Version of Record: https://www.sciencedirect.

com/science/article/pii/S0022169417300653
Manuscript_12c4291b9a3b62e8a54b7a250a8bee5c

Accepted Manuscript

Research papers

Evaluating the Performance of Remotely Sensed and Reanalysed Precipitation

Data over West Africa using HBV light

Thomas Poméon, Dominik Jackisch, Bernd Diekkrüger

PII: S0022-1694(17)30065-3
DOI: http://dx.doi.org/10.1016/j.jhydrol.2017.01.055
Reference: HYDROL 21794

To appear in: Journal of Hydrology

Received Date: 19 August 2016

Revised Date: 4 January 2017
Accepted Date: 28 January 2017

Please cite this article as: Poméon, T., Jackisch, D., Diekkrüger, B., Evaluating the Performance of Remotely Sensed
and Reanalysed Precipitation Data over West Africa using HBV light, Journal of Hydrology (2017), doi: http://
dx.doi.org/10.1016/j.jhydrol.2017.01.055

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2017 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/
 Evaluating the Performance of Remotely Sensed and Reanalysed
Precipitation Data over West Africa using HBV light
Thomas Poméon1, Dominik Jackisch2, Bernd Diekkrüger1
1) Department of Geography, University of Bonn, Meckenheimer Allee 166, 53115 Bonn,
Germany
2) Earth Observatory of Singapore, Nanyang Technological University, 50 Nanyang Avenue,
639798 Singapore, Singapore

Abstract

Water is a crucial resource in West Africa, where large parts of the population rely on rainfed

agriculture. Therefore, accurate knowledge of the water resources is of the utmost importance. Due

to the declining number of rain gauging stations, the use of satellite and reanalysis precipitation

datasets in hydrological modelling is steadily rising. However, accurate information on the benefits

and deficits of these datasets is often lacking, especially in the West African subcontinent. For

validation purposes, these products are commonly compared to freely available rain gauge data,

which has in some cases already been used to bias correct the products in the first place. We

therefore explored the possibility of a hydrological evaluation, where a model is calibrated for each

dataset using streamflow as the observed variable. In this study, ten freely available satellite and

reanalysis datasets (CFSR, CHIRPS, CMORPHv1.0 CRT, CMORPHv1.0 RAW, PERSIANN CDR, RFE 2.0,

TAMSAT, TMPA 3B42v7, TMPA 3B42 RTv7 and GPCC) were thus evaluated for six differently sized

and located basins in West Africa. Results show that while performances differ, most datasets

manage to somewhat accurately predict the observed streamflow in a given basin. Best results were

achieved by datasets which use a multitude of input data, namely infrared and microwave satellite

data, as well as observations from rain gauges (usually GPCC) for bias correction. If considering only

the Nash Sutcliffe Efficiency averaged for all six basins during the calibration phase, best results were

achieved by CMORPH CRT and PERSIANN CDR (both 0.66), followed by TAMSAT, CHIRPS and TMPA

1
3B42 (all three 0.64). Average results were achieved by RFE 2.0 (0.63), GPCC (0.61) and TMPA 3B42

RT (0.54). CMORPH RAW and CFSR performed worst (0.36 and -0.34 on average).

Keywords

Precipitation Datasets, Satellite Precipitation Estimates, Hydrological Evaluation, Remote Sensing

1 Introduction

Precipitation data is one of the most important drivers of hydrological models (Hughes, 2006;

Thiemig et al., 2013). However, especially in developing countries, rain gauge (RG) networks are

sparse or non-existent and often include erroneous data or large gaps (Behrangi et al., 2011; Bitew

and Gebremichael, 2011; Koutsouris et al., 2016). This situation is aggravated by a further decline in

the number of rainfall stations due to financial or maintenance problems (Adjei et al., 2012; Hughes,

2006). A large number of satellite-based rainfall estimate (SRFE) and reanalysis (RA) products with

high spatial and temporal resolutions are freely available and have the potential to complete gaps or

even replace rain gauge measurements (Fujihara et al., 2014; Koutsouris et al., 2016; Thiemig et al.,

2013). The uncertainties of these products over western Africa are largely unknown and only recently

scientists have started validating SRFE and RA products for the sub-continent (Awange et al., 2015).

These validations are mostly conducted by using statistical measures to compare SRFEs to RG point

data.

