Low-flow estimation beyond the mean – expectile loss and extreme gradient boosting for spatiotemporal low-flow prediction in Austria

Laimighofer, Johannes; Melcher, Michael; Laaha, Gregor

doi:10.5194/hess-26-4553-2022

Cited by 7 publications

(4 citation statements)

References 66 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The data set has already been used in a number of low-flow studies, addressing seasonality indices (Laaha and Blöschl, 2006b), catchment classification and regional regression (Laaha and Blöschl, 2006), geostatistical methods (Laaha et al, 2014), estimation methods from short records and spot gauging (Laaha and Blöschl, 2005), the link between meteorological drought indices and streamflow (Haslinger et al, 2014), and climate change (Laaha et al, 2016;Parajka et al, 2016;Karanitsch-Ackerl et al, 2019). Most recently, an updated version of the data set has been used for evaluating statistical learning methods (Laimighofer et al, 2022b) and statistical space-time models for low flow (Laimighofer et al, 2022a). In this study we use the data set of Laaha et al (2016), consisting of 329 Austrian stream gauges with measurements from the 1976 to 2010 period (Fig.…”

Section: Data Setmentioning

confidence: 99%

A mixed distribution approach for low-flow frequency analysis – Part 1: Concept, performance, and effect of seasonality

Laaha

2023

Hydrol. Earth Syst. Sci.

Self Cite

View full text Add to dashboard Cite

Abstract. In seasonal climates with a warm and a cold season, low flows are generated by different processes so that the annual extreme series will be a mixture of summer and winter low-flow events. This leads to a violation of the homogeneity assumption for all statistics derived from the annual series and gives rise to inaccurate conclusions. In this first part of a two-paper series, a mixed distribution approach to perform frequency analysis in catchments with mixed low-flow regimes is proposed. We formulate the theoretical basis of the mixed distribution approach for the lower extremes based on annual minima series. The main strength of the model is that it allows the user to estimate return periods of summer low flows, winter low flows, and annual return periods in a theoretically sound and consistent way. Using archetypal examples, we show how the model behaves for a range of low-flow regimes, from distinct winter and summer regimes to mixed regimes where seasonal occurrence in summer and winter is equally likely. The examples show in a qualitative way the loss in accuracy one has to expect with conventional extreme value statistics performed with the annual extremes series. The model is then applied to a comprehensive Austrian data set to quantify the expected gain of using the mixed distribution approach compared to conventional frequency analysis. Results indicate that the gain of using a mixed distribution approach is indeed large. On average, the relative deviation is 21 %, 39 %, and 63 % when estimating the low flow with a 20-, 50-, and 100-year return period. For the 100-year event, 75 % of stations show a performance gain of >10 %, 41 % of stations > 50 %, and 25 % of stations > 80.6 %. This points to a broad relevance of the approach that goes beyond highly mixed seasonal regimes to include the strongly seasonal ones. We finally correlate the performance gain with seasonality indices in order to show the expected gain conditional to the strength of seasonality expressed by the ratio of average summer and winter low flow seasonality ratio (SR). For the 100-year event, the expected gain is about 70 % for SR=1.0, 20 % for SR=1.5, and 10 % for SR=2.0. The performance gain is further allocated to the spatial patterns of SR in the study area. The results suggest that the mixed estimator is relevant not only for mountain forelands but to a much wider range of catchment typologies. The mixed distribution approach provides one consistent approach for summer, winter, and annual probabilities and should be used by default in seasonal climates with a cold winter season where summer and winter low flows can occur.

show abstract

Section: Data Setmentioning

confidence: 99%

A mixed distribution approach for low-flow frequency analysis – Part 1: Concept, performance, and effect of seasonality

Laaha

2023

Hydrol. Earth Syst. Sci.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Among all boosting algorithms, it is the XGBoost model that is probably most often used, also for streamflow forecasting in gauged catchments [1,16,29], but not exclusively. Laimighofer et al [2] proposes XGBoost model for predicting monthly low flow in ungauged catchments.…”

Section: Extreme Gradient Boosting (Xgboost)mentioning

confidence: 99%

“…Scientists and practitioners are interested in both extreme events, such as floods and droughts, and medium flows. In practice, however, forecasting models seem to be used and verified in relation to one specific type of task, such as prediction of low flows [2] or drought forecasting [3]. Forecasting a wide spectrum of streamflows with one model is not an easy task and few models have such a wide range of applicability [4].…”

