Assimilation of H-SAF Soil moisture products H-SAF and HEPEX workshops on coupled hydrology Assimilation of H-SAF Soil moisture products for hydrological modelling in Mediterranean catchments Christian Massari, Luca Brocca, Angelica Tarpanelli, Luca Ciabatta, Tommaso Moramarco Research Institute for Geo-Hydrological Protection (IRPI), National Research Council (CNR), Perugia, Italy http://hydrology.irpi.cnr.it

Numerical Weather Forecasting Climate Prediction Soil moisture in the hydrological cycle … Soil moisture plays a critical role in the hydrological cycle since it determines the partitioning of precipitation into runoff, evaporation and groundwater recharge) but also in atmosphere studies Numerical Weather Forecasting Climate Prediction Shallow Landslide Forecasting Agriculture and Plant Production Flood modelling and forecasting Assimilation of H-SAF SM products in Mediterranean catchments 2

Potential of soil moisture in flood modelling 1. Soil moisture used the determination of the wetness conditions of the catchment before a flood event (antecedent wetness conditions) Brocca et al., 2009 (JHE), Beck et al. 2010 (AWR) Tramblay et al. 2012 Cousteau et al. 2012 1st part of the presentation 2. Soil moisture used for rainfall correction and estimation: Crow et al. 2011 (WRR) Brocca et al. 2013, 2014 (JGR) Pellarin et al. (2013) 2nd part of the presentation In particular … SOIL MOISTURE is a key variable in rainfall runoff transformation!! And in rainfall estimation 3. Soil moisture used the improvement of the modelling of the catchment hydrological response (data assimilation): Brocca et al., 2009 (JHE), 2010 (HESS) Matgen et al., 2012 (HYP) Chen et al. 2011 (AWR)-2014(JoM) Massari et al. 2014 (a,b) HESS, AWR 3rd part of the presentation Assimilation of H-SAF SM products in Mediterranean catchments

H-SAF soil moisture products Most of the climate models projections for the Mediterranean basins have showed that the region is very sensitive to climate change. (IPCC 2012) In particular … SOIL MOISTURE is a key variable in rainfall runoff transformation!! And in rainfall estimation Assimilation of H-SAF SM products in Mediterranean catchments

Antecedent wetness conditions With this term we refer to the relative wetness or dryness of a watershed. In Mediterranean regions they are high variable and can have significant effect on the flow response in these systems during wet weather Assimilation of H-SAF SM products in Mediterranean catchments

The importance of the antecedent wetness conditions… Qp = 870 m3/s Rc = 0.34 Qp = 670 m3/s Rc = 0.17 For a given catchment the response to rainfall can be very different … 35 mm Ponte Nuovo TIBER BASIN 85 mm Assimilation of H-SAF SM products in Mediterranean catchments, Massari et al.

Rd  P Antecedent wetness conditions … having an estimate of the wetness of the catchment before a flood event is crucial for understanding how severe will be our flood SOIL CONSERVATION SERVICE METHOD (SCS-CN) soil moisture before the flood event Rd In the study of the antecedent wetness condition there is the question if some variables can be used to estimate S  API5 The concept is that … 1) A common indicator used in literature for having an estimate of the antecedent wetness condition can be obtained by inverting the relationship of the Scs 2) Sobs is an indicator of the antecedent wetness conditions at catchment scale and is an indirect measure 3) In the study of the antecedent wetness condition there is the question if some variables that can be directly measured in the catchment can be used to estimate Sobs or if there is a physical relation between Sobs and such quantity 4) A number of studies used Two common variables used are… 5) P Brocca et al. 2009 (JoH) , 2009 (JHE) Assimilation of H-SAF SM products in Mediterranean catchments

S vs Θ relation: ERS SCAT Italy Tiber River ERS SCATTEROMETER SOIL MOISTURE DATA Tevere - PN Assino Niccone Many other studies agree that there is a linear inverse relationship between soil moisture and the antecedent wetness conditions Beck et al., 2010 (JSTARS) (Australia) Tramblay et al., 2010 (JoH), 2011 (NHESS) (France) Tramblay et al., 2012 (HESS) (Morocco) 11 catchments 100-5000 km2 Cerfone Genna Caina Brocca et al., 2009 (JoH); 2009 (JHE) Timia Topino Nestore Assimilation of H-SAF SM products in Mediterranean catchments

