Although it is well known that mean annual rainfall (MAR) and rainfall seasonality have a key role in influencing the distribution of tree and grass cover in African tropical grassy biomes (TGBs), the impact of intra-seasonal rainfall variability on these distributions is less agreed upon. Since the prevalent mechanisms determining biome occurrence and distribution change with MAR, this research investigates the role of intra-seasonal rainfall variability for three different MAR ranges, assessing satellite data on grass and tree cover, rainfall and fire intervals at a sub-continental scale in sub-Saharan Africa. For MAR below 630 mm y−1, rainfall frequency had a positive relationship with grass cover; this relationship however became mostly negative at intermediate MAR (630–1200 mm y−1), where tree cover correspondingly mostly increased with rainfall frequency. In humid TGBs, tree cover decreased with rainfall intensity. Overall, intra-seasonal rainfall variability plays a role in determining vegetation cover, especially in mesic TGBs, where the relative dominance of trees and grasses has previously been largely unexplained. Importantly, the direction of the effect of intra-seasonal variability changes with MAR. Given the predicted increases in rainfall intensity in Africa as a consequence of climate change, the effects on TGBs are thus likely to vary depending on the MAR levels.
Grasslands and savannas, known together as tropical grassy biomes (TGBs), are estimated to cover ca. a third of Africa and a fifth of the world’s land area1,2. Characterised by continuous C4 grassland, and possible coexistence with trees2, they support a growing proportion of the world’s population, are home to the majority of the world’s remaining megafauna2 and are a critical store of biodiversity3. Despite their importance, TGBs have garnered little public attention or conservation effort in comparison to tropical forests1,4, and the role of grass in characterizing these biomes has only recently been studied at sub-continental level5.
The relative dominance of trees and grasses in TGBs and the extent of those systems are governed by numerous feedbacks between bottom-up factors (sensu ref.4), such as water availability and soil type, and by top-down factors, such as fire and herbivory (ref.6 and references therein). Mean annual rainfall (MAR) has been identified as the most important factor, both impacting the availability of water and seemingly influencing the relative importance of other factors5,7,8,9. Whereas tree and grass cover is limited primarily by water availability in dry TGBs, in mesic areas the high fire frequency, fostered by high grass cover, helps to limit tree cover5,7. At these higher MAR levels, both tropical forests and savannas exist10,11,12, with savannas mostly observed in areas with highly seasonal rainfall regimes, but partly also under overlapping climatic conditions with forests5,11,13. In these areas TGBs are probably maintained by a grass-fire feedback, whereby grasses provide fuel for fire, which in turn restricts the recruitment and growth of shade-tolerant and fire-intolerant forest trees, thus supporting the growth of grasses and fire-resistant savanna trees14. Generally, for mesic savannas, the ability to predict the relative dominance of tree and grass cover remains limited5.
In TGBs, competition for water by trees and grasses leads to water resource partitioning by root depth15, with root-depth separation between trees and grasses observed in dry areas16,17. Variations in rooting depth by vegetation type have been demonstrated in TGBs across differing rainfall gradients, with tree roots becoming shallower with MAR and overlapping with grass roots at higher rainfall levels18.
Soil water content depends not only on yearly rainfall levels or on rainfall seasonality, but also on the temporal distribution of rainfall within a season, described here as rainfall event frequency and intensity19. The intra-seasonal temporal distribution of rainfall can influence vegetation productivity dramatically20,21, especially in dry areas22,23,24, and it can affect the outcome of competition between plants with different water strategies25,26,27. Mirroring changes observed in global precipitation variability28, changes in rainfall patterns have been observed across Africa over several decades in the last century, with, for example, increases in rainfall intensity in Southern and Western Africa29,30, reductions in rainfall frequency specifically in Botswana31, decreases in intensity in Central Africa32, and increases in drought length across Eastern Africa33. Changes in rainfall variability have been closely associated with anthropogenic climate change34, and, with rainfall intensity in Africa expected to increase in the future35, changes in vegetation patterns may result.
Although a substantial body of research now exists in relation to TGBs and the role of various factors in explaining their occurrence and extent, the relative importance of intra-seasonal rainfall variability is less clearly understood. Recent research on the impact of intra-annual variability of rainfall on tree cover for different MAR ranges in Africa9 and across tropical areas36 has shown that intra-annual variability has an important role in determining tree cover variation, particularly within the intermediate (500–1500 mm y−1) MAR levels36. However, the influence of intra-seasonal variability on tree cover depends upon MAR throughout tropical Africa36,37, specifically for TGBs9. In drier TGBs, tree cover can increase with rainfall intensity9, in line with experimental work38 showing that increased precipitation intensity increases deep soil water, enhancing woody plant growth; this is supported by the deeper roots observed for trees than for grasses in drier savannas17,18,39. These findings are corroborated by modelling work25 that includes competition of grasses with tree seedlings, suggesting that more rare and intense rainfall events support the growth of woody plants in arid savannas, due to the out-competition of grasses on tree seedling, an effect which is especially strong at low moisture availability40,41. In contrast, areas with lower tree cover in African TGBs have been associated with greater rainfall intensity (although the precise impact depends also on soil type)9 which is analogous to that reported by ref.42 with data across all African biomes. This finding has been modelled and explained by the different relative-growth rates of grasses and trees43 (sensu ref.44), with grasses having the ability to grow faster than trees in the presence of high soil moisture, but being more negatively impacted under high water stress.
