SEARCH

SEARCH BY CITATION

Keywords:

  • index;
  • CMI;
  • SPI;
  • EDI;
  • Mali;
  • drought;
  • vulnerability;
  • Bani;
  • Niger basin

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study area and data
  5. 3. An overview of water stress using the CMI
  6. 4. Drought study using the SPI and the EDI
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgements
  10. References

As in most parts of Africa, the Bani basin (Niger River, Mali) is threatened by climate changes. This study focuses on drought hazards and its indicators as derived from the climatic moisture index using the monthly data for the period (1963–2000) in order to examine the Bani basin's vulnerability to water stress. In this study, water deficit (drought) was logically addressed using the standardized precipitation index on a 10-day time step. The inadequacies in the standardized precipitation index while estimating water deficit within the basin based on technical issues result in the usage of the effective drought index (EDI). This robust index is specially designed for daily data computation and uses daily data efficiently while estimating water deficit. Three regions were identified to be threatened by droughts (length, number, and evolution in time) within the basin, using EDI as a tool of investigation. It also shows that droughts seem to be less frequent but of a longer duration due to the increasing trend of total dry days over the study period. Copyright © 2009 Royal Meteorological Society


1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study area and data
  5. 3. An overview of water stress using the CMI
  6. 4. Drought study using the SPI and the EDI
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgements
  10. References

Recent studies have shown that climate change will have a strong impact on Africa, especially in the agro-pastoralist regions (IPCC, 2007) such as the Bani basin, which is located within the upper Niger River basin. In an attempt to address the basin's vulnerability to climate change, the RESSAC program (surface water resources vulnerability to mid-term climatic and anthropogenic evolutions in Sahelian regions) was initiated by the French Institute for Development (IRD) and funded by the French National Agency for Research (ANR).

Vulnerability is a very wide concept, with several definitions (Füssel and Klein, 2006), which demand several dataset if satisfactory assessment of a region's vulnerability to climate change is to be attained. This is evident in the water poverty index computation (and its evolution: climate vulnerability index) by Sullivan and Meigh (2005).

This study therefore entails the assessment of drought hazards, which constitute only a fraction to the understanding of the vulnerability concept, although all aspects of vulnerability to drought are not assessed due to inadequate information on the adaptation capacity (water supply system, access to irrigation, etc.) within the region. O'Brien et al. (2007) and Downing and Patwardhan's (2005) provide more useful information on vulnerability to climate change.

In the study, no attempt was made to redefine drought, but the concept remained within several drought definitions (Traore et al., 2000). Also, water stress and drought assessment within the Bani basin was carried out using vital drought occurrence indicators as, with standardized indicators, estimation of drought indicators can easily be computed over space and time (Eriksen and Kelly, 2006). Such indicators remain useful tools for stakeholders (Feitelson and Chenoweth, 2002) and for future vulnerability estimation.

This study therefore aims at providing useful drought indicators to researchers working on climate change and agricultural related issues. Eriksen and Kelly (2006) provide details on the usefulness of these indicators.

In addition, in an attempt to study the distribution and evolution of droughts within the Bani basin, an overview of drought analysis was carried out using the climatic moisture index (CMI), which was computed from the monthly data. The standardized precipitation index (SPI) was also computed on a 10-day time step using daily rainfall data for more detailed information on drought occurrence and magnitude. The inadequacies of the SPI and the need for detailed information on drought intensity and length, and classification of threatened regions, led to the estimation of the basin's effective drought index (EDI).

2. Study area and data

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study area and data
  5. 3. An overview of water stress using the CMI
  6. 4. Drought study using the SPI and the EDI
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgements
  10. References

The Bani basin is located within the upper part of the Niger River (Figure 1); the larger portion of the basin is located within Mali and extends to Burkina Faso and Ivory Coast. The surface of the basin is about 120 000 km2 (when considering an outlet located at Beneny Kegny in Mali).

thumbnail image

Figure 1. General situation of the Bani watershed. The outlet (square) in this study is located at Beneny Kegny

Download figure to PowerPoint

Two data sets were available for usage within the basin. The first is derived from the CRU (Climatic Research Unit of East Anglia University, UK) monthly data at a 0.5 square degree resolution, which are entered into a hydrological model (Paturel et al., 2003a; Dezetter et al., 2008) at the basin scale to calculate the monthly rainfall and potential evapotranspiration (PET) times series for the basin for the years 1940–1995 and 1940–1999 respectively.

The second dataset is a mix of two different daily rainfall databases; although such practice ought to be avoided, the need to reach an acceptable spatial resolution justifies our choice. The first database comes from the French Institute for Development (IRD) in Bamako (Mali) and the second from the HSM Laboratory in Montpellier (France), namely from the SIEREM database (Boyer et al., 2006; Rouche et al., 2008).

Fifty-one stations were selected (Figure 2) on the basis of two criteria: time series must extend over the 1963–2000 period and have less than 10% (Romero et al., 1998) record gaps. Deviations from this rule were the three stations namely Filamana, Zangasso, and Konseguela, with less than 15% of record gaps (Figure 3).

thumbnail image

Figure 2. The 51 selected stations. They are clustered in five groups following the global shape of annual isohyets (average on years 1963/2000)

Download figure to PowerPoint

thumbnail image

Figure 3. Percentage of daily data available per station from 1963 to 2000. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

The spatial resolution adopted for the classification of the regions provides an optimal result within the basin, except for a region without any information represented with dashes on maps.

