Assimilation of Earth Observation products and in-situ data in seasonal hydrological forecasting services
Experiment name:
Assimilation of Earth Observation products and in-situ data in seasonal hydrological forecasting services
Experiment idea:
This experiment allows assimilation of different earth observation products, assimilated both individually and in a combination, and in-situ data in a hydrological model. The model is further operating in a seasonal forecasting mode using different seasonal prediction systems. The experiment aims to quantify the added value in seasonal streamflow forecasting skill through the introduction of data assimilation in its modelling chain.
Scientific question:
How does the assimilation of different data sets impact seasonal forecasts and how long do the impacts last for in different seasons?
What is the relative importance of meteorological forcings and assimilated data in a seasonal forecasting system?
Publications:
Musuuza, J.L., D. Gustafsson, R. Pimentel, L. Crochemore, and I. Pechlivanidis. ‘Impact of Satellite and In Situ Data Assimilation on Hydrological Predictions’. Remote Sensing 12, no. 5 (2020): 811. https://doi.org/10.3390/rs12050811
Musuuza, J.L., L. Crochemore, and I. Pechlivanidis. ‘Evaluation of earth observations and in situ data assimilation for seasonal hydrological forecasting’. Water Resources Research (2023), https://doi.org/10.1029/2022WR033655
Ilias PECHLIVANIDIS
[email protected]
Jude MUSUUZA
[email protected]
Louise CROCHEMORE
[email protected]
Background information
Hydrological information for the months ahead is of great value to existing decision-making practices, particularly to those affected by the vagaries of the climate and who would benefit from better understanding and managing climate-related risks. In river systems, streamflow fluctuations are driven both by fluxes from the basin’s water storages (i.e. groundwater, snowpack, soil moisture, channel network) and by meteorological forcings. Efforts have consequently been made to apportion the role of initial hydrological conditions and meteorological forecasts (forcings) in seasonal streamflow prediction, resulting in a number of uncertainty attribution frameworks. However there is limited knowledge on the relative importance initial hydrological conditions (IHC) and meteorological forecasts have on the quality of seasonal streamflow forecasts.
Methodology
Data assimilation is a discipline that seeks to optimally combine numerical models with observations, in order, for the case of forecasting services, to determine the initial conditions for a forecast model. Here, we will apply the Ensemble Kalman Filter (EnKF) method to assimilate the following products in a seasonal hydrological forecasting service: (1) the MODIS accumulated 8-day actual and potential evapotranspiration (AET and PET, respectively), (2) the daily CRYOLAND optical satellite FSC and passive microwave snow water equivalent (SWE), (3) river discharge, (4) local inflows into hydropower reservoirs, and (5) an optimum EO combination based on the work of Musuuza et al. (2020). The aim of this experiment is to improve the seasonal skill of streamflow forecasts. Most previous studies assimilated one variable, while the assimilation of multiple data sets to achieve a practical goal has been missing until Musuuza et al. (2020).
The experiment will be conducted in the Umeälven River catchment located in northern Sweden for a 14-year period (2001-2014) during which EO, in-situ and forecast data are available. The catchment has an area of 26000 km2 with a runoff regime dominated by snow melt during the spring and summer seasons. The area is extensively used for hydropower production with 19 hydro-power plants whose inflow needs are almost the reverse of the flow regime time-wise. Consequently, the storage of water from the snow melt season to the next winter in large reservoirs high up in the head-waters is the most important constraint to take into account for reservoir management.
Precipitation and temperature forecasts are obtained from the SEAS5 Global Circulation Model of the European Centre for Medium‐Range Weather Forecasts (ECMWF). SEAS5 reforecasts for the period 2001–2014 are available at a spatial resolution of 0.5° and are initialized at the beginning of each month. Each forecast ensemble consists of 25 members covering the 7 months ahead. Forecasts are bias-adjusted at SMHI based on the Hydro‐GFD v2.0 data set (Berg et al., 2018) and using the Distribution‐based Scaling method (Yang et al., 2010). In addition we are using another predictive method named Ensemble Streamflow Prediction (ESP; Wood & Lettenmaier, 2008). The hydrologic model is initialized for the forecast date and then fed with precipitation and temperature traces starting on the forecast date and selected from historical precipitation and temperature records excluding the forecast year. The ESP ensemble takes into account the initialization of the forecast model at the forecast date and hence includes information on initial hydrologic conditions (ICs).
A sub-model covering the Umeälven river basin is extracted from the pan-European hydrological E-HYPE model (Hundecha et al., 2016) and will be the basis for the case study analysis. The forecasts performance will be assessed using the Continuous Rank Probability Score (CRPS; Hersbach, 2000) for each sub-basin, target month and lead time. We shall also compare the forecast performance to a benchmark to translate quality into gain or loss in performance. The skill in CRPS (CRPSS) will compare the performance of the forecasting systems to a benchmark based on the simulated climatology. In task 8 below, it is stated that other metrics would be used for the skill evaluation. It might be good to mention them here. The simulated climatology is based on historical simulated streamflow time series. Reference data will be observed (reality) and modelled (pseudo-reality) streamflow.
