Intensity and spatial extent of droughts in southern Africa

Authors


Abstract

[1] The standardized precipitation index allows for monitoring the intensity and spatial extent of droughts at different time scales. We used it to do a retrospective analysis of the spatial extent of droughts in Southern Africa (South of 10°S), from 1901 to 1999. Accordingly, the 8 most severe droughts at the 6-month scale (October–April) for the summer rainfall region of Southern Africa ended in 1916, 1924, 1933, 1949, 1970, 1983, 1992 and 1995. At the 2-year scale, they ended in 1906, 1933, 1983, 1984, 1992, 1993, 1995 and 1996. Areas affected by those droughts ranged from 3.4 to 2 106 km2. Eight of those 12 years are El Niño years. Preliminary data indicates that 2001/2002, 2002/2003 and 2003/2004 experienced severe droughts at a number of scales. This confirms the increase in the spatial extent of drought in Southern Africa since the 1970's due to stronger ENSO Southern African rainfall relationship

1. Introduction

[2] Drought is a regular and recurrent feature of Southern African climate. A drought is a shortage of precipitation over an extended period. The impact on society depends on its intensity and also on its duration. Over Southern Africa, the recent period (since 1970) is characterised by strong interannual rainfall variability. In particular, countries experienced more intense and more widespread droughts [Richard et al., 2001]. A common time-scale for agricultural droughts is the season when deficiency in precipitation results in damage to crop. Hydrological drought is associated with precipitation shortage on a longer time scale (6 months to 2 years or more). It takes longer for precipitation shortage to become evident in stream flow, groundwater and dam levels. It is therefore useful to define a drought index that will represent different time scales. Using that index, we have made an atlas of Southern African droughts at different time scales with the 0.5 × 0.5 degree 1901–1999 CRU2 precipitation dataset (T. D. Mitchell et al., A comprehensive set of high-resolution grids of monthly climate for Europe and the globe: The observed record (1901–2000) and 16 scenarios (2001–2100), submitted to Journal of Climate, 2005) and the 0.1 × 0.1 1961–1995 CHARM precipitation dataset [Funk et al., 2003]. This paper summarizes some of the findings. In particular it gives estimates of areas affected by the major droughts at different time scales, and, it shows a clear recent increase in drought, especially evident at the two-year scale. We also use the 1986–2004 GPCC dataset [Rudolf et al., 1994] to extend our analysis to 2004 and to assess the recent severe droughts that occurred at different time scales from 2001 to 2004.

2. Standardized Precipitation Index

[3] Hayes et al. [1999] used the standardized precipitation index (SPI) to monitor the 1996 drought in the United States of America. They show how the SPI can be used operationally to detect the start of a drought, its spatial extent and temporal progression. Although it is quite a recent index, the SPI was used in Turkey [Komuscu, 1999], Argentina [Seiler et al., 2002], Canada [Anctil et al., 2002], Spain [Lana et al., 2003], Korea [Min et al., 2003], Hungary [Domonkos, 2003], China [Wu et al., 2001], East Africa [Ntale and Gan, 2003, 2004] and Europe [Lloyd-Hughes and Saunders, 2002] for real time monitoring or retrospective analysis of droughts. T. B. McKee et al. (unpublished data, 1993) from the Colorado Climate Center formulated the SPI in 1993. The SPI allocates a single numeric value to the precipitation (−3 to 3), which can be compared across regions with different climates.

[4] The SPI was designed to show that it is possible to simultaneously experience wet conditions on one or more time scales, and dry conditions at other time scales. The SPI is based on the probability of precipitation for a given time period. Since precipitation is not normally distributed, a transformation is first applied so that the transformed precipitation values follow a normal distribution. The SPI determines the probability occurrence of dry or wet events at different time scales. The climatological rainfall distribution at any station or ensemble of nearby stations can be represented by a gamma distribution. The gamma distribution can be described by its frequency or probability distribution. The precipitation rate is fitted to a gamma distribution for different time scales. The resulting function can be used to find the cumulative probability of a rainfall event for a station for a given month of the dataset, at different time scales of interest. Details of SPI algorithm can be found in the work by Hayes et al. [1999]. The values of SPI are in standard deviations. A SPI of 2 or more happens about 2.3% of the time and a normal condition (SPI between 1 and −1) happens 68.2% of the time. This allows the establishment of classification values for SPI (Table 1). Table 1 shows that the SPI is symmetrical for the occurrence of wet and dry events.

