The suitability of a hybrid framework including data driven approaches for hydrological forecasting

Abstract

Abstract. Hydrological forecasts are important for operational water management and near future planning, even more so in light of increased occurrences of extreme events such as floods and droughts. Having a flexible forecasting framework that can deliver this information in fast and computational efficient manner is critical. In this study, the suitability of a hybrid forecasting framework, combining data-driven approaches and seasonal (re)forecasting information to predict hydrological variables was explored. Target variables include discharge and surface water levels for various stations at national scale with the Netherlands as focus. Five different ML models, ranging from simple to more complex and trained on historical observations of discharge, precipitation, evaporation and sea water levels, were run with seasonal (re)forecast data (EFAS and SEAS5) of these driver variables in a hindcast setting. The results were evaluated using the evaluation metrics Anomaly Correlation Coefficient (ACC), Continuous Ranked Probability (Skill) Score (CRPS and CRPSS), and Brier Skill Score (BSS) in comparison to a climatological reference hindcast. Aggregating results of all stations and ML models revealed that the hindcasting framework outperformed the climatological reference forecasts by roughly 60 % for discharge predictions (80 % for surface water level predictions). Skilful prediction for the first lead month, independently of initialization month, can be made for discharge. The skill extends up to 2–3 months for spring months due to snow melt dynamics captured in the training phase of the model. Surface water levels hindcasts showed similar skill and skilful lead times. While the different ML models showed differences in performance during a testing and training phase using historical observations, running the ML framework in a hindcast setting showed only minor differences between the models, which is attributed to the uncertainty in seasonal forecasts. However, despite being trained on historical observations, the hybrid framework used in this study shows similar skilful predictions as previous large scale forecasting systems. With our study we show that a hybrid framework is able to bring location specific skilful seasonal forecast information with global seasonal forecast inputs. At the same time our hybrid approach is flexible and fast, and as such a hybrid framework could be adapted to make it even more interesting to water managers and their needs, for instance a part of a fast model-predictive control framework.

Connecting hydrological modelling and forecasting from global to local scales: Perspectives from an international joint virtual workshop

“…Due to the short training time of the statistical model, these models can be updated regularly to account for changes in land use. Driving ML algorithms with the output from physics-based models is a common hybrid approach (e.g., Hauswirth et al, 2022;Hunt et al, 2022;Frnda et al, 2022). Two other presentations demonstrated the skill of hybrid forecasts created in this way.…”

Hybrid forecasting: using statistics and machine learning to integrate predictions from dynamical models

“…For Ireland, Golian et al (2022) found that MLR and ANN models applied to hindcasts of mean sea level pressure from GloSea5 and SEAS5 produced skillful forecasts of winter [DJF] and summer [JJA] precipitation for lead times of up to four months, with the ANN outperforming MLR for both seasons and all lead times. A study over the Netherlands using streamflow, precipitation, and evaporation found that the hybrid ML approach outperformed climatological reference forecasts by approximately 60% and 80% for streamflow and surface water level, respectively, using various machine learning models (Hauswirth et al, 2022). Another study employed predictions of large-scale indices by the CFSv2 model to predict the frequency of tropical cyclones in the Bay of Bengal using principal component regression (Sabeerali et al, 2022).…”

Simulating hydrological extremes for different warming levels–combining large scale climate ensembles with local observation based machine learning models

“…In the past years, several studies have shown that ML models such as Long Short-Term Memory Models (LSTM) or Random Forests (RF) are suitable for predicting hydrological variables such as discharge or groundwater levels (Kratzert et al, 2018;Hauswirth et al, 2021;Koch et al, 2021), often even outperforming the conventional physically based models (Mai et al, 2022). However, even when considering the potential of machine learning in hydrology, the challenge regarding simulating extreme events remain (Hauswirth et al, 2021(Hauswirth et al, , 2022. ML models are generally good in simulating what they have seen during training, however extrapolating to "unseen" events is not possible and as such makes it also difficult to apply to the climate change signal.…”

“…We do this by using a ML modeling framework, including several location specific ML models that are trained on historical observations, developed for the case study of the Netherlands, and GCM LE simulation data to provide outlooks for future warming scenarios. The ML modeling framework used was developed especially for this region in previous studies (Hauswirth et al, 2021(Hauswirth et al, , 2022, and has shown to be suitable for local predictions and seasonal forecasts, using both local but also larger scale input data. We approach the problem of simulating extreme events by introducing a post-processing step at the end of the ML framework, where we use the characteristics of the low flow distributions, taken from historical observations, based on the assumption that the tail of the distribution serves as baseline for extreme event extrapolation.…”