*2.4. Analysis*

Remotely-sensed proxies for terrestrial vegetation activity (SIF and NDVI) and atmospheric reanalysis data were processed via the same methodology. These data were presented as standardized anomalies (z-scores), calculated by removing climatological mean values and dividing by the standard deviation of the mean. The resulting signal was therefore corrected for seasonal variability and the differences in the magnitude of the anomaly among different variables. In this analysis, multi-year means were calculated over as long a time period as the record permitted. For zonally averaged values, the drought focus region was defined as the land area between 1.5◦ S to 4◦ N and 38.5◦ E to 46.5◦ E (see black box in Figure 1), modified from [20].

In our analysis, the 2010–2011 drought was compared to previous droughts in East Africa as recorded in the emergency events database EM-DAT: The International Disaster Database, consistent with [38]. EM-DAT provides global historical drought records. In our study we included all EM-DAT-recorded drought events in Kenya, Ethiopia, and Somalia with the exception of 1987, 1988, and 1989, which were excluded because the drought season could not be determined. Previous regional drought years in the Horn of Africa include 1980 (MAM), 1983 (OND), 1991 (OND), 1994 (MAM), 1997 (MAM), 1998 (OND), 1999 (MAM), 2000 (MAM), 2003 (MAM), 2004 (MAM), 2005 (OND), 2008 (MAM), and 2009 (MAM). We note that 1998–1999 is the only other consecutive set of drought events before the 2010–2011 drought in the past 30 years.

## **3. Results**

#### *3.1. Spatial and Temporal Patterns of the 2010–2011 Drought*

During the 2010–2011 drought, regional reductions in rainfall were evident spatially and temporally (Figures 4–6). The 2010–2011 drought, in fact, represented a significant reduction (as large as three standard deviations less) in rainfall, not only as compared to the climatology (Figures 5D and 6D), but also as compared to the region's substantial drought history (as large as 1.5 standard deviations; Figures 5F and 6F). During the previous drought years (a 'drought climatology' is presented in Figures 5B and 6B), a negative rainfall anomaly of the short rains (OND) was primarily between 40◦ E and 43◦ E, whereas the 2010 drought extended further east and with greatest intensity further westward than the average of 41.5◦ E (Figure 5E,F and Figure 6E,F). The decrease of the ensuing long rains (MAM) was one standard deviation below that of previous droughts (Figures 5B and 6B) in the southern hemisphere, exacerbating drought started in the season of short rains. In both rainy seasons, the 2010–2011 drought extended further east than in previous drought events. The 2010–2011 drought was spatially distinct from previous regional droughts.

The ITCZ does not extend over East Africa during the rainy seasons but its spatial coherence over the West Indian Ocean can be considered as a proxy for moisture transport to East Africa because the Indian Ocean temperature alters the local Walker Circulation [13]. East Indian Ocean SST is warmer than West Indian Ocean SST (Figure 7), and this east–west gradient in the SST pattern has been associated with droughts in East Africa [13]. A strong negative correlation exists between SSTs in the tropical Indian and Pacific Oceans and East African rainfall [39].

In August and September of 2010, the precipitation that eventually became the southern maximum of the West Indian Ocean double ITCZ was significantly reduced in intensity and showed a less robust spatial integrity than was evident in the climatology (Figures 6 and 7). Anomalously low precipitation over the western Indian Ocean in the months preceding the OND long rains yielded a reduced westward extension of the southern ITCZ and correspondingly reduced water vapor transport to East Africa. Ultimately, this resulted in less distinct double ITCZs over the West Indian Ocean and reduced precipitation over East Africa.

**Figure 4.** The 2010–2011 drought (red) represented a significant reduction of productivity (SIF and NDVI, Normalized Difference Vegetation Index) as compared to the mean annual cycle (thick black line) of different precipitation products averaged over the drought region.

**Figure 5.** The 2010–2011 drought was spatially and temporally unique. The left column represents monthly rainfall climatology (GPCP) (**A**), drought climatology (alternatively referred to as a drought composite—an average of previous regional droughts (**B**), and 2010–2011 rainfall (**C**). Cool colors represent high rainfall. The right column represents average rainfall anomaly (B minus A) during drought years (**D**), 2010–2011 rainfall anomaly (C minus A) (**E**), and drought anomaly (E minus D or precipitation anomaly as compared to the drought composite average anomaly) (**F**). Brown values represent anomalously low rainfall whereas blue values represent greater than expected (anomalously high) rainfall. The anomaly was averaged along the longitude between 38.5◦ E and 46.5◦ E.