Within this context, the main research question addressed in this paper is whether changes in intra-seasonal rainfall variability affect the propensity of grasses, in addition to trees, to grow in TGBs at different MAR ranges. Given the importance of grasses in determining TGB extent and existence1 and their important role in the competition for water with trees2, we explicitly included grass cover as a dependent variable, unlike previous, similar studies that analysed large scale satellite data and considered only the tree cover variable9,36,42. In so doing we consider the interaction between vegetation types in TGBs at a sub-continental scale. Following a methodological approach closely aligned to that of ref.5, which analysed grass cover for the first time at sub-continental scale, we analysed the impact of intra-seasonal temporal variability of rainfall on grass and tree cover at sub-continental scale throughout TGBs in sub-Saharan Africa, in comparison to the impact of fire, MAR and rainfall seasonality (analysed in ref.5). As in that earlier study, we assessed separately three different MAR ranges, since different explanatory variables are relevant for low (less than 630 mm y−1), intermediate (between 630 mm y−1 and 1200 mm y−1), and high MAR (above 1200 mm y−1). The analysis by MAR range enhances the ability to identify the factors, in addition to rainfall totals, that can explain differences in tree and grass cover, potentially identifying a role for intra-seasonal variability that may have otherwise been overlooked.
Material and Methods
We analysed gridded data at 0.5° resolution of observed percentages of tree (T) and grass (G) cover in areas of tropical grassy biomes in sub-Saharan Africa (between 35°S and 15°N) as a function of four different rainfall variables and average fire interval data (AFI, [y]). The rainfall variables were mean annual rainfall (MAR, [mm y−1]), average rainfall seasonality index (SI) and two variables representing intra-seasonal rainfall variability: average wet-season daily rainfall frequency (λw, [d−1]) and intensity (αw, [mm d−1]). All variables were averaged over the period 2000–2010. Vegetation cover, fire intervals, mean annual rainfall and rainfall seasonality were calculated following ref.5, and as detailed below.
We used the daily rainfall measurements of the tropical rainfall measuring mission (TRMM 3B42), with 0.25° original resolution, to derive MAR, SI45, αw and λw 24. As in ref.5 we considered grid cells with MAR up to 2500 mm y−1. SI is defined by Walsh & Lawler as a ratio with the absolute value of the sum of the differences between monthly rainfall and average monthly rainfall at the numerator and annual rainfall at the denominator. It can range from 0 (if the annual rainfall is equally distributed over the year) to 1.83 (if all the annual rainfall falls in one month). In seasonal environments, such as those associated with savannas, precipitation affects plants mainly during the growing period46,47. For this reason, the estimation of daily rainfall intensity and frequency over the wet seasons is more appropriate than over the year. We thus assumed that the wet seasons correspond to the growing seasons of both plant types (although this may not be strictly true for all savannas, see refs2,48 and references therein). For each grid cell, αw was computed as the mean, over the years, of the daily rainfall in the wet days (assumed as days with precipitation greater than 0.1 mm d−1) within the wet periods (identified as explained in the following paragraph). The small threshold was introduced to take into account observation measurement errors. For each grid cell, λw was calculated as the ratio of the average number of wet days to the average length of the wet periods (Lw [d], calculated as explained in the following paragraph). αw and λw were defined such that the following relation was verified: MARw = αw·λw·Lw + c, where MARw is the wet-season mean annual rainfall and c is the sum of the rainfall occurring in the wet-season days in which rainfall is smaller than 0.1 mm d−1; across the data analysed, c was 0.30 mm y−1 on average. Consequently, MARw, Lw, αw and λw are not independent of each other.
To calculate the wet-season length Lw, we identified the wet periods as the months in which mean precipitation was greater than a percent threshold pthr of the climatological annual average monthly rainfall. Then, by summing up the lengths in days of these wet months and dividing by the number of years, we obtained the average wet season length Lw. To decide upon the threshold pthr, we maximised the correlation of Lw with the length of the wet season calculated with an alternative method, which was related to the opposite of SI (Lw,SI), similarly to that used by ref.42, i.e.: Lw,SI = 365/12 [11·(1 − SI/1.83) + 1], so that Lw,SI is equal to ca. 30 d for SImax = 1.83 and to 365 d for SImin = 0 45. We chose pthr = 50%, as for this value the two measures of wet season length were highly correlated (R2 = 0.96).
We obtained the annual average fire interval (AFI) from the monthly MCD45A1 burnt area satellite product (Collection 5.1), with original 500 m resolution, available from April 200049,50,51. AFI is the expected return time of fire at any point in the 0.5° grid cell52, calculated as the inverse of the average annual burnt area fraction (BA, [y−1]) in each 0.5° grid cell (AFI = 1/BA). BA was obtained following the method used by ref.53. In the analysis we used log10(AFI) (equal to −log10(BA)), because AFI covered several orders of magnitude. In order to avoid infinity values of AFI (when BA = 0), we set the maximum of AFI to 10,000 years adding to BA a small constant (0.0001 y−1) (see also ref.5).
Percentages of tree cover (T) and grass cover (G) were obtained from the annual tree and non-tree vegetation cover products, respectively, of the Terra MODIS Vegetation Continuous Fields product (MOD44B, V051), with original 250 m resolution54. Since MODIS tends to underestimate tree cover in the presence of shrub55, we used the 300-m ESA global land cover map (ESA CCI-LC, v 1.6.1; 5-year-averaged dataset centred in 2010) to remove grid cells with more than 50% of the area covered by shrublands (ESA CCI-LC codes 120,122). We used the same land cover map to mask out grid cells with more than 33% of the area influenced by humans, such as cropland/urban areas, and/or covered by water (codes ≤40, 190, 210) and/or covered by permanent snow or ice (code 220). This conservative choice, motivated by the inclusion of areas with only minimal human impact, selected a representative range of areas in sub-Saharan Africa. See also ref.5 for more details.
