Abstract. The use of Generalized Linear Models (GLM) in vegetation analysis has been advocated to accommodate complex species response curves. This paper investigates the potential advantages of using classification and regression trees (CART), a recursive partitioning method that is free of distributional assumptions. We used multiple logistic regression (a form of GLM) and CART to predict the distribution of three major oak species in California. We compared two types of model: polynomial logistic regression models optimized to account for non‐linearity and factor interactions, and simple CART‐models. Each type of model was developed using learning data sets of 2085 and 410 sample cases, and assessed on test sets containing 2016 and 3691 cases respectively. The responses of the three species to environmental gradients were varied and often non‐homogeneous or context dependent. We tested the methods for predictive accuracy: CART‐models performed significantly better than our polynomial logistic regression models in four of the six cases considered, and as well in the two remaining cases. CART also showed a superior ability to detect factor interactions. Insight gained from CART‐models then helped develop improved parametric models. Although the probabilistic form of logistic regression results is more adapted to test theories about species responses to environmental gradients, we found that CART‐models are intuitive, easy to develop and interpret, and constitute a valuable tool for modeling species distributions.

Summary1. Few studies have tested the applicability of current non-equilibrium models of rangeland vegetation dynamics to a particular ecosystem, or across a range of systems that might be expected to respond di erently to grazing. This study assessed the extent to which the non-equilibrium persistent (NEP) model of rangeland vegetation dynamics applies to three distinct Mongolian rangeland ecosystems, the desert-steppe, steppe and mountain-steppe. 2. Standing biomass, vegetation cover and composition, and species richness and diversity were examined along grazing pressure gradients in ecological zones of differing productivity and interannual variability in precipitation. 3. In the desert-steppe, biomass, functional group cover, richness and diversity did not vary along grazing pressure gradients, but all vegetation variables except the cover of weedy annuals and unpalatable forbs varied signi®cantly between years. Vegetation dynamics in this zone largely conformed to the NEP model of rangeland dynamics. 4. In the mountain-steppe, grass and total biomass, total vegetative cover, the cover of grasses, weedy annuals and unpalatable forbs, and richness and diversity varied along grazing pressure gradients. With increasing grazing pressure, grasses decreased and forbs and weedy annuals increased, as the conventional range condition (RC) model predicts. Interannual variation in precipitation in¯uenced total vegetative cover, species and functional group cover, and richness and diversity. 5. In the steppe, forb biomass, grass, forb, unpalatable forb and weedy annual cover, and diversity varied along grazing pressure gradients. Grass biomass and total vegetative cover responded interactively to rainfall and grazing. Forb biomass, grass, forb and weedy annual cover and richness varied between years. Grasses decreased and forbs and weedy annuals increased with increasing grazing pressure, conforming to the RC model. 6. Ecosystem response to rainfall and grazing is complex, and interpretation of the response depends on the speci®c variables examined. The recent paradigm shift in rangeland science from the RC model to non-equilibrium models has been embraced with such enthusiasm by some that the concept of non-equilibrium rangelands may be as much in danger of being misapplied as equilibrium-based models have been.

Global climate models predict significant changes to the rainfall regimes of the grassland biome, where C cycling is particularly sensitive to the amount and timing of precipitation. We explored the effects of both natural interannual rainfall variability and experimental rainfall additions on net C storage and loss in annual grasslands. Soil respiration and net primary productivity (NPP) were measured in treatment and control plots over four growing seasons (water years, or WYs) that varied in wet-season length and the quantity of rainfall. In treatment plots, we increased total rainfall by 50% above ambient levels and simulated one early-and one late-season storm. The early-and late-season rain events significantly increased soil respiration for 2-4 weeks after wetting, while augmentation of wet-season rainfall had no significant effect. Interannual variability in precipitation had large and significant effects on C cycling. We observed a significant positive relationship between annual rainfall and aboveground NPP across the study (P 5 0.01, r 2 5 0.69). Changes in the seasonal timing of rainfall significantly affected soil respiration. Abundant rainfall late in the wet season in WY 2004, a year with average total rainfall, led to greater net ecosystem C losses due to a $ 50% increase in soil respiration relative to other years. Our results suggest that C cycling in annual grasslands will be less sensitive to changes in rainfall quantity and more affected by altered seasonal timing of rainfall, with a longer or later wet season resulting in significant C losses from annual grasslands. Abbreviations:WY 5 water year R h 5 heterotrophic respiration R s 5 soil respiration R r 5 root respiration NPP 5 net primary productivity NEP 5 net ecosystem production RC 5 root contribution

No abstract