference Sites in Victoria and period from 1970-2019 with respect to changes in the magnitude, duration, frequency, and timing of flow conditions.  In this region, we identified rivers which remained in the non-perennial regime despite rainfall totals returning to pre drought levels. Additionally, we explored attributing for the permanent streamflow regime shift and implications for water management including interacting changes to observed rainfall intensities, vegetation responses and modelled surface and sub-surface components.

How to cite: Bende-Michl, U., Bahramian, K., Thomas, S., Rudeva, I., Sharples, W., and Carrara, E.: When rivers change their tune: unraveling streamflow regime shift after multi-year droughts in south-eastern Australia, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-22387, https://doi.org/10.5194/egusphere-egu24-22387, 2024.

As a proxy for atmospheric evaporative demand, potential evapotranspiration (E_P) is usually estimated by meteorologic elements such as radiation, temperature, and vapor pressure. We investigate the controlling factors of E_P from the perspective of catchment water balance. Through analyzing the relationships and constraint conditions of the variables in the Budyko framework and the generalized proportionality hypothesis (GPH), we demonstrate that the mean annual E_P depends on precipitation (P) and runoff (Q) and the information of E_P is contained in the water balance process. Further, we propose the Budyko-based and GPH-based hydrological approaches for E_P estimation and obtain the hydrological E_P for the MOPEX catchments. Significant linear relationships exist between the hydrological E_P and the commonly used meteorological E_P, i.e., the Penman E_P (E_P-Pen), Priestley-Taylor E_P (E_P-PT), and Hargreaves-Samani E_P (E_P-HS). Specifically, four hydrological E_P are more consistent with E_P-Pen among three meteorological E_P, and the Budyko-based hydrological E_P are more closely related to meteorological E_P than the GPH-based one. This study enriches the E_P estimation methods and provides new insight into the catchment water balance from the connection between E_P and hydrologic elements. (Supported by Project 41971049 of NSFC).

How to cite: Cheng, C., Liu, W., Li, Z., Mu, Z., and Feng, H.: Potential evapotranspiration depends on precipitation and runoff in a catchment at the mean annual scale, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-6982, https://doi.org/10.5194/egusphere-egu24-6982, 2024.

The instantaneous rate of glacier melt depends strongly on the surface albedo. For example, snowfall increases surface albedo and reduces melt rate, while dark impurities deposited on the surface enhances melt. We discuss two interesting consequences of this feedback, which lead to simplifications in models describing the runoff of glacierised catchments. 

We investigated the interannual variability of modelled streamflow on two Himalayan catchments. On an excess-precipitation year, glaciers receives more snow, which reduces melt. In contrast, on a precipitation-deficient year, glaciers have less snowcover and they produce more meltwater. This behaviour makes the annual runoff contribution from the glaciers in any given catchment to be either weakly sensitive or insensitive to the interannual variability in precipitation. Further, this characteristic is shown to be independent of the climate setting of the glacier, or the models used. A general linear-response expression for the streamflow response to climatic perturbations is proposed, where the glacierised parts respond to temperature variability and the non-glacierised parts respond to precipitation variability. This simple expression reproduces several well known characteristics of the variability of the runoff of glacierised catchments, e.g., glacier-compensation effect, buffering effect, peak-water effect etc. 

The melt enhancing effects of dark supraglacial impurities lead to the formation of tiny cylindrical cryoconite holes that are commonly seen on glacier surfaces across the globe. Their contribution to glacier mass balance and runoff generation is debated. We build an idealised model of the evolution of these holes on sunny days, and show that the holes of a given diameter reach a steady depth, which scales linearly with the diameter. The predicted depth-diameter scaling is consistent with available global-scale observations. This scaling imply that the total melt contribution of the holes to glacier runoff is likely to be negligible, and that the formation of these holes provides an effective mechanism for restricting excess melt when supraglacial impurities are present.  

How to cite: Banerjee, A.: Simplifying properties of glacier runoff resulting from Ice-albedo feedback, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-5092, https://doi.org/10.5194/egusphere-egu24-5092, 2024.