In order to study the effect of the abiotic variables specifically on tropical grassy biome vegetation, we identified grid cells of TGBs as those with more than 50% of the area flagged on the ESA CCI-LC map as deciduous trees and grassland (codes 60–62,130), following ref.5. This choice was based on the fact that African savannas trees are mostly deciduous (refs56,57, and see literature review in ref.5). The use of this method for distinguishing TGBs avoided problems in identifying biomes based on tree cover discontinuities in MODIS data58. The final datasets consisted of 1692 grid cells.
We analysed the relationships between vegetation cover variables (T and G) and five abiotic variables (MAR, SI, λw, αw and log10(AFI)) using generalized linear models (GLM)59, in three different MAR ranges: 0–630 mm y−1 (R1), 630–1200 mm y−1 (R2) and 1200–2500 mm y−1 (R3). These have been determined in the study of ref.5 and highlight how MAR influences the role of the prevalent mechanisms related to water availability and fire in determining biome occurrence and tree to grass dominance in sub-Saharan Africa. See a map of their spatial distribution in Supplementary Fig. S1. Figure 1 shows the variable distributions in these three ranges. For completeness, we also repeated the GLM analysis for all the TGB data, without filtering by MAR ranges. Notice that we did not include MARw and Lw in the statistical analysis because MARw was highly correlated with MAR (R2 > 0.98 for each of the three ranges), and Lw was anti-correlated with SI (R2 > 0.94 for each of the three ranges). In R1, R2 and R3 there were 694, 783 and 215 grid cells, respectively. We tested that the medians of the variable distributions were statistically different at the 5% significance level using the two-sided Wilcoxon rank sum test.
Tree and grass cover fractions were fitted assuming a binomial error distribution with a logit link function, because they are bounded between 0 and 160,61. We investigated the relationships considering models with different terms of a single predictor up to the third order, and models with a combination of the predictors but including only linear terms. Predictor variables were standardized subtracting their mean and dividing their standard deviation, such that their coefficient magnitude was a measure of their importance in the model. Since the rainfall variables used were not independent of each other (see previous section), we decided to exclude models containing both αw and λw. We included also the intercept-only model (i.e., a constant model), obtaining a set of 34 possible models. Models were selected based on the Akaike’s information criterion (AIC)62, such that the best models were those with the lowest AIC value, and we reported the ΔAIC (i.e. the difference between the AIC value of each model with the best model). We analysed all models with AIC smaller than the model including the intercept-only. The goodness-of-fit was evaluated as the fraction of deviance explained (pseudo-R2, R2 in the following for brevity), equivalent to the explained variance in a linear least-squares regression model63. It was computed as R2 = 1 − Dm/D0, where Dm is the residual deviance, i.e. the deviance that remains unexplained by the fit, and D0 is the deviance of the intercept-only model61,63.
To take into account collinearity between explanatory variables (see Supplementary Table S1 for the Pearson’s r value of the variables in each of the three ranges), we performed a residual analysis for the cases where the best model included αw or λw. This analysis was aimed at understanding whether the influence of these predictors on the vegetation cover was independent of the other explanatory variables. To this end, we computed the GLM of each intra-seasonal rainfall variable (αw or λw) as a function of MAR, SI and log10(AFI) (see also Supplementary Note S1 for details), and computed the deviance residuals. We calculated also the deviance residuals of the vegetation cover GLM computed with respect to the same three explanatory variables (as obtained from the above-described analyses). We then calculated the R2 of the linear fit of these two residual sets, as a high R2 implied that the dependence of the vegetation cover variable on αw or λw was direct and not only mediated by the other variables.
Data analysis was performed using MATLAB R2015b. In particular, we used the Matlab function ‘fitglm’, for the GLM analysis, and ‘ranksum’ for the Wilcoxon rank sum test.
In each MAR range, one of the intra-season rainfall variables (αw or λw) always appeared in the best model for either T or G (or both). Yet, when performing the analysis without distinguishing between MAR ranges (i.e. analysing all the dataset at once), the intra-seasonal rainfall variables did not appear in any of the best models: according to these models T depended on MAR and G on both SI and log10(AFI) (see Supplementary Tables S7 and S8). In the following, we report the results of the statistical analysis computed in each MAR range separately.
MAR ≤630 mm y−1
For low precipitation areas, T was mainly determined by MAR, while λw and SI were both included in the best model for G (Table 1).
Specifically, T had a positive trend with MAR (R2 = 0.22), which is the most important factor among those we considered in explaining tree cover variations in this range. The inclusion of other rainfall variables in models with MAR only increased the explained deviance by a small amount (ca. 0.04–0.05) (Supplementary Table S2). Tree cover decreased with λw and SI, and increased with αw, however, all of the selected models had an AIC close to the intercept-only model, indicating that they were weak.
According to the best model, G, which ranges from ca. 30% to ca. 80% (Fig. 1a), decreased with SI, i.e. it was larger where precipitation was less seasonal, and increased with λw, that is, it was larger where rainfall events were more frequent (Fig. 2). The magnitude of the predictor’s standardized coefficients was similar, and this model could explain 56% of the deviance (see Table 1 and Supplementary Table S3). Although λw was present in the best model, its effective role on G variations should be analysed in the light of its collinearity with other predictors, and in particular with MAR (r = 0.73, Supplementary Table S1). This was evident from the analyses of the following models and of the residuals. The best models with SI and MAR alone (cubic logit fit, ΔAIC = 4.10, and linear logit fit, ΔAIC = 10.86, respectively, Supplementary Table S3) could explain a much larger fraction of the deviance (53% and 40%, respectively) than the best model with λw alone (parabolic logit fit, ΔAIC = 26.28, R2 = 0.23; in the following we will use ‘parabolic fit’ to indicate ‘parabolic logit fit’). The other explanatory variables (log10(AFI) and αw) had smaller importance (Supplementary Table S3).