With regard to the fact that most of the existing scientific papers only include and analyse a few

precipitation datasets (Cohen Liechti et al., 2012; Thiemig et al., 2013; Tobin and Bennett, 2014), this

study further contributes to the ongoing debate on satellite derived precipitation products’

performances by focusing on eight different state-of-the-art satellite precipitation datasets, one

primarily reanalysis product, and one rain gauge dataset.

2
SFRE and reanalysis product validation studies over West Africa were carried out, among others, by

Awange et al., 2015; Gosset et al., 2013; Nicholson et al., 2003; Pfeifroth et al., 2016 and Thiemig et

al., 2012. All of the aforementioned use at least one source of RG data in order to assess the

performance of the SRFE or RA product. A major source of RG data is the Global Precipitation

Climatology Center (GPCC) product, which has been used to validate precipitation datasets in Africa,

e.g. by Adeyewa and Nakamura (2003). However, large data gaps exist over Africa, and the number

of gauges sending data varies over time (Awange et al., 2015). Figure 1 gives an overview of GPCC

pixels including station data for at least 80 % of days for the period from 2003 to 2013. Also, gauges

connected to the Global Telecommunication System (GTS) are frequently used for bias correction of

satellite data, creating merged satellite rain gauge datasets like TMPA 3B42 or PERISANN CDR

(Ashouri et al., 2015; Huffman et al., 2007). It remains questionable whether the performance of bias

corrected SRFEs should be evaluated against data from the same rain gauges used for the bias

correction since the datasets are not independent (Awange et al., 2015; Maidment et al., 2013;

Novella and Thiaw, 2013; Thiemig et al., 2012).

A new approach which has recently gained popularity is the so called hydrological evaluation of SRFE

and RA products. In this approach, the performance of the datasets is evaluated by calibrating a

hydrological model for different precipitation products and evaluating the ability of each product to

reproduce the observed streamflow. This method on the one hand circumvents the need for reliable

RG information, but on the other hand introduces the need for streamflow information as a variable

for model calibration (Behrangi et al., 2011; Thiemig et al., 2013). Since the hydrological evaluation

takes place at the watershed scale, no problem of scale discrepancy arises as with the validation by

using point source RG data. Also, the usefulness of the SRFEs and RA products to generate

streamflow is directly evaluated within the hydrological model of choice (Bitew and Gebremichael,

2011; Thiemig et al., 2013). It should however also be considered that the model concept and

process representation introduces uncertainties into the analysis. So can the results of excess or

below average precipitation for example be dampened by the parameters governing infiltration and

3
evaporation (Seibert, 1997). Nevertheless, when using one single model concept which has been

shown to be applicable to the climatic conditions, the results may be biased but this bias would be

consistent over all precipitation products and therefore does not affect the interpretation and

evaluation.

Studies using this method in recent years are plentiful, e.g. Behrangi et al., 2011; Bitew and

Gebremichael, 2011; Bodian et al., 2016; Cohen Liechti et al., 2012; Dile and Srinivasan, 2014;

Fujihara et al., 2014 and Li et al., 2015. However, most of these studies focus on areas outside of

West Africa, where only few hydrological evaluations were conducted so far, covering a limited

number of different products and climatological regions. A need for more studies focusing on West

Africa and featuring more products is apparent.

Some of the first authors to validate SRFE and RA products in West Africa were Thiemig et al. (2013),

who evaluated CMORPH (Climate Prediction Center Morphing Technique), RFE 2.0 (African Rainfall

Estimation), TMPA 3B42 (Tropical Rainfall Measuring Mission Multi-satellite Precipitation Analysis),

PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural

Networks) and ERA-Interim over the Volta basin using the LISFLOOD model. They concluded that the

hydrological evaluation using this model is a well-suited approach to validate precipitation datasets

over sparsely or ungauged catchments, with their model producing better results in the high flow

than during the low flow periods. During this study, some products were bias corrected for a second

simulation (e.g. PERSIANN and CMORPH, which exhibited biases over lowland areas) using gauge

measurements, and generally produced better results afterwards. They concluded that the best

dataset is the one with the best intrinsic data quality needing the least amount of pre-processing and

bias correction. For the Volta basin, RFE 2.0 and TMPA 3B42 were asserted as best performing. Also

in 2013, Gosset et al. validated PERSIANN, CMORPH, TMPA 3B42 real time, GSMaP (Global Satellite