**Figure 6.** Similar to Figure 5, but the anomaly was averaged along the latitude between 1.5◦ S and 4◦ N.

**Figure 7.** The climatology of sea surface temperature (HadISST, ◦C) and TRMM rainfall (mm/day) over the Indian Ocean during both rainy seasons. The East Indian Ocean is warmer than the West Indian Ocean. Moisture transport to East Africa is linked to this temperature gradient.

Failed formation of the double ITCZ over the West Indian Ocean continued into the MAM season, a season with a typically weaker double ITCZ (Figure 7). Precipitation in the West Indian Ocean was

anomalously low during MAM of 2011, particularly over the southern maxima of the double ITCZ (Figure 8). Wind divergence from the southern ITCZ during January and February 2011, coupled with reduced long rains over the same area (Figure 9), produced a significantly weakened southern ITCZ, when compared to previous drought years. Water vapor for East African rainfall was heavily linked to this precipitation, and a relationship between rainfall in West Indian Ocean and East Africa was clear: strongly reduced rainfall in the West Indian Ocean resulted in (or was at least a strong proxy for) decreased precipitation in East Africa (e.g., [13]).

**Figure 8.** The anomaly of sea surface temperature (HadISST, ◦C) and rainfall (mm/day) during 2010–2011 East African drought.

**Figure 9.** Anomalies of 2 m surface air temperature (filled colors) and 850 hPa wind (vectors). Temperatures generally increased in East Africa during the 2010–2011 failed rains. This contributes to but may also exist as a positive feedback with decreased photosynthetic activity: photosynthesis decreases at higher temperatures, but surface temperatures increase with the resulting reduced transpiration. OND is October, November, December; MAM is March, April, May.

#### *3.2. Photosynthetic Responses to the Drought*

SIF captures the spatial distribution and severity of the 2010–2011 East African drought (Figure 10). Anomalously low photosynthetic activity spatially correlated with soil moisture anomaly over the duration of the drought. MODIS NDVI spatially (Figure 10) and temporally (Figure 11) approximated the extent of the 2010–2011 drought, particularly within 10◦ of the equator.

**Figure 10.** Anomalies of NDVI, SIF, soil moisture, and precipitation during the 2011 drought. Both SIF and NDVI approximate similar drought spatial extents. Anomalies calculated at highest available temporal resolution from the onset of the drought in October 2010 through April 2011.

The temporal response of NDVI or SIF followed soil moisture (Figure 11). The relationship was stronger with a lagged correlation (NDVI or SIF lags soil moisture) (Table 1), probably because the response of the ecosystem productivity took time. A 16-day lagged correlation has higher coefficient values (0.83 and 0.76 for NDVI and SIF) compared with no-lag correlation (0.56 and 0.35 for NDVI and SIF,). Both SIF and NDVI values were sensitive to soil moisture or water stress (Figure 11).

The small signal of SIF (1–2% of total APAR), and the large footprint of the GOME-2 sensor likely produced significant noise, as evidenced by the anomalies on the western extent of Figure 10. The considerably greater spatial resolution of the MODIS product likely yielded spatial precision in the MODIS NDVI panel of Figure 10.

As was demonstrated in previous studies [11,35], SIF scales well with GPP (Figure 3). We calculated GPP reduction using the relationship between GOME 2 SIF and Fluxnet–MTE GPP. For the long rain period, the total reduction in GPP was 117.47 gC m<sup>−</sup><sup>2</sup> (95% CI: 110.55, 124.40). For the short rain period, the total reduction in GPP was 64.4 gC m<sup>2</sup> (95% CI: 61.43, 67.38). The total reduction of GPP during this drought accounted for 32.1% of annual mean GPP (mean annual GPP in this area is 565.6 gC m<sup>−</sup>2).

The spatial distribution of surface temperature anomalies calculated from ERAI monthly averages since 1979 suggested that decreased photosynthetic activity was correlated with temperature increases of between 0.8 and 1.3 degrees Celsius in the fall of 2010. Temperature anomalies were less pronounced in MAM 2011 (Figure 9).