Table 1. SPI Classification [Hayes et al., 1999]
SPI Value OccurrenceDrought category% Occurrence
> 2.00Extremely wet2.3%
1.5 to 1.99Very wet4.4%
1. to 1.49Moderately wet9.2%
−0.99 to 0.99Near normal68.2%
−1.00 to −1.49Moderately dry9.2%
−1.50 to −1.99Severely dry4.4%
<−2.00Extremely dry2.3%

[5] The SPI has been favourably evaluated and compared with other indices [Keyantash and Dracup, 2002] and is now integrated in the set of indices used by the Drought Monitor in the USA [Svoboda et al., 2002]. The SPI is consistent with regard to the spatial distribution of rainfall that occurs with great variability in Southern Africa due to geographical location, orography and the influence of the oceans. Furthermore, the onset of the rainy season, timing of maximum rainfall or mean annual precipitation vary greatly in Southern Africa even within a single country. Hayes et al. [1999] have shown that for some regions a good rainfall for one month can create the impression that the drought is over but until the SPIs are not above a certain value (typically −1) at all scales a drought will still affect a region one way or another. However during the dry season, large negative or positive SPIs may be associated with precipitation totals not very different from the mean. Because this is a period with little rain, these mean totals will be small and relatively small deviations on either side of the mean could have large negative or positive SPI. Nevertheless, it is important to have a drought index for longer time scales. A normal 3-month period could occur in the middle of a longer-term drought that would only be visible at longer time scales.

3. Spatial Extent of Droughts in Southern Africa Since 1901

[6] We have calculated SPI for precipitation totals ranging between 3 months and 2 years at the end of each month of the 1901–1999 CRU2 precipitation. The 3-month SPI can be used to characterise a seasonal drought index well suited to describe the beginning, the heart or the end of the rainy season. The total rainy season can be described using the 6-month index and 24-month SPI can be used for long-term drought. The X-month SPI (X = 3, 6, 12 or 24) is used to compare the precipitation over a specific X-month period with the precipitation totals from the same period for all the years of the dataset. We selected area with SPI < −1 for Southern Africa, south of 10°S. To avoid misleading values of SPI with used grid cell with mean summer (October to April) rainfall >10 mm. This leaves us with a total surface of 6.65 106 km2. The area taken out corresponds mostly to the Namib Desert and western coast of South Africa. Table 2 shows the areas affected by drought in 106 km2, the percentage of the affected areas from our domain study and the El Niño Southern Oscillation (ENSO) Index for the eight worst droughts at the end of April at the 6-month and 2-year scales. Incidentally, 8 of those 12 dry years were ENSO years [Trenberth, 1997; Compagnucci et al., 2002]. El Nino is usually linked to drought in Southern Africa and La Niña with above normal rainfall [Lindesay and Vogel, 1990; Richard et al., 2000]. At the 2-year scale, the most severe droughts have occurred during the last 20 years. The strong intensity and extent of observed droughts since 1970 is associated with a reinforced influence of El Nino in the context of a warmer tropical and subtropical Indian Ocean [Richard et al., 2000; Goddard and Graham, 1999]. Figure 1 shows the dry area (SPI < −1) at the 6-month scale and Figure 2 shows the dry area at the 2-year scale at the end of April for our domain study from 1901 to 1999 with a 9-year window running mean that indicates the mean dry area per decade. The average dry area is about 1 106 km2 per year at the 6-month and 2-year scale, about 17% of our domain study. There is decadal variability in the mean dry area per decade at the 6-month scale, a feature of Southern African rainfall variability [Reason and Rouault, 2002] that makes the detection of trend difficult. Nevertheless there is a clear increase at the 2-year scale since the mid 70's. Precipitation shortage in the austral summer season of 2002, 2003 and 2004 seems to confirm that behaviour. Southern Africa experienced drought at different time scale in 2002, 2003 and 2004 with major societal implications. Dry conditions during the first three months of 2002 in Zimbabwe, Lesotho, South Africa, Swaziland, southern Mozambique, Zambia, and Botswana lead to low crop production and were partly responsible for the threat to life of 14.4 million people (International Research Institute for Climate Prediction (IRI), unpublished data, 2002). South Africa, Zimbabwe, Mozambique, and Zambia were affected by a drought in 2002/2003 with major impact on agriculture (IRI, unpublished data, 2003). In 2004, the first half of the October–March rainy season produced below-normal precipitation South Africa, Zimbabwe, Mozambique, Lesotho and Swaziland (IRI, unpublished data, 2004). Precipitation deficiency at different time scale from January 1986 to April 2004 were inferred from rain gauge observations data of the Global Precipitation Climatology Center (GPCC) a monthly analysis of surface precipitation at 1° latitude × 1° longitude resolution. Normalised anomalies (anomalies divided by standard deviation) and SPI values were calculated at each grid point at different time scale. This allowed us to do a preliminary assessment of the 2001–2004 droughts. Because the CRU encompasses 100 years and GPCC 19 years we cannot directly compare the area affected by drought with the SPI. Moreover, there are some discrepancies between the two dataset especially in 97/98 and 98/99.

