Refine
Year of publication
Document Type
- Article (132)
- Postprint (17)
- Monograph/Edited Volume (7)
- Conference Proceeding (3)
- Other (3)
- Part of a Book (2)
- Preprint (1)
- Report (1)
Keywords
- climate change (6)
- runoff (5)
- Climate change (4)
- Connectivity (4)
- meteorological drought (4)
- time-series (4)
- variability (4)
- Extreme events (3)
- Lake Malawi (3)
- Precipitation (3)
Institute
- Institut für Umweltwissenschaften und Geographie (92)
- Institut für Geowissenschaften (67)
- Extern (6)
- Zentrum für Umweltwissenschaften (4)
- Interdisziplinäres Zentrum für Musterdynamik und Angewandte Fernerkundung (2)
- Mathematisch-Naturwissenschaftliche Fakultät (2)
- Institut für Biochemie und Biologie (1)
Transferability of data-driven models to predict urban pluvial flood water depth in Berlin, Germany
(2023)
Data-driven models have been recently suggested to surrogate computationally expensive hydrodynamic models to map flood hazards. However, most studies focused on developing models for the same area or the same precipitation event. It is thus not obvious how transferable the models are in space. This study evaluates the performance of a convolutional neural network (CNN) based on the U-Net architecture and the random forest (RF) algorithm to predict flood water depth, the models' transferability in space and performance improvement using transfer learning techniques. We used three study areas in Berlin to train, validate and test the models. The results showed that (1) the RF models outperformed the CNN models for predictions within the training domain, presumable at the cost of overfitting; (2) the CNN models had significantly higher potential than the RF models to generalize beyond the training domain; and (3) the CNN models could better benefit from transfer learning technique to boost their performance outside training domains than RF models.
Transferability of data-driven models to predict urban pluvial flood water depth in Berlin, Germany
(2023)
Data-driven models have been recently suggested to surrogate computationally expensive hydrodynamic models to map flood hazards. However, most studies focused on developing models for the same area or the same precipitation event. It is thus not obvious how transferable the models are in space. This study evaluates the performance of a convolutional neural network (CNN) based on the U-Net architecture and the random forest (RF) algorithm to predict flood water depth, the models' transferability in space and performance improvement using transfer learning techniques. We used three study areas in Berlin to train, validate and test the models. The results showed that (1) the RF models outperformed the CNN models for predictions within the training domain, presumable at the cost of overfitting; (2) the CNN models had significantly higher potential than the RF models to generalize beyond the training domain; and (3) the CNN models could better benefit from transfer learning technique to boost their performance outside training domains than RF models.
Knowledge on the response of sediment export to recent climate change in glacierized areas in the European Alps is limited, primarily because long-term records of suspended sediment concentrations (SSCs) are scarce. Here we tested the estimation of sediment export of the past five decades using quantile regression forest (QRF), a nonparametric, multivariate regression based on random forest. The regression builds on short-term records of SSCs and long records of the most important hydroclimatic drivers (discharge, precipitation and air temperature - QPT). We trained independent models for two nested and partially glacier-covered catchments, Vent (98 km(2)) and Vernagt (11.4 km(2)), in the upper otztal in Tyrol, Austria (1891 to 3772 m a.s.l.), where available QPT records start in 1967 and 1975. To assess temporal extrapolation ability, we used two 2-year SSC datasets at gauge Vernagt, which are almost 20 years apart, for a validation. For Vent, we performed a five-fold cross-validation on the 15 years of SSC measurements. Further, we quantified the number of days where predictors exceeded the range represented in the training dataset, as the inability to extrapolate beyond this range is a known limitation of QRF. Finally, we compared QRF performance to sediment rating curves (SRCs). We analyzed the modeled sediment export time series, the predictors and glacier mass balance data for trends (Mann-Kendall test and Sen's slope estimator) and step-like changes (using the widely applied Pettitt test and a complementary Bayesian approach).Our validation at gauge Vernagt demonstrated that QRF performs well in estimating past daily sediment export (Nash-Sutcliffe efficiency (NSE) of 0.73) and satisfactorily for SSCs (NSE of 0.51), despite the small training dataset. The temporal extrapolation ability of QRF was superior to SRCs, especially in periods with high-SSC events, which demonstrated the ability of QRF to model threshold effects. Days with high SSCs tended to be underestimated, but the effect on annual yields was small. Days with predictor exceedances were rare, indicating a good representativity of the training dataset. Finally, the QRF reconstruction models outperformed SRCs by about 20 percent points of the explained variance.Significant positive trends in the reconstructed annual suspended sediment yields were found at both gauges, with distinct step-like increases around 1981. This was linked to increased glacier melt, which became apparent through step-like increases in discharge at both gauges as well as change points in mass balances of the two largest glaciers in the Vent catchment. We identified exceptionally high July temperatures in 1982 and 1983 as a likely cause. In contrast, we did not find coinciding change points in precipitation. Opposing trends at the two gauges after 1981 suggest different timings of "peak sediment". We conclude that, given large-enough training datasets, the presented QRF approach is a promising tool with the ability to deepen our understanding of the response of high-alpine areas to decadal climate change.