Two different methods were used to fill in the missing data in the daily rainfall time series. The first method uses a classical multi correlation using the closest stations within a homogeneous region. The homogeneous regions were defined on the basis of Soumaguel's recommendations (Soumaguel, 1996) following the general shape of isohyets (Figure 2). But R2 coefficients are frequently very low, which means that this method is frequently inaccurate. The second method uses a two-step approach: first, the same multi-correlation method is performed as before but only when the R2 coefficient is greater than 0.60. If not, we used a simple statistical approach. Daily rainfall values were, for each station and for each month, classified into categories (e.g. [2; 4 mm]) with a given probability of occurrence. Then, a uniform distribution of values between 0 and 1 was simulated. One value was attributed to each daily gap and was considered as the probability of occurrence of the category formerly created. For example, let us assume a gap in record whose corresponding value from the uniform distribution is 0.4. Then we look for the interval of rainfall with the same probability of occurrence, e.g. [2; 4 mm]. Finally, the missing rainfall value is 3 mm, the central value.

To assess which method is the best, monthly values were compared with daily values added up over a month. The first method always underestimates monthly values, although the second one performs well. Thus, the study proceeds with data filled owing to method number 2.

3. An overview of water stress using the CMI

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study area and data
  5. 3. An overview of water stress using the CMI
  6. 4. Drought study using the SPI and the EDI
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgements
  10. References

3.1. Definition

This index is expressed as the ratio of the annual precipitation to PET, with ranges between − 1 and 1. The index evaluates water scarcity with the usage of some data (Vörösmarty et al., 2005) and is defined as follows:

  • equation image

According to Vörösmarty (2005), there is a classification link between CMI values and climatic conditions (CMI < − 0.6: arid. − 0.6 < CMI < 0: semi-arid. CMI > 0: Humid)

3.2. Results

Maps depicting the rainfall patterns and trends for the periods 1940/1970 and 1971/1995 are drawn, as several works have given 1970 as the beginning of the rainfall deficit in West Africa on average (Mahe et al., 2001; L'Hote et al, 2002; Paturel et al., 2003b), as shown in the rainfall dataset plot in Figure 4, with the rupture in 1970 according to the different detection tests available in the Khronostat software (Lubes-Niel et al., 1998). The rainfall and runoff changes in 1970 are very well marked over the Bani River (Mahe et al., 2000), and on a larger scale for other rivers of West Africa (Mahe and Olivry, 1999) using the same software. It is also remarkable that for the Bani River the lowest value of CMI is recorded in 1993, and not in 1983, which has recorded the lowest rainfall and runoff in West Africa over the twentieth century (Mahe and Olivry, 1999).

thumbnail image

Figure 4. Evolution of CMI from 1940 to 1995 on a southern mesh of the basin

Download figure to PowerPoint

CMI values are represented in Figure 5 for the periods 1940/1970 and 1971/1995 together with their variability coefficients (CV) within the Bani basin, i.e. the ratio of the standard deviation to the average (of CMI values for a given period): CV = σ/mean.

thumbnail image

Figure 5. Average CMI computed for years (a) 1940/1970 and (b) 1971/1995; (c) and (d) show the CV (CV = Sdv/average) of each period. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

3.3. Discussion of CMI results

The general CMI pattern is similar to that of the annual rainfall (Figure 2), with a deformation along the central eastern area. It is also worth noting that the CMI decreases in the whole basin during the period 1971/1995 with regard to the average during the years 1940/1970. The arid zone that is located on the extreme northern part has extended greatly during the 1971/1995 period. A fraction of the humid region in the south is now semi-arid, while the wet region has disappeared.

The CVs are higher during the period 1940/1970 than during the period 1971/1995; this might have resulted from higher CMI values for the 1940/1970 period than those for 1971/1995, as noted before. Coefficient values greater than 1 might have resulted from error in data processing, as such values are not recorded by Vörösmarty (2005). Moreover, CV values here are negative almost everywhere because the CMI mean is negative.

According to CMI and CV values, we can conclude that the climate is becoming drier since 1970 but, surprisingly, also more stable on the whole basin. The southern part of the basin is also always wetter than that in the northern part but more variable. Hence, during some years, CMI values in the South can be very low showing the possibility of droughts. Agricultural vulnerability of the southern part of the basin may therefore be high. A final assessment on this point is not possible considering CMI's coarse temporal resolution and lack of data after 1995.

4. Drought study using the SPI and the EDI

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study area and data
  5. 3. An overview of water stress using the CMI
  6. 4. Drought study using the SPI and the EDI
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgements
  10. References

A review of drought analysis indices is given in Ntale and Gan (2003); Smakhtin and Hugues (2007), where the SPI that was used in this study is described (McKee et al., 1993). This choice is due to the availability of precipitation data and the non-availability of temperature and PET data for the selected stations within the region on a daily time scale. This indicator has also been frequently used in different parts of the world (Wu et al., 2001; Giddings et al., 2005; Sönmez et al., 2005) with better spatial representation than any other common indicator, for example, the Palmer drought severity index (Lloyd-Hughes and Saunders, 2002). It is also very practical because of its normalization as drought occurrence thus being associated to a return period.

However, as noted by Lloyd-Hughes and Saunders (2002), usage of SPI at short time scales (1, 2, or 3 months) within regions being characterized by low seasonal precipitation might result in the over-estimation or under-estimation of the positive or negative SPI values as seen in this study.