Figure 1.

Area (in 106 km2) affected by drought (SPI < −1) at the 6-month scale at the end of April of each year since 1901 (*) with a 9 years mean running windows (dashed line) and the overall 1901–1999 mean dry area (solid line).

Figure 2.

Area (in 106 km2) affected by drought (SPI < −1) at the 2-year scale at the end of April of each year since 1903 (*) with a 9 years mean running windows (dashed line) and the overall 1903–1999 mean dry area (solid line).

Table 2. Area and Percentage Area of Southern Africa South of 10 S for the Eight Worst Droughts (SPI < −1) at the End of April at the 6-Month Scale (Left) and 2-Year Scale (Right) for the 1901–1999 Period With ENSO Classificationa
6-Months Scale2-Years Scale
YearDry Area, 106 km2%ENSOYearDry Area, 106 km2%ENSO
  • a

    The mean dry area per year is about 106 km2.

19162.5338Neutral19061.9830El Nino
19242.4236El Nino19332.7041El Nino
19332.9244El Nino19832.6840El Nino
19492.538Neutral19841.9530Neutral
19702.9244El Nino19922.6940El Nino
19833.0947El Nino19932.7542El Nino
19923.3851El Nino19953.0648El Nino
19953.1147El Nino19962.2133La Nina

[7] According to GPCC data, 2002 was dryer than normal during the heart of the rainy season, part of southern Africa experienced a serious drought at all time scales during the 2003 rainy season. The beginning of the rainy season was very dry in 2004. Incidentally, ENSO happened in 2002/2003. GPCC data indicates that total rainfall at the 2-year scale were well below normal from December 2002 to April 2004 (in spite of above normal rainfall in February to April 2004) although the area impacted was not as severely extended as for 1992 and 1993. The dry area with 2 year SPI < −1 at the end of April 2003 and April 2004 were respectively 60% and 50% of the same quantity for 1992. This would still place those years above the maximum value of the nine year running window mean (red line in Figure 1) using the 1992 value obtained with the CRU dataset and this extends in time the trend shown in Figure 1. It is important to note that the bad start of the 2003/2004 rainy season (third lowest total dry area for October to December 2003 of the 1986–2004 GPCC dataset) that followed the bad 2002/2003 rainy season (fourth lowest total dry area for November 2003 to April 2004) lead to the third worst total dry area at the 2-year scale of December 2003 since 1986.

4. Conclusions

[8] Due to the complexity of the rainfall regime in Southern Africa, it is difficult to monitor drought with a chart showing percentage from normal or anomaly in total rainfall. The SPI is a simple index based on rainfall only that can measure drought at different time scales with spatial homogeneity. This has enabled us to observe some spatial characteristics of the droughts. Drought occurs often in Southern Africa in all climatic areas at all times of the year with different intensity, spatial extent and duration. Each drought has a different signature. There is a substantial increase in drought at the 2-year scale since the 70's. There is also considerable interdecadal variability in the spatial extent of drought since the beginning of the century. According to the CRU dataset the six strongest droughts at the 2-year scale have happened during the last two decades. Most of the severe droughts are associated with ENSO. Preliminary data indicates the 2002/2003 and 2003/2004 rainy seasons experienced a serious drought at the two year time scale mostly due to below normal rainfall in 2002, severe drought in 2002/2003 and a bad start of the rainy season in 2003/2004.

Acknowledgments

[9] Funding for this study came from the Water Research Commission, the National Research Foundation and the Centre National de Recherche Scientifique. This is a contribution to the South Africa/France co-operation program PICS. CRU, GPCP and GPCC have produced the data analysed here. Special thanks to C. Funk and G. Husak for providing assistance related to the CHARM dataset. M. Hayes provided the FORTRAN program used to calculate the SPI and gave us great assistance.

Ancillary