Leader
SMHI
Short Description
This task collects all the available data (EO, and in situ) and meteorological forcings, including seasonal forecasts, for the basin of interest.
Outcome
The EO and in situ (highlighted in grey) data sets are presented here with their sources, time coverages and spatiotemporal resolutions.
We also used the Hydrological Global Forcing Data version 2.0 (HydroGFD v2.0) product, an observation-corrected reanalysis data set providing historical meteorological information of precipitation and mean temperature for 1971-2018 at a 0.5◦ resolution (Berg et al., 2018). HydroGFD is used to drive the simulated streamflow climatology and ESP (with and without data assimilation) forecast systems.
Seasonal predictions of daily mean precipitation and temperature taken from the ECMWF SEAS5 system (Johnson et al., 2019). The system generates operationally an ensemble of 51 members initialized every month with a seven-month lead time. The SEAS5 re-forecasts (hindcasts) used here consist of 25 ensemble members available at a resolution of approximately 36 km.
Leader: SMHI
Short Description:
In this task, the seasonal meteorological forecasts from ECMWF are bias-adjusted towards the Hydro-GFD v2.0 reanalysis dataset using the DBS method.
Outcome
We bias-adjusted the data using a modified version of the Distribution Based Scaling (DBS) method (Yang et al., 2010) that accounts for drifting. Using the data for the whole analysis period, the bias adjustment parameters are conditioned on the lead month and the forecast issue date. Bias adjustment was conducted on all monthly initialized forecasts using the HydroGFD v2.0 as reference dataset. After bias adjustment, the cumulative distribution of daily meteorological forecasts follows closely the one of HydroGFD. See the results of bias-adjusting SEAS5 towards HydroGFD v2.0 in Pechlivanidis et al. (2020).
Leader: SMHI
Short Description:
The historical data and bias-adjusted forecasts are quality assured detecting missing values and potential artifacts from post-processing.
Outcome
No missing values were detected and data passed the quality checks that included detection of outliers.
Leader: SMHI
Short Description:
This task focuses on the generation of forcing files that are used by the HYPE model in a simulation and forecasting mode.
Outcome
The meteorological forcing data were transformed into the Pobs (for precipitation) and Tobs (for temperature) format that is required for the HYPE model to run. This included the conversion of the data from NetCDF (the file type of the bias-adjusted forecasts) to txt format (the file type of Pobs and Tobs forcing input).
Leader: SMHI
Short Description:
In this task the HYPE model will run in a simulation mode in order to generate the values of the state variables at the beginning of every month. This will be done with and without data assimilation of the different scenarios.
Outcome
The hydrological model was forced with the HydroGFD v2.0 meteorological data to generate initial hydrologic states. The model states are available for every month in the period 2001-2015.
Leader: SMHI
Short Description:
In this task the HYPE model will run in a seasonal forecasting mode using the state variables generated in Task 5. Two seasonal prediction approaches will be used: a dynamic one based on ECMWF SEAS5 (with and without data assimilation) and the ESP approach (without data assimilation only).
Outcome
In the case of data assimilation, the model states were generated with the four EO datasets and the two in situ data sets of streamflow and reservoir inflow. The modeled streamflow and reservoir inflow were used as target variables during the previous assimilation experiment due to their importance to the hydropower sector in the area. The hydrological model thus initialized was forced with the bias-adjusted SEAS5 and HydroGFD-based climatological forcing datasets to generate seasonal hydrological forecasts initialized with and without DA at the start of each month with a seven-month lead time.
Leader: SMHI
Short Description:
In this task the HYPE output files generated in a txt format are imported in R by calendar month to be used in the evaluation scripts written in R (Task 8).
Outcome
We extracted modeled streamflow and reservoir inflow in all gauged sub-catchments of the region to assess the model’s predictive accuracy at seasonal time scales. Seasonal re-forecasts are evaluated with respect to their forecast performance against the historical observations per initialization month and lead time. Even if the hydrological model is run at a daily time scale, the generated forecasts are aggregated to weekly means for the 2001–2015 period.
Leader: SMHI
Short Description:
This task calculates the forecast skill metrics for all different scenarios (including 2 predictive systems, one with/without data assimilation of different combinations and one without). We are also interested in addressing the first scientific question “How does the assimilation of different data sets impact seasonal forecasts and how long do the impacts last for in different seasons?”
Outcome
Our research has found that incorporating various EO products into hydrological forecasting can improve its accuracy. However, the success of assimilation depends on the quality of the assimilated data. The assimilation of various in situ and EO data sets can enhance forecast accuracy, although the extent of improvement can differ depending on the season, lead time, and data assimilated. Additionally, the persistence of the impact of data assimilation on forecast accuracy can vary across different data sets and seasons.