A novel approach for estimating precipitation patterns is developed here and applied to generate a new hydrologically corrected daily precipitation dataset, called RAIN4PE (Rain for Peru and Ecuador), at 0.1 degrees spatial resolution for the period 1981-2015 covering Peru and Ecuador. It is based on the application of 1) the random forest method to merge multisource precipitation estimates (gauge, satellite, and reanalysis) with terrain elevation, and 2) observed and modeled streamflow data to first detect biases and second further adjust gridded precipitation by inversely applying the simulated results of the ecohydrological model SWAT (Soil and Water Assessment Tool). Hydrological results using RAIN4PE as input for the Peruvian and Ecuadorian catchments were compared against the ones when feeding other uncorrected (CHIRP and ERA5) and gauge-corrected (CHIRPS, MSWEP, and PISCO) precipitation datasets into the model. For that, SWAT was calibrated and validated at 72 river sections for each dataset using a range of performance metrics, including hydrograph goodness of fit and flow duration curve signatures. Results showed that gauge-corrected precipitation datasets outperformed uncorrected ones for streamflow simulation. However, CHIRPS, MSWEP, and PISCO showed limitations for streamflow simulation in several catchments draining into the Pacific Ocean and the Amazon River. RAIN4PE provided the best overall performance for streamflow simulation, including flow variability (low, high, and peak flows) and water budget closure. The overall good performance of RAIN4PE as input for hydrological modeling provides a valuable criterion of its applicability for robust countrywide hydrometeorological applications, including hydroclimatic extremes such as droughts and floods. Significance StatementWe developed a novel precipitation dataset RAIN4PE for Peru and Ecuador by merging multisource precipitation data (satellite, reanalysis, and ground-based precipitation) with terrain elevation using the random forest method. Furthermore, RAIN4PE was hydrologically corrected using streamflow data in watersheds with precipitation underestimation through reverse hydrology. The results of a comprehensive hydrological evaluation showed that RAIN4PE outperformed state-of-the-art precipitation datasets such as CHIRP, ERA5, CHIRPS, MSWEP, and PISCO in terms of daily and monthly streamflow simulations, including extremely low and high flows in almost all Peruvian and Ecuadorian catchments. This underlines the suitability of RAIN4PE for hydrometeorological applications in this region. Furthermore, our approach for the generation of RAIN4PE can be used in other data-scarce regions.
Identifying urban pluvial flood-prone areas is necessary but the application of two-dimensional hydrodynamic models is limited to small areas. Data-driven models have been showing their ability to map flood susceptibility but their application in urban pluvial flooding is still rare. A flood inventory (4333 flooded locations) and 11 factors which potentially indicate an increased hazard for pluvial flooding were used to implement convolutional neural network (CNN), artificial neural network (ANN), random forest (RF) and support vector machine (SVM) to: (1) Map flood susceptibility in Berlin at 30, 10, 5, and 2 m spatial resolutions. (2) Evaluate the trained models' transferability in space. (3) Estimate the most useful factors for flood susceptibility mapping. The models' performance was validated using the Kappa, and the area under the receiver operating characteristic curve (AUC). The results indicated that all models perform very well (minimum AUC = 0.87 for the testing dataset). The RF models outperformed all other models at all spatial resolutions and the RF model at 2 m spatial resolution was superior for the present flood inventory and predictor variables. The majority of the models had a moderate performance for predictions outside the training area based on Kappa evaluation (minimum AUC = 0.8). Aspect and altitude were the most influencing factors on the image-based and point-based models respectively. Data-driven models can be a reliable tool for urban pluvial flood susceptibility mapping wherever a reliable flood inventory is available.
