• fire dynamics;
  • Kappa statistic;
  • leaf area index;
  • plant functional types;
  • remote-sensing;
  • tropical and arid ecosystems


Regional vegetation-climate modelling studies have typically focused on boreal or temperate ecosystems in North America and Europe, almost completely overlooking tropical ecosystems. We present the first results of simulated regional vegetation-climate dynamics in Middle America as simulated by the model, LPJ-GUESS. The Kappa statistic indicated poor agreement, with a Kappa value of 0.301. When we modified the Kappa statistic by aggregating cell sizes and using generalized biomes, the Kappa value increased to 0.543, indicating a fair agreement. Total LAI simulated from LPJ-GUESS was strongly correlated to remotely sensed LAI values (r = 0.75). Our simulations indicate that fire frequency was overestimated in tropical moist forests and underestimated in savannas. This underestimation of fire resulted in an over-simulation of dry tropical forest at the expense of savanna. We highlight additional reasons for the initially poor representation of vegetation in Middle America, including factors such as non-parameterized plant functional types (desert shrub, cacti, and other succulents), rugged topography, and an insufficient representation of soil.