A March-to-May 2022 forecast of strong “Western V Gradient” conditions suggests that another poor eastern East Africa March-to-May rainy season is likely

Chris Funk, Gideon Galu, Diriba Korecha, Laura Harrison, Shraddhanand Shukla, Greg Husak, Andrew Hoell, Diego Pedreros,  Weston Anderson, and Juliet Way-Henthorne

Executive Summary: 

In eastern East Africa, back-to-back October-November-December (OND) and March-April-May (MAM) droughts pose severe threats due to the compound shocks associated with repetitive impacts on crop production, rangeland, and household resilience. FEWS NET research has shown that these consecutive OND/MAM droughts are often associated with La Niña-like climate conditions. When the eastern Pacific is much cooler than the western Pacific, rainfall tends to be increased in the area around Indonesia, and decreased over the eastern Horn. While temperature variations in the eastern Pacific are primarily dominated by natural variability associated with the El Niño-Southern Oscillation (ENSO), western Pacific temperatures are strongly influenced by both ENSO and climate change. Since 1999, when La Niña-like climate conditions are present, western Pacific sea surface temperatures (SSTs) have typically become very warm, and these warm conditions typically reduce rainfall over eastern East Africa. Understanding these relationships helped anticipate OND/MAM droughts in 2016/2017 and 2020/21. Unfortunately, climate forecasts for 2021/22 closely resemble those observed in 2020/21, as well as many other recent dry MAM seasons. Once again, exceptionally warm west Pacific Ocean conditions are likely to appear alongside moderately cool east Pacific SSTs, raising the distinct possibility of yet another dry OND and MAM season. 

While there is strong consensus that the upcoming OND season will be below normal, there is much less agreement on the March-April-May 2022 season. Almost 20 years of FEWS NET research supports the view that an interaction of climate change and ENSO connects recent La Niña-like climate conditions to below-normal MAM rains. This link can be shown inductively by examining La Niña years, as done in a recent “Elevated Chances of Drought” post, which concludes that a below-normal eastern East African MAM season is likely. 

The key messages of the “Elevated Chances of Drought” analysis are:

1. Parts of East Africa have experienced repeated droughts and are currently facing a “crisis” level of food insecurity.
2. Current ENSO forecast-based climate analogs indicate elevated chances of below-average precipitation, yet another drought, in the March-April-May (MAM) 2022 season mainly in eastern Ethiopia, Kenya, and Somalia.
3. Below-average MAM seasonal precipitation is likely to lead to below-average crop yields (based on the analysis conducted using Kenya and Somalia crop yield data).

Here, we augment the second key message, concerns about the 2022 MAM, using a deductive approach based on our diagnostic analyses of past MAM droughts. Beginning with the best available rainfall data, we analyze SST observations and forecasts to add the following key messages:

 

1. The frequency of low MAM Eastern East Africa rainfall seasons has increased substantially following 1998 (Figures 3 and 4), and there is a strong relationship between this increase and warming in the western Pacific (Figure 7, middle). The West Pacific will almost certainly be exceptionally warm.
2. Pacific Ocean SST patterns associated with EEA dry seasons strongly resemble the MAM 2021 and 2022 SST forecasts (Figure 5 and 6). 
3. There is a very high (~90%) chance that MAM SST conditions will be associated with a strong “Western V Gradient” pattern, a pattern common to many recent dry east African rainy seasons (Figure 7, right).
4. There is >50% chance that these strong gradient conditions will lead to below-normal eastern East African MAM 2022 rains. Many areas that currently face crisis levels of food insecurity may experience yet another poor MAM 2022 rainy season (Figures 8 and 9).
5.  Considering the Western Pacific, in addition to the East Pacific, adds value,  especially during recent relatively weak La Niña events.

Figure 1. March-April-May 2021 rainfall, expressed as a percent of the 1981-2021 mean. Based on CHIRPS rainfall augmented with IRE data in Ethiopia. Eastern East Africa area of analysis is highlighted by the green polygon. A South-central “hotspot” region is noted with a blue polygon.

