**1. Introduction**

Droughts impact not only ecosystem functions but also the well-being of affected human populations [1,2]. The temporal and geospatial responses to drought are varied, and they threaten regions that are less well-adapted to water stress. A better understanding of the relationship between photosynthesis and water stress has broad reaching implications for our understanding of the global carbon and hydrological cycles.

Roughly half of the global variability in terrestrial carbon cycling can be attributed to carbon dioxide fluxes in tropical Africa, but the impact of drought on these fluxes is still modeled with significant uncertainty [3]. Further research on continental carbon fluxes is limited by data availability and characterization of regional productivity responses to drought [4,5]. The future of ecosystem productivity in the tropics is limited by how well plants cope with water stress but the influence of seasonal water stress on productivity, particularly in Africa's tropical regions, has rarely been characterized [6,7].

African climate is characterized by summer rainfall in the northern and southern tropics with an equatorial bimodal regime in-between. Regional variations notwithstanding (Figure 1), rainfall in tropical East Africa is delivered during the boreal spring and fall (Figure 2), accompanying local

solar insolation maxima. The rainy season from March to May, known as the "long rains" in Kenya, "Belg" in Ethiopia, and "Gu" in Somalia, includes the majority of annual regional rainfall. A second rainy season occurs during the boreal fall and is referred to as the "short rains" in Kenya, "Keremt" in Ethiopia, and "Deyr" in Somalia. For simplicity, Kenyan terminology is employed in this study along with March, April, May (MAM) and October, November, December (OND) to refer to respective rainy months. In the area of our interest (black boxes in Figure 1), annual mean precipitation is 439 mm/year; this is a dry region where the two growing seasons (Figure 3) are susceptible to even small decreases in precipitation.

**Figure 1.** Average annual mean precipitation from Global Precipitation Climatology Project (GPCP; left) and Tropical Rainfall Measuring Mission Multisatellite (TRMM; right) climatology data, demonstrating the variability in rainfall regimes in the area. For zonally averaged values, the drought region is defined as the land area between 1.5◦ S and 4◦ N and 38.5◦ E to 46.5◦ E (black box).

**Figure 2.** The climatology of rainfall over the drought region as presented by the two precipitation products used in this study, GPCP and TRMM.

Climate extremes on the African continent are well studied due to their potential to produce far-reaching economic and social consequences. East Africa has faced at least one major drought per decade over the last half century [8]. That said, regional droughts are typically isolated to only one of the two rainy seasons; consecutive failed rains are uncommon phenomena given the differing influences on precipitation for each rainy season. The successive failure of 2010 short rains and 2011

long rains produced the worst East African drought in the last 60 years [9], causing humanitarian crises in East Africa.

This study seeks to characterize the productivity response of tropical East Africa to water stress while identifying the unique spatial and temporal characteristics of the 2010–2011 drought. Related research on gross primary production (GPP) change in other tropical ecosystems and drought impacts relies on satellite-derived measurements because in situ measurements there remain scarce, largely due to operational constraints [10]. Here, we use relatively new satellite measurements of solar induced chlorophyll fluorescence (SIF) [11] to study the change in East African productivity during the drought, in conjunction with other atmospheric and terrestrial datasets. In Section 2, we describe the data and methods that we used. The results and discussion are presented in Sections 3 and 4. We summarize and conclude our study in Section 5.

#### **2. Materials and Methods**

#### *2.1. Study Area*

Our study area is the Horn of Africa region around the equator (Figure 1) where precipitation has two maxima (Figure 2). The Horn of Africa region, including Kenya, Ethiopia, and Somalia, collectively referred to as 'East Africa' (e.g., [12]), has been historically susceptible to droughts [13]. The El Niño Southern Oscillation (ENSO) significantly impacts OND precipitation [14]. The March and April precipitation signal, on the other hand, is more influenced by the position of the Intertropical Convergence Zone (ITCZ) (and thus the Indian Ocean warm pool temperatures [12]), and May rain is heavily influenced by the divergent low-level winds of the Indian monsoon [15].

### *2.2. Atmospheric Datasets*

We used satellite-derived rainfall estimates from the Tropical Rainfall Measuring Mission Multisatellite (TRMM) Precipitation Analysis (TMPA) 3B43 product, averaged monthly from 1998 to 2013 at a 0.25-degree spatial resolution [16]. TRMM precipitation is calculated from the radiative and emissive properties of cloud hydrometers at visible, infrared, and microwave wavelengths, which serve as proxies for rainfall rate. The TRMM 3B43 dataset combines rain gauge, infrared, passive-microwave, and precipitation radar estimates and is generally well correlated at a monthly time scale with African rain gauge measurements across the continent [17] and over East Africa's complex topography [18]. To supplement TRMM data, we used the Global Precipitation Climatology Project (GPCP 1979–2012) [19] dataset because it has a longer record, starting from 1979, and is thereby valuable to compare the 2010–2011 drought with other historic droughts. GPCP data are derived from rain gauge and satellite data. We note that the number of rain gauges has declined throughout this region (40 in 1979 to 5 in 2010) [20].

