Water cycle prediction at the regional scale on

Water cycle prediction at the regional scale: on the importance of being consistent Vincent Fortin, Pierre Pellerin Al Pietroniro, André Méthot Meteorological Research Division Meteorological Service of Canada

Canadian Meteorological Centre: more than tomorrow's weather! Page 2 – 17 March 2018

Applications of hydrological and hydrodynamic modelling • Adaptive management of watersheds • Optimization of hydropower production • Flood warning • Search and rescue • Predicting impacts on habitat of changes in water level • NWP and land-surface model verification Page 3 – 17 March 2018

Coupled modelling system for hydrological prediction GEM atmospheric model Page 4 – 17 March 2018

Coupled modelling system for hydrological prediction GEM atmospheric model 4 DVAR/En. KF data assimilation Page 5 – 17 March 2018

Coupled modelling system for hydrological prediction GEM atmospheric model 4 DVAR/En. KF data assimilation Land-surface scheme (CLASS, ISBA, SVS) Page 6 – 17 March 2018

Coupled modelling system for hydrological prediction GEM atmospheric model 4 DVAR/En. KF data assimilation Land-surface scheme (CLASS, ISBA, SVS) WATROUTE routing model Page 7 – 17 March 2018

Coupled modelling system for hydrological prediction GEM atmospheric model 4 DVAR/En. KF data assimilation Land-surface scheme (CLASS, ISBA, SVS) Ca. LDAS: En. KF data assimilation WATROUTE routing model Page 8 – 17 March 2018

Coupled modelling system for hydrological prediction GEM atmospheric model 4 DVAR/En. KF data assimilation Land-surface scheme (CLASS, ISBA, SVS) Ca. LDAS: En. KF data assimilation NEMO model for the ocean and large lakes WATROUTE routing model Page 9 – 17 March 2018

Coupled modelling system for hydrological prediction • Components can be run either coupled or offline, with prescribed forcings GEM atmospheric model Land-surface scheme (CLASS, ISBA, SVS) MESH: Modélisation Environnementale de la Surface et de l'Hydrologie NEMO model for the ocean and large lakes WATROUTE routing model Page 10 – 17 March 2018

Why not simply drive surface and hydrology models with observations? • Required observations are generally not all available • Forecasting becomes nearly impossible • Accuracy of short-term forecasts can approach or even surpasses that of observations – snowfall observations • Working within an integrated system makes it possible for hydrologists to actively contribute to the improvement of all components Page 11 – 17 March 2018

It works because weather forecasting is not so difficult • Landscape • Atmosphere Saskatchewan Northern Territories Toronto Central Quebec Page 12 – 17 March 2018 Central Quebec

Not only is weather forecasting easy, it is improving Page 13 – 17 March 2018

Not only is weather forecasting easy, it is improving • Major improvements to the data assimilation system • The ISBA land-surface model replaces the forcerestore scheme Page 14 – 17 March 2018

GEM vs reanalysis products • Many hydrologists already use reanalysis products • • (NCEP, NARR, MERRA, ERA-40, WATCH, Era-interim) For many applications where ~10 years or less of data is required, operational NWP outputs (e. g. GEM) provide higher resolution (up to 2. 5 km for GEM HRDPS) and better skill (especially for surface variables) For short-term hydrological forecasting applications, past atmospheric forcings are used only to calibrate the hydrological model and obtain initial conditions – NWP forecasts are required to obtain streamflow forecasts – by using the same data source for model calibration and forecasting, we can bypass the NWP post-processing step Page 15 – 17 March 2018

The Canadian Precipitation Analysis (Ca. PA) can be used to improve GEM precipitation • Optimal interpolation technique • • • used to merge gauges, radar and satellite data with a background provided by the GEM NWP model Fully automated quality control 6 -h and 24 -h accumulations North American domain 10 km resolution Early (T+1 h) and late (T+7 h) analyses Operational since April 2011 24 -h analysis valid 2014 -08 -15@12 Z http: //weather. gc. ca/analysis Page 16 – 17 March 2018

Great Lakes / St. Lawrence testbed • Demonstrate benefits of • • coupled numerical models WMO RFDP proposal in preparation Already included in: – Canada/Québec St. Lawrence Action Plan (SLAP): Environmental prediction working group – EC/NOAA MOU: close collaboration with the Great Lakes Environmental Research Laboratory Superior Michigan-Huron Page 17 – 17 March 2018 Ontario Erie