In this study, SPI was carried out at a 10-day short time step within a region and was characterized by high seasonal precipitation (approximately from June to September). In order to avoid the seasonality problem being associated with SPI quoted before, the analyses were carried out during the months of July to September, which constitute the core of the rainy and agricultural growing seasons.

The choice of a 10-day time step is due to the need to provide useful information about drought that can adequately support future agronomical use.

4.1. SPI definition

4.1.1. General

SPI is based on fitting a density function (and its cumulative probability distribution) with the averaged precipitation data, which is usually carried out using Gamma distribution (Guttman, 1999); the average can be calculated at different timescales. The computed distribution is later transformed into normal distribution as illustrated in Figure 6. The SPI is finally computed with zero being the centre value. The classification scheme given in Table I (Lloyd-Hughes and Saunders, 2002) is a reflection of the computed values.

thumbnail image

Figure 6. Example of an equiprobability transformation from a fitted gamma distribution to the standard normal distribution. Data are for the 3-month (DJF) average precipitation over the southeast of England (Lloyd-Hughes and Saunders, 2002)

Download figure to PowerPoint

Table I. Classification of events, according to the SPI (Lloyd- Hughes and Saunders, 2002)
SPI valueCategoryProbability (%)
2.00 or moreExtremely wet2.3
1.5 to 1.99Severely wet4.4
1.00 to 1.49Moderately wet9.2
0 to 0.99Mildly wet34.1
0 to − 0.99Mild drought34.1
− 1.00 to − 1.49Moderate drought9.2
− 1.50 to − 1.99Severe drought4.4
− 2 or lessExtreme drought2.3
4.1.2. SPI estimation

The SPI estimation involves two different steps: the first is the calculation of a 10-day average precipitation on a daily moving window. These decadal values are then used to calculate the associated density function. In the second step, we search for the fitness of this density function to a gamma distribution. The fitness was carried out using Anderson–Darling fit test, which seems to be more robust than the Chi2 and Kolmogorov–Smirnov tests (Meylan and Musy, 1996; Hamed and Rao, 2000). An R software package nsRFA was used for the quality of the fit tests (R Development Core Team, 2005).

The test was carried out for 51 selected stations using gamma distribution. At 90 and 95% confidence levels, 50 and 49 selected stations were rejected respectively.

Hence, for a better representation of the dataset, a systematic selection of the 10-day average values was carried out, on 1 day every 5 days, i.e. Days 5, 10, 15, 20, 25, and 30 for each month, instead of 10-day values everyday (i.e. days 1, 2, 3, …, 30). With this new approach, only three stations are rejected at the 95% confidence level. Although this approach provides a better data representation, the results are not as precise as working with the mean daily dataset. For example, different drought durations could be computed using the two methods. Therefore, there is a need to estimate the efficiency of this method.

4.1.3. Assessment of precision after degradation

Assessment of the efficiency of the systematic selection of 1 day every 5 days is accompanied by a comparative analysis of computed parameters such as drought duration, between the ‘normal’ and the ‘degraded’ methods for selected stations with Kouto station, the only station that accepted the use of the first approach (mean 10-day values computed everyday). It should be noted that this study focuses on the mean drought duration, maximum drought duration (MDD), number of droughts, and precipitation threshold (i.e. rainfall value for a decade to be considered in ‘drought’).

The drought duration is the number of consecutive days with a negative SPI value: the number of droughts is different from the number of dry days. For example, three consecutive days having a negative SPI value only translate to one drought event. A comparison between ‘degraded’ and ‘original’ SPI-10d values using R software package lmomco (functions pargam and cdfgam) and package stats (function qnorm) (R Development Core Team, 2005) is presented in Table II.

Table II. Comparison between ‘original’ and ‘degraded’ SPI using four parameters
 Degraded SPI-10dOriginal SPI-10dRelative difference (%)
  1. ‘Degraded’ is referring to the attempt described. Relative difference is [(degraded–original)/original] × 100.

Mean drought duration (days)13.88.265
Maximum drought duration (days)55.048.014.6
Mean number of drought3.25.5− 42.6
Précipitation threshold (mm)6.696.832

From the results, it is evident that the method fails to adequately represent the drought incidence as the mean drought duration of the ‘original’ SPI is, to a large extent, smaller than the modified one, and the mean drought occurrence is larger than the original SPI. There is a need therefore to increase the sampling size through selection of more stations, apart from Kouto (at least one for each of the five homogeneous regions) if a ‘correction parameter’ that adequately supports classical SPI is to be attained.

Meanwhile, the comparative analysis of the MDD and precipitation threshold between the original and degraded version provides a good correlation.

4.1.4. Results

The MDD and precipitation thresholds indices, the two parameters that remain the focus of this study as generated from the ‘degraded’ SPI-10d are shown in Figure 7 and Figure 8, respectively.

thumbnail image

Figure 7. MDD for all kinds of droughts (i.e. SPI-10d < 0) (a) and for extreme + severe droughts (i.e. SPI-10d < − 1.5) (b). Duration is the number of days with consecutive SPI negative values. Period studied is from 1963 to 2000, considering only July to September

Download figure to PowerPoint

thumbnail image

Figure 8. Precipitation thresholds of the category (a) mild droughts (SPI-10d < 0) and (b) extreme droughts (SPI-10d < − 2), computed owing to the SPI-10d from 1963 to 2000 (July to September). This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

4.1.4.1. Maximum drought duration

The MDD is shown in Figure 7(a) and (b) for drought magnitudes of SPI-10d < 0 and extreme + severe droughts (SPI-10d < − 1.5).