 

Background — high levels of food insecurity, repetitive droughts, low crop production

As discussed in a recent CHC post (here), “current ENSO forecast-based climate analogs indicate elevated chances of below-average precipitation, yet another drought, in the March-April-May (MAM) 2022 season mainly in eastern Ethiopia, Kenya, and Somalia.” This analysis had three key messages focused on 1) repeated East African droughts in many areas facing crisis levels of food insecurity, 2) how ENSO forecast-based climate analogs indicate elevated chances of a below-normal March-April-May (MAM) 2022 season, and 3) likely below-average crop yields (based on the analysis conducted using Kenya and Somalia crop yield data). 

At present, most of Eastern East Africa (EEA) — Ethiopia, Somalia, and Kenya (south and east of 38°E and 8°N [Figure 1]) — faces crisis levels of food insecurity “… due to conflict, weather, and economic shocks.” Given that the OND 2020 and MAM 2021 seasons were poor in many areas, and that the upcoming OND EEA 2021 rains are expected to be below-normal (here, here), the expectation for below-normal MAM 2022 rains could lead to four consecutive droughts in a region that faces substantial food-security challenges even under normal conditions.

In Ethiopia, FEWS NET reports “many southern and southeastern pastoral areas experienced consecutive poor seasons in late 2020 and early  2021, resulting in lower than normal pasture and water availability”, while “the 2021 belg harvest is expected to be delayed and below average due to a poor distribution of rainfall that led to delayed planting.” Crop Monitor analyses indicate poor Belg crop conditions due the very erratic rainfall distribution. In northern Ethiopia, “the Tigray region of Ethiopia continues to experience one of the worst food security emergencies globally” with large-area food-security emergency conditions.

In Kenya, the government has declared a national drought emergency, and the August Kenya Food Security Steering Group (KFSSG) long rains season assessment report finds that … “across the eastern and southern parts of the country the rains received were 51-75 percent of normal …. Lamu and parts of Turkana, Marsabit, Samburu, Isiolo, Wajir, Garissa, Tana River, Kilifi and Taita Taveta counties received amounts that were 26-50 percent of normal rainfall…. The rains were poorly distributed in space and time particularly where below-average rainfall was received significantly affecting cropping both livestock and crop production.”

In Somalia, “A delayed start, early end, and erratic rainfall distribution characterized the April to June 2021 Gu rainfall season. As a result, cumulative rainfall was below the 40-year average across much of the country, especially in central and southern Somalia. The poor rains led to below-average Gu crop production in southern Somalia and poor crop harvest prospects in agropastoral livelihood zones in the Northwest …. The 2021 Gu season cereal production in southern Somalia is estimated at 48,900 tons, which is 60 percent below the 1995-2020 average.” Expectations for 2021 Gu cereal production (Figure 2) appear to be only slightly higher than the very low production years of 2019 and 2011. Somalia’s Gu cereal production has fallen substantially below its 10-year average every year since 2019. It appears likely that Somalia will face a fourth low-production 2022 Gu growing season.

Figure 2. FEWS NET/FSNAU Somalia Gu season Maize+Sorghum crop production estimates.

Background – Stages of CHC Long Rains Outlooks

Here, we reinforce concerns for the MAM 2022 outcomes with a detailed “deductive” examination of 1) the 2022 March-April-May sea surface temperature (SST) forecast, 2) the uncertainties inherent in this forecast, and 3) the historical relationships between forecast MAM SSTs and the EEA long rains. These results are similar to those provided for MAM 2021 in December 2020 (here), but at a substantially greater lead time. We also provide an evaluation of historical forecast skill performance, focused on our ability to predict western and eastern Pacific SSTs. The forecasts, made in early September, can skillfully predict strong Pacific sea surface temperature gradient conditions. This skill provides the foundation for predicting likely below-normal Eastern East African (EEA) rains.

By design, this post is intended to synergize with our recent La Niña-based analysis. The La Niña analysis is inductive and has a larger regional focus, covering the entirety of East Africa, where MAM rainfall is above 30% of annual rainfall. It begins with assumptions that a La Niña will form in OND and will be followed by either La Niña or Neutral conditions in MAM. It then identifies a set of analog MAM seasons based on that assumption, and concludes that such conditions are associated with elevated chances of below-normal rains. Such analysis is clearly valuable. 