Coupled modelling system for the Great Lakes • Configuration used for recently published results GEM RDPS 15 km atmospheric model 2 integrations per day Land-surface schemes CLASS or ISBA at 15 km MESH: UU, VV, TT, HU P 0, FB, FI, PR 2 km NEMO model for the Great Lakes RFF, RCH WATROUTE routing model at 15 km Page 18 – 17 March 2018 Q, TQ

Coupled modelling system for the Great Lakes • Configuration to be implemented operationnally (sorry, no results to show yet): GEM HRDPS 2. 5 km atmospheric model 4 integrations per day Land-surface scheme SVS at 2 km UU, VV, TT, HU P 0, FB, FI, PR 2 km NEMO model for the Great Lakes RFF, RCH MESH: WATROUTE routing model at 1 km Page 19 – 17 March 2018 Q, TQ

Predicting net basin supplies to Lake Superior with GEM+ISBA • Overlake • • • evaporation (-E) Net precipitation (P-E) Net basin supplies (NBS=P-E+R) Resid: residual calculation of NBS from lake levels obs. and lake outflow Deacu et al. (2012) J. Hydromet. World's largest lake by area: - Lake area: 82 000 km² - Watershed: 128 000 km² Page 20 – 17 March 2018

Predicting net basin supplies to the Great Lakes with GEM+ISBA • REGN: from • • GEM model outputs at 15 km GLERL Lake. P: assessment by NOAA/GLERL from nearshore obs. of precip. , temperature, humidity, wind and streamflow Resid: residual calculation from lake levels obs. Deacu et al. (2012) J. Hydromet. Page 21 – 17 March 2018

Simulating Great Lakes physical behaviour using GEM+NEMO Water level change [m] Surface temperature [C] Surface currrents [m/s] Surface temperature [C] Page 22 – 17 March 2018 Dupont et al. (2012) WQRJC Ice fraction

Streamflow simulation for subwatersheds (CLASS LSS) Grand River at Iona, MI (4571 km 2) (b) (a) Black River at Watertown, NY (3000 km 2) Haghnegabar et al. (2014), Atmosphere-Ocean Page 23 – 17 March 2018

How did we get there? • Monitoring activities dedicated to improving the model • Parsimonious landscape parameterizations • Coordinated model development Page 24 – 17 March 2018

Monitoring activities dedicated to improving the model Research basins Flux towers Page 25 – 17 March 2018

Parsimonious landscape parameterizations, calibrated parameters • Grouped Response Units (Kouwen et al. , 1993) • WATDRAIN hillslope model (Soulis et al. , 2011) – identify important landscape features – within a grid cell, only keep track of areal coverage of each GRU – assign one parameter set to each GRU – takes slope into account in land-surface, hydrology and atmospheric models – influences runoff but also soil moisture and evaporation Page 26 – 17 March 2018

Coordinated model development • Working as an integrated team on atmospheric, hydrologic and ocean model development by sharing key components: – land-surface model – turbulent flux calculations – computing infrastructure • Using streamflow and water level observations for atmospheric prediction: – to verify NWP forecasts – to tune the water balance of land-surface schemes – eventually, to estimate deep soil moisture • Assessing the impacts of improvements to one component on the environmental prediction system as a whole Page 27 – 17 March 2018

Overlake evaporation prediction • Deacu, Fortin et al. (2012), Journal of Hydrometeorology Lake Superior supplies Average latent heat flux, winter 2011 (W/m²) 200 150 100 50 0 GEM 15 km GEM 10 km Page 28 – 17 March 2018 OAFlux W/m²

Conclusions • At the regional scale, feedbacks to the atmosphere cannot be ignored: if you • are using an atmospheric model product for precipitation and you want to close the water balance using a hydrological model, then you should worry about evapotranspiration computed by the atmospheric model as well Hydrologists and meteorologists have much to gain by collaborating – high-resolution land-surface modelling and data assimilation systems developed by the NWP community are evolving and improving quickly – land-surface models used by the NWP community often lack some basic hydrological processes and need to be calibrated • Be prepared: – NWP systems already provide forecasts of sufficient quality to drive hydrological models for both hindcasting and forecasting at the regional scale – NWP systems will soon provide gridded runoff fields of comparable quality – running coupled models is becoming more and more affordable: water resources engineers will soon be running such systems from their basement! • Systems like MESH offer a good starting point Page 29 – 17 March 2018