Regions with highest SPI < 0 drought duration are not the same as those for severe + extreme droughts (SPI < − 1.5) and for SPI < 0 droughts. Figure 7(a) depicts high SPI < 0 drought duration within the basin's eastern border and none in the northern region, although Figure 7(b) shows higher values in the north.

This shows that examination of the different drought occurrences apart from the SPI < 0 droughts is necessary.

4.1.4.2. Precipitation threshold

The precipitation threshold parameter is very useful due to its practical application while studying drought events and while addressing issues relating to irrigation water management. Because of this tool, it is very easy to know (for stakeholders, water managers, etc.) when a drought situation occurs through the 10-day average. The threshold values for droughts and extreme droughts are shown in Figure 8(a) and (b), respectively).

The structure of the precipitation threshold map is similar to annual isohyets, especially for the extreme drought values (Figure 8(b)), where the basin area is clearly divided following a northwest to southeast line. Meanwhile, a deformation, similar to the CMI formation, is noticed towards the eastern portion of the basin.

4.2. Effective drought index

4.2.1. Description

As shown earlier, the SPI does not provide all the information needed on drought occurrence within the Bani basin. This calls for the use of a relatively new index, namely the effective drought index (EDI). This index has been especially developed by Byun and Wilhite (1999) for daily rainfall data computation and analysis.

It should be noted that EDI is not frequently used for drought analysis and is more difficult to understand and calculate as compared to SPI. It has been proved to be more robust than the SPI (Morid et al., 2006) while analysing drought occurrence.

The EDI is based on the concept of effective precipitation (EP), which represents the ‘daily water resources depletion’. The index is (Byun and Wilhite, 1999) quantitatively expressed as

  • equation image

where i is the duration of summation (DS), Pm is the precipitation of m days before, and EPi is the EP for a day considering i days before.

The summation duration can have different values like 365 days, 15 days, etc. In this study, the same approach as that with the SPI-10d was adopted, i.e. the EP calculation was based on the 10 days before the D day. The 10-day window is the dummy DS value that is used to compute the final EDI.

The Byun and Wilhite (1999) paper provides useful information on EDI, which is expressed as

  • equation image

where MEP is the mean EP, j is the number of days over which precipitation deficit is accumulated, ST is the standard deviation, and PRN is Precipitation needed for a Return to Normal, i.e. for ‘the recovery from the accumulated deficit since the beginning of a drought’ (Morid et al., 2006).

It should be noted that EDI has similar classification, very close to the SPI code (Morid et al., 2006) with modification for the creation of ‘mild drought’ class as shown in Table III.

Table III. Classification of droughts according to the EDI
EDIDescription
− 0.69 to 0Mild drought
− 0.7 to − 1.49Moderate drought
− 1.5 to − 2.49Severe drought
< − 2.5Extreme drought

After this calculation step, which involves the use of the R script, the following parameters, namely the mean drought duration, MDD, and number of droughts, were estimated. A drought duration is defined as the number of consecutive days with EDI < 0 (or EDI < − 1.5 for extreme and severe droughts). With these three parameters, it is interesting to study the spatial drought distribution for the period 1963/2000 and drought temporal evolution between 1963/1970 and 1971/2000.

Even though it should have been more precise to use the same length of years before and after 1970, this was not possible considering the very limited number of stations with such a long dataset.

4.2.2. Results from the EDI
4.2.2.1. Drought duration

Figures 9 and 10 depict the varied droughts (EDI < 0) and extreme + severe droughts (EDI < − 1.5) with their spatial distribution of MDD, mean drought duration, and its variation, respectively.

thumbnail image

Figure 9. Study of droughts duration with the EDI, i.e. consecutive days with negative EDI value. This is for all kind of droughts (EDI < 0) and shows for years 1963/2000 (July to September) annual mean drought duration (a), MDD (b), and variation of mean drought duration (c) i.e. [(duration 1971/2000 − duration 1963/1970)/duration 1963/1970] × 100. White regions represent a diminution of droughts duration

Download figure to PowerPoint

thumbnail image

Figure 10. Same as Figure 9 but with extreme + severe droughts (EDI < − 1.5)

Download figure to PowerPoint

4.2.2.2. Number of droughts

Figures 11 and 12 show the varied drought frequency for all kind of droughts (EDI < 0) and extreme + severe droughts (EDI < − 1.5), i.e. of groups constituted of consecutive days with negative EDI value.

thumbnail image

Figure 11. Average number of droughts (a) i.e. average number of groups constituted of consecutive days with negative EDI value. This is for all kind of droughts (EDI < 0) and shows for years 1963/2000 (July to September). The variation in the number of droughts is also presented (b), with the same definition of variation as in Figure 9

Download figure to PowerPoint

thumbnail image

Figure 12. Same as Figure 11 but with extreme + severe droughts (EDI < − 1.5). The total number of droughts is drawn here

Download figure to PowerPoint

5. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study area and data
  5. 3. An overview of water stress using the CMI
  6. 4. Drought study using the SPI and the EDI
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgements
  10. References

5.1. Analysis of EDI distribution

5.1.1. Spatial drought distribution

Droughts are more frequent but of shorter duration in the southern area of the basin: this trend is especially strong for the MDD but also for the average duration (about 1 day longer).