From the residual analysis (see Supplementary Note S1) we found that the linear relationship between the residuals of λw and G was associated with a small correlation coefficient (R2 = 0.07). This indicated that the precipitation frequency variable did not have an important role in determining G without the influence of the other predictors. In fact, λw, as expected, was highly determined by MAR, SI and log10(AFI) (in order of importance), which together explained 69% of the deviance for λw (Supplementary Table S9). Therefore, the increase of G with λw could be interpreted as due to the simultaneous changes of the other factors: in particular, λw, and consequently G, was higher in areas with higher MAR (which had the largest coefficient in the model).
630 mm y−1 < MAR < 1200 mm y−1
For intermediate precipitation, the best model for T included only λw, through a parabolic fit that explained 38% of the deviance (Fig. 3a, Table 1), while G was mostly explained by log10(AFI) and λw together (R2 = 0.37, Table 1, Fig. 3b), showing an opposite dependence on λw with respect to the low rainfall range. The decreasing trend of grass cover with log10(AFI) in this range confirmed the findings of ref.5, whose results are comparable in this range as the vast majority of the areas with such MAR in sub-Saharan Africa are covered by TGBs.
According to the best model shown in Fig. 3a, T, which in this range had the larger variation with respect to the other two MAR ranges (Fig. 1a), increased with daily rainfall frequency from 0.45 d−1 (obtained from the minimum of the parabolic fit) to ca. 0.85 d−1, but varied less for λw below 0.45 d−1 (Fig. 3a). λw was present in the first 10 models (ΔAIC < 3), in which it always had greater importance than the other variables (Supplementary Table S4). Looking at the contribution of the other factors, T generally increased with MAR, log10(AFI), SI and decreased with αw (Supplementary Table S4). However, the dependence of T on these variables alone was very weak (i.e. they had an AIC close to the intercept-only model), and generally explained low deviance (R2 ≤ 0.17; Supplementary Table S4), in line with findings from ref.5 (though for MAR, log10(AFI) and SI only).
According to the best model, G decreased with rainfall frequency, and (similarly to ref.5) decreased with the average fire intervals (log10(AFI)), i.e. it was favoured by increasing fire frequency (Fig. 3b, Table 1). The model with λw alone (parabolic fit, third selected model with ΔAIC = 1.72, R2 = 0.27, Supplementary Table S5) indicated that G was disadvantaged by increasing precipitation frequency only for λw greater than 0.55 d−1 (obtained from the maximum of the parabolic fit, Fig. 3b), whereas the opposite occurred for lower λw (Fig. 3b). Interestingly, the maximum of the parabolic fit between G and λw was relatively close to the minimum of the parabolic fit of T with λw. However, the dependence of G on λw alone was quite weak (AIC close to the intercept-only model), and this was the case also for the other selected models. SI and MAR were present in models with AIC smaller than the intercept-only model but gave negligible contributions to the explained deviance (Supplementary Table S5).
Through the residual analysis (Supplementary Note S1), we verified that the tendencies of both T and G to increase or decrease in the two parts of their parabolic fits with λw were independent from MAR, SI and log10(AFI), as expected also from the smaller correlation coefficients between predictors in R2 (Supplementary Table S1). Thus, the positive dependence of T on λw (for λw > 0.45 d−1) and the negative one of G (for λw > 0.55 d−1) were valid for every value of the other explanatory variables and the correlations with the residuals of λw of the residuals of T was R2 = 0.36 and of the residuals of G was R2 = 0.31 (Supplementary Fig. S2).
MAR≥1200 mm y−1
αw was the factor determining most of the variation of T, with decreases in the variable associated with increasing rainfall intensity (Fig. 4, Table 1). This dependence could explain 32% of the deviance for T. There was also a tendency of T to decrease with log10(AFI), but this dependence was very weak (Supplementary Table S6). Models with MAR, SI and λw had AIC greater than the intercept-only model, thus they didn’t play any role.
After removing the influence of MAR, SI and log10(AFI) from both T and αw, we found that the decreasing of T with αw was preserved, although the importance of this dependence decreased (R2 = 0.20, see Supplementary Fig. S2). In fact, the variation of αw was highly determined by SI, MAR and log10(AFI) (in order of importance, R2 = 0.66, Supplementary Table S9), as expected also from the correlation coefficients between predictors (Supplementary Table S1). Therefore, T decreased with αw partially as a consequence of the simultaneously increase of the latter with MAR, SI and log10(AFI). In other words, T was smaller in areas with higher αw partly because they corresponded to areas of higher MAR, higher rainfall seasonality and rare fires.
Our results appear to indicate that intra-seasonal rainfall variability has a role to play in determining the relative predominance of grasses and trees in TGBs in sub-Saharan Africa, with that role and importance dependent on mean annual rainfall (MAR) levels. At the low and intermediate rainfall ranges, the frequency of rainfall during the rainy season was included in the best models explaining grass cover, yet with opposite effects: grasses show a positive (although noisy) trend with rainfall frequency at the dry end of the gradient, while they show a negative trend with it in mesic TGBs, which is especially evident for the highest rainfall frequencies. At this intermediate rainfall level, tree cover also increased with average rainfall frequency in the wet season for the most part, and, somewhat analogously, in the humid TGBs tree cover decreased with rainfall intensity. Importantly, average rainfall frequency in the wet season was shown to explain a reasonable share of the deviance at the intermediate rainfall level for grasses and trees for the first time. The relative importance of intra-seasonal rainfall variability is somewhat lost when investigating the vegetation responses for all TGBs across the sub-continent, with MAR the dominant factor for trees, and seasonality and average fire interval most important for grasses. This highlights the importance of disaggregating the results by MAR range, as the main mechanisms governing TGB dynamics vary with absolute rainfall values5,7,9. Differing responses of grass and tree cover to intra-seasonal rainfall variability at different rainfall levels seem to explain the contrasting findings in relation to intra-seasonal variability highlighted by previous research38,42.
