Below-surface water mediates the response of African forests to reduced rainfall

The African continent receives highly variable rainfall spatially and temporally, with an annual mean and standard deviation of 660 ± 80 mm. Spatial variability in rainfall is highest in SHR, where average annual rainfall is 600 ± 450 mm. On a temporal scale, the highest annual rainfall in the study period was 690 mm and occurred in 2010, categorized as a strong El Niño year (figure 1(a)). In 2005 and 2015, annual rainfall was low by 69 mm and 64 mm than the 2003–2017 average respectively. Annual rainfall variability in Africa shows an inconsistent response to El Niño and La Niña climate oscillations indicated by the Oceanic Niño Index (ONI; Climate Center Prediction Team 2018). In Africa, El Niño events and associated SST anomalies are not the only factors that influence rainfall variability. The Indian Ocean dipole, African and tropical easterly jets, and the Saharan heat low also influence atmospheric circulation patterns and rainfall variability in Africa (Black 2005, Parhi et al 2016).

Zoom In Zoom Out Reset image size

In 2005, the SST was lower than average over the adjacent South Atlantic and Indian Oceans, resulting in reduced rainfall across Africa (figure 1(b)). In 2010, high SSTs over the South Atlantic Ocean coincided with enhanced annual rainfall and productivity over a majority of the vegetated area of Africa. In 2015, low SSTs over the South Atlantic Ocean were associated with reduced rainfall in the southern savannas of Africa.

The L4C data record indicated that African ecosystems contribute approximately 20% (26.5 ± 0.8 Pg C yr⁻¹) to the total annual global terrestrial GPP. The TRENDY models estimated an annual GPP of 27.4 Pg C yr⁻¹, close to that of L4C GPP but with lower inter-annual variability (±0.5 Pg C). The L4C GPP inter-annual variability is closely related to rainfall patterns (r² = 0.45, p < 0.05). The TRENDY GPP models showed more agreement with annual rainfall patterns (r² = 0.57, p < 0.05) than the L4C GPP estimates, suggesting that rainfall plays a leading role in determining inter-annual variations in TRENDY model GPP for African ecosystems. Unlike TRENDY models, the L4C GPP uses the satellite-derived FPAR observations and describes the response of GPP to climate by including water (both SM and VPD) and temperature constraints to limit the LUE photosynthetic carbon assimilation rate (see Methods).

The L4C GPP anomalies showed that the largest reduction in annual GPP for Africa occurred during 2005 (−1.06 Pg C), which was a relatively weak El Niño year (Climate Center Prediction Team 2018; figure 1(a)). In contrast, the 2010 El Niño event was stronger, and coincided with positive rainfall anomalies over much of Africa; 2010 was also the most productive year of the study period (figure 1(b)). Our analysis suggests that African ecosystems respond differently to individual El Niño events (Asefi-Najafabady and Saatchi 2013), with contrasting impacts on the continental scale carbon balance. During 2010, we found widespread increases in productivity except for portions of southern Africa, the Horn of Africa and eastern Sahel. In 2005 and 2015, reduction in rainfall in southern savanna ecosystems represented by shrubs and grass in the MODIS MOD12 PFT classification substantially reduced GPP (figure 1(b)). In 2010, rainfall was on average 55 mm higher than the 2003–2017 annual mean, while TWS was higher by 21.9 mm, FPAR was 0.9% higher and PAR was 1.2% lower than corresponding 2003–2017 annual mean values. In 2004 and 2005, respective reductions in rainfall of 53 and 48 mm coincided with an average 23.3 mm reduction in TWS. This change in water availability is associated with a ∼1.5% reduction in FPAR while PAR increased by 2.5%.

Generally, GPP in African ecosystems appears to follow rainfall anomalies and there are spatial patterns in rainfall anomalies that affect the plant response to water availability or scarcity (figures S1 and S2, available online at stacks.iop.org/ERL/15/034063/mmedia). Partitioning the anomalies in L4C GPP and rainfall by PFT revealed that in tropical ecosystems, the reduction in GPP from 2003 to 2005 coincided with lower-than-average rainfall prior to the year 2005 (figures 2(a), (c)). However, the significant rainfall reduction in tropical ecosystems in 2015 had little apparent effect on GPP, which remained near average levels in the years following 2010 (figure 2(a)). In 2005, L4C GPP for African tropical evergreen broadleaf forests (EBF) was 6.39 Pg C yr⁻¹, while in 2010, EBF GPP increased to 6.82 Pg C yr⁻¹. During the 2015 low rainfall conditions, EBF GPP was 6.51 Pg C yr⁻¹, and thus still 0.1 Pg C yr⁻¹ higher than the 2003–2017 mean value. African rainforests (as defined by the EBF classification) show two rainy season peaks, one from March to May and the other from September to November (figures S3 and S4). GPP tends to increase during the wet season, consistent with previous reports that productivity of African rainforests are more water-limited than light-limited on a regional basis (Guan et al 2015, Madani et al 2017a).

