Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | erent erent ff ff Open Access , and Global Pre- 2 Discussions erences between Sciences ff Earth System Earth Hydrology and and Hydrology . The set of drought indi- , F. Wetterhall 2 ECMWF ERA – Interim reanaly- satellite monthly rainfall product 3B43, the gridded precipitation dataset, the 13408 13407 , F. Pappenberger 1 ected by drought, all the drought indicators agree on ff Global Monthly Merged Precipitation Analyses, and the , P. Barbosa 2 , E. Dutra 1 erent datasets and drought indicators on their capability to improve drought ff Tropical Rainfall Measuring Mission 1 ected area. In regions with limited rain gauge data the estimation of the di ff , the The five precipitation datasets compared are the A comparison of the annual cycle and monthly precipitation time series shows a This discussion paper is/has beenSciences under (HESS). review Please for refer the to journal the Hydrology corresponding and final Earth paper System in HESS if available. Joint Research Centre of theEuropean European Centre Commission, for JRC, Medium Ispra, Range Italy Weather Forecasts, Reading, UK tremes. Moreover, for the areas a the time of drought onsetthe and a recovery although there isdrought disagreement on indicators the is extent of characterisedgests by that a the higher mainuncertainty source uncertainty. of Further in error comparison the in sug- precipitationparameters the of datasets computation the rather of drought the than indicators. drought the indicators estimation is of the the distribution cators used includes the Standardizedtion–Evaporation Precipitation Index, Index, Soil the Moisture Standardized Anomalies Precipita- and Potential Evapotranspiration. good agreement inthe the datasets timing are of in the the ability rainy to seasons. represent the The magnitude main of di the wet seasons and ex- climatic regions (the Oum er-Rbiaper in Niger Morocco, in the Western BlueGreater Africa, Nile Horn and in of Eastern the Africa. Africa, Limpopo the in Up- South-Eastern Africa) assis well as the Global Precipitation Climatology Centre cipitation Climatology Project Climate Prediction Center Merged Analysis of Precipitation Drought monitoring is a key componentand to up-to-date mitigate impacts datasets of is droughts.gates a Lack of di common reliable challenge acrossmonitoring the in Globe. Africa. This The study study investi- was performed for four river basins located in di Abstract Establishing the dominant source of uncertainty in drought indicators G. Naumann Hydrol. Earth Syst. Sci. Discuss.,www.hydrol-earth-syst-sci-discuss.net/10/13407/2013/ 10, 13407–13440, 2013 doi:10.5194/hessd-10-13407-2013 © Author(s) 2013. CC Attribution 3.0 License. 1 2 Received: 13 October 2013 – Accepted: 23Correspondence October to: 2013 G. – Naumann Published: ([email protected]) 7 November 2013 Published by Copernicus Publications on behalf of the European Geosciences Union. J. Vogt 5 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ects ff ective ff erent indica- ff erent climate regimes across ff erences (mainly in the equatorial ff CMAP and TRMM 3B43 performed very well ◦ 13410 13409 erent river basins in Africa. From these studies it is evi- ff erent datasets and drought indicators (SPI, SPEI, PET and SMA) to ff culty in establishing a “ground truth” of precipitation in Africa also a ffi ected millions of people (Rojas et al., 2011). Since, precipitation is fundamental ff ciently answered. ffi The main goal of this study was to identify the main sources of uncertainty in the The di Liebmann et al. (2012) studied the spatial variations in the annual cycle compar- As a consequence of drought, many countries in Africa have seen recurrent famines Recently, several rain gauge and remote sensing based estimations of precipitation Horn of Africa (GHA). The size and geographical extent of the highlighted areas are Africa. 2 Data and methods 2.1 Study area The analysis was performed at continentalriver level over basins Africa (Oum with particular er-Rbia, focus Limpopo, on Niger, four and Eastern Nile) as well as the Greater indicator, system and region. computation of the droughtability indicators. Furthermore, of the an assessment di represent was the done spatio-temporal on features the of droughts in di the uncertainty in the calculationsince of the derivatives relationship of between precipitation,indicator like the is drought quality indicators, nonlinear. of Thisdampened a means precipitation when that product a errors andcompared in drought any several the drought drought index indicators precipitation (Heim, is can 2002;2011; computed. Anderson be Vicente-Serrano et et amplified Previous al., al., 2011; or works 2012). Shuklators However, et have is an al., not reviewed agreement necessarily between and observed di as the capability to detect droughts changes between area) in the qualitymate of Anomaly the Monitoring System precipitation –(CAMS-OPI) between Outgoing datasets the Longwave Radiation ERA-Interim, for Precipitation di GPCPdent Index and that the the question Cli- onsu which dataset best represents African precipitation is still not sistent. Dutra et al. (2013a) found significant di wettest parts of Africa, but the timing of the annual cycle and onset dates are con- atively dense station network overtime the scale Ethiopian and highlands a and spatial foundwith resolution that a of at 2.5 bias a monthly offound less that than the 10 Rainfall % Estimation and Algorithmtential (RFE) an in and reproducing RMS TRMM the of 3B42 interannual showed aboutand variability, a the 25 the high %. spatial timing po- Thiemig and of quantitative et rainfall distribution events. al. (2012, 2013) ing GPCP with TRMMdatasets. and They gauge-based Famine found Early that Warning GPCP System (FEWSNET) estimates are generally higher than TRMM in the real-time drought monitoring and early warning systems. became available, which exhibit discrepancies andlocal limitations and in regional representing rainfall scale. at Thisdatasets by has Dinku been et highlighted al. (2007, for 2008) daily