Soil moisture data integrate precipitation anomalies in time and were used here to depict the spatial distribution of contemporary African droughts. Soil moisture products retrieved from Advanced Scatterometer (ASCAT) and Advanced Microwave Scanning Radiometer–Earth Observing System (AMSR–E) represents soil moisture in the top 1–2 cm of soil at a spatial resolution of ~50 km, sensitive to small precipitation events [21]. Land emissivity is a function of soil moisture; AMSR–E employs a low-frequency passive microwave remote sensing approach to measure the brightness temperature at Earth's surface [21]. The ASCAT is an active microwave sensor that measures backscatter from the surface, which is a function of soil moisture. This resulting soil moisture estimate demonstrates potential for drought monitoring [22]. Remotely-sensed soil moisture measurements are used in similar studies to monitor drought, including to characterize the spatial and temporal distribution of the 2010–2011 East Africa drought [23]. This study differentiates itself from previous research in that it characterizes the vegetation response to these conditions.

Reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF) were used to characterize the atmospheric and land surface conditions that led to the successive failure of the 2010–2011 East African monsoons [24]. Variables analyzed for drought monitoring extended from January 1979 to present and were analyzed to assess different drought effects and dynamics. Two m air temperature was selected due to its relationship to surface soil drying (as in [25]), its associated role as a climate driver (as in [26]), and because higher temperatures reduce water use efficiency during photosynthesis [27]; water vapor flux data were selected to monitor regional moisture transport. Sea surface temperature (SST) data were from the HadISST [28].

#### *2.3. Terrestrial Datasets*

Data related to both canopy structure and plant physiological activities were included in this study to characterize surface-level drought impacts. We used the normalized difference vegetation index (NDVI) from the moderate resolution imaging spectrometer (MODIS) MOD13C2 product. The resolution was 250 m, and we used data from 2007–2012 at 16-day, 0.05-degree resolution to characterize the change in potential or accumulated productivity [29]. We used solar induced chlorophyll fluorescence (SIF) as an indicator of actual photosynthetic activity from the Global Ozone Monitoring Experiment-2 (GOME-2)) [11]. GOME-2 measures this fluorescence signal at 9:30 am local time and provides global coverage every 1.5 days. The nadir footprint size is 40 km × 80 km; here we used both monthly and weekly GOME-2 products from NASA at a spatial resolution of 0.5◦.

Fluorescence occurs when a solar photon is absorbed and is elevated to an excited state in the light reaction of photosynthesis. Typically, between 2 and 5 percent of photons absorbed by chlorophyll are re-emitted at longer wavelengths as fluorescence [30]. Canopy-level SIF measurements demonstrate that fluorescence capture productivity decreases even when NDVI remains constant [31]. SIF retrievals are also sensitive to seasonal dynamics of vegetation, independent from the structure of the canopy, and have been employed for stress detection in ground [32], aircraft [33], drone [34], and satellite-based instruments at wavelengths surrounding the oxygen A and B bands or Frauhoefer lines [35].

To assess the impact of the 2010–2011 drought on GPP, we scaled SIF to GPP using a linear relationship between monthly SIF and a GPP product, Fluxnet Multi-Tree Ensemble (MTE) GPP [36] during 2007–2011 (Figure 3). The Fluxnet GPP data were calculated using a machine-learning approach with eddy covariance datasets from flux towers, climatic variables, and remote sensing products. The dataset spanned from 1983 to 2011. The GOME-2 monthly SIF overlapped with Fluxnet–MTE over the 2007–2011 period. We then calculated the difference between 2010–2011 SIF and SIF climatology (2007–2012). We limited our analysis between 2007–2012 because GOME2 SIF showed a decreasing trend after 2013, likely caused by a sensor drift [37]. We calculated the GPP reduction during the drought period by using this relationship between SIF and GPP.

**Figure 3.** The relationship between solar-induced chlorophyll fluorescence (SIF) and gross primary production (GPP): (**top**) spatiotemporally-averaged SIF and GPP for the black box region in Figure 1; (**bottom**) the relationship between SIF and GPP.