Zoom In Zoom Out Reset image size

The GRACE TWS data for the year 2015 indicated that the groundwater level in tropical rainforests of Africa was 14.1 mm above the average. Meanwhile, the EBF canopy FPAR observed by the MODIS instrument, was on average about 2% higher in 2015, even though PAR was slightly lower than the average (∼1%).

Among the different biomes, grasslands in semi-arid savanna ecosystems show a relatively rapid response to changes in rainfall. In 2005, GPP of the grassland biomes declined by 8% (−0.6 Pg C yr⁻¹), while for the relatively wet year of 2010, GPP increased by 5.8% (+0.49 Pg C yr⁻¹). GRAs show the highest GPP variability, with an annual average of 7.97 ± 0.9 Pg C yr⁻¹. SHRs have higher GPP variability (4.22 ± 0.54 Pg C yr⁻¹) than the denser forest ecosystems, including EBF (6.51 ± 0.26 Pg C) and DBF (4.1 ± 0.27 Pg C yr⁻¹); whereas CROs show the lowest productivity and annual variability (3.35 ± 0.19 Pg C yr⁻¹).

The TRENDY model GPP pattern for African biomes follows regional rainfall anomalies and in general agrees with the L4C GPP record, but with less variability (figure 2(b)). However, the TRENDY models showed a 0.04 Pg yr⁻¹ reduction in EBF GPP in 2015, suggesting that there are mechanisms in ecosystem water availability modulated by groundwater that are reflected in the L4C model, but missing from the TRENDY models. The L4C and TRENDY models show similar seasonal patterns, but with smaller TRENDY GPP seasonal variation over tropical ecosystems compared to satellite observations of SIF and the modeled L4C GPP (For a comparison between seasonality in the L4C and TRENDY GPP with SIF, FPAR, PAR and environmental multipliers used in L4C LUE model, refer to figure S3 in the supplementary material).

To better understand the L4C GPP anomalies in different biomes across Africa, we focused on three major regions with different environmental conditions and climate seasonality, including (i) tropical and sub-tropical ecosystems with EBF and DBF PFTs falling between 10^o S and 10^o N; (ii) the Sahel, which is characterized by SHR and GRA PFTs; and (iii) southern savannas with GRA and SHR PFTs falling below 25^o S. We ignored CRO to eliminate the effect managed water on productivity.

In 2015, lower-than-average (2003–2017) rainfall in tropical and sub-tropical regions was offset by above-average TWS levels persisting from the net anomalously wet conditions in the 5 preceding years (figure 3). This reduced the negative impact of the 2015 rainfall deficit on GPP. While we observed a positive impact of deep soil water availability on plant productivity, it has been reported that the plants' increased photosynthetic activity can deplete the groundwater storage (Koirala et al 2017). This may explain the reduction of groundwater storage in tropical regions after the highly productive years of 2008–2010. Additionally, trees with conservative water-use behavior may maintain a low productivity level in order to conserve groundwater and maintain persistent productivity levels during drought events (Teuling et al 2010). This behavior might also explain the conservative tropical photosynthetic activity from 2014 to 2016, when TWS was above average. On the other hand, the reduction in TWS between 2010 and 2012 might be attributed to plants' increased evapotranspiration and productivity, suggesting that deep soil water has been used by plants during this rather dry period (Guan et al 2014, Tian et al 2018).