and and Hirpa monthly et al. precipitation (2010). The authors studied a rel- plete historical rainfall records may, however,of lead water to significant supply errors indices in derived the from estimation precipitation time-series. that a for rain-fed crops in theseand drought-prone early regions, improvements warning in will drought(Wilhite improve monitoring et our capacity al., to 2000;date detect, climatological Rowland anticipate, data et in and al., many mitigate regions 2005). famine of However, Africa the hinders the lack development of of e reliable and up-to- Assessment of drought impacts requiresas understanding well of as regional the historical bearingsods on droughts for human activities drought during assessmentprecipitation their time-series are occurrences. alone. Traditional mainly A meth- sparse based distribution on of rain water gauges and supply short indices or incom- derived from 1 Introduction 5 5 25 15 20 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | - ), ff ◦ res- ◦ 0.25 × ◦ 0.25 × ◦ ), “Tropical Rain- ◦ 0.5 × ◦ latitude/longitude grid and grid. man et al., 2009) combines in the grid-point space) with ◦ ◦ ff ◦ 2.5 0.5 0.7 × × × ◦ ◦ ◦ ) and the CPC Merged Analysis of Precipita- ◦ 13412 13411 2.5 × ◦ , bilinear interpolation to 0.5 ◦ 0.7 × ◦ erent thresholds but is only shown for the values below the ff erent thresholds of the standardized indicators. The duration ff ) (Table 2). ◦ grid. ◦ 2.5 × 2.5 ◦ ), the Global Precipitation Climatology Project (GPCP) Global Monthly ◦ × ◦ 0.5 × ◦ The individual drought episodes from the time series of all indicators were deter- The CPC Merged Analysis of Precipitation (“CMAP”) is a technique which produces The ECMWF ERA-I reanalysis, the latest global atmospheric reanalysis produced by The Global Precipitation Climatology Project (GPCP, Hu This work uses the TRMM Multisatellite Precipitation Analysis (TMPA) estimation 1.0 threshold. GPCP and GPCC for the SPI,from ERA-I ERA-I and GPCP simulations. for the SPEI while the SMA are derived mined by considering di of each dry event wasvalues determined (positive as for the number PET) ofarea over consecutive was the months computed period with for negative 1998–2010. di − The monthly drought fractional 2.3 Drought indicators The set of hydro-meteorologicalcipitation indicators Index (SPI), analysed Standardized includedEvapotranspiration Precipitation–Evaporation (PET) the Index and (SPEI), Standardized Soil Potential Pre- Moisturecators Anomalies (SPI, (SMA). SPEI, The SMA) three were drought computed indi- at continental scale based on ERA-I, TRMM, of GPCP v2.2 thata regular was 2.5 used is available since Januarypentad 1979 and to monthly December analysesgauges 2010 are of merged in global with precipitation precipitation estimates(infrared from in and several which satellite-based microwave). algorithms The observations analysisextend from are back rain to on 1979. a For 2.5 further information refer to Xie and Arkin (1997). ily from geostationary satellites,Infrared low-Earth Sounder orbit (AIRS) estimatesOperational Television including Vertical Infrared Sounder the Observation (TOVS), and Atmospheric Satellite OutgoingIndex Longwave (OPI) Program Radiation data Precipitation (TIROS) from theanalyzed NOAA by series the satellites. The Global gauge Precipitation data Climatology are assembled Centre and (GPCC). The latest version (DMSP,United States) satellites, infrared (IR) precipitation estimates computed primar- the precipitation information availablesor from Microwave/Imager several (SSM/I) data sources from such the as Defence Meteorological the Satellite Special Program Sen- ECMWF extends from 1detailed January descriptions 1979 of the to atmospheric the modeltem, used present the in observations date. ERAI, used, the See and data Deehas various assimilation a performance et sys- spectral aspects. al. T255 The (2011) horizontal ERAI resolution60 configuration for (about model 0.7 vertical levels.were For spatially the interpolated present (bilinear) application, to the a monthly regular precipitation 0.5 means man et al., 2007).other This satellite product products combines (3B-42) the withgridded estimates the rain generated gauge Climate by data Anomaly the and/or Monitoringolution TRMM the System The and GPCC (CAMS) GPCC global full rainto reanalysis gauge 2010. version data at 5 This 0.25 (Rudolpha dataset et large is number al., of based 1994) stations was on (up used to quality-controlled for 43 000 precipitation 1979 globally) observations with irregular from coverage in time. analysis” (approximately 0.7 fall Measuring Mission” (TRMM) satellitethe monthly “Global rainfall product Precipitation 3B43V.5 (0.25 Climatology (0.5 Centre” (GPCC)Merged Precipitation gridded Analyses precipitation (2.5 dataset tion (CMAP, 2.5 computed at monthly intervals as TRMM 3B-43 dataset for the period 1998–2010 (Hu climates and socio-economic systems in Africa. 2.2 Precipitation data The five precipitation datasets used were the “ECMWF ERA-INTERIM (ERA-I) Re- provided in Table 1 and Fig. 1. The regional study areas selected cover a range of 5 5 25 15 20 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ects of ff erent scores ff erent time-scales ff cult due to the lack of an ffi score, a classification scheme is z erent time scales, and an adjustment ff erent natural and human needs. Since score (Dutra et al., 2008). In the current ff z 13414 13413 score following the considerations depicted in Dutra et al. z ), Mean Absolute Error (MAE), Percent Bias (PBIAS) and the Index r ). Details of the Evaluation scores are listed in the Appendix. d eld et al., 2004) or just a simple ffi cient ( ffi erent time dimensions (like the SPI) with the capacity to include the e ff erent drought properties. This analysis does not assume that a single dataset Evapotranspiration (PET) ff A direct quantitative assessment at continental scale is di Typically, the gamma distribution is the parametric CDF chosen to represent the A persistent negative anomaly of precipitation is the primary driver of drought, re- over di or