Exploiting the soil moisture importance … Why not embed directly the relation between the parameter S and the soil moisture into the hydrological model? Assimilation of H-SAF SM products in Mediterranean catchments

“Observed” Soil Moisture Soil moisture estimate inside an event based model… “Observed” Soil Moisture S-θ relationship Rainfall excess calculation Routing IUH SCS-CN Advantages: 1) No need of continuous rainfall and evapotranspiration datasets. Good in poorly gauged areas. 2) Parsimony and simplicity. Good for operational purposes. Rainfall P Observed rainfall “Simplified Continuous Rainfall Runoff” model (SCRRM, Massari et al. 2014, HESS) Assimilation of H-SAF SM products in Mediterranean catchments

Initial conditions from MISDc model (Brocca et al. 2011) Application of the model in Attica (Greece)... Area 109 km2 Initial conditions from NS MISDc model (Brocca et al. 2011) 0.79 ERA-Land 0.80 ASCAT 0.73 AMSR-e MODEL PERFORMANCES Early Warning System for Flood and Fire forecasting Mean Nash-Sutcliffe Index Lumped version of the model NS=1 perfect model NS<0 bad performance 16 flood events 2009-2013 Event 24-Feb-2011 Era-Land AMSR-e ASCAT The model has provided performance comparable with those obtained by a classical continuous model MISDc MISDc (Brocca et al. 2011) Assimilation of H-SAF SM products in Mediterranean catchments

Initial conditions from Application of the model in Attica (Greece)... Area 109 km2 RESULTS IN VALIDATION MODEL PERFORMANCES Early Warning System for Flood and Fire forecasting Initial conditions from NS MISDc model (Brocca et al. 2011) 0.55 ERA-Land 0.56 ASCAT 0.48 AMSR-e 0.47 In situ 0.43 Mean Nash-Sutcliffe Index 16 flood events 2009-2013 NS=1 perfect model NS<0 bad performance Lumped version of the model The model has provided performance comparable with those obtained by a classical continuous model MISDc Assimilation of H-SAF SM products in Mediterranean catchments

Application of the model to 35 Italian catchments pixels inside the catchments were averaged for obtaining a single time series To obtain the root zone soil moisture we used the exponential filter for H07 The model was applied to 35 Italian catchments areas ranging from 800 to 7400 km2 Period 2010-2013 593 flood events We use H07 and H14 Wagner et al., 1999 (RSE) and a depth parameter z which account for the information given by the layers of H14 Assimilation of H-SAF SM products in Mediterranean catchments

H14 H07 Application of the model to 35 Italian catchments: results Results in terms of Nash-Sutcliffe efficiency index Good Poor Mean NS equal to 0.58 Mean NS equal to 0.61 H14 H07 Massari et al. 2014 (Hydrology) (Under review) MISDc model used as benchmark NS=0.66 Assimilation of H-SAF SM products in Mediterranean catchments

Application of the model to 35 Italian catchments Po River at Carigliano Area=3570 km2 MISDc NS =0.84 H07 NS =0.79 H14 NS=0.81 Sebbene il bacino sia molto strumentato il contenuto d’acqua consente di fornire una buona stima della portata questo sottolinea come esso possa essere usato con successo per la previsione delle piene Assimilation of H-SAF SM products in Mediterranean catchments

Exploiting soil moisture derived rainfall in flood modelling: SM2RAIN Another important topic which we try to respond is that if we can use soil moisture for estimating rainfall and integrating rainfall with observed for improving flood prediction that has to do with soil moisture is if can have information of rainfall from soil moisture observation and then use this information for flood modelling Assimilation of H-SAF SM products in Mediterranean catchments

BOTTOM-UP PERSPECTIVE Rainfall estimation: Top down vs bottom up perspective radar raingauge Remote sensing of rainfall Remote sensing of soil moisture Is it raining? TOP-DOWN PERSPECTIVE The concept of the method is that instead of looking at the sky to see how much rainfall we look at the ground so we used the soil as natural raingage BOTTOM-UP PERSPECTIVE CAN WE USE SOIL MOISTURE DATA TO INFER THE AMOUNT OF WATER FALLING INTO THE SOIL??? Assimilation of H-SAF SM products in Mediterranean catchments 17

