Below normal forecast for the 2017 East African long rains

Chris Funk, Greg Husak, Diriba Korecha, Gideon Galu and Shraddhand Shukla

Here we present an empirical forecast of below normal rainfall for the East African 2017 March-June (MAMJ) long rains. This forecast is based on statistical relationships between March-June rainfall in eastern East Africa (38-50°E, 5°S-8°N) and October-November Sea Surface Temperature (SST) observations in the Indo-Pacific. FEWS NET research has shown that dynamic models in this region and season lack skill, we therefore advocate using an outlook based on FEWS NET’s numerous statistical and diagnostic analyses focused on understanding and predicting the long rains (a, b, c, d, e, f).

As predicted based on September SSTs and October Kenya-Somalia rainfall, much of eastern East Africa received very poor October-November 2016  short rains and in many regions vegetation conditions are extremely poor. For the southeastern highlands region of Ethiopia, current dry conditions follow four poor rainy seasons during the 2015 and 2016 Belg and Kiremt rainy  seasons. For southern Somalia and eastern Kenyan crop growing areas, poor short rain harvests in 2016 follow a poor 2016 March-June cropping season. Current vegetation conditions in our focus region are very degraded (Figure 1). A poor 2017 March-June long rains could lead to yet another poor harvest in eastern Kenya, south-eastern Ethiopia and southern Somalia while delaying any substantial respite for pastoralists.

Figure 1. eMODIS NDVI anomalies and focus region.
Figure 1. eMODIS NDVI anomalies and focus region.

Eastern East Africa has experienced more frequent MAMJ droughts since 1995 (Figure 2) due to warm conditions in the Indo-Pacific warm pool, cool La Nina-like conditions in the eastern Pacific, and Pacific Decadal Oscillation-like SST variations in the northwest Pacific. FEWS NET research has suggested that warming in the Eastern Indian and Western Pacific Ocean has led to stronger March-June droughts during La Nina-like seasons (a, b).

In this study we focus on recent years with cool or neutral October-November eastern equatorial Pacific SSTs. During these years strong predictive relationships allow us to anticipate below normal rains for the following March-June rainy season. During El Niño-like years, predictability for the long rains tends to be low. While El Niño can bring above normal rains, some El Niño events, like 1992, were associated with drought and famine conditions. Excluding these El Niño seasons produces more coherent prediction results.

Beginning in 1980, we identify 16 cool-neutral October-November seasons, based on Niño 4 SSTs: 1980, 1981, 1983, 1984, 1985, 1988, 1989, 1995, 1996, 1998, 1999, 2000, 2007, 2008, 2010, and 2011. However, as shown below (Table 1), the selection of years has relatively little impact on our prediction of below normal rains. Selecting cool seasons or using all the years after 1980/81 produce similar pessimistic (below normal) forecasts for 2017.

Figure 2. Eastern East Africa standardized rainfall anomalies [Z scores]. Based on a 1981-2014 baseline.
Figure 2. Eastern East Africa standardized rainfall anomalies [Z scores]. Based on a 1981-2014 baseline.
Figure 3 shows a map of the correlation between October-November SSTs during neutral-Nina years and March-June East African rainfall the following spring. This is a familiar pattern we have seen associated with recent long rains droughts. A curve of negative correlations (~-0.7) begins in the north-central Pacific, curves through the Indo-Pacific Warm Pool and continues into the southeastern Pacific. A region of positive correlation in the north- central Pacific provides a counterpoint, and the combination of these warm and cool SSTs creates a pattern of low and high pressure that intensifies moisture convergence and rainfall over the eastern Indian Ocean and the Western Pacific – an area called the ‘Warm Pool’ because of its high average ocean temperatures. Increased Warm Pool precipitation leads to below normal East African rainfall.

Figure 3. Correlation between EA long rains and October-November SSTs.
Figure 3. Correlation between EA long rains and October-November SSTs.

Motivated by a desire to understand, and better anticipate, back-to-back droughts such as those associated with the 2010/2011 East African food security crisis, FEWS NET scientists have published numerous papers describing how warm Warm Pool and cool central Pacific SSTs combine to produce dry long rain conditions (a, b, c, d, e, f), while other scientists have emphasized the important role played by the north Pacific (g, h). All three regions (Warm Pool, North Pacific, East Pacific) are currently warm or cool in ways that are conducive to poor East African long rains. In 2012 and 2014, FEWS NET scientists used these relationships to make successful forecasts of below normal rains based on February or March SSTs.  Here, we extend this work to examine levels of predictability based on October-November SSTs, finding reasonable levels of skill and a forecast for poor March-June 2017 rainfall.