indicator is better than the other but highlights their temporal and spatial coherency. The precipitation datasets and drought indicatorsavailable were in assessed the using di hydroGOFlation R-Package coe (Zambrano-Bigarini, 2013): Spearman’sof corre- Agreement ( actual validation dataset that representsor the temporal ground resolution. truth The withand index adequately performance of high metrics agreement) spatial were (mean used absolute to diagnose error, the relative relative reliability bias of each indicator (2008). By normalizing theobtained soil that moisture is with similarSerrano the et and al. comparable (2012). to that of2.4 McKee et al. Evaluation metrics (1993) and Vicente 2.3.3 Soil Moisture Anomalies (SMA) Soil moisture anomalies wereannual derived cycle. from Further ERA-I simulations standardization,distribution by to could a removing be probability the distribution achieved mean bution (similar (She by to SPI fitting or, the SPEI)work such we soil as compare moisture the the SMA Beta distri- to a log-logistic probabilityperature distribution. impact SPEI via is the similarThornthwaite potential to (1948). evapotranspiration SPI, In (PET) the but that current itPET, work, is includes and we calculated the used the following ERAI tem- multiscalar 2-meterand temperature index normalized to is (like derive the calculated SPI) as using P-PET the log-logistic over probability the distribution. di the accumulation of a water deficit/surplus at di bining di temperature variability on drought. The calculation combines a climatic water balance, ative values correspond toto drier wetter periods periods than normala than probabilistic and normal. measure positive of The values the magnitude correspond severity of of a the wet2.3.2 or departure dry from event. Standardized the Precipitation–Evaporation Index median (SPEI) is and Potential The Standardized Precipitation Evapotranspiration Index (SPEI,2010) Vicente-Serrano is et based al., on precipitation and temperature data, and it has the advantage of com- the Maximum-Likelihood Estimation (MLE)gamma method distribution. to estimate the parameters of the sulting in a successive shortageSPI of values water are for given di in units of standard deviation from the standardised mean, neg- tion time-series and applying anvariable. equi-probability This transformation gives to the the SPI standard in normal units of number ofprecipitation standard time-series deviations from (e.g. the McKee median. et2002; al., Husak 1993, et 1995; al., Lloyd-Hugheszero 2007) and and since Saunders, positively it skewed (Thom, hasand 1958; the Naumann Wilks, et advantage 2006). al. of Moreover,els (2012) being Husak have precipitation bounded et shown time-series al. on that in (2007) the the most left gamma of at distribution the adequately locations mod- over Africa. In this study we use The Standardized Precipitation1995) Index to provide (SPI) a spatially was and(or developed temporally surplus) invariant by measure for of McKee any thefitting accumulation et precipitation a deficit parametric timescale al. Cumulative (e.g. Distribution (1993, Function 3, (CDF) 6, to a 12 homogenized precipita- months). It is computed by 2.3.1 Standardized Precipitation Index (SPI) 5 5 20 25 15 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1 − . The 1 − values greater than 0.6 for almost all the erences are observed in the tropical area d ff 13416 13415 ) shows that there is a good correspondence be- d erences were observed for ERA-I in the Blue Nile and GHA regions. cients which is usually greater than 0.8 (Table 3). The CMAP dataset ff ffi The biggest di Overall, the index of agreement ( The monthly datasets show a reasonable agreement over all regions in terms of the Apparently the density of rain gauges plays a role in determining the agreement be- For all the datasets and regions analysed the mean annual cycle of precipitation tween indicators in all regions with mean and the bias can reach 90 % in the3.2 Blue Nile and around Comparison 50 of % drought in indicators theThe GHA. monthly patterns ofshow drought that over dry Africa areas forthan (indicators January one with 2000, indicator, negative 2003, butspatial values) 2006 their and are and consistency temporal generally 2009 varies scale depictedbetween (Fig. with in all 4). the the more There drought indicators is over type, a the as generally study well good period. spatial as correspondence the the best agreement betweenbias in datasets those two with regions MAE isare below compared values 20 (30 % %). below in all 10 mmmonth the cases except when TRMMIn and CMAP these regions the overestimation of monthly precipitation reached 40 mmmonth precipitation variability is mainlywhile controlled in by the large-scalethese tropical synoptic regions, region model weather uncertainties systems, small-scale (for exampleties convective land-atmosphere in events coupling), satellite uncertain- play retrievals an asthe well mean important as annual role. poor cycles. gauge In cover contribute to thecorrelation large coe spread in deviates with values below 0.7 in some regions. Oum er-Rbia and Limpopo areas show Basin. tween datasets. The bestthose gauged with the regions lowest (Oumregions dispersion er-Rbia (Oum in er-Rbia and terms and Limpopo; of Limpopo) Table annual are 1) cycle. located are outside By the coincidence, tropical these region, two and their timation of ERA-I in the Blue Nile Basin and GHA and an underestimation in the Niger for the basins located between the tropics the discrepancies are higher with an overes- other datasets. ERA-I overestimateslikely the to rainfall be in associated the withmation central of a African aerosol substantial optical region warm depth which bias in in is the the region model (Dee dueshows et al., to good 2011). an agreement underesti- withis respect true to also the forHowever, with onset the respect GHA and to region end intensity the whichLimpopo of results is and the are characterized Oum more rainy heterogeneous. by er-Rbia season. Although two Basins in This rainy there the seasons is (Fig. a 3). good agreement between the datasets, south-western part of Africa.and over The un-gauged main areas. di estimations In transition can regions exceed from GPCCthan the more Sahel the than to other twofold the whileet estimations Sahara al., TRMM along TRMM 2012). is the There substantiallyin southwestern is lower GPCP also coast (Liebmann a et of tendency al., 2012) West of and higher Africa ERA-I precipitation (Dutra (Liebmann in et al., the 2013a, tropical b) rainforest compared with the completely independent. For example, TRMMsensing and data GPCP and are GPCP mainly uses basedcipitation GPCC on for over remote the land. ERA-I, Figure GPCC,is 2 GPCP, an CMAP shows overall the and agreement mean TRMM betweengeneral annual datasets the pre- spatial over datasets Africa. patterns with There respect ofsouth to annual gradient the precipitation. mean from as TheseSahel, the well datasets as followed desert agree the by areas on thethe in the precipitation location of north- the maximum the North over Inter-tropical the Convergence to Zone African the (ITCZ) rainforests and tropical related the savannahs drier to in climate in the the 3.1 Comparison of global precipitation datasets The datasets analysedtions are (TRMM, based GPCP) on and in-situ a data global (GPCC), circulation remote model sensing (ERA-I). estima- The datasets are not 3 Results and discussion 5 5 25 15 20 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1.0 are − ected and ff erent SPI esti- ff ected by drought (e.g. 1.0. In general there is ff − ciently, at the same time ffi erent thresholds of SPI and ect the agreement between ff ff 0.5 (Fig. 10). Below this threshold − is often below 0.5. d ered the most (Fig. 7). Overall, dry periods erent SPI’s is higher for thresholds closer to ff ff 13418 13417 erence of more than 50 % between SPI and SPEI esti- ect on the computations of the SPEI is not major, since ff ff erences between the indicators appear in the extreme events. ff ected by droughts in each region. Here the results of the di . However the e ff 1), therefore, will reduce the disagreement between indicators. However it d cult to monitor in areas with scarce in-situ rain gauges. − 1. For Niger and GHA regions there is only a reasonable agreement between ffi − erent factors. The main common factor is the remarkable absence of observations ff Generally in the Oum er-Rbia and Limpopo Basins, both extra-tropical regions, the For the basins located between the tropics a greater disagreement is observed due For dryer regions, such as the inner Niger delta and the GHA, the estimation of the Also, the bias between estimations indicates an acceptable departure between esti- In order to define how the selected threshold could a Regarding the areas that are under drought, all the datasets agree with the time of The average duration of dry episodes lasted between 2 to 6 months for all indicators, Figure 6 shows the evolution of drought areas in 2000, 2003, 2006 and 2009 charac- 0.8 or distribution parameters needed for the computation of the standardized indicators can importance of large-scale synoptic weather systems in these areas. to di to calibrate and test themountainous datasets. These areas deficiencies such are also astorial more the regions evident in Eastern are complex Nile mainlyseason. driven Basin. These by mesoscale Furthermore, dimension the droughts events absence areeven in di hard of equa- to convective be events reproduced during by models the and rainy the bias increases exponentiallyues surpassing of quickly aERA-I bias and of GPCC estimations. 100 % around SPI val- agreement is high, possibly due to the greater number of in-situ observations and the shown). For almost all regions (exceptconstant) for Oum the er-Rbia where correlation this between relationshipzero is the almost (Fig. di 9). To consider− a higher thresholdputs to a define limit areas to a thehighlight detection that of the main the di significance and severity of a drought.mations These from results normal conditions until values near period 2000/2002 in the inner Niger delta. datasets a correlation analysis wasthe performed areas between a di mations are presented, however similar results were found for the other indicators (not to the other indicators, the soil moisture shows a good correspondence except for the under dry conditions. However, even if the magnitude of the area is smaller with respect onset and recovery butthis there are disagreement sometimes tends disagreements tothe on be monthly the dependent fractional area area on a undera the better standardised threshold agreement values selected. if below Figure theconsidered. areas 8 In covered this shows by analysis anydiscrepancies the standardised Niger reaching indicator Basin below a and di mations Greater during Horn the of 2009/2010 Africaanomalies and present tend 2005/2006 to more define periods less respectively. generalised The droughts soil as it moisture is hard to reach half the region and stakeholders to better handle uncertainties in early warningwhere systems. the Niger Basinmeasured and with SPEI the tend GHAother to estimations, di while be PET 1 is or the indicator 2 with months less more memory. persistent if compared with the is a good agreement, withper usually grid more than cell. 3fined However, indicators there threshold, reporting mostly are drought over conditions some Centraldiscrepancies Africa. areas and There agreements with is and only scope propose(Svoboda to et one the al., take construction 2002; indicator advantage of Sepulcre-Canto of below et adevelopment these al., of the composite 2012; a indicator Hao de- single and composite AghaKouchak, droughtonset 2013). indicator The of could a improve the drought detectionproviding and of information the help to on monitor the its uncertainty evolution more in e the data. This will allow decision makers values of the agreement of this indicatorcharacterized with by the a others weaker is agreement, still where high. Only the inner Nigerterized Delta by is the number of indicators below a certain threshold. In almost all areas there comparisons (Fig. 5). PET seems to be uncoupled with the other indicators with low 5 5 25 15 20 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | . 1 − 1 in Niger and GHA). The − erent drought indicators (including SPI, ff erent combinations may have to be chosen. ff cult to get an additional value with respect to 13420 13419 ffi erences are observed in the ability to represent ff with a bias of up to 92 %. Also in the GHA region the 1 − 2 it is di . The bias in those two regions is below 20 %. The worst − 1 − erent precipitation datasets (TRMM, ERA-Interim, GPCC, GPCP and ff cult task, not only due to the nature of the phenomenon, but also due to the lim- ffi The comparative analysis between TRMM, ERA-I, GPCP and GPCC datasets sug- The comparison between drought indicators suggests that the main discrepancies A comparison of the annual cycle and monthly precipitation time series shows a good The monthly mean precipitation datasets agree over all regions with the only ex- The development of a combined indicator based on a probabilistic approach (e.g. Du- The above discussion underlines the fact that drought monitoring and assessment is Our results further underline the value of maintaining an operative monitoring net- than to the estimation of the distribution parameters. This is why the SPI estimations gests that it is feasibledrought to monitoring use TRMM over parts time series ofmainly with Africa. at high It spatial regional is resolution level possible forcies due to reliable in take to SPI advantage estimations its are of high shownstation this in spatial dataset density. mountainous On resolution. areas and the However, areas higher otherperforms with better hand, discrepan- a outside sparse drought in the monitoring situ areas at influenced continental by the level ITCZ. withare ERA-I due to the uncertaintiescertainties in in the the datasets estimation algorithms (driven or by the a parameterization lack of of the ground convection) rather information, un- errors below 10 mmmonth performance of ERA-I was observed inprecipitation the up Blue to Nile 40 mmmonth Basin, overestimatingbias the is monthly around 50 % with an overestimation of up to 17 mmmonth focused on four river basins and on the Greater Hornagreement of in Africa. the timingare of the two peaks, rainy including seasons.the the The magnitude Greater main of Horn the di of wet Africa seasons were and the there extremes. ception of the CMAPLimpopo dataset Basins that there shows is a a lower good agreement. agreement In between the the Oum datasets er-Rbia with mean and absolute This study evaluatedSPEI, the PET capabilities and SMA) ofusing in di detecting five the di timingCMAP). and The extension analysis of was drought performed across on Africa, a Pan-African scale and on a regional scale 4 Conclusions calibration. Without constant calibration, model-inherent errorssame can magnitude propagate of up to the the phenomenauncertainties (or can indicator) be to be socorresponding analysed. to big In an that fact, SPI the forstandard resulting of climatologies. certain events such as droughts withtra a et severity al., 2013c)case. could However, be at useful localchosen as carefully scale a taking the monitoring into account product kind their at of limitations. continental indicator scale and in the this source of data must be quence no definite conclusion canDepending be on drawn for the the region use to of be a studied, single di dataset orwork indicator. at country, continental orintrinsic even uncertainties global linked level since to indirect the observations availability have and their reliability of “ground truth” for their gauge data is characterised bythe the uncertainties main of sources extreme of values.propagated This error in suggests the are that estimation the of the uncertainties distribution in parameters of thea the precipitation drought di datasets indicators. thatitations are inherent in theregions. availability The of meteorological long-term datasets and asbe high well selected quality as carefully datasets the and for indicators their extended and limitations models need used to must be taken into account. As a conse- observations. As depicted in Wudensity et function al. and (2007), the the limitedSPI estimation sample of values. size the In in gamma these dry probability detect cases, areas some reduce the drought the SPI occurrences confidencediscrepancies (e.g. may of SPI never between the always attain indicators above very for negative lower values, thresholds failing to over regions with limited rain be biased (or lower bounded) by the large amount of zero or near null precipitation 5 5 25 15 20 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | (A4) (A2) (A3) (A1) erent drought ff cient is applied ffi cient computed us- ffi erences in the observed ff ’s. i R -th pair of data values. In cases of ties, i 13422 13421 2 ) | erent indicators and accumulation periods in the form ff obs ) developed by Willmott (1981) as a standardized mea- d − 2 | ) i obs sim) | obs) erences (Legates and McCabe, 1999). − + ff − | obs (obs) − erence in ranks between the i P ff obs (obs (sim cient (Wilks, 2006). Computation of the Spearman rank correlation can − P ffi P 1) 2 i (sim | R − 1 sim 2 erent estimations of a certain indicator. It measures accuracy for continuous | n 100 = P ( i ff X is the di n represents the number of pairs of the simulated (sim) and observed (obs) ( 6 = i P 1 n n n R − = cient is a more robust and resistant alternative to the Pearson product-moment − 1 ffi 1 The Index of Agreement ( The percent bias (PBIAS) measures the average tendency of the simulated values It is proposed to integrate di Regarding the areas that are under drought, all the indicators agree with the time of The Mean Absolute Error (MAE) measures the average magnitude of the errors in = = observations (obs) as shown in Eq. (2) MAE used to measure how close simulated forecasts or predictions (sim) are to the eventual fected, and the level of disagreement tends to be dependent onof the a threshold multivariate combined selected. indicator