Simulated dynamics of succession in a North American subtropical Prosopis savanna

Abstract

Abstract. A transition matrix model was used to explore the dynamics, rate and potential extent of changes in landscape vegetation patterns on a southern Texas Prosopis savanna. Transitions between seven vegetation classes were determined for the periods 1941–1960 and 1960–1983 on aerial photographs of three sites. During these periods, the sites were heavily grazed by cattle and were fire‐free. Vegetation states assessed in grids of 20 m x 20 m cells superimposed on photographs ranged from grass‐dominated to woody plant‐dominated. The 1941–1960 period (denoted DRY) was characterized by prolonged drought, whereas annual rainfall during the 1960–1983 period (denoted WET) was typically normal to above‐normal. The 1941 landscape consisted of herbaceous zones (6% of cells), woodland (50% of cells) and savanna parkland (44% of cells with grass/woody plant mixtures). The woodland state was the most stable, with probabilities of no change being 0.970 and 0.873 in WET and DRY periods, respectively. The herbaceous state was least stable, with corresponding values of 0.074 and 0.353.Past and future landscape structure was modelled by randomly selecting DRY or WET transitions at 20 year time steps. The model was run under a series of rainfall scenarios where the probability of selecting the WET transition matrix (P[WET]) ranged from 0 (DRY always chosen) to 1 (WET always chosen). Historical records indicate P[WET] has approximated 0.3 to 0.4 in the region. The rate of succession to states of greater woody cover increased as P[WET] increased. Forward simulations based on P[WET] > 0.2 suggest the present landscape is unstable and will develop into a closed‐canopy woodland within the next 180 years, assuming the processes operating between 1941 and 1983 continue (e.g. grazing by cattle and lack of fire). Reverse simulations concur with historical observations and projections derived from woody plant growth rates in other studies and suggest that 200 to 300 yr BP these landscapes contained a substantially greater proportion of cells dominated by grassland or grassland with scattered woody plants (43 to 74%) than was present in 1983 (19%). Based upon elapsed time between predicted past and future steady states, succession from open savanna to closed‐canopy woodland may occur in ca. 400 to 500 yr for P(WET) ≥ 0.33. Arresting or reversing the projected trend may require changes in climate and/or changes in livestock grazing and land management practices.The approaches employed in this study illustrate how time series maps, aerial photographs and satellite imagery can be analyzed and used to interpret, project and reconstruct local and regional changes in ecosystem structure. Difficulties and limitations associated with the use of Markov chains to model succession are identified and discussed.