Recent disasters have highlighted a concerning possibility: the escalation of climate extremes leading to megafloods and megadroughts, commonly known as "black swans". This phenomenon, called "flood and drought risk amplification", is giving rise to impactful disasters at an increasing frequency, which is beyond our current understanding and modelling capabilities. The lack of comprehension poses a scientific challenge in pinpointing the drivers behind these amplifications.

This scenario prompts a crucial research question about the unexpected increase in flood and drought risks due to climate variability — why, where, and when it may occur. In response, this work aims to create a new modelling framework capable of unravelling the complex interplay of processes and factors contributing to flood and drought risk amplification.

Grounded in the hypothesis that local conditions play a pivotal role in risk amplification, this study focuses on identifying specific elements, known as "leverage points", where a small change or action can significantly impact the investigated system. These leverage points can be quantitatively analysed and classified based e.g. on the level of complexity of the implementation of such changes and their potential for sustainability transformation. To achieve this, this work adopts an integrated modelling approach, combining System Dynamics (SD) modelling with elements from Graph Theory and stochastic methods. SD modelling, increasingly applied in water resources planning and management, supports the mapping of the system’s feedback structure and has the potential to describe and analyse its complexity. As SD models can be represented as a directed graph of variables and their connections, Centrality Measures (e.g., degree centrality, eingenvector centrality, etc.) based on Graph Theory can help quickly and objectively pinpoint important mechanisms regardless of the size or complexity of the map. In addition, to handle uncertainty arising from the incomplete understanding of processes that contribute to risk amplification (especially the counterintuitive ones), the recent concept of Process Based Stochastic modelling will be introduced. This novel approach to uncertainty assessment involves converting the deterministic SD model into a stochastic formulation.

The proposed modelling framework will be applied to different case studies in Europe and other continents, relevant for unexplained flood and drought risk amplification, at regional and local scales. To compensate for the absence of quantitative data on some technical and non-technical factors, local stakeholders will be actively involved throughout various stages of the modelling process. Their engagement not only supplements the data gap but also aids modellers in identifying critical system components, feedback loops, and vulnerabilities.

How to cite: Coletta, V. R., Pluchinotta, I., Pagano, A., Giordano, R., Fratino, U., and Montanari, A.: An integrated modelling framework for exploring the root causes of flood and drought risk amplification by climate change, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-19889, https://doi.org/10.5194/egusphere-egu24-19889, 2024.

Aspects of landscape morphology including slope, curvature, and drainage dissection are important controls on runoff generation in upland landscapes. Over long timescales, runoff plays an essential and modifying role in shaping these same features through surface erosion. These feedbacks suggest that modeling long-term landscape evolution while accounting for hydrology could provide insight into hydrological function. Here we examine the hydrological features that emerge when runoff is equilibrated with topography, focusing particularly on the emergence and persistence of saturated areas. We use a new coupled hydro-geomorphic model that captures saturated and unsaturated zone storage and water balance partitioning between surface flow, subsurface flow, and evapotranspiration, but has numerical efficiency sufficient to drive a landscape evolution model over millions of years. Our results reveal the emergence of perennial and ephemeral stream networks, variable source areas, and even non-dendritic drainage networks under certain circumstances. When capacity for water storage and lateral drainage relative to climate are low, lower relief landscapes emerge with greater variability in the extent of saturated areas, while greater relief and less variability in saturated areas emerge as soil storage and lateral drainage capacity increase. Results from a case study suggest that emergent topography and runoff generation patterns reflect this coevolution in some places.

How to cite: Litwin, D., Tucker, G., Barnhart, K., and Harman, C.: Catchment coevolution and the geomorphic origins of variable source area hydrology, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-7612, https://doi.org/10.5194/egusphere-egu24-7612, 2024.

Numerous studies have been performed to assess climate change impacts on hydrological processes. However, a number of recent studies have pointed to some generic shortcomings of actual approaches, e.g.

• Many models have severe problems to map observed trends in groundwater head (Scanlon et al. 2018);
• Rainfall-runoff models often need to be re-calibrated after extended drought periods (Peterson et al. 2021, Fowler et al. 2022);
• Models tend to disregard the positive relationship between groundwater head and flood risk (Berghuijs and Slater 2023).