inproperties. order The to probabilistic take advantage nature ofcision of their makers such di and an for approach the would combined be analysis very of helpful multiple risks. for de- Appendix A The Spearman correlation represents theing Pearson correlation the coe ranks ofto the the data. ranks Conceptually, thecoe of Pearson the correlation data coe correlation rather coe than tobe the described data as: values themselves. The Spearman r between estimations. While for theare in other many regions cases the acceptable,Basin discrepancies greater when between discrepancies comparing datasets are ERA-I observed estimations with for the the other inner datasets. Niger onset and recovery but there are sometimes disagreements with respect to the area af- for the Oum er-Rbia and Limpopo regions are those that exhibit a better agreement d indicates a perfect match,index and of 0 agreement indicates canand no detect simulated agreement additive means at and and alldue variances; proportional to (Willmott, however, di it the 1981). squared is The di overly sensitive to extreme values The optimal value ofresentation PBIAS of is 0, droughtwhereas with indicators. negative low-magnitude values values Positive indicating indicatecount values accurate an that rep- underestimation this indicate metric bias. an dependsvations. It on overestimation must which be dataset bias, is taken considered into to ac- represent thesure obser- of the degree of model prediction error varies between 0 and 1. A value of 1 indicators. to be larger or smaller than the observed ones. Bias(%) where where where a particular dataassigned value their appears average rank more before than computing once, the all ofa these set equal of di values are variables without considering the direction of the error. Also, this quantity is usually 5 5 20 10 15 10 15 20 25 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , 2009. 10.1029/2009GL040000 , 2013a. 13424 13423 , 2008. erent metrics to compare indicators and to make available the ff This work was funded by the European Commission Seventh Framework , 2011. 10.5194/hess-17-2359-2013 , D. B.: The TRMM Multi-satellite Precipitation Analysis: Quasi-Global, Multi- ff 10.1029/2008GL035381 10.1002/qj.828 man, G. J., Adler, R. F., Bolvin, D. T., and Gu, G.: Improving the global precipitation record: man, G. J., Adler, R. F., Bolvin, D. T., Gu, G., Nelkin, E. J., Bowman, K. P.,Hong, Y., Stocker, ff ff Proc. Ninth Conf. on Applied Climatology, Amer. Meteor. Soc., Dallas, TX, 233–236, 1995. to time scales, in:184, Proc. 1993. 8th Conf. on Appl. Clim., 17–22 January 1993, Anaheim, CA, 179– Funk, C.: Seasonality of African2012. precipitation from 1996 to 2009, J. Climate, 25, 4304–4322, 1571–1592, 2002. rainfall in Africa for drought monitoring applications, Int. J.hydrologic Climatol., and 27, hydroclimatic 935–944, model 2007. validation, Water Resour. Res., 35, 233–241, 1999. E. F., and Wol Year, Combined-Sensor Precipitation Estimates at2007. Fine Scale, J. Hydrometeorol., 8, 38–55, GPCP Version 2.1, Geophys. Res. Lett., 36, L17808, doi: in African basins using2359–2373, doi: the Standardized Precipitation Index, Hydrol. Earth Syst. Sci., 17, tion products over very complex2010. terrain in Ethiopia, J. Appl. Meteorol. Clim., 49, 1044–1051, Sens., 28, 1503–1526, 2007. resolution satellite rainfall products4110, 2008. over complex terrain, Int. J. Remotetions Sens., 29, in 4097– doi: the characterization of regional drought, Geophys. Res. Lett., 35, L19402, Pappenberger, F.: Global meteorological drought: Partmitted I to – Hydrol. probabilistic Earth monitoring, to Syst. be Sci., sub- 2013c. index model, Adv. Water Resour., 57, 12–18, 2013. Meteorol. Soc., 83, 1149–1165, 2002. Validation of satellite rainfall products over East Africa’s complex topography, Int. J. Remote berger, F.: The 2010–2011 drought inforecast the products, Horn Int. of J. Africa Climatol., in 33, ECMWF 1720–1729, reanalysis 2013b. and seasonal Balmaseda, M. A., Balsamo,L., G., Bidlot, Bauer, J., P., Bormann,L., Bechtold, Healy, N., P., S. Beljaars, Delsol, B., A. C., Hersbach,M., Dragani, C. H., McNally, R., Hólm, M., A. E. Fuentes, van P.,nay, V., Monge-Sanz, P., M., de Tavolato, Isaksen, C., B. Geer, Thépaut, Berg, L., J.-N. A. M., Kållberg, andand J., P., Morcrette, Vitart, Köhler, performance Haimberger, M., J.-J., F.: of The Matricardi, Park, the ERA-Interimdoi: data B.-K., reanalysis: assimilation configuration Peubey, system, C., Q. de J. Ros- Roy. Meteorol. Soc., 137, 553–597, Evaluation of drought indicesthe based continental on United States, thermal J. remote Climate, sensing 24, 2025–2044, of 2011. evapotranspiration over McKee, T. B., Doesken, N. J., and Kleist, J.: Drought monitoring with multiple time scales, in: McKee, T. B., Doesken, N. J., and Kleist, J.: The relationship of drought frequency and duration Lloyd-Hughes, B. and Saunders, M. A.: A drought climatology for Europe, Int. J. Climatol., 22, Legates, D. R. and McCabe Jr.,Liebmann, G. B., Bladé, J.: I., Evaluating Kiladis, G. the N., Carvalho, use L. M. of B., Senay, G., “goodness-of-fit” Allured, D., measures Leroux, S., in and Hu Husak, G. J., Michaelsen, J., and Funk, C.: Use of the gamma distribution to represent monthly Hu Dutra, E., Di Giuseppe, F., Wetterhall, F., and Pappenberger, F.: Seasonal forecasts of droughts Hirpa, F. A., Gebremichael, M., and Hopson, T.: Evaluation of high-resolution satellite precipita- Heim Jr., R. R.: A review of twentieth-century drought indices used in the United States, B. Am. Dutra, E., Viterbo, P., and Miranda, P. M. A.: ERA-40 reanalysis hydrological applica- Hao, Z. and AghaKouchak, A.: Multivariate Standardized Drought Index: a parametric multi- Dinku, T., Chidzambwa, S., Ceccato, P.,Connor, S. J., and Ropelewski, C. F.: Validation of high- Dutra, E., Wetterhall, F., Di Giuseppe, F., Naumann, G., Barbosa, P., Vogt, J., Pozzi, W., and Dutra, E., Magnusson, L., Wetterhall, F., Cloke, H. L., Balsamo, G., Boussetta, S., and Pappen- Dinku, T., Ceccato, P., Grover-Kopec, E., Lemma, M., Connor, S. J., and Ropelewski, C. F.: Dee, D. P., Uppala, S. M., Simmons, A. J., Berrisford, P., Poli, P., Kobayashi, S., Andrae, U., Anderson, M. C., Hain, C., Wardlow, B., Pimstein, A., Mecikalski, J. R., and Kustas, W. P.: References Acknowledgements. Programme (EU FP7) in the frameworkto of the Strengthen Improved Preparedness Drought Early and WarningGrant Adaptation and Agreement Forecasting to 265454. Droughts in Gustavoportant Africa Naumann discussions (DEWFORA) thanks about project Mauricio under hydroGOF di Zambrano-Bigarini R-package. for the im- 5 5 30 25 20 15 30 25 10 20 15 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , http://CRAN. 10.1029/2004JD005182 , 2012. 