Section: Introduction 1streamflow Forecastingmentioning

confidence: 99%

Daily Streamflow Forecasting in Mountainous Catchment Using XGBoost, LightGBM and CatBoost

Szczepanek

2022

Hydrology

View full text Add to dashboard Cite

Streamflow forecasting in mountainous catchments is and will continue to be one of the important hydrological tasks. In recent years machine learning models are increasingly used for such forecasts. A direct comparison of the use of the three gradient boosting models (XGBoost, LightGBM and CatBoost) to forecast daily streamflow in mountainous catchment is our main contribution. As predictors we use daily precipitation, runoff at upstream gauge station and two-day preceding observations. All three algorithms are simple to implement in Python, fast and robust. Compared to deep machine learning models (like LSTM), they allow for easy interpretation of the significance of predictors. All tested models achieved Nash-Sutcliffe model efficiency (NSE) in the range of 0.85–0.89 and RMSE in the range of 6.8–7.8 m3s−1.A minimum of 12 years of training data series is required for such a result. The XGBoost did not turn out to be the best model for the daily streamflow forecast, although it is the most popular model. Using default model parameters, the best results were obtained with CatBoost. By optimizing the hyperparameters, the best forecast results were obtained by LightGBM. The differences between the model results are much smaller than the differences within the models themselves when suboptimal hyperparameters are used.

show abstract

“…These are the existing log-linear model for median flood estimation in Norway and a gradient-boosted tree ensemble (XGBoost). XGBoost has established use in hydrology (Zounemat-Kermani et al, 2021) and is applied in, for example, Laimighofer et al (2022a) and Ni et al (2020). As part of what distinguishes the GAM from the log-linear model is the flexible, data-driven nature of the response relationship, it is useful to have a comparison point from a fully data-driven model such as XGBoost.…”

Section: Introductionmentioning

confidence: 99%

Regional index flood estimation at multiple durations with generalized additive models

Barna,

Engeland,

Kneib

et al. 2023

Preprint

View full text Add to dashboard Cite

Abstract. Estimation of flood quantiles at ungauged basins is often achieved through regression based methods. In situations where flood retention is important, e.g. floodplain management and reservoir design, flood quantile estimates are often needed at multiple durations. This poses a problem for regression-based models as the form of the functional relationship between catchment descriptors and the response may not be constant across different durations. A particular type of regression model that is well-suited to this situation is a generalized additive model (GAM), which allows for flexible, semi-parametric modeling and visualization of the relationship between predictors and the response. However, in practice, selecting predictors for such a flexible model can be challenging, particularly given the characteristics of available catchment descriptor datasets. We employ a machine learning-based variable pre-selection tool which, when combined with domain knowledge, enhances the practicality of constructing GAMs. In this study, we develop a GAM for index (median) flood estimation with the primary objective of investigating duration-specific differences in how catchment descriptors influence the median flood. As the accuracy of this explainable approach is dependent on the fitted GAM being adequate, the secondary objective of our study is prediction of the median flood at ungauged locations and multiple durations, where predictive performance and reliability at ungauged locations are used as proxies for adequacy of the GAM. Predictive performance of the GAM is compared to two benchmark models: the existing log-linear model for median flood estimation in Norway and a fully data-driven machine learning model (an extreme gradient boosting tree ensemble, XGBoost). We find that the predictive accuracy and reliability of the GAM matched or exceeded that of the benchmark models at both durations studied. Within the predictor set selected for this study, we observe duration-specific differences in the relationship between the median flood and the two catchment descriptors effective lake percentage and catchment shape. Ignoring these differences results in a statistically significant decline in predictive performance. This suggests that models developed and estimated for prediction of the index flood at one duration may have reduced performance when applied directly to situations outside of that specific duration.

show abstract

Low-flow estimation beyond the mean – expectile loss and extreme gradient boosting for spatiotemporal low-flow prediction in Austria

Cited by 7 publications

References 66 publications

A mixed distribution approach for low-flow frequency analysis – Part 1: Concept, performance, and effect of seasonality

A mixed distribution approach for low-flow frequency analysis – Part 1: Concept, performance, and effect of seasonality

Daily Streamflow Forecasting in Mountainous Catchment Using XGBoost, LightGBM and CatBoost

Regional index flood estimation at multiple durations with generalized additive models

Contact Info

Product

Resources

About