In addition, however, for a targeted region of EEA (rather than EA), it is also useful to pursue a deductive process where we begin with observed EEA MAM rainfall, and then diagnose large-scale forcing regions that have historically been important for abnormally dry seasons. With these, we can build statistical models that are explicitly tailored to EEA. For EEA, prior FEWS NET research has combined statistical prediction schemes with accompanying diagnostic analyses using climate model simulation results. 

There have been three major stages in this work. The first stage explored the first and second modes of Indo-Pacific climate variability, and concluded that a combination of climate change and La Niña-related variability increased EEA drought impacts during recent La Niñas (here and here). This research supported a 2010 alert, and corresponds with the post-1998 maps shown in Figure 5 of our “Elevated Chances” post. One clear advantage of this inductive approach is its ability to function at very long lead times. 

The next stage focused on more precise assessments, at shorter leads, based on statistical regressions or constructed analog approaches. In late 2016, regression approaches were used to translate observations of December SSTs into reasonably accurate predictions of low MAM EEA rainfall. Forming part of a special joint alert, these forecasts helped motivate early and effective responses when the 2017 rains failed.

The third stage (which we still occupy) involves combining statistical approaches with the ability of the climate model forecasting systems to make accurate long-lead predictions of western and eastern Pacific SSTs. We are using methods similar to those used previously, but powering them with long-lead SST forecasts from climate model ensembles. Deductive analyses, such as those described in a 2018 Quarterly Journal of the Royal Meteorological Society (QJRMS) paper and a 2019 Bulletin of the American Meteorological Society paper (BAMS), relate EEA MAM droughts to two well-predicted regions — the western Pacific “Western V” region (described below) and the equatorial east Pacific Niño4 and Niño3.4 regions. As we have described in several papers, and a 2021 blog that analyzed and described conditions last March, Warm Western V and cool east Pacific Ocean conditions collaborate to reduce rainfall over EEA. The robust links between SST conditions in these regions and EEA MAM rainfall creates opportunities for effective long-lead predictions. Deductive statistical forecasts, such as those presented here, can operate synergistically with inductive outlooks, such as those presented in our “Elevated Chances” post. Table 5 in our QJRMS paper documents the recent increase in EEA MAM/Niño3.4 correlations, while also describing a mechanism explaining this increase—a positive interaction between exceptionally warm west Pacific SST conditions and cool east Pacific La Niña SST anomalies.

In our December 2020 post, we started with a scatterplot (Fig. 2 here) showing observed MAM EEA Standardized Precipitation Index (SPI) values along with forecast-based estimates derived using the observed MAM “Western V Gradient” (WVG)—defined as the difference between equatorial east Pacific SST and “Western V” SST values. While this relationship is relatively weak, and fails to capture droughts in 2004 and 2019, it does have the ability to effectively capture many recent dry seasons. Furthermore, when there is a strong observed gradient, this approach has had a low false alarm rate. Numerous peer-reviewed studies have established the empirical link between Western V Gradient conditions and EEA MAM dry seasons via statistical regressions and composites (QJRMS, BAMS). 

In this post, we 1) recreate some composite-based results using the most up-to-date and highest quality EEA rainfall data, 2) explicitly evaluate the predictability of MAM WVG gradient conditions, and 3) use empirical data and WVG forecasts to show that the WVG forecasts can effectively anticipate most EEA below-normal seasons in September, and finally 4) use the 2022 WVG forecast to identify a set of analogs indicating that EEA in MAM 2022 will likely receive below-normal rains.

The best-available precipitation estimates

The quality of our deductive analyses depends on the quality of our empirical data. Here we examine a combination of high-quality CHIRPS data for East Africa and even-higher quality Improved Rainfall Estimates (IRE) for Ethiopia. In 2020 and 2021, collaborative efforts by the Ethiopian National Meteorological Agency (NMA) and FEWS NET produced a series of high-quality agro-meteorological special reports. These reports are based on IRE—precipitation grids that are produced by combining the Climate Hazards Center InfraRed Precipitation with Stations (CHIRPS) product with 40-to-140 additional stations from Ethiopia. To provide the best possible synoptic description of the 2021 MAM season, we have combined Ethiopian IRE data in Ethiopia with CHIRPS outside Ethiopia. A map of 2021 MAM rainfall, expressed as a percent of the long-term (1981-2021) mean, reveals the relatively poor performance. Two regions of interest are also marked in the figure — EEA (green polygon) and “South Central Ethiopia.” The latter region has been highlighted in recent Ethiopia agrometeorological reports as a densely populated area that has received very poor 2021 rains (here). The EEA region has been singled out because prior FEWS NET research has documented the substantial MAM rainfall declines in this region, while also highlighting teleconnections between this region and potentially predictable MAM Indo-Pacific SSTs. 