The mean duration of droughts is longer in the north for both classes of EDI values, but the highest values of mean or maximum values of drought duration are not observed in the extreme north, but slightly southward, along an area crossing the basin from the central northwest to the central northeast. Locally, high values of EDI are also observed in the south of the basin, but without overall geographical organization. It is worth noting that, for two stations in the central southwest of the basin, the mean duration of drought has decreased (EDI < 0) over the period 1963–2000 during July to September months (Figure 9(b)). This is confirmed with extreme + severe droughts (EDI < − 1.5), but with more stations concerned in the south, and even some in the central north band.

As shown in Figures 9–12, MDD is about 40 days longer in the northern portion of the basin, as is apparent in the north-western quarter of the basin, while the extreme northern portion has a drought duration that is similar to the southern mid-area.

A comparative analysis of the varied droughts being studied (EDI < 0 and EDI < − 1.5) shows no significant spatial differentiation in their frequencies as shown in Figures 11(a) and 12(a). However, regarding drought duration, Figure 9(b) shows much higher MDD in the north than in the south, while Figure 10(b) presents a more homogeneous situation for extreme + severe droughts only.

It should be noted that regions with less frequent droughts experience longer drought days; this is unsuitable for agricultural activities.

However, for a small region located in the central west (Bougouni, Dogo Bougouni, etc.), both parameters depict a decreasing trend.

5.1.2. Temporal variability

A comparative analysis of the various droughts studied (EDI < 0 and EDI < − 1.5) shows varied temporal variation, while areas with less frequent droughts usually experience longer drought duration; this pattern is evident for EDI < 0 droughts (Figures 11(b) and 12(b)). Also, extreme + severe drought frequency seems to increase with an increasing trend after 1970 in regions located within the northern part of the basin.

Figure 13 shows the categorization of drought days frequency for the ‘drought conditions’ and ‘extreme + severe conditions’ (i.e. number of days with negative (and < − 1.5, respectively) EDI value) within the Bani catchment.

thumbnail image

Figure 13. Evolution of the total number of days classified in drought (EDI < 0, (a)) and extreme + severe drought (EDI < − 1.5 (b)), sum on the whole basin. The Y-axis represents standard normal value and the X-axis represents years 1963 to 2000

Download figure to PowerPoint

The number of drought days was higher during the 1980s, which was the driest decade since the beginning of the century over the area (L'Hote et al., 2003). High values were also observed during the 1990s, but with more frequent low values of EDI for extreme + severe droughts during this decade (Figure 13(b)).

5.2. Summary map of drought analysis

One of the problems associated with the drought indicator is that large amount of information needs to be provided; hence, it is difficult to have a global view of droughts in a general way. Hence, in this study, the stations are categorized from the less to the most threatened by droughts using EDI. This classification provides an instant virtual assessment of drought occurrence within the Bani basin. It is also very important to study other associated maps like MDD and the number of droughts, among others, as each map is a part of the final general assessment.

The drought classification map is based on four main criteria namely number of droughts, mean drought duration, MDD, and temporal evolution. ‘Temporal evolution’ is the sum of the relative evolutions between years 1963/1970 and years 1971/2000 of the three previous criteria.

  • equation image

For each of the four criterions described above, the 51 stations are ranked from the worst to the best situation and the final rank is the average of these four parameters.

Moreover, in order to provide a more quantitative assessment to extreme and severe droughts, weights were attached to each criterion. For example, the MDD estimation ranking involves the consideration of various droughts (EDI < 0) for each station, while a second rank considers only droughts with EDI < − 1.5. These two ranks are then averaged and the average value remains the rank for ‘MDD’ for the station. This is summarized below:

  • equation image

With, for example:

  • equation image

where

FR: final rank, calculated for each station.

RX: rank for category ‘X’. R = {1, 2, …, 51}, 1 being the worst situation i.e. highest number of droughts, longest duration, and strongest positive evolution of these parameters.

It is important to note that these two parameters that address the drought duration have been used in order to provide a better weighted assessment for each station. Finally, Figure 14 shows the station ranking within the Bani basin, which depicts the three regions that seem to be threatened by droughts: the southern quarter (Filamana, Manankoro, Odienne, etc.), the northeast (Dionkele, Zangasso, Klela, etc.), and the central west (Fana, Dioila, Beleko, etc.). The extreme northern part is surprisingly not in a bad situation, like the centre of the basin. The whole central south of the basin (Bougouni, Kolondieba, Sikasso) seems to have been less affected by the drought hazard during the 1963–2000 period.

thumbnail image

Figure 14. General assessment of drought hazard on the Bani watershed using EDI with daily data, from 1963 to 2000 (July to September). The description of the calculation is given in Section 5.2

Download figure to PowerPoint

Such a geographical structure might be a coincidence. It may also reflect some long-term climatic organization. To what could this be related has to be searched, for instance, into atmospheric and climatologic features. This is not the topic of this paper. But we can nevertheless put forward a hypothesis, which is that this structure could be linked to the average location of the rainiest area of the inter-tropical convergence zone during the rainy season. On this subject, Lienou et al. (2008) have recently shown a southward shift of the climatic areas during boreal summer since 1970 in central south Cameroon, which they tentatively connected to the rainfall reduction over West Africa since 1970.