13426 13425 , last access: 4 October 2013. 10.5194/nhess-12-3519-2012 hydroGOF = eld, J., Goteti, G., Wen, F., and Wood, E. F.: A simulated soil moisture based drought ffi the standardized precipitation index65–79, in 2007. arid locations and dry seasons, Int.vations, J. satellite Climatol., estimates, 27, and numerical2558, model 1997. outputs, B. Am. Meteorol. Soc., 78, 2539– observed hydrological timeR-project.org/package series, R package version 0.3-7, available at: satellite-based rainfall estimates over the338, Volta and 2013. Baro-Akobo Basin, J. Hydrol., 499, 324– R., Palecki, M., Stooksbury, D.,Meteorol. Miskus, Soc., D., 83, and 1181–1190, Stephens, 2002. S.: The Drought Monitor,satellite-based B. precipitation Am. products over sparselyeteorol., gauged 13, 1760–1783, African 2012. river basins, J. Hydrom- Press, Burlington, Massachusetts, 627 pp., 2006. indicators and applications, J. Hydrometeorol., 12, 66–83, 2011. paredness and mitigation, in: Early WarningManagement, Systems for edited Drought Preparedness by: andno. Wilhite, Drought 1037, D. World Meteorological A., Organization, Sivakumar, Geneva, Switzerland, M. 1–21, V. 2000. K., and Wood, D. A., WMO/TD analysis for the United States,2004. J. Geophys. Res., 109, D24108, doi: sitive to global warming: the23, standardized 1696–1718, precipitation 2010. evapotranspiration index, J. Climate, Kenawy, A., López-Moreno, J., Nieto, R.,Challenges Ayenew, T., for Konte, drought D., Ardö, mitigation J., ininformation and Africa: systems, Pegram, Appl. the G. Geogr., potential G.: 34, use 471–486, of 2012. geospatial data and drought 55–94, 1948. bined Drought Indicator toSci., detect 12, 3519–3531, agricultural doi: drought in Europe, Nat. Hazards Earth Syst. and required density ofedited point by: measurements, Desbois, in: M.1994. Global and Precipitation Desahmond, and F., Climate Springer-Verlag, Change, Berlin, Heidelberg, 173–186, Africa with coarse resolution352, remote 2011. sensing imagery, Remote Sens. Environ., 115.2, 343– early warning systems in Africa,Study, edited in: by: Monitoring Boken, V. and K., Predicting Cracknell,New A. Agricultural York, P.,and 252–265, Drought: Heathcote, 2005. a R. Global L., Oxford University Press, and their uncertainties in Africa2012. using TRMM data, J. Appl. Meteorol. Clim., 51, 1867–1874, Xie, P. and Arkin, P. A.: Global precipitation: a 17-year monthly analysis based on gauge obser- Zambrano-Bigiarini, M.: hydroGOF: goodness-of-fit functions for comparison of simulated and Thom, H. C.: A note on the gamma distribution, Mon. Weather Rev., 86, 117–122, 1958. Wu, H., Svoboda, M. D., Hayes, M. J., Wilhite, D. A., and Wen, F.: Appropriate application of Willmott, C. J.: On the validation of models, Phys. Geogr., 2, 184–194, 1981. Thiemig, V., Rojas, R., Zambrano-Bigiarini, M., and De Roo, A.: Hydrological evaluation of Thiemig, V., Rojas, R., Zambrano-Bigiarini, M., Levizzani, V., and De Roo, A.: Validation of Wilks, D. S.: Statistical Methods in the Atmospheric Sciences, 2nd Edn., Elsevier Academic Svoboda, M., LeComte, D., Hayes, M., Heim, R., Gleason, K., Angel, J., Rippey, B., Tinker, Wilhite, D. A. and Svoboda, M. D.: Drought early warning systems in the context of drought pre- Shukla, S., Steinemann, A. C., and Lettenmaier, D. P.:Drought monitoring for Washington State: Vicente-Serrano, S. M., Beguería, S., Gimeno, L., Eklundh, L., Giuliani, G., Weston, D., El She Vicente-Serrano, S. M., Beguería, S., and López-Moreno, J. I.: A multiscalar drought index sen- Thornthwaite, C. W.: An approach toward a rational classification of climate, Geogr. Rev., 38, Sepulcre-Canto, G., Horion, S., Singleton, A., Carrao, H., and Vogt, J.: Development of a Com- Rudolf, B., Rueth, W., and Schneider, U.: Terrestrial precipitation analysis: operational method Rowland, J., Verdin, J., Adoum, A., and Senay, G.: Drought monitoring techniques for famine Rojas, O., Vrieling, A., and Rembold, F.: Assessing drought probability for agricultural areas in Naumann, G., Barbosa, P.,Carrao, H., Singleton, A., and Vogt, J.: Monitoring drought conditions 5 5 20 25 30 15 25 10 20 15 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | NCEP/ S] 54 (56, 44) N] 36 (52, 65) N] 100 (23, 75) ◦ N] 180 (15, 85) + ◦ N] 120 (23, 70) ◦ ◦ ◦ S–20 N–35 ◦ N–17 S–12 ◦ N–18 ◦ ◦ ◦ [7 [26 [2 [31 [6 × × × × × E] E] E] E] E] ◦ ◦ ◦ ◦ ◦ W–0 W–0 E–40 E–34 E–52 ◦ ◦ ◦ ◦ ◦ SSM/I emission and MSU NCAR Reanalysis) and sounder data and precipitation gauge analyses (GPCC). CAMS and/or GPCC) 13428 13427 ) Longitude–Latitude GPCC Grid cells 2 km × 6 1979–2009 RSE (GPI, OPI,S SM/I scattering, irregular 1979–2010 RSE (merged from microwave, infrared irregular 1901–2010 In-situ data 1 month 1979–present1998–present ECMWF Reanalysis RSE (combination 3B-42, 1 or 2 months 1/2 month ◦ ◦ ◦ ◦ ◦ 0.25 ) (–present) ◦ 2.5 2.5 0.5 0.5 × 1 ◦ × × × × × ◦ ◦ ◦ ◦ ◦ Description of global datasets available in near-real time that could be used for moni- Geographical extent of the African regions and number of grid cells analysed for each RegionA – Oum er-Rbia 0.49 Area (10 [10 B – Niger 1.48 [10 C – Eastern Nile 1.23 [30 D – Limpopo 0.94 [25 E – GHA 2.22 [40 CMAP 2.5 (Combined)GPCP v.2.2 (1 2.5 GPCC v.5 0.5 TRMM 3B-43 v.6 0.25 DatasetsERA INTERIM 0.5 resolution period Source Update Table 2. toring precipitation conditions at continental level. Table 1. dataset. For GPCC, thestations percentage are of respectively shown stations between per brackets. grid and the percentage of