Mapping of Precipitation) MVK and real time, GPCP (Global Precipitation Climatology Project), TMPA

3B42, RFE 2.0 and EPSAT-SG (Estimation of Precipitation by SATellite Second Generation) using high

resolution gauge data from two AMMA-CATCH sites in Niger and Benin as well as two hydrological
4
models. The Soil Conservation Service method (SCS) was used to generate runoff for the Niamey site

and the lumped conceptual Modèle du Génie Rural à 4 paramètres Journalier (GR4J) was used to

generate streamflow for the Benin site. For the statistical evaluation, the authors concluded that

TMPA 3B42 performed best on both sites with very low biases. RFE 2.0 exhibited low biases but also

low correlations to gauge data with an overestimation of low rain rates. Concerning the hydrological

evaluation in the Niamey area, the two products best able to generate streamflow are TMPA 3B42

(slight overestimation) and RFE 2.0 (low bias but underestimated streamflow). The evaluation

conducted for the Ouémé area revealed RFE 2.0 and EPSAT to perform best with a slight

underestimation of the discharge. TMPA 3B42 was found to have a high inter-annual variability of the

biases and did not provide adequate results.

In 2014, Fujihara et al., using a conceptual hydrological model, assessed that a bias-corrected SRFE

dataset outperforms RA datasets over the White Volta basin. Results showed that while the

reanalysis products exhibited acceptable seasonal accuracy, annual variations were not reproduced.

The only merged SRFE/RG product used (GPCP) delivered good results and proved to be superior.

Very recently, Bodian et al. (2016) conducted a hydrological evaluation of TMPA 3B42 precipitation

estimates over the Upper Sénégal river basin, also using the GR4J model. Results showed good

agreement with measured discharge.

As stated before, the current study is a contribution to the ongoing debate on satellite derived and

reanalysis precipitation products’ performances. Specifically, we aim at exploring the differences

between the considered datasets for basins of different locations and sizes and evaluating the

robustness of the products in accurately simulating streamflow during a hydrological evaluation. Ten

state of the art datasets which were generated according to different methods were chosen for the

evaluation. For the first time, the conceptual HBV model is used in a hydrological evaluation,

generating knowledge in its applicability in exploring the different capabilities of precipitation

datasets to generate streamflow. HBV was chosen due to it being user-friendly, easily accessible and

free of charge while having been proven to deliver robust results (Seibert and Vis, 2012), in order to
5
entice a maximum of researchers to adopt this validation approach. Conceptual models like HBV

light were already successfully applied in simulating discharge in West Africa, e.g. by Cornelissen et

al., 2013.

2 Materials and Methods

2.1 Study Area

The basins used to evaluate the performance of the selected precipitation estimates in generating

daily streamflow were chosen within the COAST (Studying changes of sea level and water storage for

coastal regions in West-Africa, using satellite and terrestrial data sets) project study area in West

Africa. Large (about 100,000 km²) as well as small (~ 2,500 km²) sized basins were identified both in

the south, which is characterized by a bimodal rainy season, and the north, which is subject to a

unimodal rainy season, in order to evaluate differences arising from the diverse rainfall distributions

as well as the increase in aridity from south to north (Fink et al., 2010). The research area (5°N-15°N

and 6°W-4°E) is depicted in Figure 1. Specific basins were chosen based on the availability and

reliability of discharge data, an overview of the chosen basins is given in Table 1. Subwatersheds

within the selected watershed were delineated using the ArcSWAT model with a threshold area of at

least 2,500 km². This seems a reasonable subbasin size if one wants to model the whole of the

subcontinent, as is planned for the future. Schuol et al. applied the SWAT model to estimate the

freshwater availability in West Africa in 2008 and suggested a minimum drainage area for the

subbasins of 10,000 km² (Schuol et al., 2008).

Figure 1: Research area. Pixels of precipitation products including the subbasin centroids were used

for data extraction. Highlighted GPCC pixels have station data for at least 80 % of the period 2003-

2013

Table 1 Selected basins

6
2.2 Products

Ten precipitation products were identified for evaluation during this study. While this study focuses

on satellite precipitation datasets, the GPCC product, which relies on station data, was included since

it is often used to compare remotely sensed data to observations. We furthermore included the CFSR

reanalysis product because it is often used for SWAT simulations as data can be downloaded in SWAT

format. Although SWAT-ready CFSR is widely used by modelers in Africa in our experience, we

remain convinced that an evaluation of the data quality is necessary before use. Each product will be

briefly described in this section with an overview being given in Table 2.