5.3. SPI versus EDI

The SPI-10d calculation provides less accurate results than expected for some values but is quite accurate for MDD. Hence, it is interesting to compare MDD results from the SPI and EDI maps, which are very similar. It should be noted that the temporal resolution of the degraded SPI-10d is 5 days (the 10-day wide temporal window is moving every 5 days)

Owing to the differences in temporal resolution (1 day for the EDI vs 5 days for the SPI), further comparison between these indicators is quite difficult. Nevertheless, it is possible to try to compare them with a degraded EDI resolution. This was carried out using the calculated mean value every 5 days in order to assess the similarity of each 5-day period and their categorization with the SPI and the EDI.

The result of the classification is given in Table IV: if both categories (one assessed by the SPI, the other by the EDI) are the same, then the classification is ‘good’, if there is just one category between both, the classification is ‘intermediate’, and if there is more than one category, the classification is ‘bad’.

Table IV. Comparison of each 5-day group classification (SPI vs EDI)
ClassificationAverage (%)Sdv (% of average)
  1. ‘Average’ is the average of each station's percentage of days classified as ‘good’ or ‘bad’. A degraded version of EDI is used here.

Good58.57.6
Intermediate31.83.6
Bad9.74.6

Results show that both indicators give results that are quite close but the difference is nevertheless noticeable. This confirms our previous view concerning SPI degradation, which is not a bad estimator but it is not good enough.

The comparison of the SPI and EDI global map (Figure 15(a) and (b)) provides a similar final assessment because the classification is based on ranks and not on values. But, even if similarities do exist between the SPI and EDI global assessment, they differ for some regions. Figure 15(b) depicts the final drought assessment but with a less complicated drought definition as days without any rainfall.

thumbnail image

Figure 15. Comparison of three assessments of droughts hazard using SPI (a), dry days (b), and EDI (c). Dry days are days without rainfall. This figure is available in colour online at wileyonlinelibrary.com/journal/joc

Download figure to PowerPoint

Moreover, the role played by these indicators is well depicted by these three maps (Figure 15(a–c)). Without giving an elaborate definition of droughts (Figure 15(b)), the critical situation in the southern part of the basin could not have been explained well.

Finally, SPI-10d is still interesting for some parameters, even if EDI seems better to us, as it also gives valuable information for MDD and precipitation thresholds. Summary of both indicators' advantages is given in Table V.

Table V. Summary of advantages and disadvantages of SPI and EDI
 SPIEDI
+Standardized (each drought category has a given return period) and well knownCreated for daily data
 Relatively easy to computeGood use of daily data
 Free access routines already existNo station rejected
Very hard to compute on a daily time stepDifficult to understand quickly
 Limitation due to gamma distribution (stations rejected)Time of calculation
 Loss of informationFew studies on it
  No free access routines

5.4. Link with CMI

EDI estimation is globally coherent with CMI. It is also evident that water stress has increased within the entire basin after 1970. This is confirmed by the temporal evolution of droughts as noted in Section 3.3. There is a trend in the south-eastern part of the basin, where CMI values decrease more than those in the rest of the basin. The final EDI assessment map provides similar information that there exists a threatened region near the south-eastern border.

6. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study area and data
  5. 3. An overview of water stress using the CMI
  6. 4. Drought study using the SPI and the EDI
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgements
  10. References

6.1. Summary

The aim of this work was to study droughts that occur in the Bani basin (Mali) using indexes at different time steps. Indexes were computed because they are useful in comparing different situations over time and space and also because they give a standardized definition of droughts.

The computed CMI using monthly PET and rainfall data provides the first indication of the water stress with an increasing trend all over the basin after 1970, the critical year in this region. Indeed, the part of the basin classified as humid for the years 1940–1970 disappeared for the years 1971–1995 and the arid zone got much larger.

Further drought analysis was provided using the SPI at a 10-day time step. Although it was totally unsuccessful when computed with the ‘classical’ definition, it provided good information with a ‘degraded’ definition: MDD and precipitation threshold for each drought category. Unfortunately, important information such as mean drought duration was very imprecise.

The inadequacies of the SPI call for the usage of another index, namely the EDI, which is more complicated but more robust with daily data analysis and computation. Moreover, the EDI does not use a statistical distribution like the gamma distribution for the SPI. Thus, the EDI does not reject any station that increases its accuracy. This index also gives very interesting data for agriculture purposes like mean and MDD. In order to have a quick overview of regions threatened by increasing drought indices, we produced a drought assessment rank map of the basin,

Three main threatened regions identified are as follows: the whole southern part of the basin (i.e. south of the 1200 mm annual isohyet), the north-eastern part of the Bani Basin (Dionkele, Zangasso, Klela, etc.) and the central western part (Fana, Dioila, Beleko, etc.).

Owing to this drought assessment rank, it is now possible to focus first on most threatened regions for creating coping strategies such as irrigation, weather insurances, or seasonal forecasts.

Moreover, it is evident that the drought situation is getting worse after 1970 and that there is an increase in the number of days being classified as drought days. Indeed, if the number of events is decreasing, their duration is increasing. This tendency is especially obvious considering all types of droughts (EDI < 0). This supports the findings made with the CMI.

Finally, the EDI provides a better tool for studying droughts than SPI-10d. The EDI also allows a more accurate description of the spatial extension of the drought than other indices like, for instance, the variable ‘days without any rainfall’, which seems too simple and with its use we would have to skip the extension of the drought in the south especially. Indeed, this simple approach is not ‘site located’, which means that a drought is defined relatively to the rest of the basin. From this point of view, a 5-day drought has the same impact in the south and in the north. This is obviously wrong, as crops are very different, especially considering their drought tolerance (e.g. maize vs millet).

