UFZ -CENTRE FOR ENVIRONMENTAL RESEARCH LEIPZIG-HALLE
Extracting Signals from Satellite retrieved Land Surface Temperature for the Calibration of a Hydrological Model
Matthias Zink, Luis Samaniego, Juliane Mai, and Matthias Cuntz
Helmholtz Centre for Environmental Research - UFZ, Leipzig, Germany (matthias.zink@ufz.de)
1. Motivation
Hydrological models are usually calibrated against discharge measurements, and thus are only trained on information of a few points within a catchment.
This procedure does not take into account any spatio-temporal variability of fluxes or state variables. Satellite data may help to account for this spatial variability. The objective of this study is to calibrate a hydrological model with satellite derived land surface temperature Ts. These data have the advantage to be broadly available even in regions where discharge measurements are barely on hand.
i
z3 z2 z1
ET P
K E
Rn
I
C
q1
q2
q4 q3 x1
x3
x5 x4
x2
4 km
ra
ra
rs
ra H
ET ... evapotranspiration [mm d−1]
H ... sensible heat flux [W m−2]
P ... precipitation [mm d−1]
Q ... observed discharge [mm d−1] Qb ... simulated discharge [mm d−1] ra ... aerodynamic resistance [s m−1]
Rn ... net radiation [W m−2]
Ta ... air temperature [K]
Ts ... satellite land surface temperature [K]
Tcs ... simulated Ts [K]
∆S ... change in soil moisture [mm d−1] λ ... latent heat of vaporization [kJ kg−1]
Pattern Similarity (PS)
6 5
13 7 14
11 10
11 12
Measurement
7 9
13 7 5
12 11
13 11 Simulation
XOR
O
6 O
P P
P P
P P
8
2. Methodology
Mesoscale Hydrologic Model (mHM) To incorporate satellite data into the hy-
drological model mHM [1] an additional module has been developed to estimate Tbs using the energy balance. By closing the water balance with mHM the evapo- transpiration is estimated by
ET = P − Q − ∆S .
Land Surface Temperature Model
Tbs was derived using mHM’s ET estimation for solving the energy balance and the sensible heat equation. Assuming that the soil heat flux is negligible at the daily time scale, we get:
H = Rn − λ · ET
H = ρ · cp · Tcs − Ta ra
Tcs = ra · Rn − λ · ET
ρ · cp + Ta
Optimization Objectives
Q : kE1 + E2k Ts : kE3 + E4k Q & Ts: 23 Q + 13 Ts
E1 = NSE(Q, Q)b
E2 = NSE(ln Q, ln Q)b E3 = PS(Ts, Tbs)
E4 = ρ(Ts, Tbs)
14°E 12°E
10°E 8°E
54°N
52°N
50°N
48°N
0 100km
Ems
Saale
Main
Mulde Weser
Danube Rhine
Neckar
Oder Elbe
3. Input Data & Study Domain
• LSA SAF [2]: Ts, long- and shortwave radiation, albedo, emissivity
• German Weather Service [3]: air temperature, precipitation
• NCEP-CFSR [4]: wind data
• German authorities [5][6]: DEM, pedological and geological data
ET estimation for 2009-07-01 in the Neckar
4. Optimization Regarding Q & T
sEvaluation regarding Evapotranspiration ET The evapotranspiration estima-
tion which has been derived by the calibration against discharge and land surface temperature (Q & Ts) has lower spatial vari- abilities compared to calibrations against only discharge (Q). This behavior is observed especially during summer.
50 100 150 200 250 300 350 Day of year 2009
8 10 12 14 16
SNRET=µET/σET
JF MAM JJA SON D
Q Q & Ts
50 100 150 200 250 300 350 Day of year 2009
8 10 12 14 16
SNRET=µET/σET
JF MAM JJA SON D
Q Q & Ts
Signal to Noise ratio of the catchments Ems (left) and Neckar (right)
Mulde Ems Neckar Saale Main Weser 7
8 9 10 11 12 13 14 15
SNR
Q Q & Ts
Mulde Ems Neckar Saale Main Weser 0.70
0.75 0.80 0.85 0.90 0.95 1.00
NashSutcliffeEfficiency[-] Q Q & Ts
Temporal average of SNR (left), and model performance regarding discharge (right)
1 2 3 4 5 6 7
ET related parameter γi 0.0
0.2 0.4 0.6 0.8 1.0
normalizedparameter
Q Q & Ts
Parametric Uncertainty
γ1 ... infiltration shape parameter
γ2 ... max. capacity of surface water reservoir
γ3 ... permanent wilting point γ4 ... field capacity
γ5 ... fraction of roots (forest) γ6 ... fraction of roots (pervious) γ7 ... fraction of roots (impervious)
5. Predictive Skill of T
s(T
sonly Calibration)
50 100 150 200 250 300 350
Day of Year 2009 0
50 100 150 200 250 300
DischargeQ[m3 s−1 ]
0 20 40 60 80 100 120 140
Precipitation[mmd−1 ]
Observed
Median Precipitation
Percentile range [p5,p95]
50 100 150 200 250 300 350
Day of Year 2009 0
100 200 300 400 500 600
DischargeQ[m3 s−1 ]
0 20 40 60 80 100
Precipitation[mmd−1 ] Observed
Median Precipitation
Percentile range [p5,p95]
Estimation of discharge for Ems (left) and Neckar (right).
Mulde Ems Neckar Saale Main Weser
−0.2 0.0 0.2 0.4 0.6 0.8 1.0
NashSutcliffeEfficiency[-]
Ts
Mulde Ems Neckar Saale Main Weser
−0.2 0.0 0.2 0.4 0.6 0.8 1.0
NashSutcliffeEfficiency[-]
Q Ts
Model performance regarding Q for site specific (left) and transferred parameters (right).
6. Conclusions
• The spatial variability of ET is reduced by incorporating Ts in the calibration.
• The modification of the spatial fields of ET are evoked by a reduced uncer- tainty in the estimation of ET related parameters.
• The dynamics and high flows in Q are well captured by a Ts only calibration.
References
[1] L. Samaniego, R. Kumar, and S. Attinger, “Multiscale parameter regionalization of a grid-based hydrologic model at the mesoscale,” Water Resources Research, 2010.
[2] Satellite Application Facility for Land Surface Analysis (LSA SAF), EUMETSAT.
[3] Deutscher Wetterdienst (DWD), Offenbach, Germany.
[4] National Centers for Environmental Prediction (NCEP), National Weather Service (NOAA), USA.
[5] Bundesanstalt f¨ur Geowissenschaften und Rohstoffe (BGR), Hannover und Berlin, Germany.
[6] Bundesamt f¨ur Kartographie und Geod¨asie (BKG), Frankfurt am Main, Germany.