SM2RAIN concept RAINFALL SOIL MOISTURE The soil moisture variations are strongly related to the amount of rainfall falling into the soil. Therefore, we can use soil moisture observations for estimating rainfall by considering the “soil as a natural raingauge”. Assimilation of H-SAF SM products in Mediterranean catchments

SM2RAIN algorithm + Inverting for p(t): Assuming: Brocca et al 2013, GRL relative saturation surface runoff precipitation drainage Z< Surface runoff: r(t) Evapotranspiration: e(t) Drainage: g(t) Precipitation: p(t) Soil water balance equation Soil water capacity soil water capacity = soil depth X porosity evapotranspiration Inverting for p(t): Assuming: + during rainfall Assimilation of H-SAF SM products in Mediterranean catchments

SM2RAIN Performance: satellite soil moisture data (H07) Snow-Freeze Mountain Desert Forest GPCC global precipitation climatology center PCC Global Precipitation Climatology Centre monthly precipitation dataset from 1901-present is calculated from global station data. Correlation map between 5-day rainfall from GPCC and the rainfall product obtained from the application of SM2RAIN algorithm to ASCAT data (VALIDATION period 2010-2011) Assimilation of H-SAF SM products in Mediterranean catchments

SM2RAIN Performance: in situ data 12 sites, 5 countries At least one year of rainfall and soil moisture data at hourly time scale. Most of the sites are located in Italy Cerbara Colorso Ingegneria Torano Bagnoli Cordevole Ressi Bibeschbach Valescure Remedhus Yanco Salento # Site Country Data Period Depth (cm) Climate 1 Bagnoli Southern Italy 2007-2008 25-35 Arid 2 Torano 2007-2010 3 Salento 2010-2014 15-25 4 Cerbara Central Italy 2011 5-15 Semiarid 5 Colorso 2002-2004 6 Ingegneria 2009-2010 2.5-7.5 7 Cordevole Northern Italy 2009 0-30 Humid 8 Ressi 2012-2013 9 Bibeschbach Luxembourg 4-7 10 Remedhus Spain 2007-2009 11 Valescure France 2008-2010 12 Yanco Australia 2001-2011 GPCC global precipitation climatology center PCC Global Precipitation Climatology Centre monthly precipitation dataset from 1901-present is calculated from global station data. Mean R = 0.88 Assimilation of H-SAF SM products in Mediterranean catchments

Can we use the SM2RAIN algorithm for improving flood modelling? Exploiting the SM2RAIN derived rainfall… Can we use the SM2RAIN algorithm for improving flood modelling? Assimilation of H-SAF SM products in Mediterranean catchments

“Observed” Soil Moisture SCRRM model … “Observed” Soil Moisture S-θ relationship Rainfall excess calculation Routing IUH SCS-CN The model uses external source of soil moisture only for initialization Rainfall P Observed rainfall Assimilation of H-SAF SM products in Mediterranean catchments

“Observed” Soil Moisture We considered a modified configuration of the previous model … Initialization “Observed” Soil Moisture S-θ relationship Rainfall excess calculation Routing IUH SCS-CN Rainfall estimation SM2RAIN rainfall Integrated rainfall via Nudging Rainfall P Observed rainfall Soil moisture in this case is used twice, first for initialize the model and then indipendely to correct rainfall K=0  Pint=Pobs K=1  Pint=PSM2RAIN Assimilation of H-SAF SM products in Mediterranean catchments

France - Valescure 30 minutes interval recorded data of Rainfall Catchment of Valescure - France Area: 3.83 km2 Elevations: from 244 m to 815m ASL Mean slope: 56 % France - Valescure (Tramblay et al. 2010) 30 minutes interval recorded data of Rainfall Temperature Discharge Soil moisture (30 cm depth from a representative location) Assimilation of H-SAF SM products in Mediterranean catchments

we obtained an increase in performance from NS=0.49 to NS=0.83 Results … we obtained an increase in performance from NS=0.49 to NS=0.83 when soil moisture is used also for correcting rainfall during the flood event Improving Assimilation of H-SAF SM products in Mediterranean catchments