Zoom In Zoom Out Reset image size

In the dry regions of the Sahel and southern savannas, GPP inter-annual variability is strongly correlated with rainfall patterns (Pearson correlation of 0.67 and 0.84, respectively). The southern savannas showed a strong GPP response to inter-annual SM and rainfall variability, even though TWS was also highly correlated with the GPP anomalies. GPP inter-annual variability in the southern savannas shows higher correlation with SM compared to the Sahel region (r = 0.51 versus r = 0.20), suggesting that in southern savannas below-surface water has a great influence on ecosystem GPP. As shrubs in water-limited environments can be deeply rooted (Jackson and Jackson 2002), GPP seasonality in southern savannas is also well correlated (r = 0.7) with groundwater seasonality. As TWS includes both SM and groundwater (Tapley et al 2004), the high correlation of savanna inter-annual variability of GPP with TWS may be more related to variations in SM than to deeper groundwater variations. Even though shrub dominated southern savannas have access to groundwater, the GPP in this region has more variability than that of the grasslands of the Sahel. One explanation is that plants in more arid climates are closer to their limits of climatic tolerance and show higher drought stress as compared to vegetation in wetter environments (Young et al 2017). In 2015 in the southern savannas, the strong rainfall reduction by 145 mm coincided with on average 3.5% less FPAR, while PAR increased by about 0.4%. In 2010, FPAR in the Sahel was ∼2.4% above-average and at its highest level for the study period. The FPAR increase coincided with 70 mm higher than average rainfall in this region, which led to an approximate 4% reduction in PAR.

To analyze how GPP is affected by water availability memory, we performed a lagged-correlation analysis between water availability indicators and the L4C GPP. Our results revealed that in tropical ecosystems, annual GPP anomalies directly correlate with inter-annual variation in TWS and SM (figure 4(a)). However, the tropical GPP-rainfall correlations are greater over a 16–20 months time-lag, which may explain the 2015 increase in productivity despite significant reduction in rainfall, as well as reduced productivity in 2006 and 2017 despite greater rainfall in those years. On the other hand, TWS shows direct correlation with GPP, but there is a time-lag between rainfall and TWS recharge (figure 3(a)). This time lag may be attributed to two-way interactions between GPP and TWS and loss of TWS as a result of increased plant activity (Tian et al 2018). It has been shown that plant use below surface SM and TWS to maintain their photosynthetic activity during the entire dry season and rain-free periods (Guan et al 2014, Tian et al 2018). This water use strategy among tropical trees has been documented, where they use the available water to flush new leaves during the dry periods (Reich 1995, Chapotin et al 2006, Tian et al 2018).

Zoom In Zoom Out Reset image size

Savanna ecosystems in the Sahel, on the other hand, show significant response to rainfall and SM with no time lag. Southern savanna GPP also corresponds to rainfall patterns with little or no time lag, even though the variability in SM and groundwater also plays a major role in explaining the annual variation in GPP (figures 4(b), (c)). Semi-arid lands play a significant role in the global carbon cycle, as they control inter-annual variability in global productivity (Poulter et al 2014, Ahlström et al 2015, Haverd et al 2016). Higher GPP variability and sensitivity to rainfall in savanna ecosystems is consistent with the strong sensitivity of southern savannas to soil water content (Madani et al 2017a).

The long-term variability in annual GPP relative to annual variations in SM, groundwater and rainfall, and the relative spatial pattern of the correlations, is specified in an RGB map (figure 5). The partial correlation results reveal that in the wet tropical regions including central Africa, most of the variation in annual GPP is related to changes in groundwater storage. In southern Africa, variability in GPP is strongly related to changes in annual rainfall, even though groundwater and SM also have strong influence on GPP variability. In the Horn of Africa, eastern coasts and central Sahel, GPP variation is highly correlated with SM. The resulting pattern of regional influence shows that SM, groundwater, and rainfall interact in complex ways to control the inter-annual variation in productivity in Africa, even though the amount of variation in productivity is higher and more pronounced in semi-arid GRA and SHR regions.

Zoom In Zoom Out Reset image size

Our analysis of the effect of rainfall on GPP demonstrated that every ecosystem in Africa shows a positive response to rainfall, except in some wet tropical regions covered by EBF where productivity declines with additional rainfall (when rainfall exceeds 1800 mm yr⁻¹), likely due to reduced light availability under greater cloud cover (figure S5). The total area of the vegetated regions receiving more than 1800 mm annual rainfall represents approximately 6.7% of the total land area of Africa and falls within the rainforest biome. The average rain use efficiency (GPP/rainfall) in Africa is 1.44 ± 0.08 gC mm⁻¹ and varies conservatively across the different land cover types, whereby dryland shrubs have the greatest efficiency (1.55 gC mm⁻¹) and moist EBFs have the lowest efficiency (1.39 gC mm⁻¹; figure S6).

