Résumé
Dehesa ecosystems of the southwestern Iberian Peninsula are characterized as a savanna-like rangeland dominated by scattered mediterranean evergreen oak trees. We investigated whether isolated trees modify the water balance of this ecosystem and if so, what implications this finding might have on models that assume homogeneity of soil water resources. The water balance of the two ecological components of the dehesas--(1) the tree-grass component, and (2) the open areas between the tree canopies with unshaded grass vegetation--was studied for three consecutive years in three locations in the Sierra Norte de Sevilla region of Andalusia in southern Spain. In this region, annual rainfall was generally between 600 and 800 mm, and the summer drought lasted @?130 d. Soil water storage was measured with a neutron moisture gauge outside and under the tree canopy. Deep drainage between two consecutive census dates was calculated using field-measured drainage characteristics. Evapotranspiration (Ea) and surface runoff were computed from the water balance equation assuming that Ea is limited by Penman potential evapotranspiration. Monthly Ea by annual species in open areas was poorly correlated with rainfall levels in the autumn and was limited during the spring by availability of water in the top 40 cm of soil. During summer, monthly Ea by trees ranged from 30 to 50 mm. Mean annual Ea was 400 mm outside and 590 mm under the tree cover. In open areas, water yield (WY), defined as the sum of deep drainage and surface runoff, ranged from 65 to 100% of total Ea, whereas under the tree canopy WY was only 20 to 40% of the Ea. Under the tree canopy, when annual precipitation was <570 mm, WY was negligible and all precipitation was lost by evapotranspiration. Outside the tree canopy, WY occurred as soon as annual precipitation exceed 250 mm. Models of competition between trees and grass generally assume a spatial homogeneity of soil hydrodynamic properties. Our results, however, show that both soil water storage and evapotranspiration are greater for the tree-grass component. Consequently, these models must account for this spatial variability in water resources according to species.