Do Grasslands Have a Memory: Modeling Phytomass Production of a Semiarid South African Grassland

Abstract

We analyzed data sets on phytomass production, basal cover, and monthly precipitation of a semiarid grassland in South Africa for good, medium, and poor rangeland condition (a) to investigate whether phytomass production per unit of basal cover differed among rangeland conditions, (b) to quantify the time scales of a carryover effect from production in previous months, and (c) to construct predictive models for monthly phytomass. Finally, we applied the best models to a 73-year data set of monthly precipitation data to study the long-term variability of grassland production. Our results showed that mean phytomass production per unit of basal cover did not vary significantly among the rangeland conditions—that is, vegetated patches in degraded grassland have approximately the same production as vegetated patches in grassland in good condition. Consequently, the stark decline in production with increasing degradation is a first-order effect of reduced basal area. Current-year precipitation accounted for 64%, 62%, and 36% of the interannual variation in phytomass production for good, medium, and poor condition, respectively. We found that 61%, 68%, and 33%, respectively, of the unexplained variation is related to a memory index that combines mean monthly temperature and a memory of past precipitations. We found a carryover effect in production from the previous 4 years for grassland in good condition and from the previous 1 or 3S month for grassland in medium and poor condition. The memory effect amplified the response of production to changes in precipitation due to alternation of prolonged periods of dry or wet years/months at the time scale of the memory. The interannual variability in phytomass production per unit basal cover (coefficient of variation [CV] = 0.42–0.50 for our 73-year prediction, CV = 0.57–0.71 for the 19-year data) was greater than the corresponding temporal variability in seasonal rainfall (CV = 0.29).