Using such an index seems to be very useful for agricultural purposes. For example, insurance systems based on weather information are now being used more frequently in African countries (e.g. in Malawi) and, according to Berg et al. (2008), there is a real need to use a robust daily or decadal index.

6.2. Limitations

It is important to be aware of the study limitations, which are mainly due to lack of data as there is no precise way to fill lacks of data on a daily time step. Hence, the results, especially values, must be taken with caution.

Another limitation is the rainy period chosen, which spreads over the months of July, August, and September. Although it is obvious that these 3 months constitute the rainy period, the rainy season can last longer than 3 months in some regions, especially in the south. Hence, the result is more for the months of July, August, and September rather than for ‘the rainy season’.

6.3. Prospective

It is interesting to study the relative usefulness of the new indicator (EDI) more precisely, as it has been done with the SPI. As noted, one of the main limitations of the EDI was the time required to compute it and also its definition, with a lot of parameters. It is therefore necessary to write a script with free access for better computation of the index in a more efficient way (a package capable of computing several drought indicators on a monthly time step already exists: SPATSIM, but it is not free (Smakhtin and Hughes, 2005). These kind of indicators have to be simple and user friendly.

Regarding the rainy period, it could be useful to use a variant of the EDI, the available water resource index (Han and Byun, 2006), in order to compute the beginning and the end of the rainy period for each station. This could also show if this period is changing over time. It has been found that there was no change (Traore et al., 2000) in this region, but this does not support the local farmers' point of view (Roncoli et al., 2001)

Finally, a summary of all methods to fill the daily data gap is necessary and the development of a precise method is indeed very important.

Acknowledgements

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study area and data
  5. 3. An overview of water stress using the CMI
  6. 4. Drought study using the SPI and the EDI
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgements
  10. References

