Why CHC is Confident the 2022 March-April-May Drought in Somalia, Ethiopia, and Kenya was the Worst on Record

Why CHC is Confident the 2022 March-April-May Drought in Somalia, Ethiopia, and Kenya was the Worst on Record

Chris Funk

Background: Unprecedented Drought and Food Insecurity

On June 9 2022, the Famine Early Warning Systems Network (FEWS NET), in partnership with FSNWG, ICPAC, WFP, FAO, and JRC, released an exceptional collaborative multi-agency alert describing the alarming combination of sequential droughts, agro-pastoral shocks, and food insecurity in Somalia, Ethiopia, and Kenya. The alert was unparalleled in terms of the breadth of the coverage and the level of partnership among the collaborators. These extraordinary efforts were motivated by one shared goal: participating organizations wanted to speak with one clear and coherent voice about the still-accelerating humanitarian tragedy evolving in the eastern Horn of Africa. Now, in early September, it is evident that crop harvests in Kenya, Somalia, and southern Ethiopia have been and will remain very poor, more than 9 million livestock have perished, water resources have become extremely scarce, and millions of people face crisis or worse levels of food insecurity. In Somalia, “agropastoral populations in Baidoa and Burhakaba districts and displaced populations in Baidoa town of Bay Region in Somalia are projected to face Famine (IPC Phase 5) between October and December 2022,” as described in a multi-partner technical release, joint FEWS NET-FSNAU alert, and an IPC Famine Review Committee (FRC) report. Looking forward, there is a broad consensus that a fifth dry season in October-November-December 2022 is likely to occur, followed by elevated chances of a sixth drought in March-April-May of 2023. Given these perilous conditions, it is important that the depth of the March-April-May 2022 drought be taken very seriously. To that end, this short post describes the data and methods used to produce Figure 1A and B from the June multi-agency alert (Figure 1A and B below). 

For 20 years, FEWS NET has been working to develop the best-available rainfall data sets for East Africa. These data sets have been designed for the explicit purpose of monitoring extreme droughts. Here, we briefly describe Figure 1, the supporting data, and explain the statistical analyses that supports the statement “We can say with a 95% degree of confidence that the 2022 March-April-May drought in Somalia, Ethiopia, and Kenya was the Worst on Record (i.e., in the last 73 years)”.

Figure 1A shows rainfall ranks derived from the Climate Hazards center InfraRed Precipitation with Stations (CHIRPS) data set. With over 2,800 citations, CHIRPS is widely used for drought monitoring and crop and hydrologic modeling. While CHIRPS has global coverage, its development was motivated by a need to monitor droughts in East Africa. CHIRPS combines three components: a high-resolution, satellite-enhanced climatology; satellite-based rainfall estimates; and a reasonably dense set of rainfall observations. This blog will focus on describing the latter two inputs as a means of describing temporal variations in the eastern Horn of Africa.

How CHC Develops CHIRPS Data Sets

CHIRPS combines a satellite-only product (CHIRP) with station data (S).

Long-term (1981 to present) continuous weather satellite observations rest at the core of the satellite-only CHIRP. These “geosynchronous” satellites reside over the equator, orbiting around the Earth at speeds that allow them to maintain a fixed longitude. They sustain a fixed location over our planet, observing thermal infrared temperatures at a high spatial resolution several times per hour. While they cannot directly see rain, these satellites are very good at observing clouds, especially clouds high in the atmosphere, which are much colder than the Earth’s surface. This is what these weather satellites were designed to do. The University of California, Santa Barbara’s Climate Hazards Center (CHC) uses these observations to estimate rainfall, and these estimates, without stations, have been shown to be very accurate in East Africa. There are two reasons for this. First, it is very easy to see when it is not cloudy, and if it is not cloudy, it cannot be raining. Second, when it does rain in Ethiopia, Kenya, and Somalia, these rains tend to fall from high towering clouds with very cold cloud tops, which makes them easy to see from space.

In very dry seasons, like March-April-May 2011 or 2022, weather satellites looked down from space, day after day, reporting back the absence of clouds, and hence, an absence of rain. While CHIRP has trouble estimating extreme rainfall amounts, it can see an absence of rain with considerable precision.