Our primary focus is on EEA, but we include some discussion of south-central Ethiopia because of high levels of concern in this area.

Figures 3 and 4 show areally-averaged MAM rainfall totals expressed as “Standardized Precipitation Index” (SPI) time-series. As anticipated in late 2020 (here), rainfall totals were below normal, on average, across this region. We also note from these time-series that there has been a substantial increase in the frequency of dry seasons since after 1998. Many of these seasons exhibited La Niña conditions in the prior OND and La Niña or neutral conditions in the following MAM season—conditions which our recent “elevated chances” post links to higher probabilities of below-average or below-normal rainfall (top panels, Figure 4, here).

Figure 3. Eastern East African MAM SPI time series. Below-normal (bottom tercile) seasons are noted in red.

Figure 4. South-central Ethiopia MAM SPI time-series. Below-normal (bottom tercile) seasons are noted in red.

Deducing SST forcing regions

Following standard deductive analytical practices, we can gain insights into at least some of these dry seasons by examining composites of NOAA Extended Reconstructed SSTs (Figure 5). These results are similar to results shown in QJRMS and BAMS, but include the most up-to-date data available. To produce Figure 5, the 1981 to 2021 SST values were standardized using a 1981-to-2021 baseline. Observed MAM SSTs were then grouped into two composites — one composite corresponding to normal to above-normal MAM rainfall in EEA (i.e., the years with blue bars in Figure 3)  — and one composite corresponding to below-normal MAM rainfall in EEA (i.e., the years with blue bars in Figure 3). Then, differences in composite mean SST were computed and screened for statistical significance using a p-value of 0.1. These differences are shown in Figure 5.

In the Pacific, these results resemble composites from our QJRMS and BAMS papers. “Western V” and Niño3.4 Pacific Ocean forcing regions are identified with yellow boxes: Tropical West Pacific (120-16E, 15S-20N), Western North Pacific (160-150W, 20-35N), Western South Pacific (155E-150W, 30S-15S), and Niño3.4 (170W-120W). A peer-reviewed discussion of how these regions interact is provided here; a less-technical discussion based on February 2021 conditions is available here. Two central themes of the peer-reviewed QJRMS paper were that a) exceptionally warm west Pacific SSTs are enhancing La Niña-related risks in MAM, increasing the potential for back-to-back OND/MAM shocks, and b) the increasing ENSO-related extreme SSTs are likely to produce more intense climate shocks in eastern and southern Africa. Since 2014/15, we have experienced three El Niño seasons and four La Niña seasons, with another La Niña season likely to be on the way. During the one neutral OND season (2019), we experienced an exceptionally strong Indian Ocean Dipole event. If Indo-Pacific SST volatility is increasing, then analyses such as those provided here will help provide long-lead forecasts.

Significant forcing regions are also identified in the Indian Ocean—off the coast of southern Africa and northeastern Somalia. The north-south structure of these anomalies could help pull the main location of MAM African rains south, inhibiting onset and intensity. Here, however, we focus on the Pacific, because of our extensive prior research and because Pacific SSTs are better predicted at long-lead times.

Figure 5. Standardized observed SST composites for 1981-2021 dry versus wet-to-normal EEA MAM seasons. Results screened at an 0.1 significance level. 

The MAM 2022 forecast appears similar to MAM 2021

We next present the MAM 2022 SST forecast (Figure 6, left) which appears very similar to the MAM 2021 forecast from early September (Figure 6, middle) and the actual observed NOAA Extended Reconstruction SSTs (Figure 6, right). These SSTs are expressed as standardized anomalies based on a 1982-2021 baseline. The NMME forecasts begin in 1982. These predictions are derived from the September update of the North American Multimodel Ensemble (NMME). Five models were used to produce the NMME SST forecast—the CanCM4i, COLA-RSMAS-CCSM4, GEM-NEMO, NASA-GEOSS2S, and the NCEP-CFSv2. The MAM 2022 conditions are similar to the forecast from last year. The Niño3.4 SST forecast is -0.6Z. The Western V MAM forecast is +1.8Z. The WVG forecast is also quite extreme—1.5Z.

