Global streamflow modelling using process-informed machine learning
We present a novel hybrid framework that incorporates information from the process-based global hydrological model PCR-GLOBWB, to reduce prediction errors in streamflow simulations. In addition to catchment attributes and meteorological data, our methodology employs simulated streamflow and state va...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IWA Publishing
2023-09-01
|
| Series: | Journal of Hydroinformatics |
| Subjects: | |
| Online Access: | http://jhydro.iwaponline.com/content/25/5/1648 |
| _version_ | 1797661684389642240 |
|---|---|
| author | Michele Magni Edwin H. Sutanudjaja Youchen Shen Derek Karssenberg |
| author_facet | Michele Magni Edwin H. Sutanudjaja Youchen Shen Derek Karssenberg |
| author_sort | Michele Magni |
| collection | DOAJ |
| description | We present a novel hybrid framework that incorporates information from the process-based global hydrological model PCR-GLOBWB, to reduce prediction errors in streamflow simulations. In addition to catchment attributes and meteorological data, our methodology employs simulated streamflow and state variables from PCR-GLOBWB as predictors of observed river discharge. These outputs are used in a random forest, trained on a global database of streamflow measurements, to improve estimates of simulated river discharge across the globe. PCR-GLOBWB was run for the years 1979–2019 at 30 arcmin and its inputs and outputs were upscaled from daily to monthly time steps. A single random forest model was trained with these state variables, meteorological data and catchment attributes, as predictors of observed streamflow at 2,286 stations worldwide. Model performance was evaluated using Kling–Gupta efficiency (KGE). Results based on cross-validation show that the model is capable of discerning between a variety of hydroclimatic conditions and river flow dynamics, improving KGE of PCR-GLOBWB simulations at more than 80% of testing locations and increasing median KGE from −0.03 in uncalibrated runs to 0.51 after post-processing. Performance boosts are usually independent of the availability of streamflow data, making our method a potential candidate in addressing prediction in poorly gauged and ungauged basins.
HIGHLIGHTS
A hybrid framework for global streamflow modelling is developed, connecting PCR-GLOBWB with random forest.;
The framework enables the correction of global-scale streamflow predictions with parsimonious parametrization.;
Random forests improve streamflow predictions better when additionally fed with outputs from the hydrological model, as opposed to only using meteorological forcing and catchment attributes.; |
| first_indexed | 2024-03-11T18:49:13Z |
| format | Article |
| id | doaj.art-ebdecca69e364f198cb406bee23e5759 |
| institution | Directory Open Access Journal |
| issn | 1464-7141 1465-1734 |
| language | English |
| last_indexed | 2024-03-11T18:49:13Z |
| publishDate | 2023-09-01 |
| publisher | IWA Publishing |
| record_format | Article |
| series | Journal of Hydroinformatics |
| spelling | doaj.art-ebdecca69e364f198cb406bee23e57592023-10-11T15:09:17ZengIWA PublishingJournal of Hydroinformatics1464-71411465-17342023-09-012551648166610.2166/hydro.2023.217217Global streamflow modelling using process-informed machine learningMichele Magni0Edwin H. Sutanudjaja1Youchen Shen2Derek Karssenberg3 Department of Physical Geography, Faculty of Geosciences, Utrecht University, Utrecht, The Netherlands Department of Physical Geography, Faculty of Geosciences, Utrecht University, Utrecht, The Netherlands Department of Physical Geography, Faculty of Geosciences, Utrecht University, Utrecht, The Netherlands Department of Physical Geography, Faculty of Geosciences, Utrecht University, Utrecht, The Netherlands We present a novel hybrid framework that incorporates information from the process-based global hydrological model PCR-GLOBWB, to reduce prediction errors in streamflow simulations. In addition to catchment attributes and meteorological data, our methodology employs simulated streamflow and state variables from PCR-GLOBWB as predictors of observed river discharge. These outputs are used in a random forest, trained on a global database of streamflow measurements, to improve estimates of simulated river discharge across the globe. PCR-GLOBWB was run for the years 1979–2019 at 30 arcmin and its inputs and outputs were upscaled from daily to monthly time steps. A single random forest model was trained with these state variables, meteorological data and catchment attributes, as predictors of observed streamflow at 2,286 stations worldwide. Model performance was evaluated using Kling–Gupta efficiency (KGE). Results based on cross-validation show that the model is capable of discerning between a variety of hydroclimatic conditions and river flow dynamics, improving KGE of PCR-GLOBWB simulations at more than 80% of testing locations and increasing median KGE from −0.03 in uncalibrated runs to 0.51 after post-processing. Performance boosts are usually independent of the availability of streamflow data, making our method a potential candidate in addressing prediction in poorly gauged and ungauged basins. HIGHLIGHTS A hybrid framework for global streamflow modelling is developed, connecting PCR-GLOBWB with random forest.; The framework enables the correction of global-scale streamflow predictions with parsimonious parametrization.; Random forests improve streamflow predictions better when additionally fed with outputs from the hydrological model, as opposed to only using meteorological forcing and catchment attributes.;http://jhydro.iwaponline.com/content/25/5/1648global hydrologyhybrid streamflow modellingmachine learningpost-processingrandom forests |
| spellingShingle | Michele Magni Edwin H. Sutanudjaja Youchen Shen Derek Karssenberg Global streamflow modelling using process-informed machine learning Journal of Hydroinformatics global hydrology hybrid streamflow modelling machine learning post-processing random forests |
| title | Global streamflow modelling using process-informed machine learning |
| title_full | Global streamflow modelling using process-informed machine learning |
| title_fullStr | Global streamflow modelling using process-informed machine learning |
| title_full_unstemmed | Global streamflow modelling using process-informed machine learning |
| title_short | Global streamflow modelling using process-informed machine learning |
| title_sort | global streamflow modelling using process informed machine learning |
| topic | global hydrology hybrid streamflow modelling machine learning post-processing random forests |
| url | http://jhydro.iwaponline.com/content/25/5/1648 |
| work_keys_str_mv | AT michelemagni globalstreamflowmodellingusingprocessinformedmachinelearning AT edwinhsutanudjaja globalstreamflowmodellingusingprocessinformedmachinelearning AT youchenshen globalstreamflowmodellingusingprocessinformedmachinelearning AT derekkarssenberg globalstreamflowmodellingusingprocessinformedmachinelearning |