6. Evaluation of SHTM Accuracy and Uncertainty
6.3. Influence of Different Model Configurations
6.3.3. Influence of Meteorological Input Data
Meteorological input variables for catchment scale simulations with PROMET are provided by an algorithm that supplies hourly time series by temporal interpolation of climate station records. Subsequently, at each time step, a spatial interpolation scheme prepares the meteorological input for each proxel on the basis of the surrounding climate stations data. As a consequence, the meteorological drivers of the land surface process modules already differ from the true conditions found at the investigated site. To assess the deviation of the most influential inputs, Fig. 6-11 presents scatter plots of air temperature and global radiation. Notably, simulated air temperature has a lower daily amplitude (gain < 1), but is generally higher than the true air temperature at Gut Hüll (offset = 2.7), especially at night. Warmer air temperatures that have been mostly recorded at midday are reproduced much better by the interpolation procedure of the model. Overall, the difference between the arithmetic means of the simulated and measured time series for the recorded period (30.04. to 15.07.2003) is +0.52 °C, mainly caused by an overestimation of night time air temperatures by the temporal interpolation of the meteorological module.
Global radiation from the radiation model is actually on average 14 W/m2 lower than the measured one, but average net radiation above canopy is about 40 W/m2 higher in the model than measured in the field. Three reasons seem to lead to this overestimation of net radiation: Especially during night time, the long wave net radiation is biased by overestimated air temperatures in the model. Additionally, too high simulated air temperatures reduce the sensible heat flux from the soil surface into the atmosphere and therefore adding a positive bias on simulated soil temperatures.
Fig. 6-11: Differences between simulated and measured air temperature (left) and
above canopy global radiation (right) at Gut Hüll.
Secondly, reflection and absorption of incoming radiation are strongly related to vegetation and soil parameters that can not be exactly known. Finally, standard
measurements of net radiation generally underestimate true net radiation. HODGES & SMITH (1997) found, that at 15 of 21 measuring sites they investigated, net radiation
was underestimated by about 5% in the daytime and about 45% at night. TWINE et al
(2000) also estimate the daytime accuracy of net radiation measurements to be about 6 %, mainly caused by the heating of their measurement devices by solar radiation. Spatial interpolation of precipitation introduces even more uncertainty, due to the spatial and temporal non-continuous nature of precipitation. But since SHTM does not consider convective heat transport due to infiltrating water, the main effect of precipitation taken in consideration is the increase in soil moisture during and after rain events. Graphs of simulated and recorded soil moisture time series (Fig. 6-12) reveal that too little precipitation is computed in the beginning and during the second half of the validation period. Substituting the interpolated precipitation with the one recorded at the EF site produces a better fit of simulated and recorded soil moisture in the second layer.
Fig. 6-12: Measured soil moisture at different depths (Gut Hüll 2003) vs. simulated
In some way, the performance of the model is improved during the validation period, by using the recorded values of air temperature, global radiation above canopy and precipitation as driving input (Fig. 6-13). The fit of the top layer temperature with the 7.5 cm measurement improves (Table 6-3), but the measures of deviation between the simulated second layer temperature and the associated measurement are not improved. Only the coefficient of determination (R2) points out, that the non- systematic errors, due to the meteorological inputs, are reduced. As expected, the average simulated daily amplitude is greater than before for both soil layers, as a result of increased input from global radiation, and the coefficients of determination for the regression of daily amplitudes increase too, at both soil depths (Fig. 6-14 and Table 6-4), when using the recorded meteorological inputs.
Fig. 6-13: Performance of PROMET with measured meteorological input at Gut Hüll:
Measured vs. simulated hourly soil temperature of the second soil layer.
Fig. 6-14: Performance of PROMET with measured meteorological input at Gut Hüll:
Scatter plots of measured vs. simulated daily soil temperature amplitudes of the upper two soil layers.
This investigation shows that interpolated climate data leads to non-systematic errors in soil temperature simulations that can be reduced for point-scale model runs, when using recorded meteorological data. It reveals also, that in this case, the implemented combination of canopy model, explicit energy balance module and SHTM introduces a systematic overestimation of daily and long-term soil temperature amplitudes in the upper layers. Though, the deviation of the modelled temperatures from the measured ones is rated low. This is because a simplified, mesoscale model, run with standard parameterisations, is compared to point scale measurements of a sensitive variable, that is affected by highly complex conditions.
Table 6-3: Summary of statistical criteria related to different model runs (standard =
standard model run; meteo input = model run with recorded Tair, Rglobal and
precipitation).
Table 6-4: Summary of linear regression between measured and different simulated