For grasses, the positive relationship between rainfall frequency and grass cover at the lower MAR level is similar to earlier modelling results for dry savannas25. Given that increases in rainfall frequency can enhance water availability in surface soils19,64, our finding is consistent with observations regarding the ability of grasses to extract shallow soil water in comparison to trees65,66,67. At lower rainfall levels, the difference between grass root depth and tree root depth is large18,68, suggesting a more shallow soil moisture profile would favour grasses, as they also have a more aggressive uptake strategy with respect to trees in dry savannas16,69. This result is in line with earlier experimental work38, based on a study site with MAR of 544 mm y−1, where it was found that increases in intensity and, by virtue of the experimental design, reduction in rainfall frequency, push soil moisture deeper into the soil, benefitting trees with deeper roots at the expense of shallow rooted grasses. In line with this, we also observed, as in other studies9,36, that in dry TGBs trees, which are deeper rooted than grasses at this MAR range, increased in cover with more intense, less frequent rainfall. Trees have also access to deeper water during the dryer periods of the year66, and increased rainfall intensity with reduced rainfall frequency would assist in enhancing the competitive advantage of trees by increasing deep water recharge. However, it must be noted that intra-seasonal rainfall variables explained a very small part of the deviance for tree cover, which mainly depended on MAR, possibly indicating that tree growth is generally water-limited in these areas. Furthermore, our residual analysis showed that part of the deviance for grass cover explained by rainfall frequency at the lower MAR range is in fact due to its collinearity with the other independent variables, especially with rainfall seasonality and annual amounts. Rainfall seasonality plays a large role at the low MAR range, explaining a large share of the variation for grass cover on its own (R2 = 0.53). This implies that of the two variables included in the best model for grasses, seasonality is by far more important than rainfall frequency. It is also important to note that rainfall frequency is also closely correlated with MAR (Pearson’s r = 0.73), and a model including MAR alone explains 40% of the deviance for the grass cover; this highlights the importance, in this low range, of the absolute levels of the water resource for both grasses and trees, as already observed for dry sub-Saharan biomes in general5. This may also help to explain why it is rainfall frequency, as opposed to rainfall intensity, that is included in the best model for grass cover, as rainfall intensity shows a much lower correlation with MAR in this range (Pearson’s r = 0.14).
In contrast to the findings for grass cover in the low MAR range, in the intermediate rainfall range, increases in rainfall frequency in the rainy season were associated with a reduction in grass cover for average rainfall frequencies above 0.55 d−1. Contrasting responses along a rainfall gradients have been observed previously for grasses in North American prairies70, and recently for trees in African TGBs9 and across the global tropics36. This tendency of grassland to prefer less frequent rainfall at this range is similar to that observed in few experimental field sites in Africa71 as well as in other parts of the globe (e.g. ref.72). In this intermediate rainfall range, we also observe a positive relationship between tree cover and rainfall frequency (above 0.45 d−1), confirming previous results9,36,42. While TGB grass cover has been largely unexplained so far in this intermediate range (Supplementary Table S5), more than a quarter of the deviance for grass cover is explained by rainfall frequency alone (R2 = 0.27). As previously mentioned, in these mesic savannas there is greater overlap between tree and grass root depth compared to xeric savannas16,18, hypothetically implying competition between grasses and mature trees for shallow soil water. The broadly positive relationship between tree cover and rainfall frequency may indicate that trees can exploit surface soil water better than grasses at this MAR range, leading to tree dominance and out-competition of shade-intolerant grasses. This is an interesting finding in our opinion, suggesting that, differently from dry savannas, tree access to deep water is not as relevant in mesic savannas, as also suggested by field observations of ref.18, although it is important to temper this finding with reference to the relatively lower explanatory power of rainfall frequency on vegetation cover at this range.
Whereas rainfall frequency in the wet season is included as a predictor in the best models in the low and intermediate ranges, in the higher MAR range increasing rainfall intensity appears to be negatively correlated with tree cover, aligned with earlier findings in relation to rainfall intensity and tree cover9,42. Indeed, rainfall intensity is the only variable included in the best model for TGB tree cover in this range. As there is a negative correlation between rainfall frequency and intensity in this MAR range (Pearson’s r = −0.41), this result is also in line with what has been observed at intermediate MAR (for the most part), and with recent findings by ref.36, i.e. increases in daily rainfall frequency are associated with increases in tree cover. However, although rainfall intensity remains the best predictor for tree cover, its decrease with rainfall intensity is also partially due to the simultaneous increase of rainfall intensity with both MAR, SI and log10(AFI), as shown by the residual analysis. Hence the signal identified linking rainfall intensity with tree cover is partly noisy.