The authors would like to thank the financial support of this project, the French National Agency for Research (ANR) through the RESSAC program, and all reviewers of this paper's English version, especially Olusegun Adeaga, from Lagos University, Nigeria.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Study area and data
  5. 3. An overview of water stress using the CMI
  6. 4. Drought study using the SPI and the EDI
  7. 5. Discussion
  8. 6. Conclusion
  9. Acknowledgements
  10. References
  • Berg A, Quirion P, Sultan B. 2008. Can Weather Index Drought Insurance Benefit to Least Developed Countries' Farmers? A Case Study on Burkina-Faso. Congrès de l'Association française de science économique: Paris, Septembre 2008.
  • Boyer JF, Dieulin C, Rouche N, Cres A, Servat E, Paturel JE, Mahe G. 2006. SIEREM: an environmental information system for water resources. IAHS Publ. 308, IAHS Publications: USA; 1925.
  • Byun HR, Wilhite DA. 1999. Objective quantification of drought severity and duration. Journal of Climate 12: 27472756.
  • Dezetter A, Girard S, Paturel JE, Mahe G, Ardoin-Bardin S, Servat E. 2008. Simulation of runoff in West Africa: is there a single data-model combination that produces the best simulation results? Journal of Hydrology 354(1–4): 203212.
  • Downing TE, Patwardhan A. 2005. Assessing vulnerability for climate adaptation. In Adaptation Policy Frameworks for Climate Change: Developing Strategies, Policies and Measures, LimB, Spanger-SiegfriedE, BurtonI, MaloneE, HuqS (ed). Cambridge University Press: Cambridge.
  • Eriksen SH, Kelly PM. 2006. Developing credible vulnerability indicators for climate adaptation policy assessment. Mitigation and Adaptation Strategies for Global Change 12: 495524, DOI: 10.1007/s11027-006-3460-6.
  • Feitelson E, Chenoweth J. 2002. Water poverty: towards a meaningful indicator. Water Policy 4: 263281.
  • Füssel HM, Klein RJT. 2006. Climate change vulnerability assessments: an evolution of conceptual thinking. Climate change 75: 301329, DOI: 10.1007/s10584-006-0329-3.
  • Giddings L, Soto M, Rutherford BM, Maarouf A. 2005. Standardized precipitation index zones for Mexico. Atmosfera 18: 3356. ISSN: 0187–6236.
  • Guttman NB. 1999. Accepting the Standardized Precipitation Index: a calculation algorithm. Journal of the American Water Resources Association 35(2): 311322.
  • Hamed K, Rao R. 2000. Flood Frequency Analysis. CRC press: Boca Raton.ISBN 0-8493-0083-5.
  • Han SU, Byun HR. 2006. The existence and the climatological characteristics of the spring rainy period in Korea. International Journal of Climatology 26: 637654, DOI: 10.1002/joc.1274.
  • International Panel on Climate Change (IPCC). 2007. Climate Change 2007: Synthesis Report. Summary for Policymakers. An Assessment of the Intergovernmental Panel on Climate Change.
  • L'Hote Y, Mahe G, Some B, Triboulet JP. 2002. Analysis of a Sahelian rainfall index from 1896 to 2000: the drought continues. Hydrological Sciences Journal 47(4): 563572.
  • L'Hote Y, Mahe G, Some B. 2003. The 1990s rainfall in the Sahel: the third driest decade since the beginning of the century. Reply to discussion. Hydrological Sciences Journal 48(3): 493496.
  • Lienou G, Mahe G, Paturel JE, Servat E, Sighomnou D, Ekodeck GE, Dezetter A, Dieulin C. 2008. Changes of hydrological regimes in the equatorial area of Cameroon: an impact of climate change in equatorial Africa? Hydrological Sciences Journal 53(4): 789801.
  • Lloyd-Hughes B, Saunders MA. 2002. A drought climatology for Europe. International Journal of Climatology 22: 15711592, DOI: 10.1002/joc.846.
  • Lubes-Niel H, Masson JM, Paturel JE, Servat E. 1998. Variabilite climatique et statistiques. Etude par simulation de la puissance et de la robustesse de quelques tests utilises pour verifier l'homogeneite de chroniques. Revue des Sciences de l'Eau 11(3): 383408.
  • Mahe G, Olivry JC. 1999. Assessment of freshwater yields to the ocean along the intertropical Atlantic coast of Africa. Comptes Rendus de l'Academie des Sciences, Serie IIA 328: 621626.
  • Mahe G, Olivry J, Dessouassi R, Orange D, Bamba F, Servat E. 2000. Relations eaux de surface—eaux souterraines d'une riviere tropicale au Mali. Comptes rendus de l'Academie des Sciences, Seire IIA 330: 689692.
  • Mahe G, L'Hote Y, Olivry JC, Wotling G. 2001. Trends and discontinuities in regional rainfall of west and central Africa—1951–1989. Hydrological Sciences Journal 46(2): 211226.
  • McKee TB, Doesken NJ, Kleist J. 1993. The relationship of drought frequency and duration to time scale. Report of the Eighth Conference on Applied Climatology. Anaheim (California).
  • Meylan P, Musy A. 1996. Hydrologie frequentielle, version provisoire. Departement de Genie rural, Institut d'Amenagement des terres et des Eaux: Lausanne.
  • Ntale HK, Gan TY. 2003. Drought indices and their application to east Africa. International Journal of Climatology 23: 13351357, DOI: 10.1002/joc.931.
  • Morid S, Smakhtin VU, Moghaddasi M. 2006. Comparison of seven meteorological indices for drought monitoring in Iran. International Journal of Climatology 26: 971985, DOI: 10.1002/joc.1264.
  • O'Brien K, Eriksen S, Nygaard LP, Schjolden A. 2007. Why different interpretations of vulnerability matter in climate change discourses. Climate Policy 7: 7388.
  • Paturel JE, Ouedraogo M, Mahe G, Servat E, Dezetter A, Ardoin S. 2003a. Influence of the spatialization of data on the modelling of monthly river regimes in West Africa. Hydrological Sciences Journal 48(6): 881890.
  • Paturel JE, Ouedraogo M, Servat E, Mahe G, Dezetter A, Boyer JF. 2003b. The concept of rainfall and stream flow normals in West and Central Africa in a context of climatic variability. Hydrological Sciences Journal 48(1): 125137.
  • R Development Core Team. 2005. R: A Language and Environment for Statistical Computing, Reference Index version 2.6.2. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-0. http://www.R-project.org.
  • Romero R, Guijarro J, Alonso S. 1998. A 30 year (1964–1993) daily rainfall data base for the Spanish Mediterranean regions: first exploratory study. International Journal of Climatology 18: 541560, DOI: 10.1002/(SICI)1097-0088(199804)8:5 < 541::AID-JOC270 > 3.0.CO;2-N.
  • Roncoli C, Ingram K, Kirshen P. 2001. The costs and risks of coping with drought: livelihood impacts and farmers' responses in Burkina Faso. Climate Research 10: 119132.
  • Rouche N, Ardoin-Bardin S, Boyer JF, Brissaud B, Cres A, Dieulin C, Mahe G. 2008. Constitution d'une grille de pluies mensuelles periode 1900–2000 pour l'Afrique. CD-ROM, Actes du 13eme congres mondial de l'eau. Montpellier, France, septembre 2008.
  • Smakhtin VU, Hughes DA. 2007. Automated estimation and analyses of meteorological drought characteristics from monthly rainfall data. Environmental Modelling and Software 22: 880890, DOI: 10.1016/j.envsoft.2006.05.013.
  • Sönmez FK, Kömüscü AU, Erkan A, Turgu E. 2005. An analysis of spatial and temporal dimension of drought vulnerability in Turkey using the Standardized precipitation Index. Natural Hazards 35: 243264, DOI: 10.1007/s11069-004-5704-7.
  • Soumaguel A. 1996. Elaboration des fichiers Opérationnels pour le calcul Régionalisé des pluies par la méthode du vecteur Régional (MVR) sur le bassin versant du Niger. Rapport d'activité FRIEND-AOC, Bamako, ORSTOM, Laboratoire d'hydrologie, 35 p.
  • Sullivan C, Meigh J. 2005. Targeting attention on local vulnerabilities using an integrated index approach: the example of the Climate Vulnerability Index. Water Science and Technology 51: 6978.
  • Traore SB, Reyniers FN, Vaksmann M, Kone B, Sidibe A, Yorot A, Yattara K, Kouressy M. 2000. Adaptation a la secheresse des ecotypes locaux de sorghos au Mali. Secheresse 11(4): 227237.
  • Vörösmarty CJ, Douglas EM, Green PA, Revenga C. 2005. Geospatial indicators of emerging water stress: an application to Africa. AMBIO 34(3): 230237.
  • Wu H, Hayes MJ, Weiss A, Hu Q. 2001. An evaluation of the Standardized Precipitation Index, the China Z-index and the statistical Z-score. International Journal of Climatology 21: 745758, DOI: 10.1002/joc.658.