We hypothesize that these problems are related to a current underrating of the soil-groundwater-stream continuum, particularly of the role of the deep vadose zone. We combined principal component analysis, spectrum analysis and vadose zone modelling applied to more than 500 time series of groundwater head, lake level, and stream discharge in Germany, covering an area of about 90,000 km² in total, and up to 42 years.

Principal component analysis confirmed that groundwater head, lake water level and stream discharge were closely interrelated. Thus the data were merged for subsequent analyses. First order control of the spatial variability of the temporal dynamics was the degree of damping of the hydrological input signal within the vadose zone (Lischeid et al. 2021). Contrary to common assumptions especially the deeper soil layers underneath the rooting zone played an outstanding role in his regard. The degree of damping in groundwater head time series was very closely related to direction and strength of long-term trends. In contrast, there was no clear correlation, e.g., with climatic or land-use trends.

Spectrum analysis allowed to draw generalizable conclusions. From the spectrum analysis perspective the vadose zone acts as a low-pass filter of the input signal. The degree of low-pass filtering can be quantified, e.g., via spectrum analysis of the respective time series. That approach does neither require any additional site specific information nor does it depend on the specific calibration of single models. Damping of a time series via low-pass filtering inevitably results in increasing probability of significant trends even for very long periods, thus explaining the increase of probability of significant trends with thickness of the vadose zone. Another consequence of low-pass filtering is an increase of hydrological memory. For example groundwater head at a depth of 20 m below surface exhibited a memory of roughly 50 years which is massively underestimated by current models.

Groundwater head dynamics had a clear effect on discharge dynamics as well. Similarly as observed for groundwater dynamics hydrographs differed primarily with respect to the degree of damping of the hydrological input signal. Moreover, the degree of damping varied over time and reflected local groundwater head dynamics. It is remarkable that the shape coefficient of the extreme value distribution depends on the degree of damping. Consequently, not only drought risk assessment but flood risk assessment as well needs to consider explicitly deep vadose zone processes and groundwater head dynamics.

References:

Berghuijs and Slater (2023), Environ. Res. Lett., https://doi.org/10.1088/1748-9326/acbecc

Fowler et al. (2022), WRR, https://doi.org/10.1029/2021WR031210

Lischeid et al. (2021), JHyd, https://doi.org/10.1016/j.jhydrol.2021.126096

Peterson et al. (2021), Science, https://doi.org/10.1126/science.abd5085

Scanlon et al. (2018), PNAS, http://ww.pnas.org/cgi/doi/10.1073/pnas.1704665115

How to cite: Lischeid, G., Steidl, J., Weyers, J., and Kröcher, J.: Making hydrological science fit for climate change: The underrated soil-groundwater-stream continuum, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-3583, https://doi.org/10.5194/egusphere-egu24-3583, 2024.

Allometric scaling relations are widely used to link biological processes in nature. They are typically expressed as power laws, postulating that the metabolic rate of an organism scales as its mass to the power of an allometric exponent, which ranges between 2/3 and 3/4. Several studies have shown that such scaling laws hold also for natural ecosystems, including individual trees and forests, riverine metabolism, and river network organization. Here, we focus on allometric relations at watershed scale to investigate “catchment metabolism”, defined as the set of ecohydrological and biogeochemical processes through which the catchment maintains its structure and reacts to the environment. By revising existing plant size-density relationships and integrating them across large-scale domains, we show that the ecohydrological fluxes (representative of metabolic rates of a large and diverse vegetation assemblage) occurring at the catchment scale are invariant with respect to its average above-ground biomass, while they scale linearly with the basin size. We verify our theory with hyper-resolution ecohydrological simulations across the European Alps, which represent an ideal case study due to the large elevation gradient affecting the availability of energy and water resources. Deviations from the isometric scaling are observed and ascribable to energy limitations at high elevations. Remote sensing data from semiarid and tropical basins are also used to show that the observed scaling of water and carbon fluxes with size holds across a broad spectrum of climatic conditions.

