Coupling ecosystem and landscape models to study the effects of plot number and location on prediction of forest landscape change

A fundamental but unsolved dilemma is that observation and prediction scales are often mismatched. Reconciling this mismatch largely depends on how to design samples on a heterogeneous landscape. In this study, we used a coupled modeling approach to investigate the effects of plot number and location on predicting tree species distribution at the landscape scale. We used an ecosystem process model (LINKAGES) to generate tree species response to the environment (a land type) at the plot scale. To explore realistic parameterization scenarios we used results from LINKAGES simulations on species establishment probabilities under the current and warming climate. This allowed us to design a series of plot number and location scenarios at the landscape scale. Species establishment probabilities for different land types were then used as input for the forest landscape model (LANDIS) that simulated tree species distribution at the landscape scale. To investigate the effects of plot number and location on forest landscape predictions, LANDIS considered effects of climate warming only for the land types in which experimental plots were placed; otherwise inputs for the current climate were used. We then statistically examined the relationships of response variables (species percent area) among these scenarios and the reference scenario in which plots were placed on all land types of the study area. Our results showed that for species highly or moderately sensitive to environmental heterogeneity, increasing plot numbers to cover as many land types as possible is the strategy to accurately predict species distribution at the landscape scale. In contrast, for species insensitive to environmental heterogeneity, plot location was more important than plot number. In this case, placing plots in land types with large area of species distribution is warranted. For some moderately sensitive species that experienced intense disturbance, results were different in different simulation periods. Results from this study may provide insights into sample design for forest landscape predictions.