Our interpretation of Figure 6 will depend on our understanding of what drives (most) below-normal EEA MAM rainy seasons. If we make the assumption that La Niña, as represented by the Niño3.4 region, is the best discriminator, then we might conclude that the MAM 2022 outlook is uncertain, with perhaps a weak tilt towards below-normal rains. This assumption is not well-supported by the data (Figure 5), and prior peer-reviewed literature, which identifies an interaction between the “Western V” region and the eastern Pacific. From this perspective, Figure 6 is quite concerning. We find a very strong MAM 2022 WVG gradient (-1.5), similar to forecasts and observations from last year, and the composite structure associated with below-normal EEA MAM outcomes (Figure 5). 

This post will conclude with a formal assessment of how well September WVG forecasts can discriminate between EEA below-normal and normal to above-normal MAM seasons. But, before examining that question, we assess the September NMME’s ability to forecast observed MAM Niño3.4 and Western V SSTs.

Figure 6. Standardized SST forecasts for MAM 2022 and 2021 (left and middle). Observed MAM 2021 SST conditions (right).

Assessing the September NMME’s ability to forecast MAM Niño3.4, Western V and WVG

We next examine the September NMME’s ability to forecast MAM Niño3.4, Western V, and WVG values. Scatterplots of predicted NMME values (x-axes) and observed MAM SST values (right axes) are shown in Figure 7. This figure also allows us to place the 2022 forecast in historical context. Overall, all three indices are predicted fairly well, with R2 values ranging from 0.65 to 0.75. Standard errors were calculated based on the observed Extended Reconstruction version 5 SSTs and NMME ensemble averages estimates (i.e., the data shown in each scatterplot). The associated Niño3.4, Western V, WVG standard errors are 0.6Z, 0.4Z, and 0.6Z. These standard errors were used to derive 80% confidence intervals. The 2022 forecasts are shown along with these intervals. The y-axis 2022 value is assumed to be the same as the 2022 NMME forecast value.

Consistent with our visual interpretation of the 2022 MAM forecast, shown in the left panel of Figure 7, we see that the Niño3.4 and Western V SST patterns correspond, respectively, with modestly cool and exceptionally warm ocean conditions. Taken alone, the Niño3.4 index (left panel) does not do a good job of discriminating below-normal EEA MAM seasons (orange circles), except in the case of forecasts for strong (<-1Z) Niño3.4 conditions. When such conditions were forecast there were four hits (1984, 1999, 2000, 2011) and two misses (1985, 1989). 

The Western V SSTs (middle panel), on the other hand, does a better job of discriminating dry events. Focusing on the relationship between the forecast Western V values occurrence of dry seasons, we can see that there is marked increase in the number of dry seasons when forecasts start exceeding ~+0.6Z. During the nine MAM seasons when Western V forecasts were predicted to be greater than +0.6Z, six seasons were below normal (1999,  2011,  2012,  2017, 2019, and 2021), while two were exceptionally wet seasons (2018, 2020). It should be noted that the NMME’s ability to forecast Western V SSTs is very good, and the confidence intervals for this forecast suggest that there is a 90% chance that Western V SSTs will exceed a value of +1.2Z. As discussed in Figure 10 QJRMS, this exceptional warmth should be interpreted as an interaction between climate change and natural ENSO variability. A dry outcome, however, is far from certain, as indicated by the very wet 2018 and 2020 rainfall seasons (Figure 3).  EEA can still receive copious rains, even when the Western V region is very warm. Nevertheless, the Western V forecasts clearly capture EEA’s increased frequency of dry events. Decision-makers should not ignore this and only focus on Niño3.4 conditions. 