How to cite: Bassani, F., Fatichi, S., and Bonetti, S.: Towards a metabolic theory of catchments: scaling of water and carbon fluxes with size, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-8242, https://doi.org/10.5194/egusphere-egu24-8242, 2024.

The quest for generalizable insights in hydrology begins with quantifying hydrological behavior in ways that are widely applicable (and thus potentially generalizable) and that also reflect key characteristics of hydrological processes.  Rainfall-runoff data sets are widely available, but runoff responses to individual precipitation events are rarely generalizable, because each mm of rain may affect streamflow differently, depending on how it fits into the sequence of past and future precipitation.  A longstanding approach to this problem is the unit hydrograph and its many variants, but these typically assume linearity (runoff response is proportional to precipitation) and stationarity (runoff response to a given unit of rainfall is identical, regardless of when it falls).  By contrast, landscape responses to precipitation are typically nonlinear and nonstationary, and quantifying this nonlinearity and nonstationarity is essential to unraveling the mechanisms and subsurface properties controlling hydrological behavior.

Here I show how the nonlinearity and nonstationarity of rainfall-runoff behavior can be quantified, directly from data, using Ensemble Rainfall-Runoff Analysis (ERRA), a data-driven, model-independent method for quantifying rainfall-runoff relationships across a spectrum of time lags.  ERRA combines least-squares deconvolution (to un-scramble each input's temporally overlapping effects) with demixing techniques (to separate the effects of individual inputs, or inputs occurring under different antecedent conditions) and broken-stick regression (to quantify nonlinear dependencies).

Applications of ERRA to experimental catchments and large multi-catchment data sets reveal that some catchments exhibit substantially greater nonstationarity and nonlinearity than others do.  ERRA also reveals that some catchments exhibit strong spatial heterogeneity in their response to precipitation, resulting from spatial heterogeneity in land use and subsurface characteristics. Results from this approach may be informative for catchment characterization and runoff forecasting; they may also lead to a better understanding of short-term storage dynamics and landscape-scale connectivity. 

How to cite: Kirchner, J.: Generalizable insights for nonlinear, nonstationary hydrological behavior using Ensemble Rainfall-Runoff Analysis (ERRA), EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-12219, https://doi.org/10.5194/egusphere-egu24-12219, 2024.

The Budyko hypothesis provides a useful framework for comprehending the behaviour of long-term water balance for a natural and closed catchment. It is widely used for partitioning precipitation into other water-cycle components, characterising hydrological response, and assessing long-term water availability.

Since, Budyko framework was developed for natural catchments assuming no storage change, its suitability for studying catchments with human-intervention in a changing climate is debated. This study aims to contribute to this debate by assessing appropriateness of Budyko framework for studying and predicting long-term water cycle changes stemming from climate change and human-driven storage changes. We simulate various climate and anthropogenic scenarios in a closed-loop environment using SWAT (Soil and Water Assessment Tool) model across three climate zones (humid, semi-arid, and arid) for more than 70 years. The scenarios reflect secular changes in climatic variables such as precipitation and temperature, and anthropogenic changes such as change in storage. The long-term time-series data for climatic variables were synthetically generated by obtaining the best-fit probability distribution, which were used to create various scenarios by trend-injection method. The model outputs were used for both steady and unsteady-state conditions to get the Budyko plots and to understand the deviation from the Budyko curve for various climate and storage change scenarios from the reference scenario.

We found that for small changes in precipitation and temperature, catchments translate along the reference Budyko curve but deviates away from the curve for a large climatic change and even a small storage change. The Budyko framework is more sensitive to precipitation change compared to temperature change. For realistic long-term storage change, particularly in arid or semi-arid regions, Budyko points in the traditional framework breach water limit. We observed that when the storage change is considered in the Budyko framework, the points again constrain itself in the Budyko space. Therefore, we developed a generalised Budyko framework by incorporating storage change in a mathematical equation considering water and energy balance. The traditional Budyko framework is a special case within it. The novel generalised Budyko framework proposed here could prove to be an indispensable tool for effective water resources management and studying catchment response to various climate change projections.

How to cite: Shaw, B., Sharma, P., and Vishwakarma, B. D.: Towards a novel water budget partitioning framework to better characterize the impact of climate and storage change on water fluxes., EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-958, https://doi.org/10.5194/egusphere-egu24-958, 2024.