**Figure 11.** Zonally averaged SIF and NDVI response to soil moisture in East Africa (a black box in Figure 1. This figure was generated from 16-day average SIF from Global Ozone Monitoring Experiment-2 (GOME-2) data (red), NDVI from moderate resolution imaging spectrometer (MODIS) data (dark green), and daily soil moisture (blue) data.


**Table 1.** The relationship between soil moisture and NDVI or SIF from Figure 11. Lagged correlation is calculated as NDVI or SIF is 16 days lagging soil moisture.

## **4. Discussion**

The mean annual cycle of precipitation in East Africa is already difficult to characterize in modeled results and interpolated data sets [40]. A dearth of regional climate data contributes to this problem: precipitation estimates are limited by a low density of rain gauges and the short record of high spatial resolution soil moisture data (SMOS data only dates back to 2010) [41]. Long term reductions to the long rains are linked to rising SSTs in the western Indian Ocean, a trend likely to continue as a consequence of anthropogenic climate change [12]. If the West Indian Ocean becomes warmer, long rains are projected to fail with increased frequency, as is evident since 1999.

Intergovernmental Panel on Climate Change (IPCC) projections for the region sugges<sup>t</sup> a future characterized by increased episodic, extreme precipitation events, meaning there will be more rain and more droughts [42]. Other studies project an increase of rainfall in East Africa during the short rains in response to a large scale weakening of the Walker circulation [13]. Possible land-cover/land-use change as a consequence of agricultural technology adaption further exacerbates this uncertainty and are relevant to both natural ecosystems and agricultural regions, especially in food-insecure areas with high population densities [43].

Additionally, water stress is certain to have a substantial influence on natural ecosystems of East Africa in the coming decades. Previous research in the Amazon basin documented the impacts of drought on tropical ecosystems and was partially corroborated by this study [27]. Decreased water availability causes plant stomata to close, effectively shutting down photosynthesis. Increased temperature reduces the enzyme activity that enhances photosynthesis. Temperatures in East Africa are projected to rise with global warming and therefore the positive temperature anomaly during the OND 2010 drought offers insight into the productivity response of East Africa to this inevitable trend. Rising temperatures increase potential evaporation, thereby decreasing effective moisture. This mechanism is correlated with the drought stress influencing regional photosynthesis and suggests that water stress exhibits a first order control on ecosystem productivity in East Africa. A temporal analysis of surface temperature would be needed to corroborate this hypothesis and future work decoupling the relationship between temperature, water stress, and regional productivity is critical to an accurate projection of East Africa's future amidst climate change.

Previous research established the impact of rising global sea surface temperatures on the short rains: the dominant mode of variability in the tropical warm pool is related to global temperatures [39]—increased temperatures are likely to further dry East Africa due to the strong relationship between "short rains" and Indian Ocean sea surface temperatures. Efforts to link East African spring rains with sea surface temperature indices in the Pacific and Indian Ocean [44] continue, thus far suggesting that MAM rains are most sensitive to January sea surface temperatures (SSTs).

Both SIF and MODIS NDVI spatially approximated the extent of the 2010–2011 drought as compared to soil moisture and previous analysis [45], and SIF captured the response almost as well as did the NDVI, even if it has lower spatial resolution and much smaller energy, suggesting that SIF can also serve as an early indicator of drought in the future. However, state-of-the-art retrievals from GOME-2 are still limited by the sensor's design; the satellite was intended to monitor ozone, not SIF. Further, GOME-2 measurements occur at 0930 local time, well before the midday insolation and accompanying light saturation that causes stomata to close and thus shut down photosynthesis; vegetation may not be fully stressed at the time of SIF measurements. Taken together, these caveats sugges<sup>t</sup> that GOME-2 SIF data may have under-reported water stress during the 2010–2011 drought. Measurements by an optimized sensor at peak insolation could serve as an improved record of vegetation stress.

The spatial extent of the failed 2011 long rains was amplified by the reduced availability of water resources following the failed short rains in the previous season. Failed rainy seasons amplify the consequences of subsequent reductions in rainfall. The converse, however, was not true in 2010–2011: anomalously high rainfall in MAM 2010 did not protect the region from the failed rains of OND 2010.