Figure 4. January-November 2016 Standardized Precipitation Index values. Based on CHIRPS data.
Figure 4. January-November 2016 Standardized Precipitation Index values. Based on CHIRPS data.

This forecast activity has been prompted by two factors – the very poor performance of 2016 January-November rainfall (Figure 4), and the emergence of a very strong West Pacific SST gradient (WPG) (Figure 5). What we see in Figure 4 is an area of extreme sustained dryness in the eastern Horn that appears associated with both the current poor vegetation conditions (Figure 1 below) and also the sustained low Normalized Difference Vegetation Index values throughout the year. Figure 5 suggests that this dryness has been associated with the development of an extremely steep West Pacific Gradient, caused by a combination of very warm West Pacific SSTs and moderately cool La Niña SSTs. What is interesting about Figure 5 is that some (but not all) recent strong El Niños rapidly transitioned into strong West Pacific Gradient events: the 1987/88, 1997/98 and 2008/09 Niños were followed by strong gradient conditions when cool La Niña SSTs combined with warm conditions in the Western Pacific. Following the 1997/1998 and 2008/09 Niño events, strong gradient conditions also tended to persist for several years while long rains performance tended to be below normal (Figure 2).

Figure 5. Time series of standardized 6-month running average Nino 3.4 and West Pacific SSTs, and their difference (WPG = Nino3.4 minus West Pacific).
Figure 5. Time series of standardized 6-month running average Nino 3.4 and West Pacific SSTs, and their difference (WPG = Nino3.4 minus West Pacific).

Focusing more closely on the Warm Pool region shown in Figure 3, which we selected because of its high correlation with March-June long rains in East Africa, we see very warm conditions (Figure 6, top), suggestive of below normal March-June rains. Several previous post-El Niño years (1998, 2008, and 2010) exhibited similar levels of warmth. Focusing just on years with neutral or cool El Niño-Southern Oscillation conditions, we find a strong negative relationship between October-November Warm Pool SSTs and East African (EA) long rains in the following year (Figure 6, bottom), with a +1°C increase in Warm Pool SSTs associated with a ~-3 standardized anomaly decrease in EA precipitation. Note, however, that a +1°C anomaly is unprecedented in this region (Figure 6, top). The high slope coefficient for this warm pool region is likely to be related to the strong relationship between increases in Warm Pool SSTs and rainfall over the Warm Pool. This region is very warm, and a +0.5°C increase in SST can lead to cyclones and very heavy Warm Pool convection. In turn, this increased precipitation can drive an overturning circulation associated with subsidence and reduced rainfall over East Africa.  Atmospheric model simulations support this relationship, showing a strong (~-0.8) anti-correlation between ensemble average Warm Pool and East African precipitation. A simple bivariate regression with Warm Pool SSTs suggests that we might expect a -1.5Z standardized March-June rainfall anomaly given the current observed SST conditions (-1.5Z = +0.5°C * -3 Z/°C + 0.05Z).

Figure 6. Top - October-November 2016 Warm Pool SST anomalies, derived using a 1981-2010 baseline. Bottom – Scatterplot showing October-November Warm Pool SST anomalies and standardized East African March-June rainfall anomalies.
Figure 6. Top – October-November 2016 Warm Pool SST anomalies, derived using a 1981-2010 baseline. Bottom – Scatterplot showing October-November Warm Pool SST anomalies and standardized East African March-June rainfall anomalies.

Figure 7 shows similar results but for the North Pacific region from Figure 3. We identify a strong negative relationship between North Pacific SST and East African long rains.  October-November North Pacific SSTs are very warm (~+0.9°C), similar to 1998, 2007, 2010, and 2011, seasons preceding below normal long rains (Figure 2). A simple bivariate regression prediction based on North Pacific SSTs would be: -1.4Z = +0.9°C * -1.8Z/°C + 0.2Z.

Figure 7. Top - October-November 2016 North Pacific SST anomalies, derived using a 1981-2010 baseline. Bottom – Scatterplot showing October-November North Pacific SST anomalies and standardized East African March-June rainfall anomalies.
Figure 7. Top – October-November 2016 North Pacific SST anomalies, derived using a 1981-2010 baseline. Bottom – Scatterplot showing October-November North Pacific SST anomalies and standardized East African March-June rainfall anomalies.

Figure 8 shows a similar scatterplot and time series for the eastern Pacific region identified in Figure 3. Here we find a modest predictive relationship (R2=0.34). East Pacific SSTs do a fairly good job of stratifying dry events. Just using East Pacific SSTs to predict MAMJ rains we would arrive at below normal conditions -0.7Z = 1.8Z/°C * -0.5°C *+ 0.2Z.