Estimation of future streamflows is generally done using rainfall-runoff models to generate streamflow projections based on future climate inputs. Unfortunately, the performance of these models degrades significantly when predicting values outside of their calibration range, which undermines the credibility of projected scenarios. This abstract presents a method to analyze and improve the equations constituting a rainfall-runoff model structure in the context of climate change scenario modelling demonstrated with an application to the GR2M model and 201 catchments in South-East Australia. The method, termed "Data Assimilation Informed model Structure Improvement" (DAISI), enhances a rainfall-runoff model by combining data assimilation with polynomial updates of the state equations. The method is generic and modular, and consistently improves model performance across various metrics, including KGE, NSE on log-transformed flow, and flow duration curve bias. The updated model exhibits higher elasticity of runoff to rainfall, indicating potential significance for climate change simulations. The DAISI diagnostic identifies a reduced number of update configurations in the GR2M structure, with distinct regional patterns in three sub-regions (Western Victoria, central region, and Northern New South Wales). We suggest potential improvements for DAISI, such as incorporating additional observed variables like actual evapotranspiration to better constrain internal model fluxes.

How to cite: Lerat, J., Chiew, F., Robertson, D., Andreassian, V., and Zheng, H.: Data Assimilation Informed model Structure Improvement (DAISI) to improve prediction under climate change, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-184, https://doi.org/10.5194/egusphere-egu24-184, 2024.

Recently published literature has confirmed time and time again that machine learning (ML) algorithms (including LSTMs, GRUs, and Transformers) and conceptual lumped hydrological models (such as SAC-SMA and HBV) perform more reliably in hindcast and forecast flood prediction intercomparison experiments than more sophisticated high-resolution hydrological models. These provocative results have challenged decades of development of physics-based hydrological models for streamflow prediction, which seem more sensitive to the errors in forcing precipitation data, and the spatial description of landscape attributes. Thus, the long-standing promise that a better and more detailed understanding and description of hydrological processes would yield better predictions of streamflow fluctuations (including floods, droughts, etc.) is yet to be fulfilled. In a recently published study by our research group, we proposed and tested a methodology to benchmark ML algorithms using artificially generated data using physically-based hydrological models under very controlled conditions. Our approach combined the implementation of the hillslope-link distributed hydrological model (HLM) on a 4,500 km² basin driven by precipitation fields created using the stochastic storm transposition (SST) framework. We demonstrated that ML algorithms could effectively identify the input-output relations between the average rainfall over a basin and streamflows (as time series) at multiple sub-basin outlets under very general conditions of space-time variability of flood-generating storm systems. This result matches the reported performance by ML algorithms under a great variety of conditions.

We are extending our work to ask a new question: How reliable are trained ML algorithms and calibrated lumped hydrological models at predicting floods that have never been observed in the “historical” record? This question goes to the heart of what these black/grey-box and conceptual types of tools represent mathematically: a deterministic estimate for the input-output relationship between rainfall and streamflow. Therefore, when any of these black-box models predicts a flood there are two possible scenarios, 1) interpolation, which means that the hydrograph and peak flow being predicted are within the range of floods observed in the past, and 2) extrapolation, the case when the event being predicted is significantly larger than anything observed in the past.  In this study, we will present the results of controlled experiments to investigate this question and show which class of algorithms are less susceptible to over or under-estimation when extrapolating beyond the range of the “historical record”. We will present results for hourly and daily prediction timescales. This investigation is very relevant in the current environment of climate change where the water-holding capacity of the atmosphere increases with every degree of warming leading to storms that seem to constantly break every record in terms of intensity, duration, and spatial coverage.

How to cite: Mantilla, R., Barco, J., Gurbuz, F., Xiao, S., Muñoz, D., Lewkebandara, K., and Sharma, V.: Interpolation vs. Extrapolation in Flood Forecasting: Exploring the Predictive Capability of Conceptual and Machine Learning Tools in Non-Stationary Scenarios, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-13400, https://doi.org/10.5194/egusphere-egu24-13400, 2024.