Finally, the right panel in Figure 7 presents WVG forecasts and observations, along with confidence intervals. Overall, this combined index does quite a good job of discriminating dry versus normal to above-normal outcomes. Based on the 2022 forecast and historical standard errors, there is a 90% chance of the 2022 MAM observed WVG value being less than -0.7Z. During these nine years (1989, 1999, 2000, 2008, 2011, 2012, 2017, 2018, 2021) there were six below-normal seasons (1999, 2000, 2008, 2011, 2017, 2021), two normal seasons (1989, 2012) and one wet season (2018). When the WVG is expected to be above +0.5Z, there were only two dry seasons (1983, 1992). So, while September WVG forecasts can clearly not capture all dry seasons, they do clearly indicate a substantial tilt towards drier conditions in 2022, for EEA in MAM.

Figure 7. Forecast and observed Niño3.4, Western V, and WVG index values. Forecasts are based on the early September NMME ensemble. Observations are based on the MAM NOAA extended reconstruction version 5. Orange circles note below-normal EEA MAM seasons (red bars in Figure 2). The red circles and red horizontal line denote the 2022 forecasts and 80% confidence intervals. The associated Niño3.4, Western V, WVG R2 values are 0.65, 0.75, and 0.76. 

Probability distributions and percent anomalies associated with analog EEA and South Central Ethiopia seasons

Finally, we plot PDFs based on the observed SPI values (Figures 3 and 4) for the analog seasons identified via our WVG analysis (1989, 1999, 2000, 2008, 2011, 2012, 2017, 2018, 2021). To support both regional decision-making and early action—early warning, more specifically, in Ethiopia, we present results both for the EEA region, based on Figure 3, and for the densely populated region of south-central Ethiopia (Figure 4). The medians associated with the analogs for these regions were -0.7 SPI and -0.9 SPI, so we find slightly more pessimistic outlooks for South-Central Ethiopia. Using the associated analog-based PDFs (Figure 8), we can calculate the probability of below-normal (SPI < -0.44Z) MAM 2022 rainfall. For EEA and South-Central Ethiopia, the associated probabilities (58% and 64%) indicate that a fourth below-normal season for these regions is likely. 

We can also use our selected analogs to make a spatial map of the likely spatial distribution (Figure 9), with the caveat that there is a high degree of spatial variability, even within dry years. Much of the eastern Horn can be expected to be moderately dry (<85% of normal) or dry (<70% of normal). To facilitate interpretation in applied impact assessment settings, we have used the past ten years (1992-2021) to represent “normal” conditions. Maps based on a 1981-2021 period of record are quite similar. 

Figure 8. PDFs for EEA and South-Central Ethiopia, based on the selected analog years (1989, 1999, 2000, 2008, 2011, 2012, 2017, 2018, 2021).

Figure 9. Median analog precipitation expressed as a percentage of the recent 2012-2021 mean. 

Conclusion: both inductive and deductive risk assessments support an assumption for below-normal MAM 2022 EEA rains, with associated disruptions in crop production

The CHC advocates a “defense-in-depth” approach to early warning, in which multiple approaches are used throughout the growing season to provide advance notice of potential climate shocks. At very long leads (>8 months), ENSO-based assessments, such as those presented in our “Elevated Chances” post can provide a very valuable first line of defense. These assessments are inductive. We begin with assumptions about the large-scale climate, and induce likely outcomes. Such assessments support a below-normal outlook for MAM 2022. 

Once we enter the 8-month forecast window provided by climate model ensembles like the NMME, we can employ deductive estimation procedures. Here, building on numerous prior empirical and diagnostic analyses, we begin by simply compositing observed SSTs during dry versus normal to above-normal seasons (Figure 5). These composites look like forecasts for 2021 and 2022 (Figure 6), reinforcing concerns raised by our “Elevated Chances” analysis.

Since now, in September, we can explicitly assess the skill of the NMME Niño3.4 Western V and WVG forecasts, we do so. These results indicate we are almost certainly going to see exceptionally warm Western V SSTs, and that past September forecasts for very warm Western V presaged many below-normal EEA rainy seasons (Figure 7, center). Conversely, we find that NINO3.4 forecasts are poor predictors of low EEA MAM rains (Figure 7, left). Yet, combining the Western V and NINO3.4 forecasts (Figure 7, right) does a reasonable job at discriminating droughts, even at an 8-month lead time. 

Using WVG forecasts to identify analogs allows us to generate MAM EEA PDFs and anomaly maps (Figures 8 and 9). These results help bolster concerns for MAM 2022.