The study found that the forecasting system performed well at short lead times, even without data assimilation, although its accuracy was somewhat lower during summer months. However, by integrating various in situ and EO data sets throughout the year, the skill of the system was generally enhanced. The highest levels of skill were observed during winter and spring following data assimilation. Nonetheless, during spring and summer, the skill displayed a significant spatial variability for most assimilation experiments. Additionally, we found that the assimilation of streamflow in situ data had the highest impact on forecast skill, but the observed streamflow may represent past headwater snowmelt events due to memory effects in the river system.
More details can be found in: Musuuza, J.L., L. Crochemore, and I. Pechlivanidis. ‘Evaluation of earth observations and in situ data assimilation for seasonal hydrological forecasting’. Water Resources Research (2023), doi:10.1029/2022WR033655
Leader: SMHI
Short Description:
This task compares the different systems to generate seasonal forecasts with and without data assimilation and is hence focusing on addressing the second scientific question: “What is the relative importance of meteorological forcings and assimilated data in a seasonal forecasting system?”
Outcome
The assimilated datasets had more control on forecast quality than meteorological forcing data, highlighting the importance of accurately quantifying the initial hydrological conditions for snow-dominated river systems. The positive effects of data assimilation were observed to persist over extended periods, although the duration varied based on the assimilated data and season. Notably, the impact of assimilating snow water equivalent (SWE) data during winter persisted for over three months, underscoring the critical role of precise and timely snow information. Furthermore, data assimilation was found to be more influential than meteorological forcing in improving the accuracy of hydrological forecasts over short lead times during all seasons except autumn.
The results highlight the benefits of improved data sets for assimilation and the need for in situ measurements to validate and adjust EO datasets. We emphasize the importance of efficient assimilation algorithms for improving the hydrological forecast skill, during all seasons except autumn. The results are very promising for enhancing hydro-climate services for snow-dominated river systems, which can lead to high economic benefits. Therefore, we suggest improving the quality of EO by enhancing the spatiotemporal resolution of the products and obtaining in situ measurements of snow and evapotranspiration. It would also be beneficial to investigate the assimilation of the same variable from different providers.
More details can be found in: Musuuza, J.L., L. Crochemore, and I. Pechlivanidis. ‘Evaluation of earth observations and in situ data assimilation for seasonal hydrological forecasting’. Water Resources Research (2023), doi:10.1029/2022WR033655
Berg, P., C. Donnelly, and D. Gustafsson. ‘Near-Real-Time Adjusted Reanalysis Forcing Data for Hydrology’. Hydrology and Earth System Sciences 22, no. 2 (2018): 989–1000. https://doi.org/10.5194/hess-22-989-2018.
Hersbach, Hans. ‘Decomposition of the Continuous Ranked Probability Score for Ensemble Prediction Systems’. Weather and Forecasting 15, no. 5 (2000): 559–570. https://doi.org/10.1175/1520-0434(2000)015<0559:DOTCRP>2.0.CO;2
Hundecha, Yeshewatesfa, Berit Arheimer, Chantal Donnelly, and Ilias Pechlivanidis. ‘A Regional Parameter Estimation Scheme for a Pan-European Multi-Basin Model’. Journal of Hydrology: Regional Studies 6, no. Supplement C (2016): 90–111. https://doi.org/10.1016/j.ejrh.2016.04.002.
Johnson, S. J., Stockdale, T. N., Ferranti, L., Balmaseda, M. A., Molteni, F., Magnusson, L., . . . Monge-Sanz, B. M. (2019). Seas5: The new ECMWF seasonal forecast system. Geosci. Model Dev., 12 (3), 1087–1117. doi:10.5194/gmd-12-1087-2019
Musuuza, J.L., D. Gustafsson, R. Pimentel, L. Crochemore, and I. Pechlivanidis. ‘Impact of Satellite and In Situ Data Assimilation on Hydrological Predictions’. Remote Sensing 12, no. 5 (2020): 811. https://doi.org/10.3390/rs12050811.
Pechlivanidis, I. G., Crochemore, L., Rosberg, J., & Bosshard, T. (2020). What are the key drivers controlling the quality of seasonal streamflow forecasts? Water Resour. Res., 56 (6), 2019WR026987. https://doi.org/10.1029/2019WR026987
Wood, Andrew W., and Dennis P. Lettenmaier. ‘An Ensemble Approach for Attribution of Hydrologic Prediction Uncertainty’. Geophysical Research Letters 35, no. 14 (30 July 2008): L14401. https://doi.org/10.1029/2008GL034648.
Yang, Wei, Johan Andréasson, L. Phil Graham, Jonas Olsson, Jörgen Rosberg, and Fredrik Wetterhall. ‘Distribution-Based Scaling to Improve Usability of Regional Climate Model Projections for Hydrological Climate Change Impacts Studies’. Hydrology Research 41, no. 3–4 (2010): 211–229. https://doi.org/10.2166/nh.2010.004.
Leave a Reply
You must be logged in to post a comment.