CFSR

The CFSR (Climate Forecast System Reanalysis) coupled atmosphere-ocean-land system is developed

at NOAA-NCAR and provides a variety of atmospheric parameters including precipitation,

temperature, wind speed, relative humidity and radiation, available from 1979 to present (Blacutt et

al., 2015; Dile and Srinivasan, 2014; Saha et al., 2010). CFSR is based on historical and operational

archives and incorporates various data sources such as radiosondes, surface observations including

rain gauge information or data from satellite instruments (Saha et al., 2010). Furthermore,

precipitation estimates are updated every 6 hours in near real time (Fuka et al., 2013).

CHIRPS

CHIRPS (Climate Hazards Group Infrared Precipitation with Station data) is a dataset developed

especially for drought monitoring purposes. The data is generated from 1981 to present (Funk et al.,

2015). Compared to other precipitation products, its main characteristic is its very fine spatial

resolution of 0.05° (Katsanos et al., 2016). CHIRPS uses several data sources, such as the monthly

precipitation climatology CHPclim, infrared measurements from geostationary satellites, and

information from the TMPA 3B42 product. The precipitation estimates are merged with in-situ gauge

data from several archives including GTS to reduce biases (Funk et al., 2015).
7
CMORPHv1.0 RAW and CRT

CMORPH (Climate Prediction Center Morphing Technique) is developed by NOAA-CDC and is a global

precipitation analysis algorithm, with data available from 1998 to present 18 hours after real time

observation. Originally providing precipitation estimates for the period 2002 onwards, two new

versions of CMORPH have been released recently and labeled as CMORPHv1.0 RAW and CRT. The

former CMORPH dataset has been renamed to CMORPHv0.x (Koutsouris et al., 2016). Two versions

are available: CMORPH RAW derives half hourly precipitation estimates which are based on satellite

microwave (MW) and infrared (IR) information. The precipitation estimates are morphed through a

time-weighted linear interpolation (Joyce et al., 2004; Ramarohetra et al., 2013). CMORPH RAW is a

satellite-only product, since no RG information is incorporated (Joyce et al., 2004; Thiemig et al.,

2013). CMORPH CRT is similar to CMORPH RAW, but has been bias-corrected using historical and

real-time gauge data (Xie et al. 2011

(ftp://ftp.cpc.ncep.noaa.gov/precip/CMORPH_V1.0/REF/EGU_1104_Xie_bias-CMORPH.pdf)).

PERSIANN CDR

PERSIANN CDR (Precipitation Estimation from Remotely Sensed Information using Artificial Neural

Networks-Climate Data Record) is maintained by the University of California together with NOAA and

provides data from 1983 to present (Ashouri et al., 2015). Precipitation estimates are calculated from

IR and MW satellite data (Ashouri et al., 2015; Jobard et al., 2011). Unlike the PERSIANN product (Hsu

et al., 1997), which is available in near real time and solely based on satellite measurements (Awange

et al., 2015; Jobard et al., 2011; Ramarohetra et al., 2013), PERSIANN-CDR incorporates GPCP gauge

data into its precipitation estimates (Ashouri et al., 2015).

RFE 2.0

The RFE (African Rainfall Estimation) 2.0 algorithm was developed by the NOAA CPC. Daily rainfall

estimates are available from late 2000 to present (The NOAA Climate Prediction Center, 2002). The

dataset is based on IR and MW data and incorporates rain gauge information from the GTS network,

8
which is used to correct biases (Cohen Liechti et al., 2012; Thiemig et al., 2012). Up to 1000 GTS

stations are merged to derive the final precipitation product (Ramarohetra et al., 2013). The product

was originally developed for drought monitoring purposes on the African continent and is available

within 24 hours after real time (Jobard et al., 2011).