Figure 8. Top - October-November 2016 East Pacific SST anomalies, derived using a 1981-2010 baseline. Bottom – Scatterplot showing October-November East Pacific SST anomalies and standardized East African March-June rainfall anomalies.
Figure 8. Top – October-November 2016 East Pacific SST anomalies, derived using a 1981-2010 baseline. Bottom – Scatterplot showing October-November East Pacific SST anomalies and standardized East African March-June rainfall anomalies.

A difficult question to answer, because we have so few samples, is whether a very warm Western and Northern Pacific, acting alone, can produce substantial East African drying. Figure 9 chracterizes the current situation and identifies two modestly cool East Pacific analogs (2000 and 2008). This scatterplot shows the unique conditions we face at present. The x-axis represents the average Warm Pool/North Pacific SSTs. The y-axis shows our East Pacific SST anomalies. The average of our Western and Northern Pacific time series indicate the warmest conditions on record. The East Pacific shows modest La Niña-like cooling. Because of the non-stationarity of the climate system it is hard to find a large number of analogs. The two closest seasons with ~-0.5°C eastern Pacific SST appears to be October-November of 2000 and 2008, although 2008 is much closer. The right panels of Figure 9 show the longs rains for the following years (2001, 2009). These results could indicate a range of outcomes from slightly below normal to the worst on record (see Figure 2). We do not have to have a strong La Nina to have a severe long rains drought.

Additional evidence supporting the important role of Western and Northern Pacific SSTs comes from our statistical analyses (Figures 6-7), published forecast models (references cited above) and the dynamic response of climate models to Western and Northern Pacific forcing (i), West Pacific forcing (c), and the West Pacific Gradient (d), which is driven by both the western and eastern Pacific, and is currently very strong (Figure 5).

Figure 9. Assessing risks during modestly cool East Pacific seasons. Left - scatterplot of the average WP/NP SST anomalies and East Pacific Anomalies. Right - CHIRPS MAMJ anomalies for modestly cool East Pacific analog seasons.
Figure 9. Assessing risks during modestly cool East Pacific seasons. Left – scatterplot of the average WP/NP SST anomalies and East Pacific Anomalies. Right – CHIRPS MAMJ anomalies for modestly cool East Pacific analog seasons.

We conclude by combining all three predictors using a standard statistical technique – take-one-away cross-validated regression – to evaluate the robustness of our prediction scheme. Cross-validation is carried out by sequentially removing a year’s data, recalculating regression coefficients without this data, and then comparing the forecasted values with the real observations. These values are shown in the scatterplot in Figure 10. The cross-validated R2 is modest (0.36), but the model appears to do a good job of capturing most of the dry events in 1999, 2000, 2008, 2009, 2011 and 2012 but not 1984. Two false alarms appear in 1989 and 2001, although 2001 rainfall performance was dry in some current crisis areas (Figure 9).

Figure 10. Forecasts of March-June Eastern East African rains based on October-November SSTs. Green circles show cross-validated hindcasts. Red star denotes the 2017 forecast (a -1 standardized anomaly ± 1.2Z).
Figure 10. Forecasts of March-June Eastern East African rains based on October-November SSTs. Green circles show cross-validated hindcasts. Red star denotes the 2017 forecast (a -1 standardized anomaly ± 1.2Z).

While our regression-based forecast for 2017 MAMJ rains is very low (-1.5Z), we advocate a more modest outlook this far ahead of the long rains. We assume that MAMJ 2017 will be similar to the other 8 years for which our forecast was below normal. These analog years correspond to the dots to the left of the y-axis in Fig. 10.

Table 1. Forecast statistics for various sets of model years. Years
Table 1. Forecast statistics for various sets of model years.

This approach seems robust to different selections of years (Table 1).  In all cases we find a prediction for below normal MAMJ rainfall. Please also note that the strong negative western and northern slope coefficients remain fairly robust across all the year selections. Knowing the true value of these coefficients is difficult, but the data converges on a pessimistic outlook.  Table 1 also suggests that during neutral-Nina seasons East Africa’s sensitivity to Warm Pool and North Pacific SSTs increases.

