Assimilation of satellite soil moisture observations in ...

2 downloads 0 Views 5MB Size Report
Christian Massari(1), Luca Ciabatta(1), Luca Brocca(1), Stefania. Camici(1), Angelica Tarpanelli(1), Ivan Marchesini(1) and Silvia Puca(2). (1) Research Institute ...
Assimilation of satellite soil moisture observations in hydrological modelling: an intercomparison study between different sensors in Europe Christian Massari(1), Luca Ciabatta(1), Luca Brocca(1), Stefania Camici(1), Angelica Tarpanelli(1), Ivan Marchesini(1) and Silvia Puca(2) (1) Research Institute for Geo-Hydrological Protection, National Research Council, Perugia, Italy ([email protected]) (2) Italian Civil Protection Department, Rome, Italy ([email protected])

State of the soil moisture sensors ESA CCI Soil Moisture

PASSIVE MICROWAVE (L-,C-,X-band)

Dorigo et al. (2017)

a long-term consistent soil moisture time series, based on active and passive data, suitable for climate change studies and not only Passive, Active, Combined

ACTIVE MICROWAVE (C-band)

Coverage 1978-2015 - 0.25°, - daily www.esa-soilmoisture-cci.org/

Data assimilation of soil moisture - where we are … Assimilation Satellite soil moisture Pauwels et al., 2001, 2002 (JoH, HYP) Crow et al. 2005 (GRL)

Sensor ERS1/2 TMI X band

Francois et al., 2003 (HSJ)

ERS1

Brocca et al., 2010, 2012, 2013 (HESS, IEEE TGRS, IGARSS)

ASCAT

Draper et al., 2011 (HESS)

ASCAT, AMSRE

Chen et al., 2011, 2014 (AWR, JHM)

ASCAT, SMOS

Matgen et al., 2012 (AWR)

ASCAT

Alvarez-Garreton et al., 2014, 2015 (JoH, HESS)

ASCAT, AMSRE, SMOS

Wanders et al., 2014 (HESS)

ASCAT, AMSRE, SMOS

Lievens et al., 2015 (RSE)

SMOS

Massari et al., 2015 (RS)

ASCAT

Laiolo et al. 2015 (JAEG)

ASCAT, SMOS

Leroux et al. 2016 (HESS)

SMOS

Liu et al. 2017 (JHH)

ESA CCI

There is not a well accepted opinion whether hydrological assimilation of satellite soil moisture has a benefit or not and there are still many controversial issues to be solved …

Data assimilation of soil moisture a complex recipe? MISDc model 1 Layer Ensemble Kalman Filter ASCAT 2007-2013 Area 150-2000 km2 Model error Observation error Rescaling technique Seasonality Basin size Soil type and basin hydrology Flow magnitude

Massari et al. 2015 (Remote Sensing)

What we need … Some limits of the previous studies (not all): - Local studies (effect of local conditions: climate, topography, catchment hydrology) and few catchments analysed - Results can be model dependent (model structure is important, Chen et al. 2011, Brocca et al. 2012) - Short analysis period Then … ü 1. Studies involving more statistically significant samples ü 2. Long-period experiments > 5-6 years ü 3. Testing different models within the same DA experiments (Louizu et al. 2017 in review)

The study goal

Exploring the capacity of active and passive soil moisture products to improve hydrological simulations Data -

-

GRDC and EWA datasets: discharge data from 1950 to now (we selected 2003-2011, 9 years of data containing sufficient observations of the most used sensors in the last decade, i.e., AMSRE and ASCAT) Rainfall and Temperature data from the European high-resolution gridded data sets, E-OBS (Haylock et al., 2008) CCI soil moisture Active, Passive and Combined v03.2 (available from 1978 to now)

226 catchcments

The study goal

Hydrological model and parameter calibration -

Model: MISDc 2 layers Calibration (SCEUA algorithm, Duan et al. 1993) Score KGE (Kling-Gupta efficiency index, Gupta et al., 2009) KGE = 1 - ED ED = (r - 1) + (a - 1) + ( b - 1) 2

2

2

-¥ < KGE £ 1

r=correlation coefficient α= relative variability between observed and simulated discharge β=Bias normalized by the standard deviation 12 parameters in total Wmax1=100 mm storage Wmax2=parameter

Data assimilation: pre-processing of the observation CCI soil moisture observations were pre-processed before the assimilation 1. Masking: removing flags (>0 e.g., dense vegetation, snow cover, frozen soil)

CCISM Active Passive Combined

ESA CCI SM v03.2 (0.25°, daily) 2. Averaging: mean CCI soil moisture calculation - improved weighted merging based on signalinside the catchment to-noise-ratio estimates - Data record 1978-2015 3. Filtering: using the Exponential filter (Wagner et - improved temporal sampling for most periods. al. 1999). Match of the temporal variability - improved estimates of random errors and enhanced metadata descriptions

4. Rescaling: matching mean and variance between the model and the CCI observations.

CCI*SM Active Passive Combined

Data assimilation We update the surface layer and leave the model to update the second layer Assimilation scheme very simple * SM 1+ (t ) = SM 1- (t ) + G × (CCI SM (t ) - SM 1- )

G=0 We trust in the model (no assimilation) G=1 We trust in the CCI*SM G was calibrated by maximizing the KGE during the data assimilation period G=arg(MAX(KGE))

Expetiment setup – Assimilation periods We considered two eras: 1. AMSRE-Era 2003-2006 2. ASCAT-Era 2007-2011 Sensors used within the CCI SM project

For each era we assimilated ACTIVE, PASSIVE and COMBINED products

Active: AMI-WS (ERS1/2) Passive: AMSRE 2003-2006

Active: ASCAT Metop A Passive: AMSRE & Windsat & SMOS (2 years) 2007-2011

Modified from Dorigo et al. 2017

Results: Calibration KGE

NS

Low flows

High flows Pa

NS = 1 -

å ( Qsim - Qobs )

2

å (Q

2

t =1 Pa

t =1

0.8

obs

- Qobs )

Pa

NS(log Q ) = 1 -

+ e ) - log ( Qobs + e ) ùû

2

å éëlog ( Q

+ e ) - log ( Qobs + e ) ùû

2

sim

t =1 Pa

obs

t =1

0.75 0.52

0.6

å éëlog ( Q

−∞ < NS ≤ 1 −∞ < NSlnQ ≤ 1

0.4 0.2

0.03

0

KGE

NS

NSlnQ

We removed catchments with KGE