An ecohydrological approach to predicting regional woody species distribution patterns in dryland ecosystems
Introduction
Drylands are extensive, covering 30% of the Earth’s land surface and 50% of Africa [62], [66]. Furthermore, dryland ecosystems support a large fraction of the human population and most pastoralist societies [62]. Two challenges facing the sustainability of pastoralist societies are land use and climate change [32]. Although changes in land use are an important and a major concern, we focus our efforts on predicting natural ecosystem responses to climate change. In this study, we present a modeling framework for addressing the impacts of rainfall on the distribution of woody vegetation species. We apply our framework to an African dryland watershed in central Kenya.
Prior studies demonstrate that dryland ecosystems are sensitive to shifts in rainfall climatology [12], [25], [59], [66], [69], [70]. Regional rainfall patterns are primarily governed by larger scale phenomena, such as the Intertropical Convergence Zone (ITCZ), Hadley cells [21], and global teleconnections [46]. Recent changes and predicted future changes in rainfall patterns vary greatly around the globe [32]. For example, in the southwest United States, rainstorms have become more frequent but less intense, with decreasing storm depths caused by changes in the Hadley cell over the Pacific Ocean [21], [28]. Using the observed changes in rainfall patterns, climate model predictions indicate a decrease of 15% in water availability (annual rainfall minus evapotranspiration) over the next two to three decades [69]. Comparatively, rainstorms in sub-Saharan Africa are predicted to become more intense and less frequent [29]. Explanations for these changes include increases in surface albedo and increases of dust particle concentrations and drop nucleation [77]. Decreased infiltration and expansion of bare soils, amplified by changes in the distribution of rainfall for sub-Saharan Africa, will lead to greater runoff and decreased soil moisture.
Dryland vegetation responds strongly and rapidly to changes in soil moisture through shifts in stomatal conductance [33], and soil moisture dynamics are governed by the daily arrival of storms and the dynamics of plant water use between storm arrivals. Despite the importance of daily rainfall processes in determining dryland soil moisture and plant water use, a review of 41 recent models on arid and semi-arid grazing found only five models that included daily rainfall as a forcing variable [72]. The vegetation response to highly temporally variable rainfall pulses has been found to be nonlinear [2], [33] and temporal averaging can lead to erroneous solutions of the vegetation response. Previous studies have found daily rainfall to be well represented by stochastic processes [33], [59]. In addition to temporal variations, rainfall in many dryland ecosystems is highly heterogeneous in space, which is caused by the occurrence of localized convective storms [12].
Besides rainfall, the representation of vegetation needs to be further refined. Although fractional woody cover is the most common description of dryland vegetation structure, the ubiquity of water limitation indicates that differential water use across functional groups or individual species may also be critical. Plant water use is species-dependent [73], which suggests that differences in dryland species composition can result in substantial differences in basin water use [39], [53]. Given the complexity of ecological and environmental factors that together determine the spatial patterns of species occurrence, a general framework for describing factors that govern geographical patterns of species extent has not yet been developed. However, the field of biogeography has addressed species distributions and local inter-species competition using the niche concept [18], [24], [27], [30]. Here, we adopt Hutchinson’s definition, which defines a niche as the subset of environmental conditions that affect a particular organism and determine its absolute fitness [30], [34]. We refer to fitness as the ability of an individual to grow, reproduce, and survive and will provide a formal definition later in the paper. Modeling of species distributions using niche-based approaches has been primarily studied in two ways. The first method uses correlation between abiotic drivers (e.g. soil, temperature, rainfall) and species occurrence to statistically represent the niches in multidimensional space [3], [50]. While this method obtains satisfactory correlation with observed vegetation patterns, it fails to address specific causal mechanisms governing species distribution patterns. In addition, this framework is difficult to extrapolate to novel groups of species [3], [34]. The second method adopts a mechanistic approach that describes plant fitness, explicitly modeling growth, reproduction, and survival. However, at regional and continental scales, these models require many input parameters making them difficult to parameterize [43], [44], [54].
This paper is motivated by a need to develop more detailed frameworks of coupled water and vegetation systems in drylands while avoiding overparameterization and model complexity in order to address issues such as climate change. The two main challenges we address with our proposed framework are the representation of rainfall as a daily stochastic process and the refinement of fractional woody cover into predictions of the spatial distribution of species.