Overall, although our results support a role for intra-seasonal rainfall variability for trees and grasses in different MAR ranges, our understanding of the mechanisms that lead to vegetation responses is still necessarily limited and speculative. Soil hydrology is complex and an approach taking into account its dynamics explicitly would be needed to proceed further. To understand the benefits to plants of the temporal distribution of rainfall, soil water availability is key, and this depends on a series of factors that also feedback to the vegetation itself, forming a complex system. Although more rare events can be beneficial for vegetation at low precipitation and less beneficial at high precipitation22, the picture here is complicated by water competition between trees and grasses, which should be properly disentangled by including the changes in dynamics along a rainfall gradient, for example including the MAR-dependent relative rooting depth of tree and grasses18. Tree access to deep water during the dry season may generate long-term memory effects73, and contribute to different phenology of trees and grasses74, which can lead to a temporal niche-partitioning48, a known mechanism of species coexistence (the so-called storage-effect)44. Soil texture influences water availability, and it thus mediates the response of tree cover to intra-seasonal rainfall variability, with a role that can be almost as important as total yearly rainfall9,19. Finally, we would like to acknowledge that the scale of the hydrological dynamics of tree-grass water competition is much finer than the scale of the data used in the present study (0.5°). Given this resolution, the intra-seasonal rainfall variables here considered displayed very small variations in their mean values. However, these variations are seemingly relevant for vegetation, with local relationship possibly scaling up and emerging at the coarser scale of this analysis.
TGBs are likely to show contrasting responses to anthropogenic climate change, which is expected to increase rainfall intensity across Africa35, depending on the absolute values of mean annual rainfall. If this increase will be accompanied by a decrease of rainfall frequency, our results suggest that grasses will decrease in dry TGBs possibly favouring further woody encroachment, while woody cover will be limited in mesic and humid savannas. Our findings suggest that intra-seasonal rainfall variability is especially important for grasses and trees in mesic TGBs. For these areas in particular, our research serves to enhance the so-far-limited understanding of the factors determining the relative balance of these two vegetation life-forms.
The observational datasets used in this study are all freely available. The TRMM 3B42 dataset is available at https://mirador.gsfc.nasa.gov/. The ESA CCI-LC, v 1.6.1 dataset is available at http://maps.elie.ucl.ac.be/CCI/viewer/download.php. The MODIS datasets (MOD44B and MOD45A1) are available at https://earthdata.nasa.gov. Post-processed data are available upon request to the authors.
Parr, C. L., Lehmann, C. E. R., Bond, W. J., Hoffmann, W. A. & Andersen, A. N. Tropical grassy biomes: Misunderstood, neglected, and under threat. Trends Ecol. Evol. 29, 205–213 (2014).
Scholes, R. J. & Archer, S. R. Tree-Grass Interactions in Savannas. Annu. Rev. Ecol. Syst. 28, 517–544 (1997).
Murphy, B. P., Andersen, A. N. & Parr, C. L. The underestimated biodiversity of tropical grassy biomes. Philos. Trans. R. Soc. B Biol. Sci. 371, 20150319 (2016).
Bond, W. J. Ancient grasslands at risk. Science 351, 120–122 (2016).
D’Onofrio, D., von Hardenberg, J. & Baudena, M. Not only trees: Grasses determine African tropical biome distributions via water limitation and fire. Glob. Ecol. Biogeogr. 27, 714–725 (2018).
Oliveras, I. & Malhi, Y. Many shades of green: the dynamic tropical forest–savannah transition zones. Philos. Trans. R. Soc. B Biol. Sci. 371, 20150308 (2016).
Lehmann, C. E. R., Archibald, S. A., Hoffmann, W. A. & Bond, W. J. Deciphering the distribution of the savanna biome. New Phytol. 191, 197–209 (2011).
Sankaran, M. et al. Determinants of woody cover in African savannas. Nature 438, 846–849 (2005).
Case, M. F. & Staver, A. C. Soil texture mediates tree responses to rainfall intensity in African savannas. New Phytol. 219, 1363–1372 (2018).
Dantas, V., de, L., Hirota, M., Oliveira, R. S. & Pausas, J. G. Disturbance maintains alternative biome states. Ecol. Lett. 19, 12–19 (2016).
Staver, A. C., Archibald, S. & Levin, S. A. The global extent and determinants of savanna and forest as alternative biome states. Science 334, 230–232 (2011).
Hirota, M., Holmgren, M., Van Nes, E. H. & Scheffer, M. Global Resilience of Tropical Forest and Savanna to Critical Transitions. Science 334, 232–235 (2011).
Yin, Z., Dekker, S. C., van den Hurk, B. J. J. M. & Dijkstra, H. A. The climatic imprint of bimodal distributions in vegetation cover for western Africa. Biogeosciences 13, 3343–3357 (2016).
Staver, C. A. & Levin, S. Integrating Theoretical Climate and Fire Effects on Savanna and Forest Systems. Am. Nat. 180, 211–224 (2012).
Walker, B. H. & Noy-Meir, I. In Tropical savannas (eds Huntley, B. J. & Walker, B. H.) 556–590 (Springer-Verlag, 1982).
Ward, D., Wiegand, K. & Getzin, S. Walter’s two-layer hypothesis revisited: Back to the roots! Oecologia 172, 617–630 (2013).
Holdo, R. M. & Nippert, J. B. Transpiration dynamics support resource partitioning in African savanna trees and grasses. Ecology 96, 1466–1472 (2015).
Holdo, R. M., Nippert, J. B. & Mack, M. C. Rooting depth varies differentially in trees and grasses as a function of mean annual rainfall in an African savanna. Oecologia 186, 269–280 (2018).
Noy-Meir, I. Desert Ecosystems: Environment and Producers. Annu. Rev. Ecol. Syst. 4, 25–51 (1973).
Medvigy, D., Wofsy, S. C., Munger, J. W. & Moorcroft, P. R. Responses of terrestrial ecosystems and carbon budgets to current and future environmental variability. Proc. Natl. Acad. Sci. 107, 8275–8280 (2010).
