Effects of consecutive dry and wet days on the forest–savanna boundary in north‐west South America

Abstract

AbstractAimMultiple abiotic and biotic factors, mainly the amount and seasonality of precipitation, fire regime, soil properties and species adaptation strategies, have been invoked to explain the existence of the tropical savanna–humid forest transition. We explored the rainfall variables that influence the probability of forest or savanna occurring in north‐west South America, specifically monthly precipitation and the mean and maximum duration of wet and dry spells, defined as the number of consecutive wet (dry) days.LocationNorth‐west South America.Time period1980–2017.Major taxa studiedSavanna and humid forest.MethodsWe used a set of logistic generalized linear models to assess the effect of multiple variables on the probability of occurrence of forest. Predictor variables included monthly precipitation, mean monthly dry and wet spells and maximum monthly dry and wet spells.ResultsThe probability of forest (savanna) occurrence depends mostly on the average duration of dry and wet spells and the interaction between these two variables, whereas the effect of total monthly precipitation is marginal. Dry spell duration had the largest effect on the probability of forest occurrence, which decreased by 7.5% for each additional consecutive dry day. For wet spells, the probability of forest occurrence increased by 4.1% for each additional consecutive wet day. Our model shows that the savanna–forest transition takes place under an average combination of dry spells lasting 8–10 days and wet spells lasting 1–3 days during the dry season (December–February). This combination changes to dry spells lasting 1–3 days and wet spells lasting 4–6 days during the wet season (May–July).Main conclusionsThis study reveals the importance of using precipitation data with a fine temporal scale to investigate forest–savanna distribution and indicates the limited explanatory power of aggregated precipitation variables (e.g., monthly and annual values). Models that assess the effect of climate change should therefore incorporate precipitation variables acquired at finer temporal resolution.