The study examined the potential future changes of drought characteristics in the Greater Lake Malawi Basin in Southeast Africa. This region strongly depends on water resources to generate electricity and food. Future projections (considering both moderate and high emission scenarios) of temperature and precipitation from an ensemble of 16 bias-corrected climate model combinations were blended with a scenario-neutral response surface approach to analyses changes in: (i) the meteorological conditions, (ii) the meteorological water balance, and (iii) selected drought characteristics such as drought intensity, drought months, and drought events, which were derived from the Standardized Precipitation and Evapotranspiration Index. Changes were analyzed for a near-term (2021-2050) and far-term period (2071-2100) with reference to 1976-2005. The effect of bias-correction (i.e., empirical quantile mapping) on the ability of the climate model ensemble to reproduce observed drought characteristics as compared to raw climate projections was also investigated. Results suggest that the bias-correction improves the climate models in terms of reproducing temperature and precipitation statistics but not drought characteristics. Still, despite the differences in the internal structures and uncertainties that exist among the climate models, they all agree on an increase of meteorological droughts in the future in terms of higher drought intensity and longer events. Drought intensity is projected to increase between +25 and +50% during 2021-2050 and between +131 and +388% during 2071-2100. This translates into +3 to +5, and +7 to +8 more drought months per year during both periods, respectively. With longer lasting drought events, the number of drought events decreases. Projected droughts based on the high emission scenario are 1.7 times more severe than droughts based on the moderate scenario. That means that droughts in this region will likely become more severe in the coming decades. Despite the inherent high uncertainties of climate projections, the results provide a basis in planning and (water-)managing activities for climate change adaptation measures in Malawi. This is of particular relevance for water management issues referring hydro power generation and food production, both for rain-fed and irrigated agriculture.
The study examined the potential future changes of drought characteristics in the Greater Lake Malawi Basin in Southeast Africa. This region strongly depends on water resources to generate electricity and food. Future projections (considering both moderate and high emission scenarios) of temperature and precipitation from an ensemble of 16 bias-corrected climate model combinations were blended with a scenario-neutral response surface approach to analyses changes in: (i) the meteorological conditions, (ii) the meteorological water balance, and (iii) selected drought characteristics such as drought intensity, drought months, and drought events, which were derived from the Standardized Precipitation and Evapotranspiration Index. Changes were analyzed for a near-term (2021–2050) and far-term period (2071–2100) with reference to 1976–2005. The effect of bias-correction (i.e., empirical quantile mapping) on the ability of the climate model ensemble to reproduce observed drought characteristics as compared to raw climate projections was also investigated. Results suggest that the bias-correction improves the climate models in terms of reproducing temperature and precipitation statistics but not drought characteristics. Still, despite the differences in the internal structures and uncertainties that exist among the climate models, they all agree on an increase of meteorological droughts in the future in terms of higher drought intensity and longer events. Drought intensity is projected to increase between +25 and +50% during 2021–2050 and between +131 and +388% during 2071–2100. This translates into +3 to +5, and +7 to +8 more drought months per year during both periods, respectively. With longer lasting drought events, the number of drought events decreases. Projected droughts based on the high emission scenario are 1.7 times more severe than droughts based on the moderate scenario. That means that droughts in this region will likely become more severe in the coming decades. Despite the inherent high uncertainties of climate projections, the results provide a basis in planning and (water-)managing activities for climate change adaptation measures in Malawi. This is of particular relevance for water management issues referring hydro power generation and food production, both for rain-fed and irrigated agriculture.
