PILPS semi-arid experiment: preliminary results

Two four-year long data sets from the USDA Experiment Watershed in the Walnut Gulch, Arizona, are used for an intercomparison analysis of Land Surface Models (LSM) in the PILPS Semi-Arid Experiment (PILPS San Pedro). In the present work we focus on evaluating the consistency of one of the most important functions of a LSM, the net energy partition into latent, sensible, and ground heat components. Additionally, because the data sites correspond to two different semi-arid environments (shrub and grass) it is possible to verify if standard semi-arid representation of the eight participating models is adequate.

The interannual energy closure check showed that all of the schemes have energy balance residuals of less than ± 2 wm-2. The mean annual ground heat flux is zero for three models while other three models lose ground heat flux and the remaining two have positive mean annual ground heat fluxes.

Large differences in the surface energy partitioning between latent and sensible heat fluxes have been observed among models and between sites. The Bowen ratio for the Lucky Hills site (shrub) is 4.86 as opposed to 2.23 for the Kendall site (grass). The among-models range of Bowen ratios for the grassland site goes from 1.38 to 5.30. In the shrub site it ranges between 2.65 and 6.27. The annual mean latent heat flux in Kendall is almost 75% larger than in Lucky Hills. Four models present the same partitioning independently of the site, suggesting that the standard semi-arid parameter classification used in the models may not be adequately representing the different biomes existing in such environments. Only one model presents a relationship between sites in the Bowen ratio that resembles the observed one.

In Lucky Hills all of the models (except two) represent well the annual latent heat flux but underestimate on the sensible by more than 15%. In Kendall, only one model represents the mean annual latent heat adequately while the remaining seven underestimate it. Three models overestimate the sensible heat and the underestimation of the others is not larger than 20%.

Lucky Hills receives annually 10% more net radiation than Kendall. All the models underestimate the net radiation in Lucky Hills but only six do so in Kendall. One model is an outlier; the computed net radiation is less than 50% of the observed. Accordingly models who report low latent heat, also present relative warmer surfaces and lower net radiation.

The observed mean soil temperatures are 293.3 and 292.9 K in Lucky Hills and Kendall respectively. The among-models range for both sites of about 10 K is the same order that the observed standard deviation.

In general the models represent net radiation and sensible heat flux with high correlation and small differences in Root Mean Square Error among models. The diurnal cycle of net radiation and sensible heat is well reproduced by the models although the amplitude is larger in Kendall. Four models overestimate by up to 37% the sensible heat flux close to noon in both sites in the dry season. The night time sensible heat flux loss is overestimated consistently by all models except one the year around.

There are significant differences among schemes in latent and ground heat fluxes, especially in the diurnal cycles. One model performs better in representing the annual ground heat flux in both sites.

During the monsoon season (July to September), where more than 2/3 of the annual precipitation occurs, the increase in observed latent heat is about 100% and the model error of all models is in general smaller than for the whole period.