The next component of CHIRPS is a reasonably dense network of carefully quality-controlled rainfall station observations, and a very well-considered method for blending these observations with the satellite-only CHIRP estimates. In addition to all the globally available sources of station data, the CHC incorporates about 140 and 90 additional gauge observations into CHIRPS each month. These observations are graciously provided by the Ethiopian National Meteorology Agency and the FAO SWALIM program in Somalia. More detailed Ethiopian analyses are available here. After an automated screening, a group of “reality checkers” at the CHC performs a final visual evaluation that involves comparisons with neighboring stations and cross-checking with independent data sources. 

After screening, the monthly station totals are blended with CHIRP to produce a final version of CHIRPS. Because the CHIRP and station observations have long-term averages that are very similar to each other, discontinuities associated with changes in the composition of the gauge network are minimized. Furthermore, CHIRPS uses satellite rainfall observations to estimate a “decorrelation distance” that is used to limit the spatial influence of a station observation. The resulting final version of CHIRPS is consistent, high-resolution, rapidly updated, and well-suited for drought monitoring. While there is uncertainty in all rainfall estimates, especially at the pixel level, we should be confident in the rank map shown in Figure 1A. Many regions in East Africa experienced an extraordinarily dry March-April-May rainy season in 2022. Since June, the magnitude of these rainfall deficit shocks has been confirmed by the observation of time-lagged indicators including very poor harvests, livestock destruction, and extremely low water hole levels.

Why CHC Made the Centennial Trends Data Set

From a climate perspective, CHIRPS’s ~40-year period of records is relatively short. Because of this, CHC used station observations to create the longer 1900-2014 “Centennial Trends” gridded rainfall archive. The Centennial Trends archive uses the same long-term mean fields as CHIRPS, so the two products are relatively “interoperable”. The geostatistical interpolation process used for the Centennial Trends archive (kriging) supports the estimation of maps of standard errors, and these maps indicate reasonably good levels of skill, especially back through the 1950s when gauge observations in the Greater Horn of Africa were relatively plentiful. CHC’s Centennial Trends paper has a detailed discussion of the changing station distributions over time. Note that, as with CHIRPS, the Centennial Trends interpolation process is done using anomalies space, which diminishes discontinuities associated with shifting observational networks.

How CHC Combined CHIRPS and Centennial Trends Data

While all rainfall estimates are uncertain, spatial averaging should reduce these uncertainties, as pixel-by-pixel random errors tend to cancel each other out. Regional averages, such as those shown in Figure 1B, should therefore be a mainstay of effective drought monitoring. The 1981-2022 values in Figure 1B are based on the CHIRPS data, which incorporates numerous stations in the worst-affected areas of Ethiopia and Somalia. According to this data set, 2022 was substantially drier than recent severe droughts in 2000, 2009, and 2011. The 1984 value seems the closest, but 2022 was drier.

The 1950-2014 values in Figure 1B are based on Centennial Trends data, transformed with a regression to match the CHIRPS time-series. The R2 on this regression is 0.88, and the associated standard error was 15 mm. Figure 2 shows the regression-adjusted Centennial Trends and CHIRPS regional averages. In the combined data set, the 2022 value is the lowest on record. In the 1950-1980 period, the lowest value of 162 mm was recorded in 1965. Using our regression standard error, we can say with 95% confidence that the 1965 value was greater than 132 mm, the observed 2022 CHIRPS value. The 2022 value was also drier than Centennial Trends estimates in 1925 and 1933. Our confidence in Centennial Trends during this period is much lower, however, due to substantial reductions in the number of stations, especially in Ethiopia and Somalia. Figure 1B, therefore, begins in 1950. 

Considering the history and availability of Centennial Trends data, CHC is confident that 2022 was very likely the “worst drought on record, since reliable reporting began in the 1950s”. It should be noted, however, as was discussed in the June alert, that in addition to being exceptionally intense, the March-April-May 2022 drought was also more extensive than the 1984 drought. Furthermore, this massive hydrologic shock followed three antecedent dry seasons, and was accompanied by much warmer air temperatures than those present in the 1980s.  

Figure1 . A. MAM 2022 rainfall ranks indicate that most of the Horn of Africa received extremely low rainfall amounts, based on 42 years of CHIRPS rainfall. The purple polygon in Panel A denotes the area of exceptional dryness in MAM 2022. B. Time-series of dry region MAM CHIRPS/Centennial Trends rainfall, expressed as Standardized Precipitation Index (SPI) values). Also noted with yellow circles are strong negative Western V Gradient (WVG) seasons.
Figure 2. Regionally averaged regression-adjusted March-April-May Centennial Trends and CHIRPS rainfall data.