TAMSAT

TAMSAT (Tropical Applications of Meteorology using Satellite data and ground-based observations)

was developed at the University of Reading specifically for continental Africa with a spatial resolution

of 0.0375°, which makes it the finest resolution among all the datasets considered. Rainfall estimates

from TAMSAT are available from 1983 to present. The TAMSAT method uses IR imagery from

METEOSAT (Grimes et al., 1999; Maidment et al., 2014; Tarnavsky et al., 2014). Rain rates are then

calculated based on a cloud temperature threshold, which varies according to time and location

(Jobard et al., 2011). Contrary to other merged products, TAMSAT does not use GTS data but

historical data of about 4000 stations from various African agencies mostly acquired since the early

1990s (Maidment et al., 2014).

TRMM TMPA 3B42v7

The TRMM (Tropical Rainfall Measuring Mission) Multi-satellite Precipitation Analysis (TMPA)

product is produced at NASA’s Goddard Space Flight Center (Huffman et al., 2007), with TMPA

3B42v7 (hereafter named TMPA) being the latest version. Originally developed for rainfall retrievals

in the tropics, it has been extended to a quasi-global coverage. Rainfall estimates are available for

the period of 1998 to present. The TMPA algorithm integrates rainfall estimates from multiple

satellites and a variety of sensors. Passive MW data is derived from several low earth orbiting

satellite sensors additionally to satellite IR data (Jobard et al., 2011). Rain gauge observations from

the GPCC are used to bias correct the precipitation estimates (Moazami et al., 2013; Worqlul et al.,

2014).

TRMM TMPA 3B42v7 RT

9
Similar to the TMPA 3B42v7 dataset, the TMPA 3B42v7RT (Real-Time, hereafter named TMPA RT)

product is produced by NASA. It provides daily precipitation estimates from March 2000 to the

present day. The product is based on the same algorithm as TMPA 3B42v7, which combines MW and

IR estimates. Unlike 3B42v7, no rain gauge data is used for bias-correction due to the real-time

nature of the product. However, a climatological adjustment is performed (Huffman et al., 2007). Its

precipitation estimates are available nine hours after overflight (Huffman et al., 2007; Jobard et al.,

2011).

GPCC FDD

The GPCC FDD product (Global Precipitation Climatology Centre Full Data Daily 1°) developed at the

Deutscher Wetterdienst is a global land-surface precipitation dataset that covers the period from

1988 to 2013. It is based on station data from up to 29,000 gauge stations per month, including GTS

stations with automated quality control. The GPCC FDD precipitation estimates provide a temporal

resolution of one day, while the spatial resolution is 1°, making it the coarsest precipitation dataset

among the analyzed products. Interpolation through a block Kriging scheme with a global variogram

is carried out for each station, producing the daily rainfall totals (Schamm et al., 2015; Schneider et

al., 2015).

Table 2: Selected precipitation products

2.3 Extraction of Precipitation Data

The chosen SRFEs and RA rainfall estimates were acquired via various online sources, all being freely

available to the general public. With the exception of CFSR, all datasets were distributed in the

common netCDF data format. In order to extract the precipitation data for the research basins, we

followed the approach utilized by the semi-distributed, physically based Soil and Water Assessment

Tool (SWAT) hydrological model (Arnold et al., 1998), which uses weather data from the climate

station located closest to the center of each subbasin. This was accomplished by deriving the

centroid for each subbasin and extracting the data from the precipitation pixel covering this point

10
(Winchell et al., 2010). For larger basins, namely Atchérigbé, Gbasse and Lawra, precipitation data

was likewise derived for each subbasin and in a second step area-averaged to generate the data for

the whole basin. This way, the performance of the approach for single-cell as well as area-averaged

data can be evaluated. When we compared the SWAT centroid method to calculating the area-

average of all pixels in the respective basins for the TMPA dataset from 1998 to 2013, the correlation

of the daily data for both methods was high, ranging from r = 0.87 to r = 0.99 with a mean r of 0.91. A

comparison of the total rainfall sums generated by both methods revealed an r value of 0.99. The

average monthly rainfall sums were also highly correlated with an r of between 0.998 and 0.999 for

all six considered basins. However, when we considered the amount of rainy days, the basin-

averaged method included between 10.6 and 29.4% (average: 17.8%) more rainy days than the

distributed method. Therefore, the distributed approach was chosen in order to preserve a

representative account of the number of rainy days while at the same time not increasing or

decreasing total rainfall sums. This method was also chosen by Bitew and Gebremichael for basins of

299 and 1,656 km² size over complex topography in Ethiopia; Dile and Srinivasan for 5 to 30 km²

basins, also in Ethiopia and Tobin; and Bennett for basins of between 175 and 200 km² in Texas

(Bitew et al., 2012; Bitew and Gebremichael, 2011; Dile and Srinivasan, 2014; Tobin and Bennett,

2009).