Assuming 1989, 1999, 2000, 2001, 2008, 2009, 2011 and 2012 as analog years leads to a forecast of -1Z±1.2Z, with the range depicting 80% confidence intervals. This model gives a 75% chance of below normal rains and has an historic hit rate of predicting 6 out of 7 recent droughts (86% success rate). The red star in Figure 10 shows our 2017 forecast. Different selections of years were tested (Table 1), and all selections produced similar results: below normal longs rains are the most likely outcome, primarily due to negative teleconnections with western and northern Pacific SST. Our forecast range (-1Z±1.2Z) also fits with the outcomes spanned by our two modest-Nina analogs (2000/2001, 2008/2009, Figure 9).

Note that the closest analog (2008/2009) was a very poor rainy season, and that our WPG time series (Figure 5) and regression estimates all suggest SST conditions similar to those preceding some very poor rainy seasons (1999, 2000 and 2011).  In Figures 11 and 12 we highlight this similarity by showing 2016 and 2010 October-November SST and rainfall anomalies. 2016 SSTs are warmer in the western Pacific and northern Pacific than in 2010, but the eastern Pacific is warmer as well. Milder La Niña conditions in 2016 are accompanied by exceptionally warm conditions in the western and northern Pacific. The corresponding ‘Western V’ structure contrasting the western/northern Pacific with cool eastern Pacific SSTs tends to produce dry conditions in the eastern Horn (Figure 3). The observed 2016 SSTs (Figure 11) look a lot like the inverse of our correlation map (Figure 3), indicating a good chance for below normal rains.

Figure 11. Standardized October-November SSTs. A 1981-2010 baseline was used to derive anomalies.
Figure 11. Standardized October-November SSTs. A 1981-2010 baseline was used to derive anomalies.

October-November rainfall anomalies (Figure 12) also suggest conditions similar to, but not identical with, 2010. During October-November of 2010 La Niña-like conditions were better developed, with deeper rainfall reductions near the dateline and more enhanced precipitation over the warm pool. On the other hand, in 2016, rainfall deficits over the Indian Ocean, indicative of the strength of the overurning circulation, are actually larger than those present in 2010. 2016 conditions over the Warm Pool and eastern Pacific appear La Niña-like, with a vigorous overturning circulation. This circulation, however, seems weaker than in 2010, consistent with weaker 2016 La Nina conditions. Going forward, monitoring maps like Figure 11 and 12 will help us assess whether the climate is maintaining a dangerous state or transitioning to a more normal conditions.

Figure 12. October-November CHIRP anomalies.
Figure 12. October-November CHIRP anomalies.

Increased confidence in our outlook is provided by building a MAMJ rainfall composite using atmospheric global circulation model simulations for our analog years. When our statistical model predicted below normal rains based on October-November SSTs, the climate model simulations, driven with observed MAMJ SSTs the following year, indicated a strong overturning circulation with drying over the Horn of Africa (Figure 13). These results suggest that the observed overturning circulation (Figure 12) may persist into 2017. These simulations provide convergent support for our statistical analyses.

Figure 13. Standardized March-June precipitation from ECHAM5 simulations for analog years.
Figure 13. Standardized March-June precipitation from ECHAM5 simulations for analog years.

We can produce a spatial map of the expected drying by using the composite function in the GeoCLIM tool (Figure 14). This map is based on 1989, 1999, 2000, 2001, 2008, 2009, 2011 and 2012. Spatially, these results are broadly similar to rainfall estimates produced by the ECHAM5 model (Figure 13), which also shows below normal rainfall across much of eastern Africa for our analog years. Southern Ethiopia, Somalia and Central and Eastern Kenya appear likely to receive below normal rains. While anticipated conditions in northern Somalia are normal, this could be due to deficiencies in our CHIRPS data – this region has very few stations, and satellite-based rainfall estimates can have trouble seeing low stratus-related rainfall over the easternmost areas of East Africa. Low confidence in our data in this region tranlates into low confidence in our forecast for northern Somalia. On the other hand, this region can also be influenced by extra-tropical disturbances (see below). For Ethiopia, Kenya and Southern Somalia, we have good station coverage as well as high confidence in our satellite estimates.

Figure 14. March-June CHIRPS rainfall anomalies for selected analog seasons - 1989, 1999, 2000, 2001, 2008, 2009, 2011 and 2012.
Figure 14. March-June CHIRPS rainfall anomalies for selected analog seasons – 1989, 1999, 2000, 2001, 2008, 2009, 2011 and 2012.

This analog analysis indicates that the belg growing season in the eastern highlands of Ethiopia may be below normal, resulting in the fifth poor season in parts of this region, which had below normal belg and kiremt rains in 2015 and 2016. Crop growing conditions in central/eastern Kenya and southern Somalia may also be below normal. For many Eastern Kenya and Southern Somalia maize growing regions this would mean a third poor season, since MAMJ and October-November 2016 rains were low. Pastoral areas in these areas may receive little respite.

