Plant identity and shallow soil moisture are primary drivers of stomatal conductance in the savannas of Kruger National Park

Abstract

Our goal was to describe stomatal conductance (gs) and the site-scale environmental parameters that best predict gs in Kruger National Park (KNP), South Africa. Dominant grass and woody species were measured over two growing seasons in each of four study sites that represented the natural factorial combination of mean annual precipitation [wet (750 mm) or dry (450 mm)] and soil type (clay or sand) found in KNP. A machine-learning (random forest) model was used to describe gs as a function of plant type (species or functional group) and site-level environmental parameters (CO2, season, shortwave radiation, soil type, soil moisture, time of day, vapor pressure deficit and wind speed). The model explained 58% of the variance among 6,850 gs measurements. Species, or plant functional group, and shallow (0-20 cm) soil moisture had the greatest effect on gs. Atmospheric drivers and soil type were less important. When parameterized with three years of observed environmental data, the model estimated mean daytime growing season gs as 68 and 157 mmol m-2 sec-1 for grasses and woody plants, respectively. The model produced here could, for example, be used to estimate gs and evapotranspiration in KNP under varying climate conditions. Results from this field-based study highlight the role of species identity and shallow soil moisture as primary drivers of gs in savanna ecosystems of KNP.

Fig 1. Average daily values of (a) relative humidity, (b) shortwave radiation, (c) temperature, (d) vapor pressure deficit (VPD), and (e) wind speed over two years at the Pretorioskop study site, Kruger National Park, South Africa.

Fig 2. Stomatal conductance (gs) of grasses (a) and woody plants (b) measured over five years in four sites in Kruger Park that varied in precipitation regime from 400 mm mean annual precipitation (MAP; Dry) to 750 mm MAP (Wet) on either clay or sand soils.

Fig 3. Variable importance in random forest models of stomatal conductance for (a) grasses and (b) woody plants.

For both grasses and woody plants shallow (0–20 cm) soil moisture and soil type (i.e., clay or sand) were the most and least important variables describing g_s, respectively. Variable importance is the difference in prediction error before and after a predictor variable is randomly permutated. Large variable importance values indicate that specifying the variables incorrectly increases prediction error. See text for further variable descriptions.

Fig 4. Partial dependence of grass and woody plant stomatal conductance (g_s) on environmental drivers.

Values shown are predicted grass and woody plant g_s as a function of each variable when all other variables are held at their mean value. Drivers listed in descending order of importance to grass g_s: shallow soil moisture (0–20 cm; a and b), vapor pressure deficit (VPD; c and d), atmospheric CO₂ (e and f), time of day (g and h), deep soil moisture (50–150 cm; i and j), wind speed (k and l), and shortwave radiation (m and n). Note that axis values are different for grasses and woody plants.

Fig 5. An example of diurnal variation in soil water potential (MPa) at 20 cm over six days at the Letaba study site, 2010.

Soil water potentials were greatest each day between 8 am and 11 am and smallest each day between 2 pm and 7 pm. Consistent with hydraulic redistribution, this indicated that shallow soils gradually became ‘wet’ each night and quickly dried between roughly 9 am and 5 pm.