Guan, K. et al. Simulated sensitivity of African terrestrial ecosystem photosynthesis to rainfall frequency, intensity, and rainy season length. Environ. Res. Lett. 13, 025013 (2018).
Baudena, M., Boni, G., Ferraris, L., Von Hardenberg, J. & Provenzale, A. Vegetation response to rainfall intermittency in drylands: Results from a simple ecohydrological box model. Adv. Water Resour. 30, 1320–1328 (2007).
Chesson, P. et al. Resource pulses, species interactions, and diversity maintenance in arid and semi-arid environments. Oecologia 141, 236–253 (2004).
Rodriguez-Iturbe, I. & Porporato, A. Ecohydrology of water-controlled ecosystems: soil moisture and plant dynamics. (Cambridge University Press, 2004).
D’Onofrio, D., Baudena, M., D’Andrea, F., Rietkerk, M. & Provenzale, A. Tree-grass competition for soil water in arid and semiarid savannas: The role of rainfall intermittency. Water Resour. Res. 51, 169–181 (2015).
Yuan, C. & Chesson, P. The relative importance of relative nonlinearity and the storage effect in the lottery model. Theor. Popul. Biol. 105, 39–52 (2015).
Verwijmeren, M. Interspecific facilitation and critical transitions in arid ecosystems. (Utrecht University, 2016).
Medvigy, D. & Beaulieu, C. Trends in daily solar radiation and precipitation coefficients of variation since 1984. J. Clim. 25, 1330–1339 (2012).
New, M. et al. Evidence of trends in daily climate extremes over southern and west Africa. J. Geophys. Res. Atmos. 111 (2006).
Mason, S. J., Waylen, P. R., Mimmack, G. M., Rajaratnam, B. & Harrison, J. M. Changes in extreme rainfall events in South Africa. Clim. Change 41, 249–257 (1999).
Batisani, N. & Yarnal, B. Rainfall variability and trends in semi-arid Botswana: Implications for climate change adaptation policy. Appl. Geogr. 30, 483–489 (2010).
Aguilar, E. et al. Changes in temperature and precipitation extremes in western central Africa, Guinea Conakry, and Zimbabwe, 1955–2006. J. Geophys. Res. Atmos. 114 (2009).
Nicholson, S. E. Climate and climatic variability of rainfall over eastern Africa. Rev. Geophys. 55, 590–635 (2017).
Min, S. K., Zhang, X., Zwiers, F. W. & Hegerl, G. C. Human contribution to more-intense precipitation extremes. Nature 470, 378–381 (2011).
Nikulin, G. et al. The effects of 1.5 and 2 degrees of global warming on Africa in the CORDEX ensemble. Environ. Res. Lett. 13, 065003 (2018).
Xu, X., Medvigy, D., Trugman, A. T., Guan, K. & Rodriguez-iturbe, S. P. G. I. Tree cover shows strong sensitivity to precipitation variability across the global tropics. Glob. Ecol. Biogeogr. 27, 450–460 (2018).
Guan, K. et al. Continental-scale impacts of intra-seasonal rainfall variability on simulated ecosystem responses in Africa. Biogeosciences 11, 6939–6954 (2014).
Kulmatiski, A. & Beard, K. H. Woody plant encroachment facilitated by increased precipitation intenstity. Nat. Clim. Chang. 3, 833–837 (2013).
February, E. C. & Higgins, S. I. The distribution of tree and grass roots in savannas in relation to soil nitrogen and water. South African J. Bot. 76, 517–523 (2010).
Baudena, M., D’Andrea, F. & Provenzale, A. An idealized model for tree–grass coexistence in savannas: the role of life stage structure and fire disturbances. J. Ecol. 98, 74–80 (2010).
February, E. C., Higgins, S. I., Bond, W. J. & Swemmer, L. Influence of competition and rainfall manipulation on the growth responses of savanna trees and grasses. Ecology 94, 1155–1164 (2013).
Good, S. P. & Caylor, K. K. Climatological determinants of woody cover in Africa. Proc. Natl. Acad. Sci. 108, 4902–4907 (2011).
Xu, X., Medvigy, D. & Rodriguez-iturbe, I. Relation between rainfall intensity and savanna tree abundance explained by water use strategies, Proc. Natl. Acad. Sci. 112, 12992–12996 (2015).
Chesson, P. Mechanisms of maintenance of species diversity. Annu. Rev. Ecol. Syst. 31, 343–366 (2000).
Walsh, R. P. D. & Lawler, D. M. Rainfall seasonality: description, spatial patterns and change through time. Weather 36, 201–208 (1981).
Rodriguez-Iturbe, I. Ecohydrology: A hydrologic perspective of climate-soil-vegetation dynamics. Water Resour. Res. 36, 3–9 (2000).
Knapp, A. K. et al. Consequences of More Extreme Precipitation Regimes for Terrestrial Ecosystems. Bioscience 58, 811–821 (2008).
Sankaran, M., Ratnam, J. & Hanan, N. P. Tree-grass coexistence in savannas revisited - Insights from an examination of assumptions and mechanisms invoked in existing models. Ecology Letters 7, 480–490 (2004).
Roy, D. P., Boschetti, L., Justice, C. O. & Ju, J. The Collection 5 MODIS Burned Area Product - Global Evaluation by Comparison with the MODIS Active Fire Product. Remote Sens. Environ., 112, 3690–3707 (2008).
Roy, D. P., Jin, Y., Lewis, P. E. & Justice, C. O. Prototyping a global algorithm for systematic fire-affected area mapping using MODIS time series data. Remote Sens. Environ. 97, 137–162 (2005).