The study examined the potential future changes of drought characteristics in the Greater Lake Malawi Basin in Southeast Africa. This region strongly depends on water resources to generate electricity and food. Future projections (considering both moderate and high emission scenarios) of temperature and precipitation from an ensemble of 16 bias-corrected climate model combinations were blended with a scenario-neutral response surface approach to analyses changes in: (i) the meteorological conditions, (ii) the meteorological water balance, and (iii) selected drought characteristics such as drought intensity, drought months, and drought events, which were derived from the Standardized Precipitation and Evapotranspiration Index. Changes were analyzed for a near-term (2021–2050) and far-term period (2071–2100) with reference to 1976–2005. The effect of bias-correction (i.e., empirical quantile mapping) on the ability of the climate model ensemble to reproduce observed drought characteristics as compared to raw climate projections was also investigated. Results suggest that the bias-correction improves the climate models in terms of reproducing temperature and precipitation statistics but not drought characteristics. Still, despite the differences in the internal structures and uncertainties that exist among the climate models, they all agree on an increase of meteorological droughts in the future in terms of higher drought intensity and longer events. Drought intensity is projected to increase between +25 and +50% during 2021–2050 and between +131 and +388% during 2071–2100. This translates into +3 to +5, and +7 to +8 more drought months per year during both periods, respectively. With longer lasting drought events, the number of drought events decreases. Projected droughts based on the high emission scenario are 1.7 times more severe than droughts based on the moderate scenario. That means that droughts in this region will likely become more severe in the coming decades. Despite the inherent high uncertainties of climate projections, the results provide a basis in planning and (water-)managing activities for climate change adaptation measures in Malawi. This is of particular relevance for water management issues referring hydro power generation and food production, both for rain-fed and irrigated agriculture.
Identifying urban pluvial flood-prone areas is necessary but the application of two-dimensional hydrodynamic models is limited to small areas. Data-driven models have been showing their ability to map flood susceptibility but their application in urban pluvial flooding is still rare. A flood inventory (4333 flooded locations) and 11 factors which potentially indicate an increased hazard for pluvial flooding were used to implement convolutional neural network (CNN), artificial neural network (ANN), random forest (RF) and support vector machine (SVM) to: (1) Map flood susceptibility in Berlin at 30, 10, 5, and 2 m spatial resolutions. (2) Evaluate the trained models' transferability in space. (3) Estimate the most useful factors for flood susceptibility mapping. The models' performance was validated using the Kappa, and the area under the receiver operating characteristic curve (AUC). The results indicated that all models perform very well (minimum AUC = 0.87 for the testing dataset). The RF models outperformed all other models at all spatial resolutions and the RF model at 2 m spatial resolution was superior for the present flood inventory and predictor variables. The majority of the models had a moderate performance for predictions outside the training area based on Kappa evaluation (minimum AUC = 0.8). Aspect and altitude were the most influencing factors on the image-based and point-based models respectively. Data-driven models can be a reliable tool for urban pluvial flood susceptibility mapping wherever a reliable flood inventory is available.
Glaciated high-alpine areas are fundamentally altered by climate change, with well-known implications for hydrology, e.g., due to glacier retreat, longer snow-free periods, and more frequent and intense summer rainstorms. While knowledge on how these hydrological changes will propagate to suspended sediment dynamics is still scarce, it is needed to inform mitigation and adaptation strategies. To understand the processes and source areas most relevant to sediment dynamics, we analyzed discharge and sediment dynamics in high temporal resolution as well as their patterns on several spatial scales, which to date few studies have done.
We used a nested catchment setup in the Upper Ötztal in Tyrol, Austria, where high-resolution (15 min) time series of discharge and suspended sediment concentrations are available for up to 15 years (2006–2020). The catchments of the gauges in Vent, Sölden and Tumpen range from 100 to almost 800 km2 with 10 % to 30 % glacier cover and span an elevation range of 930 to 3772 m a.s.l. We analyzed discharge and suspended sediment yields (SSY), their distribution in space, their seasonality and spatial differences therein, and the relative importance of short-term events. We complemented our analysis by linking the observations to satellite-based snow cover maps, glacier inventories, mass balances and precipitation data.
Our results indicate that the areas above 2500 m a.s.l., characterized by glacier tongues and the most recently deglaciated areas, are crucial for sediment generation in all sub-catchments. This notion is supported by the synchronous spring onset of sediment export at the three gauges, which coincides with snowmelt above 2500 m but lags behind spring discharge onsets. This points at a limitation of suspended sediment supply as long as the areas above 2500 m are snow-covered. The positive correlation of annual SSY with glacier cover (among catchments) and glacier mass balances (within a catchment) further supports the importance of the glacier-dominated areas. The analysis of short-term events showed that summer precipitation events were associated with peak sediment concentrations and yields but on average accounted for only 21 % of the annual SSY in the headwaters. These results indicate that under current conditions, thermally induced sediment export (through snow and glacier melt) is dominant in the study area.
Our results extend the scientific knowledge on current hydro-sedimentological conditions in glaciated high-alpine areas and provide a baseline for studies on projected future changes in hydro-sedimentological system dynamics.