pixels without Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 34 9.6 1.5 − 47.8 11.6 48.1 32.3 33.1 30.9 35.1 40.6 − − − − − − − − − − MAE BIAS MAE r 1 0.95 22.2 8.3 − 0.7 –0.6 0.93 43.9 – – r − − MAE BIAS MAE r r 22 – – – 0.86 43.4 2.1 0.79 12.8 10.6 – – – 3.3 0.91 8.3 1.8 0.98 8.1 8.5 – – – 0.61 22.7 − 11.5 – – – 0.79 12.8 25.315.5 – 0.6814.514.1 0.8 7.0 – 0.97 13.1 13.920.3 – 6.9 20.9 – 7.2 – 0.78 0.94 0.6822.4 25.8 0.82 23.2 – 7.0 – 15.3 8 – – 0.78 25.8 0.7 − − − − − − − − − − − MAE MAE BIAS r r ), mean absolute error (MAE) and percent bias r 13430 13429 4 – – – 0.73 6.5 33.9 0.95 5.7 18.4 − 1.8 0.79 9.9 1.9 0.98 13.6 7.7 0.95 25.8 1.9 – – – 0.854.2 14.3 0.88 28.26.6 6.6 0.977.8 – 30.1 0.7 1.7 0.72 9.6 – 9.2 11.2 – 0.84 0.7 17.8 9.6 9.3 0.92 16.4 18.719.6 0.73 0.95 6.5 5.7 23.7 0.93 17.4 18.4 0.85 14.3 − − − − − − − − − − − cient ( MAE ffi MAE BIAS ), Mean absolute error (MAE) and percent bias (%) between r TRMM GPCC GPCP ERAI r r cient ( 17 0.91 8.3 30 0.94 4.6 8.2 0.98 8.1 1.5 0.97 8.8 6.77.4 0.97 0.95 6.9 22.2 1 0.82 16.8 6.3 0.99 5.1 3.41.7 – 0.9 8.2 – – 0.79 9.9 13 0.97 8.8 2.1 8.2 – – – 0.99 5.1 2.66.2 – 0.99 2.5 – – 0.99 2.5 4.2 0.94 4.7 23.1 0.95 6.7 24.4 − − ffi 20.8 0.95 6.6 10.1 0.84 9.4 − − − − − − − − − − GPCCGPCP 0.85ERAI 0.26 0.79 0.38 0.71 – 0.91GPCC 0.5 0.29GPCP 0.54 – 0.72ERAI 0.54 0.53 – 0.46 0.55 0.91 0.6 – 0.67 0.92 0.29GPCC – 0.65 0.5 0.27 0.72GPCP 0.91 – – 0.46 0.67ERAI 0.28 0.84 0.57 – 0.65 0.39 0.41 0.92 0.8 – – 0.92 0.67 0.27GPCC – 0.46 0.27 0.46 0.57GPCP 0.58 – 0.91 0.41 0.67ERAI – 0.4 0.65 – 0.33 0.46 0.44 0.92 0.61 0.88 – – 0.86 0.27 0.44 – 0.35 0.29 0.91 0.58 – 0.33 0.88 – 0.42 – 0.35 0.68 – 0.86 0.45 0.29 – – 0.58 0.68 0.42 0.45 – GPCPERAI 0.81 0.38 0.84 0.81 0.37 0.35 0.81 0.34 – 0.74 0.5 – – 0.74 0.5 – erent SPI-3 estimations averaged over each region for the common period ff MAE BIAS TRMM GPCC GPCP CMAP ERAI r Spearman correlation coe Correlation coe Niger TRMM –Blue Nile – TRMM – 0.85 – 0.26Limpopo 0.79 TRMM 0.38 – 0.54 0.71 0.54 0.5 – 0.53GHA 0.55 0.6 0.91 TRMM 0.28 – 0.5 0.84 0.39 – 0.8 0.46 0.58 0.4 0.65 0.44 0.61 0.44 Oumer-Rbia TRMM GPCC – 0.89 0.28 – – 0.89 – 0.28 0.81 0.81 0.38 0.35 0.84 0.81 0.37 0.34 erent precipitation datasets averaged over each region for the common period 1998– ERAI 0.93 43.9 92.8 0.97ERAI 29.9 0.96 47.6 10.4 0.97 30.1 49.5 0.86 43.4 91.7 – – – ERAI 0.95GPCC 7.3 GPCP 0.99CMAP 0.98 5.8ERAI 0.8 13.6 0.94 13.8 1.9GPCC 17 23.1 –GPCP 0.94 0.99CMAP 0.93 17.6 11.5 0.82 17.4 – 16.4 15.3 31 – 28.9 – 1 0.6GPCP – 0.97CMAP 0.99 0.97 2.66 – 12.1 0.76 11.6 – 8.4 12.6 GPCC – –GPCP 0.82 0.82 1CMAP 0.88 16.7 9.8ERAI 0.72 25.4 6.6 0.84 0.95 2.7 9.2 4.4 17.8 25.8 – 1.9 51.5 26.4 0.97 0.83 12.1 – 17.1 22.5 44.7 0.97 – 0.92 29.9 0.9 16.4 54.1 8.2 0.61 22.7 7.1 0.84 68.4 – 9.4 8.4 0.83 – 17.1 – CMAP 0.74 7.8 GPCC 0.98 7.0 GPCCGPCP 0.99 0.99 2.5 2.9 ff ENL TRMM –LIM – TRMM – – 0.94 17.6 GHA – TRMM – – 0.98 7.03 – 8.9 – 0.97 0.82 8.4 9.8 6.7 0.76 12.6 20.6 0.96 10.4 9 NIG TRMM – – – 0.99 5.8 OER TRMM – – – 0.99 2.5 2.7 0.99 2.9 6.7 0.74 7.8 42.8 0.95 7.3 26.3 Table 4. (%) between the di 1998–2010. Table 3. the di 2010. All correlations are significant at 99 %. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | erent datasets for the common ff E of mean annual precipitation. ◦ ) from di 1 − 13432 13431 longitudinal cross section at 25 (F) Mean annual precipitation (mmyr (E) – Annual mean precipitation from the GPCC dataset and African regions used in this Fig. 1. analysis as defined in TableLIM: 1 Limpopo (OER: Basin Oum and er-Rbia; GHA: NIG: Greater Inner Horn Niger of Delta; Africa). ENL: Eastern Nile, Fig. 2. (A) period 1998–2010, Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | erent datasets averaged over the five ff 13434 13433 Mean annual cycle of precipitation from the di Monthly anomalies in SPI-3 (ERAI, GPCP, TRMM), SPEI (ERAI and GPCP) and Soil Fig. 4. Moisture Anomalies (SMA) forzero January contour. 2000, 2003, 2006 and 2009. Solid lines indicates the Fig. 3. regions defined in Fig.Limpopo 1 Basin (OER: and Oum GHA: er-Rbia, Greater Horn NIG: of Inner Africa) Niger for Delta, the NIL: common Eastern period 1998–2010. Nile, LIM: Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1.0. Values are − 13436 13435 ) between SPI, SPEI, SMA and PET for the five case studies d Month by month evolution of droughts in 2000, 2003, 2006 and 2009 according to Index of agreement ( ranged between 0 (no dataset withthreshold). SPI-3/SPEI-3 below the threshold) and 5 (all datasets below Fig. 6. grid cells with SPI-3/SPEI-3 computed using ERA-I GPCP, and TRMM below Fig. 5. and the whole continent.Limpopo (OER: Basin Oum and er-Rbia, GHA: NIG:centile Inner Greater of Niger Horn estimations, Delta, of boxes NIL: Africa.)within extend Eastern each Dashed box from Nile, lines indicate LIM: 25th extend the to from mean for 5th 75th each to percentile region. 95th and per- middle horizontal lines Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1.0 − scores below z 13438 13437 Duration of dry spells for the standardized indicators below zero in the common period Fractional area of each region under SPI, SPEI and SM and PET for the period 1998–2010. (OER:Limpopo Oum Basin er-Rbia, and NIG: GHA: Inner Greater Niger Horn Delta, of NIL: Africa.) Eastern Nile, LIM: Fig. 8. Fig. 7. 1998–2010. (OER: Oum er-Rbia, NIG: Innerand Niger GHA: Delta, Great NIL: Horn Eastern of Nile,boxes Africa). LIM: extend Limpopo Dashed Basin from lines extend 25th from tothe 5th 75th to mean 95th percentile for percentile and each of middle region. estimations, horizontal lines within each box indicate Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | erent erent ff ff 13440 13439 cient of fractional areas under drought between di ffi Relative bias between the estimation of fractional areas under drought for di Pearson’s Correlation coe Fig. 10. Fig. 9. datasets and thresholds. Theare horizontal considered axis to represents be the under SPI drought. threshold below which areas datasets and thresholds. Theare horizontal considered axis to represents be the under SPI drought. threshold below which areas