Roy, D., Lewis, P. & Justice, C. Burned area mapping using multi-temporal moderate spatial resolution data-a bi-directional reflectance model-based expectation approach. Remote. Sens. Environ. 83, 263–286 (2002).
Johnson, E. A. & Wagner, C. E. Van. The theory and use of two fire history models. Can. J. For. Res. 15, 214–220 (1985).
Lehsten, V., Harmand, P., Palumbo, I. & Arneth, A. Modelling burned area in Africa. Biogeosciences 7, 3199–3214 (2010).
Townshend, J. R. et al. Vegetation Continuous Fields MOD44B, 2000–2010 Percent Tree Cover, Percent Non-Tree Vegetation, Collection 5, Version 051, University of Maryland, College Park, Maryland. (Digital data accessed 08/12/2016 from, https://earthdata.nasa.gov/) (2011).
Bucini, G. & Hanan, N. P. A continental-scale analysis of tree cover in African savannas. Glob. Ecol. Biogeogr. 16, 593–605 (2007).
Bowman, D. M. J. S. & Prior, L. D. Why do evergreen trees dominate the Australian seasonal tropics? Aust. J. Bot. 53, 379–399 (2005).
Shorrocks, B. The biology of African savannahs. 1–9 (Oxford: OUP Premium, 2007).
Gerard, F. et al. MODIS VCF should not be used to detect discontinuities in tree cover due to binning bias. A comment on Hanan et al. (2014) and Staver and Hansen (2015). Glob. Ecol. Biogeogr. 26, 854–859 (2017).
McCullagh, P. & Nelder, J. A. Generalized Linear Models, Second Edition. (Taylor & Francis, 1989).
Dobson Annette J. An introduction to generalized linear models/Annette J. Dobson. (Chapman & Hall/CRC Boca Raton, 2002).
Schwarz, M. & Zimmermann, N. E. A new GLM-based method for mapping tree cover continuous fields using regional MODIS reflectance data. Remote Sens. Environ. 95, 428–443 (2005).
Akaike, H. A new look at the statistical model identification. IEEE Trans. Automat. Contr. 19, 716–723 (1974).
Guisan, A. & Zimmermann, N. E. Predictive habitat distribution models in ecology. Ecol. Modell. 135, 147–186 (2000).
Fay, P. A., Carlisle, J. D., Knapp, A. K., Blair, J. M. & Collins, S. L. Productivity responses to altered rainfall patterns in a C 4-dominated grassland. Oecologia 137, 245–251 (2003).
Hipondoka, M. H., Aranibar, J., Chirara, C., Lihavha, M. & Macko, S. Vertical distribution of grass and tree roots in arid ecosystems of Southern Africa: niche differentiation or competition? J. Arid Environ. 54, 319–325 (2003).
Kulmatiski, A. & Beard, K. H. Root niche partitioning among grasses, saplings, and trees measured using a tracer technique. Oecologia 171, 25–37 (2013).
Cabal, C. & Rubenstein, D. I. Above- and below-ground allocation and functional trait response to soil water inputs and drying rates of two common savanna grasses. J. Arid Environ. 157, 1–12 (2018).
Mazzacavallo, M. G. & Kulmatiski, A. Modelling Water Uptake Provides a New Perspective on Grass and Tree Coexistence. 10, e0144300 (2015).
Walter, H. K. Vegetation of the earth in relation to climate and the eco-physiological conditions. (London: English universities press, 1973).
Heisler-White, J. L., Blair, J. M., Kelly, E. F., Harmoney, K. & Knapp, A. K. Contingent productivity responses to more extreme rainfall regimes across a grassland biome. Glob. Chang. Biol. 15, 2894–2904 (2009).
Swemmer, A. M., Knapp, A. K. & Snyman, H. A. Intra-seasonal precipitation patterns and above-ground productivity in three perennial grasslands. J. Ecol. 95, 780–788 (2007).
Knapp, A. K. Rainfall Variability, Carbon Cycling, and Plant Species Diversity in a Mesic Grassland. Science 298, 2202–2205 (2002).
Richard, Y. et al. Multi-month memory effects on early summer vegetative activity in semi-arid South Africa and their spatial heterogeneity. Int. J. Remote Sens. 33, 6763–6782 (2012).
Guan, K. et al. Terrestrial hydrological controls on land surface phenology of African savannas and woodlands. J. Geophys. Res. Biogeosciences 119, 1652–1669 (2014).
D. D’Onofrio and J. von Hardenberg acknowledge support from the European Union Horizon 2020 research and innovation programme under grant agreement No. 641816 (CRESCENDO) and from the Project of strategic Interest NextData of the Italian Ministry of Education, University and Research (MIUR) (http://www.nextdataproject.it).
The authors declare no competing interests.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
About this article
Cite this article
D’Onofrio, D., Sweeney, L., von Hardenberg, J. et al. Grass and tree cover responses to intra-seasonal rainfall variability vary along a rainfall gradient in African tropical grassy biomes. Sci Rep 9, 2334 (2019). https://doi.org/10.1038/s41598-019-38933-9
Contrasting responses of woody and herbaceous vegetation to altered rainfall characteristics in the Sahel
Linking Vegetation-Climate-Fire Relationships in Sub-Saharan Africa to Key Ecological Processes in Two Dynamic Global Vegetation Models
Frontiers in Environmental Science (2020)
Spatial transitions in tree cover are associated with soil hydrology, but not with grass biomass, fire frequency, or herbivore biomass in Serengeti savannahs
Journal of Ecology (2020)