2.4 Model

The hydrological evaluation was carried out using the conceptual HBV light model (Seibert, 2000).

HBV light is a development of the HBV model developed in the 1970s by the Swedish Meteorological

and Hydrological Institute (SMHI) and its current version was made available in 2010 (Seibert and Vis,

2012; Singh, 1995). Compared to other hydrological models, the relatively low demand for input data

and the limited number of parameters are strengths of the HBV model (Rientjes et al., 2013; Rusli et

al., 2015). The structure of HBV consists of four main routines (snow-, soil-, routing- and response

11
routine) and simulations of daily discharge are based on time series of precipitation and

temperature, as well as potential evapotranspiration (Seibert and Vis, 2012). Since no snowfalls occur

in the research area, the snow routine was not considered in this study. As shown in Figure 2, liquid

precipitation contributes to soil moisture and runoff depending on the actual moisture status.

Figure 2: HBV Model structure (Seibert 2000)

The remaining precipitation contributes to the soil moisture storage, which can still evaporate as long

as there is enough soil water available. This means that in the soil routine, groundwater recharge and

actual evaporation are simulated as functions of actual water storage (Seibert and Beven, 2009).

Three components, namely runoff, interflow and baseflow contribute to the discharge at the outlet

of the modelled basin (Aghakouchak and Habib, 2010). The response routine uses three linear

reservoir equations and the final routing routine uses a triangular weighting function to compute

discharge (Seibert and Beven, 2009). The Genetic Algorithm and Powell (GAP) optimization was used

to calibrate the model. The default model parameter ranges were chosen for the calibration, as they

are described to be realistic by the developer (Seibert, 2000; Uhlenbrook et al., 1999) and similar sets

have been used in a multitude of studies (Radchenko et al., 2014; Seibert, 1999, 1997). Increasing the

parameter range bears the risk to generate a good simulation by using unrealistic parameter sets.

This is especially the case when rainfall is extremely over- or underestimated and a good discharge

simulation is obtained at the expenses of a realistic simulation of evapotranspiration. Since the goal

was to compare the simulation results of the precipitation products among each other and not to

reach the absolute highest model efficiency, these ranges were accepted for every model run. The

calibrated parameters and their initial ranges are shown in Table 3.

Table 3: Parameters and their initial (default) ranges in the gap optimization runs

The model was calibrated for each precipitation product and basin separately in 10,000 GAP runs,

calibration and validation periods are supplied in Table 4. This method was applied to explore the

independence of the results from reference data (see also Behrangi et al., 2011; Bodian et al., 2016;

12
Tobin and Bennett, 2009). However, many studies evaluating the efficiency of SRFE and RA products

initially calibrate a model using observed rain gauge data (Bitew et al., 2012; Gourley et al., 2011;

Thiemig et al., 2013). Because no rain gauge data was available for this study, the only alternative

was to use GPCC data. Since some of the evaluated products were bias-corrected using GPCC data,

the question arises whether this method does not discriminate products not including GPCC data in

their estimates. The same holds true for a classical comparison between GPCC and GPCC-corrected

datasets. To test this hypothesis, the model was also initially calibrated with GPCC data for the Lawra

basin and the initial parameters were applied to the other precipitation products.

Table 4: Calibration and validation periods. A: CFSR, CHIRPS, PERSIANN CDR, TAMSAT, TMPA, GPCC;

B: CMORPH; C: RFE 2.0, TMPA RT

As the objective functions, the Nash Sutcliffe efficiency (NSE) and Percent BIAS (PBIAS) were chosen.

The NSE is calculated as shown in equation 1:

మ
∑೙
೔సభ೔
೚್ೞ
೔ೞ೔೘ 
మ (Eq. 1)
∑೙
೔సభ೔
೚್ೞ
 ೘೐ೌ೙ 

where ௜௢௕௦ is the ith observation of the variable to be evaluated, ௜௦௜௠ is the ith simulation of the

variable to be evaluated,  ௠௘௔௡ is the mean of the observed variables and  is the number of

observations. The NSE can take values of between – ∞ and 1. The performance ratings proposed by

Moriasi et al. (2007) were applied in this study. The result of a simulation is unsatisfactory if the NSE

is ≤ 0.5, satisfactory if between > 0.5 and ≤ 0.65, good if between > 0.65 and ≤ 0.75 and very good if

between > 0.75 and ≤ 1.