A similar approach with satellite data Runoff simulation by using as input 1. Observed rainfall 2. SM2RAINASCAT 3. SM2RAINASCAT + Obs. Rainfall Tiber River catchment @ Ponte Felcino central Italy Ciabatta et al. (2014 in prep.) MISDc rainfall-runoff model Assimilation of H-SAF SM products in Mediterranean catchments 27

Data assimilation of soil moisture for improving flood predictions Assimilation of H-SAF SM products in Mediterranean catchments

Satellite soil moisture … Data assimilation of soil moisture - is there a real benefit? In rainfall-runoff modelling … In the last decades many studies performed data assimilation experiments and tested different techniques and approaches for soil moisture assimilation within rainfall-runoff modelling … In situ soil moisture … Satellite soil moisture … Loumagne et al., 2001 (HSJ) Houser et al. (1998) (WRR) Aubert et al., 2003 (JoH) Pauwels et al., 2001, 2002 (JoH, HP) Anctil et al., 2008 (JoH) Crow et al. 2005 (GRL Matgen et al., 2006 (IAHS) Lee et al. 2011 (AWR) Reichle et al (2008) Matgen et al. (2011) (AWR) Crow and Ryu (2009) (HESS) Brocca et al., 2010 (HESS) Chen et al. (2011) (AWR) Montzka et al (2011) (JoH) Brocca et al. 2012 (IEEE TGRS) Chen et al. 2014 (JoH) Garreton et al. 2014 (HESSD) However, few studies (to our knowledge) demonstrated the real value of assimilating true soil moisture data for improving runoff prediction and there are still many controversial issues to be solved …. Assimilation of H-SAF SM products in Mediterranean catchments

The problem is often not the ingredients but the cooking technique … Data assimilation of soil moisture - a complex recipe? Data Assimilation ingredients Observations In situ Satellite data Land surface model data Rainfall runoff model Lumped distributed Single layer Multiple layers Assimilation technique Variational Sequential The problem is often not the ingredients but the cooking technique … “Cooking” techniques Bias Handling Variance matching Least square rescaling Cdf matching Triple collocation Filtering Soil water index (Swi) Others No filtering Observation error Temporal variability of the obs. error Spatial correlation between the observations Masking Given etc… one should obtain the same resutls however.. Depending how one cooks the ingredients… Model error Model error covariance estimation (i.e. EnkF: ensemble size) What to perturb. (parameters, inputs, states etc …) How to perturb (amount of perturbation) Assimilation of H-SAF SM products in Mediterranean catchments

Data assimilation of soil moisture - a complex topic? Bias handling Filtering Temporal variability Spatial variability What to perturb Bias correction Ensemble size Ensemble verification Given a model and the EnKF as a assimilation technique we can obtain 2300 different results The task can be even more difficult if we consider different catchment sizes and climatic conditions …. Assimilation of H-SAF SM products in Mediterranean catchments

Toward a systematic study: Tiber River (Central Italy) Tiber River Basin Basin Area (km2) Tevere at Ponte Felcino 2080 Nestore at Marsciano 725 Chiani at Morrano 457 Topino at Bevagna 440 Marroggia at Azzano 258 Niccone at Migianella 137 6 sub-catchments (140-2080 km²) Rainfall-runoff data from 1989 at hourly time resolution Assimilation of H-SAF SM products in Mediterranean catchments 32

assimilation of H07 during 2010 - 2013 Hydrological model RR model with 1 layer (Brocca et al., 2010 HESS) Lumped version calibration period: 1989 - 2009 assimilation of H07 during 2010 - 2013 Assimilation of H-SAF SM products in Mediterranean catchments

Ensemble Kalman Filter yk Nonlinearly propagates ensemble of model trajectories. Can account for wide range of model errors (incl. non-additive). Reichle et al., 2002 (MWR) xki state vector (eg soil moisture) Pk state error covariance Rk observation error covariance Ensemble size N=50 members Perturbing parameters and inputs ENSEMBLE TEST De Lannoy et al. 2006, JGR We assume observation error constant in time Assimilation of H-SAF SM products in Mediterranean catchments

