Comparison of cross-validation and bootstrap aggregating for building a seasonal streamflow forecast model

Abstract

Abstract. Based on a hindcast experiment for the period 1982–2013 in 66 sub-catchments of the Swiss Rhine, the present study compares two approaches of building a regression model for seasonal streamflow forecasting. The first approach selects a single "best guess" model, which is tested by leave-one-out cross-validation. The second approach implements the idea of bootstrap aggregating, where bootstrap replicates are employed to select several models, and out-of-bag predictions provide model testing. The target value is mean streamflow for durations of 30, 60 and 90 days, starting with the 1st and 16th day of every month. Compared to the best guess model, bootstrap aggregating reduces the mean squared error of the streamflow forecast by seven percent on average. Thus, if resampling is anyway part of the model building procedure, bootstrap aggregating seems to be a useful strategy in statistical seasonal streamflow forecasting. Since the improved accuracy comes at the cost of a less interpretable model, the approach might be best suited for pure prediction tasks, e.g. as in operational applications.

Monthly streamflow forecasting at varying spatial scales in the Rhine basin

“…In combination with the moderate sample size n = 31 for model fitting, perturbations in the training set can lead to large changes in the predictors time lengths a i,j and regression coefficients. In order to reduce model variance, we draw 100 non-parametric bootstrap replicates of the training set, fit the model to these replicates, and combine the predictions by unweighted averaging (Breiman, 1996;Schick et al, 2016).…”

Enhancing monthly streamflow forecasting for Brazilian hydropower plants through climate index integration with stochastic methods

“…In the existing literature, it's also common to encounter machine learning techniques capable of integrating climate data into streamflow forecasts (Pham et al, 2021;Li et al, 2020;Ribeiro et al, 2020;Fu et al, 2019;Botsisa et al, 2018;Schick et al, 2016). However, many of these models are often labeled as "black box" due to their reduced interpretability compared to other methods.…”

Supplementary material to "Monthly streamflow forecasting at varying spatial scales in the Rhine basin"

“…transitions between wet and dry or cold and warm seasons) and catchmentspecific hydrological storages (e.g. surface water bodies, soils, aquifers, and snow) and can vary from 0 up to several months (van Dijk et al, 2013;Shukla et al, 2013;Yossef et al, 2013). Indeed, this source of predictability is the rationale behind the application of the ESP approach in operational forecast settings, and it can be further exploited by conditioning on climate precursors (e.g.…”