PBIAS expresses the deviation of the data to be evaluated in percent and is calculated according to

equation 2:

∑೙
೔సభ೔
೚್ೞ
೔ೞ೔೘ 
(Eq. 2)
∑೙
೔సభ೔
೚್ೞ


13
Where ௜௢௕௦ is the ith observation of the variable to be evaluated and ௜௦௜௠ is the ith simulation of the

variable to be evaluated. It should be noted that positive values indicate an underestimation and

negative values an overestimation (Moriasi et al., 2007).

Since no data on the potential evapotranspiration was available for the chosen basins, ETO was

calculated from the CFSR reanalysis data using the Penman-Monteith approach (Allen et al., 1998)

shown in equation 3:

వబబ
. ∆೙     ೌ
೅శమళయ మ ೞ
∆  .మ

(Eq. 3)

The parameters refer to “A hypothetical reference crop with an assumed crop height of 0.12 m, a

fixed surface resistance of 70 s m-1 and an albedo of 0.23.” (Allen et al., 1998). Where ௢ is the

reference evapotranspiration in mm/day, ∆ is the slope vapor pressure curve in kPa per °C, ௡ is the

net radiation at the crop surface in MJ/m²/day, is the soil heat flux density in MJ/m²/day, is the

psychrometric constant in kPa/°C,  is the mean daily air temperature at 2m height in °C, ଶ is the

wind speed at 2m height in m/s and ௦ ௔ is the saturation vapor pressure deficit in kPa.

Since ten precipitation products were evaluated for six basins, 60 models were calibrated and a

validation run was performed for each product and basin. In the last step of the evaluation, a ranking

system was developed in order to gain a one-glance overview of the performance of each product

using NSE and PBIAS values. Products received one point for a very good NSE, two points for a good

NSE, three points for a satisfactory NSE, and four points for an unsatisfactory NSE. The same was

applied for PBIAS with one point being given for a very low PBIAS (< 5 %), two points for a low PBIAS

(5 < 15 %), three points for medium PBIAS (15 < 30 %) and four points for a high PBIAS (> 30 %).

Products were considered as performing very good if the score was between two and three, good if

between four and five, satisfactory if between six and seven and unsatisfactory if the score was

higher than seven.

14
3 Results and Discussion

First, the GPCC Full Data Daily (FDD) 1° product gauge coverage was evaluated for the research area

(5°N-15°N and 6°W-4°E, see Figure 1). Since the number of stations reporting for each grid cell are

supplied in the GPCC dataset on a daily basis, cells with actual station data for the period of interest

can be derived, as previously shown by Nikulin et al. in 2012. The percentage of pixels with reporting

station per observation days (in percentage) over the period of 2003-2013 is shown in Figure 3. While

a maximum of 48 % of pixels include a station for 5 % of observation days, the pixel count quickly

reduces if longer timelines are considered. For 50 % of the period, only 33 % of the pixels have

reporting data included and for 80 % of days, this reduces to 23 %. The results show large areas of

irregular gauge measurements and interpolation. Therefore, the GPCC product should by no means

be considered as a high resolution gauge dataset in this region.

Figure 3: Overview of GPCC station data in percent of pixels with reporting station for the research

area of 5°N-15°N and 6°W-4°E (see Figure 1)

3.1 Comparison of Precipitation Estimates

In a preliminary data exploration, the precipitation estimates derived from all ten products for the six

basins were compared. This was conducted in two steps. In the first step, the average yearly and

monthly sums for the eleven-year period from 2003 to 2013 (the only period were data from all ten

precipitation estimates was available) were calculated for each product and basin and compared in

Figure 4.