Data assimilation setup FLAGGING AND AVERAGING OF THE OBSERVATION Data removed when quality flags of H07 >1 Averaging pixels fallen inside the catchment using the weighted linear inverse method FILTERING SWI: Soil Water Index t: time SSMti: relative Surface Soil Moisture [0,1] ti: acquisition time of SSMti T: characteristic time length Wagner et al., 1999 (RSE) RESCALING 1) Variance matching (Brocca et al 2010, HESS) 2) Linear least square rescaling (Yilmaz and Crow 2012, JM) 3) CDF matching (Reichle and Koster, 2004) Assimilation of H-SAF SM products in Mediterranean catchments 35

Centroid of the catchment Chiani catchment: results Ascat Pixel Centroid of the catchment H07 pixel Centroid Chiani catchment Area 457 km2 The satellite measure only the first centimeters of the soil while the model simulate the soil moisture of the root zoneWe have an observation and the outcome of the model whi Assimilation of H-SAF SM products in Mediterranean catchments 36

Chiani catchment Calibration 1989-2009 NS=0.73 Validation 2010-2013 NS=0.72 Calibration 1989-2009 NS=0.73 Assimilation of H-SAF SM products in Mediterranean catchments 37

Linear least square rescaling Chiani catchment: observation handling CDF rescaling (Reichle and Koster, 2004) Model Filtering SWI T=67 days Model SSM H07 Linear least square rescaling Yilmaz and Crow (2012) Variance matching Brocca et al (2010) The satellite measure only the first centimeters of the soil while the model simulate the soil moisture of the root zoneWe have an observation and the outcome of the model whi Assimilation of H-SAF SM products in Mediterranean catchments 38

Results: Chiani river basin data assimilation Rescaling: Linear least square sobs= 0.05 NSNoASS=0.72 NSASS=0.85 Assimilation of H-SAF SM products in Mediterranean catchments 39

Results: Chiani river basin most important flood events Improving Rescaling linear least square sobs= 0.05 NSval=0.63 NSNoASS=0.66 NSASS=0.82 Assimilation of H-SAF SM products in Mediterranean catchments 40

Variance matching Linear least square CDF EFF NS Results: Chiani river basin Variance matching Linear least square CDF Effe=38.3 Effv=43.0 Effe=41.2 Effv=45.7 Effe=33.8 Effv=38.9 EFF sobs=0.05 no ass sobs=0.05 no ass sobs=0.03 no ass NS=0.84 NS=0.85 NS=0.83 NS Deteriorations Improvements no ass no ass no ass Assimilation of H-SAF SM products in Mediterranean catchments 41

Results for all cathcments Assimilation Basin NS Validation (Ass -LS) (Ass - VM) (Ass - CDF) T (days) sobs Tevere Ponte Felcino 0.61 0.81 0.78 56 0.03 Nestore Marsciano 0.87 0.91 101 0.07 Chiani Morrano 0.72 0.85 0.84 0.83 67 0.05 Topino Bevagna 0.41 0.40 34 0.09 Marroggia Azzano 0.69 0.71 62 0.30 Niccone Migianella 0.65 0.79 0.76 68 DETERIORATION IMPROVING Improving Marginal or no improvement Assimilation of H-SAF SM products in Mediterranean catchments 42

A lot of work remain to be done for: Conclusions H-SAF soil moisture products may offer a great benefit in rainfall-runoff modelling in Mediterranean catchments and can be used directly both in RR modelling for operational purposes and as additional information for improving flood modelling in the contest of data assimilation A lot of work remain to be done for: a better characterization of the errors to associate to the satellite observations (there should be a stronger connections between users and developers) the understanding the effect of the RR model structure, rescaling, filtering on the results of the data assimilation Exploring the potentiality for a larger range of climatic conditions and catchment characteristics Assimilation of H-SAF SM products in Mediterranean catchments 43

Thanks for your attention Christian Massari christian.massari@irpi.cnr.it http://hydrology.irpi.cnr.it/people/c.massari 44