Further investigation and improvement of the FoGSS model

[摘要] After a number of years of development, the Forecast Guided Stochastic Scenarios (FoGSS) seasonal streamflow forecasting model is able to produce reliable streamflow forecasts to lead times of 12 months that are as skilful as possible. FoGSS works well for a wide range of catchments, typically producing skilful forecasts at lead times of 1-5 months, and then transiting to neutrally skilful stochastic scenarios as skill recedes at longer lead times. FoGSS does not, however produce skilful or reliable forecasts in months or catchments that are extremely dry (50 % of flows are zero).In this study, we have reformulated FoGSS to treat both simulations and observations as censored data when we infer parameters and produce forecasts. This made little overall difference to forecast skill (i.e., it did not always remove negatively skilful forecasts) but it did ensure that forecasts offers. This report documents our attempts to improve the FoGSS model and expand its application to a greater range of catchments.The work and conclusions of this study are:1. Highly ephemeral catchments remain difficult to forecast, and producing even neutrally skilful forecasts remains a formidable technical challenge2. The existing FoGSS model treats only observations as censored data. With this model, it is theoretically impossible to get reliable forecasts in catchments where zero flows occur more than half the time3. We presented a new reformulated FoGSS model that treats both simulations and observations as censored data.4. The reformulated FoGSS model did not materially improve forecast skill5. The reformulated FoGSS model greatly improved the reliability of highly ephemeral catchments, where flows in some months are zero more than half the time6. We introduced a method for scaling data before transformation as part of the reformulated FoGSS error model7. Scaling data before transformation is a practical and effective way to impose a consistent set of parameter bounds on very different catchments, and we recommend that it be implemented on the existing FoGSS model to aid operationalisation.

[发布机构] CSIRO

[学科分类] 地球科学(综合)