Validation of single-angle model over Barrax (REFLEX 2012 Campaign)

TSEB model output for Rn, G, H and LE as derived from the AHS overpass at 09:28 UTC on 25th _July,

2012, was validated against ground observations. For this purpose the so-called field-of-view of the local sensors needs to be determined. This is especially important when dealing with high-resolution imagery as is the case in the underlying study.

For the net radiation sensor, 99% of the observations originate from a circle whose diameter is 10 times the sensor height (i.e. 5 meter), although ground surfaces closer to the sensor have a higher weighting. A window of 10 x 10 pixels (i.e. 40 x 40 m) was selected around the location of the observation. The same was done for the soil heat flux observations, which are characterized by a high spatial variation. To at least take this effect into account we chose a similar window as for the net radiation observations. For the turbulent fluxes a different strategy is followed. The “field-of-view”, or footprint (Vesala et al., 2008), of these sensors depend on terrain characteristics, wind speed and wind

direction. The procedure outlined in Timmermans et al. (2009) is used to calculate the footprints of the observation towers at the moment of airborne overpass. Footprint-weighted averages of the model output for H and LE were then compared to the ground observations. Results for the individual sites for all four fluxes are shown in Figure 4.2.

Figure 4.2: Observed versus estimated fluxes with the TSEB model over the Barrax site for 25th _{July 2012 (REFLEX}

campaign).

Model performance was evaluated using difference statistics comprising of the mean absolute difference (MAD), the mean absolute error (MAE) and the root mean square difference (RMSD), Table 4.2.

Table 4.2. Difference statistics for the four observation sites.

H LE G Rn

MAD [Wm-2_{] 22.5 8.7 85.0 51.5}

MAE [%] 13.9 29.4 51.2 13.6

Although only a limited ground observations were available for this particular study, a general good agreement between observed and modelled fluxes is noted from Table 4.2 although performance for Rn and G is less than what is observed in other studies (French et al., 2005; Timmermans et al., 2007). Despite that the focus of the current contribution is on the turbulent fluxes, we explain the results for all four fluxes below.

Generally, modelled net radiation estimates are slightly lower than the observed values. However, the somewhat high difference between observed and modelled net radiation estimates is mainly due to the difference over the vineyard site. This is attributed to the position of the sensor relative to the geometry of the vineyard. At a sensor height of 5 m over a row crop of 2 m height and a sensor field of view of 150° the canopy will be more dominant than it is in the airborne observations. This phenomenon has a greater effect in the shortwave region than in the longwave region under the circumstances during the overpass. Therefore the, locally observed lower albedo resulted in a higher local observation of net radiation. Leaving out this observation results in a far better match between modelled and observed radiation values, that is comparable to previous studies (MAD=36.3 Wm-2_;

MAE=10.3%; RMSD=37.4 Wm-2_).

As mentioned before, soil heat flux quantities may be spatially highly variable. Despite the attempt to position the limited number of available soil heat flux plates at representative locations, this makes validation slightly difficult. Moreover, local calibration of the model coefficient cg in Eq. (4.7) linked the

model soil heat flux estimates to the model estimates of net radiation, which may reach up to 50% thereof in semi-arid ecosystems like the study area. Although this results in a slight underestimate of the soil heat fluxes, the effect on the available energy (i.e. net radiation minus soil heat flux) is partially cancelled out by this phenomenon.

The results for the turbulent fluxes show a good similarity to local observations, with RMSD for H and LE equal to 28 Wm-2 _{(MAE 14%) and 10 Wm}-2 _{(MAE 29%) respectively. The relatively high value of MAE}

for the LE fluxes is due to the low absolute magnitude of this flux. In this semi-arid climate during the summer, over non-irrigated areas, this flux will rarely exceed 5% to 10% of net radiation rates. Not surprisingly, the vineyard observations of LE show the highest values of the observation sites, which is reflected by the model results. However, observations at this location are influenced by the neighbouring fields. During the overpass, the prevailing wind direction was from the South-East. For the camelina, nursery and wheat stubble sites, the footprint analysis revealed that observed fluxes

originated almost solely from the land cover where the observations were made. For the vineyard site however, 64.9% of the observed flux originated from the nearby dry barley stubble field. For validation purposes this effect is taken into account, but “pure” vineyard rates for LE will be higher than those observed by the flux tower.

Model estimates for sensible heat flux show very good agreement with local observations for all sites. When compared with the error obtained by other studies (~30 Wm-2_{) for relatively homogeneous}

canopies (French et al., 2005; Gonzalez-Dugo et al., 2009; Kustas and Norman, 1997; Timmermans et

al., 2007) the results obtained over the current area are even more favorable. Therefore, we regard the

overall model performance of TSEB1 with respect to the estimation of both radiative and especially turbulent fluxes over the heterogeneous Barrax site as reliable.

4.3.2. Comparison between single-angle and dual-angle model (EODIX 2011