Summary: Ocean conditions appear similar to 2010/11 and 2008/09, and for our target region below normal rains in MAMJ appears to be the most likely outcome given October-November SST conditions. We are certainly facing an elevated chance of very poor rainfall in at least parts of our target region, though we find normal years within our span of analogs, and SST conditions may change between now and 2017. Spatially, many of the areas expected to receive poor 2017 March-June rains are already very dry (Figures 1 and 4), and the hydrologic impact of these consecutive droughts could be similar in magnitude to 2010/11. We will be able to provide more information as the season progresses. We may see the ‘Western V’ SST pattern cool while eastern Pacific SSTs warm. If not, a continued below normal outlook seems warranted. Continued monitoring of large scale Indo-Pacific SST and rainfall anomalies will help us guage the persistence of the current overturning circulation that produced an exceptionally dry short rains season.

Spatial and Seasonal Forecast Limitations

In closing, we would like to emphasize the spatial and seasonal limitations of this forecast, which has been targeted on Eastern Equatorial East Africa (Figure 1), where prior FEWS NET research has noted strong negative links between increased precipitation over the eastern Indian and Western Pacific Ocean during the March-June season. Deep convection over the Indo-Pacific warm pool almost always produces subsidence and drying over at least part of this focus region. This response, however, tends to be strongest near the equator, and our outlook should not be considered as a forecast for all or even most of the Greater Horn of Africa. Many different climate drivers play different roles in different locations, creating a complex climatic tableau.

In October-May, in the northern parts of the Horn, we know that Ethiopia, Sudan, Eritrea, Djibouti, Yemen, Saudi Arabia and other middle east countries, and northern Somali (Somali Land) are all significantly affected by the southward penetration of north-south or back-hanged mid-latitude troughs. These troughs often originate from polar low pressure cells and penetrate into the tropical regions, where they are associated with short-lived (few days to a week) rainfall events across northern Africa, the Mediterranean, the Middle East and other countries bordering or confined within the Great East Africa Rift Valley (north of the Equator). These wet anomalies are very common between October and May, and current ECMWF forecasts for December indicate substantial trough activity (Figure 15). Whenever these troughs approach or lie along the Rift Valley, the Sahara and Arabian High pressure system disintegrates and is partly pushed over the Arabian Sea, which usually leads to the formation of easterly moisture flows from the Arabian Sea and north Indian Ocean.  These troughs are typically well-established in the mid-troposphere (500 hPa) but sometimes deepens up to 200 hPa, depending on the westerly frontal systems from which they emerge. Some years with anomalously wet October-February conditions were aligned with times when more westerly fronts and the associated trough systems penetrated in to the tropics. The wet anomaly over Ethiopia between 23-30 November 2016 appears to be related to such a trough. The number of troughs so far developed and that will be developing in the coming days seem likely to be more frequent (Figure 15). This anomaly is expected to enhance unseasonal rains over northern sectors of GHA and the Middle East in months of December 2016 as well as January 2017.

We would like to emphasize that in addition to large scale global teleconnection systems associated with the El Niño/Southern Oscillation, West/North Pacific and Indian Ocean Dipole there are a number of other key regional factors that are just as important as ENSO in predicting rains during the October-May semi-dry seasons over the northern sectors of the GHA and the Middle East. These include the pressure gradient and sea surface temperature gradient between the Indian Ocean and Arabian Sea (including the land surface), the oscillatory nature of the North Atlantic Oscillation, the west-east pressure/sea surface temperature gradient over the subtropical Atlantic Ocean, the number and frequency of circumpolar troughs, the position and intensity of the semi-permanent high/low pressure systems (the Azores, Siberia, Sahara, Arabia highs) and middle and upper-tropospheric dynamics and the Madden-Julien Oscillation. Here, we do not take these complexities into account, so our forecast is only appropriate for the March-June over eastern equatorial areas of East Africa.

Figure 15. ECMWF forecast of 850 hPa temperature and 500 hPa geopotential heights. Trough analysis by Diriba Korecha.
Figure 15. ECMWF forecast of 850 hPa temperature and 500 hPa geopotential heights. Trough analysis by Diriba Korecha.

Acknowledgements:  We would like to thank James Verdin and Andrew Hoell for constructive and insightful discussion and analysis involving the current climate situation. We would also like to thank FEWS NET and the USGS for supporting the research that supports this assessment.

Leave a Reply

Your email address will not be published. Required fields are marked *