To further understand the response of African ecosystems to plant-available water, we separated the effect of rainfall on GPP variability by regressing the residuals of the rainfall-GPP relationship with GRACE TWS and FLDAS SM (figures S7(a), (b)) used as independent proxies for available water. We also looked at the same relationships between rainfall, SIF, TWS and SM (figures S7(c), (d)). We found that in central Africa (tropical and sub-tropical regions) annual GPP variability, when separated from a rainfall effect, strongly correlated with TWS and SM, while groundwater still has a major role in explaining some of the GPP variability in the southern savannas. The SIF-TWS relationship shows similar patterns as GPP-TWS even though there is some mismatch between the two datasets, which may be due to the different periods of record between GOME-2 SIF and GPP, and lower resolution and higher noise in the SIF signal. The spatial pattern examined indicated that below-surface water and rainfall interact in complex ways to control the inter-annual variation in GPP in Africa, with more pronounced variability in the semi-arid regions (GRA and SHR).

In 68% of the vegetated area, the GPP inter-annual variation is highly correlated with groundwater variability, making it the most important factor controlling GPP in Africa. This is in line with previous reports that the atmospheric CO₂ growth rate is partly controlled by anomalies in TWS (Humphrey et al 2018).

The direct relationship between annual anomalies in rainfall and TWS with GPP and SIF in tropical and sub-tropical regions indicate that the GPP in these regions is enhanced by greater rainfall and groundwater availability (figure S8). However, when rainfall is below average a positive signal in GPP is related to higher than average TWS. SIF follows the same pattern as GPP with the exception of reduction in SIF signal as a result of high precipitation, highlighting the role of groundwater storage in moderating the effect of drought in the dense forests of Africa.

There have been increasing attempts to add groundwater information to Earth system models (Barbeta and Peñuelas 2017). Remote sensing-driven LUE models can capture temporal information about plant phenology and canopy photosynthetic capacity with relatively high spatial resolution, which can indirectly reflect the influence of available water on photosynthesis. The anomalous 2015 El Niño year in Africa highlights the importance of soil water memory, as it was detected by changes in remote sensing vegetation indices and the fluorescence signal observed by GOME-2 (figure S9(a)). The TRENDY ensemble models describe the effect of rainfall reduction on GPP, but do not show the effect of available soil water as inferred from the satellite record (figure S9(b)). This may also be due to oversensitivity of the models to atmospheric CO₂ concentrations (Smith et al 2016); once we removed the seasonality and trend, the inter-annual variability in TRENDY GPP was lower compared to L4C GPP results.

Plants in humid tropics are less sensitive to climate anomalies (Reich 1995, Zhang et al 2016). However, significant changes in water balance can influence plant phenology and productivity (Reich 1995, Guan et al 2014), and here L4C GPP shows higher sensitivity to changes in water balance than do Earth system models. Our results also demonstrate differences in the response to variations in water availability between savanna and tropical forest ecosystems. While GPP inter-annual variability in tropical forests is more correlated with deep soil water, savanna ecosystems exhibit a mixed response to all water sources from SM to groundwater (figure 5), which may be due to species with diverse functional traits.

Savanna ecosystems of Africa are covered by grass and tree species, while woody cover tends to decline with more frequent yet less intense rainfall events (Good and Caylor 2011). Here, we analyzed ecosystems at regional scales; however, to better understand different plant water use strategies, we recommend monitoring plant species based on their key life history traits. Plants exhibit different strategies in the face of dry conditions. For example, anisohydric species are at risk of hydraulic failure as a result of xylem cavitation while isohydric species regulate stomata and thereby maintain plant water potentials to avoid cavitation, but are at risk of carbon starvation from depletion of carbohydrate reserves (Sala et al 2012, Klein 2014, Sankaran 2019). Here, we focused on periods where annual rainfall was lower than its annual average and not specifically on drought. Future studies could focus on the response of ecosystems in savanna and tropical regions during dry conditions with respect to total available water and their life history traits and drought tolerance strategies.

Overall, our results indicate that the anomaly in TWS determines the net balance between water availability and water loss and highlight the critical role of TWS in buffering rainfall shortages and ensuring continued availability of near-surface water to plants through dry spells.