In this work, we use a previously-developed stochastic soil water balance model [35], [36], [55], [58] to represent the interactions between climate, soils, and plants. Such models have been used on a precipitation gradient in the Kalahari desert [63], [64], [65], [74], [75], [76], the Upper Rio Salado basin of New Mexico [7], and a California oak savanna [9]. The model is nonspatial and determines average, steady-state growing season values of runoff, leakage, interception, and evapotranspiration of a single species. We expand the model into a spatial context by applying an optimality trade-off hypothesis which states that dryland vegetation patterns are constrained by maximization of water use and simultaneous minimization of water stress [8]. The model is forced by daily rainfall, which is represented as a marked Poisson process described by the mean depth of daily rainfall and the mean arrival rate of storms. For the basin presented in this study, we generate spatial estimates of each parameter and analyze temporal trends in the variables using a long-term daily precipitation dataset. The stochastic soil water balance model is used to determine each component of a fitness vector for all species, where growth and reproduction are estimated as the ratio of evapotranspiration to growing season rainfall, and plant survival is estimated with the dynamic water stress over the growing season. The contribution of each of these components to overall plant fitness is unknown. The skill and accuracy of our model results are tested against two separate model cases that represent the end members on a continuum of possible model skill and accuracy. The lower bound case is described by selecting a species at random. The upper bound case, a neural network, is a powerful predictive model for linear and nonlinear data [26], [37].
This research has three objectives. The first is to extend an existing mechanistic approach in order to predict changes in the distribution of woody plant species in a central Kenyan watershed caused by specific climate change scenarios based on historical rainfall observations. Our second objective is to predict changes in species patterns and relative abundance in response to changes in total growing season rainfall or changes in the variability of growing season rainfall. Our third objective is to determine the relative importance of plant water use and plant water stress in determining the overall distribution of three Acacia tree species, Acacia drepanolobium, Acacia tortilis, and Acacia xanthophloea.
The remainder of the paper is organized as follows: first, in the methods section, we describe our study site and key meteorological data necessary to apply the stochastic water balance model. We also introduce our modeling framework and calibration procedure we use to apply the fitness vector framework we have adopted for predicting individual species distributions. The methods section ends with a brief examination of model sensitivity to key parameters. Our results provide model predictions of species distribution patterns based on observed trends in rainfall variability, as well as the differing effects of changing either the mean or variability of growing season rainfall. Our results conclude with an examination of our fitness vector framework’s performance as well as the inferred relative importance of water use and water stress in governing species distribution within our study basin. We conclude with a discussion section that explores the implications of our fitness vector results, the potential for our approach to explain observed shifts in dryland species composition, and a discussion of model limitations and plans for future refinement.
Section snippets
Basin description
Our study basin is the Upper Ewaso Ng’iro river basin (15,200 km2), which is located in the central Kenya highlands. The basin spans gradients of elevation, temperature, precipitation, and contains eight different soil texture classes (Fig. 1, Fig. 2). Because the basin is located on the equator, the annual climate consists of two rainy seasons caused by the Intertropical Convergence Zone (ITCZ). Temperature and precipitation patterns are heavily influenced by elevation (Figs. 2a and 4a). Soil
Modeled changes in species patterns
We investigated changes in the species distribution patterns using two sets of rainfall scenarios. With the first set of scenarios we explored changes in the species patterns driven by the mean slope of linear trends in time for the basin rainfall parameters, α = 0.04 mm yr−1 and λ = −0.0018 day−1 yr−1. The changes in rainfall were applied uniformly across the basin and the new species patterns were estimated with the model and fitness vector (Fig. 5). The modeled changes in species cover for the Upper
Water use versus stress avoidance
While early studies on optimality regarding vegetation patterns focused solely on maximization of resource use [16], [17], more recent approaches have begun to consider the additional costs of chronic resource scarcity that accompany over-consumption of limited resources [8], [68]. However, we are not aware of any prior uses of ecohydrological optimality approaches to predict basin-scale patterns of individual species distribution. We address the issue of species distribution through the
Acknowledgements
We thank the Princeton Institute of International and Regional Studies (PIIRS), Princeton Environmental Institute (PEI), Princeton Grand Challenges, and the Mpala Research Conservation Center for their support. K.K.C. also acknowledges the support of the NSF through Grants DEB-742933 and EAR-0847368. The authors would also like to thank Natural Resource Monitoring, Modeling and Management Project (NRM3) of Nanyuki, Kenya for providing the extensive climatic dataset for the study. A special
References (78)
Vegetation response to rainfall intermittency in drylands: results from a simple ecohydrological box model
Adv Water Resour
(2007)- et al.
On the coupled geomorphological and ecohydrological organization of river basins
Adv Water Resour
(2005) The use of plant water relations to characterize tree species and sites in the drylands of northern Ethiopia
J Arid Environ
(2005)Biogenic hydrocarbon emissions and landcover/climate change in a subtropical savanna
Phys Chem Earth Pt B: Hydrol Oceans Atmos
(1999)- et al.
Predictive habitat distribution models in ecology
Ecol Model
(2000) Plants in water-controlled ecosystems: active role in hydrologic processes and response to water stress – IV. Discussion of real cases
Adv Water Resour
(2001)Plants in water-controlled ecosystems: active role in hydrologic processes and response to water stress – II. Probabilistic soil moisture dynamics
Adv Water Resour
(2001)Application of neural networks to modelling nonlinear relationships in ecology
Ecol Model
(1996)Physiological tradeoffs in the parameterization of a model of canopy transpiration
Adv Water Resour
(2003)The nature of rainfall variability over Africa on time scales of decades to millenia
Global Planet Change
(2000)