Evaluating post-processing approaches for monthly and seasonal streamflow forecasts

Julien Lerat,Dmitri Kavetski,Daehyok Shin,Fitsum Woldemeskel,Narendra Tuteja,George Kuczera,David Mcinerney,Mark Thyer

doi:10.5194/hess-22-6257-2018

Abstract

Abstract. Streamflow forecasting is prone to substantial uncertainty due to errors in meteorological forecasts, hydrological model structure, and parameterization, as well as in the observed rainfall and streamflow data used to calibrate the models. Statistical streamflow post-processing is an important technique available to improve the probabilistic properties of the forecasts. This study evaluates post-processing approaches based on three transformations – logarithmic (Log), log-sinh (Log-Sinh), and Box–Cox with λ=0.2 (BC0.2) – and identifies the best-performing scheme for post-processing monthly and seasonal (3-months-ahead) streamflow forecasts, such as those produced by the Australian Bureau of Meteorology. Using the Bureau's operational dynamic streamflow forecasting system, we carry out comprehensive analysis of the three post-processing schemes across 300 Australian catchments with a wide range of hydro-climatic conditions. Forecast verification is assessed using reliability and sharpness metrics, as well as the Continuous Ranked Probability Skill Score (CRPSS). Results show that the uncorrected forecasts (i.e. without post-processing) are unreliable at half of the catchments. Post-processing of forecasts substantially improves reliability, with more than 90 % of forecasts classified as reliable. In terms of sharpness, the BC0.2 scheme substantially outperforms the Log and Log-Sinh schemes. Overall, the BC0.2 scheme achieves reliable and sharper-than-climatology forecasts at a larger number of catchments than the Log and Log-Sinh schemes. The improvements in forecast reliability and sharpness achieved using the BC0.2 post-processing scheme will help water managers and users of the forecasting service make better-informed decisions in planning and management of water resources. Highlights. Uncorrected and post-processed streamflow forecasts (using three transformations, namely Log, Log-Sinh, and BC0.2) are evaluated over 300 diverse Australian catchments. Post-processing enhances streamflow forecast reliability, increasing the percentage of catchments with reliable predictions from 50 % to over 90 %. The BC0.2 transformation achieves substantially better forecast sharpness than the Log-Sinh and Log transformations, particularly in dry catchments.

Highlights

Hydrological forecasts provide crucial supporting information on a range of water resource management decisions, including flood emergency response, water allocation for various uses, and drought risk management (Li et al, 2016; Turner et al, 2017)
To determine the “summary skill” of the forecasts in each catchment, we evaluate the total number of months in which forecasts are reliable and sharper than the climatology (i.e. IQR99 < 100 %)
To determine whether the differences in metrics are consistent over multiple catchments, the Log and Log-Sinh schemes are compared to the BC0.2 scheme

Summary

Introduction

Hydrological forecasts provide crucial supporting information on a range of water resource management decisions, including (depending on the forecast lead time) flood emergency response, water allocation for various uses, and drought risk management (Li et al, 2016; Turner et al, 2017). Sub-seasonal and seasonal streamflow forecasting systems can be broadly classified as dynamic or statistical (Crochemore et al, 2016). F. Woldemeskel et al.: Evaluating post-processing approaches a hydrological model is usually developed at a daily time step and calibrated against observed streamflow using historical rainfall and potential evaporation data. Rainfall forecasts from a numerical climate model are used as an input to produce daily streamflow forecasts, which are aggregated to the timescale of interest and post-processed using statistical models A statistical model based on relevant predictors, such as antecedent rainfall and streamflow, is developed and applied directly at the timescale of interest (Robertson and Wang, 2009, 2011; Lü et al, 2016; Zhao et al, 2016). Hybrid systems that combine aspects of dynamic and statistical approaches have been investigated (Humphrey et al, 2016; Robertson et al, 2013a)

Results

Discussion

Conclusion