Figure 4: Average yearly precipitation (2003-2013) for all products and basins in mm/y

Results show that in all basins except for Prestea, the average yearly sums of PERSIANN-CDR, TMPA,

CHIRPS, CMORPH CRT, RFE 2.0 and GPCC are similar. Average deviations from the mean are 0.1 % for

15
PERSIANN-CDR, +2.2 % for TMPA, -0.8 % for CHIRPS, -5.9 % for CMORPH CRT, -6.1 % for RFE 2.0 and

+1.1 % for GPCC. In the Prestea Basin, RFE 2.0 and TAMSAT estimate less precipitation than the

average. The CFSR reanalysis product generally estimates low precipitation rates in the central basins

of Aval Sani, Kaboua and Gbasse with an average of 25.8 % less than the mean. In the northernmost

basin of Lawra, the average annual precipitation is decidedly lower with a deviation from the average

of -63.9 %, the lowest estimation of all products. In the southern basins however, CFSR tends to

deliver higher precipitation amounts with values for Atchérigbé being close to the mean (-5.6 %). In

the southernmost Prestea basin, CFSR seems to overestimate the annual precipitation, deviating

from the mean by 31.1 %.

For the CMORPH RAW product annual averages seem high, with the highest yearly precipitation of all

products being reached in all basins except Prestea, where it comes second after CFSR. Deviations

from the mean are between +18.0 % in Prestea and +48.3 % in Lawra, with an average

overestimation of 35.8 %. TAMSAT generally shows below average annual precipitation, with values

for Aval Sani, Kaboua, Atchérigbé and Gbasse being similar to CFSR values. In the Lawra basin, values

are also below the average (-23.2 %) but higher than CFSR. In Prestea, TAMSAT estimates a lower

than average precipitation of -22.2 %, which is similar to RFE 2.0 but -760 mm/y lower than CFSR.

The yearly sums generated by the non-bias corrected, real time TMPA RT product are higher than the

mean estimates by an average of 16.1 % except for the Kaboua basin, where the sums are similar.

The overestimation of CMORPH RAW in this region as well as the similar performance of TMPA and

RFE 2.0 have also been observed by Thiemig et al. (2012) for the Volta basin, and Gosset et al. (2013)

for Benin and Niger. (Pfeifroth et al. (2016) and Thiemig et al. (2012) both observe an overestimation

of the precipitation by the PERSIANN product, which is not observed in this study, most probably

because the PERSIANN CDR product used is created using different input data as well as bias

adjustment.

16
When comparing the average monthly rainfall distribution for the same period, as done in Figure 5,

CHIRPS, CMORPH CRT, PERSIANN-CDR, RFE 2.0 and TMPA again perform similarly. While CFSR

predicts higher precipitation than the other products in the southern Prestea basin, it generally

predicts lower precipitation in the other basins located further to the north, the extreme case being

the Lawra basin, as well as a distinctly late onset of the rainy seasons. In the Kaboua basin, CFSR

rainfalls are very high during the beginning of the rainy season between July and August and in the

Atchérigbé basin, a peak can be observed during August. CMORPH RAW estimates very high peaks

for the first rainy season in the southern Prestea and Atchérigbé basin as well as the central Kaboua

basin, while the rainfall distribution of the Lawra, Aval Sani and Gbasse basins follows the average,

although with a clear overestimation.

Figure 5: Average monthly precipitation (2003-2013) for all products and basins in mm/y

In the Kaboua and Atchérigbé basins, CMORPH RAW predicts higher than average precipitation from

March to May. TAMSAT rainfall estimates follow the mean distribution but exhibit lower than

average peaks in all basins. TMPA RT precipitation is close to the average in the Prestea, Kaboua and

Gbasse basins but overestimates the peaks in the Lawra and Aval Sani basins. In the Atchérigbé basin,

TMPA RT predicts a peak in May, which is not observed by other products with the exception of

CMORPH RAW. GPCC generally performs very close to the mean values of all products. With the

exception of GPCC, products relying on only two out of the three possible input data sources (IR,

MW, RG), display the highest deviations from the mean, notably CFSR (RA, RG), CMORPH RAW (MW,

IR), TAMSAT (IR; RG) and TRMM RT (IR, MW).

3.2 Results of the Hydrological Evaluation

The influence of the model calibration strategy on the SFREs products’ performances for the Lawra

basin is shown in Figure 6. Simulations using GPCC-based model optimum parameters (initial

calibration) are compared to simulations using default model parameter ranges for individual SFRE

17
products (each product). The results show that if the model is calibrated using GPCC data, the

performance of products using GPCC and/or independent RG data for bias-correction (namely

CHIRPS, CMORPH CRT, PERSIANN CDR, RFE 2.0, TAMSAT and TMPA) only changes slightly as opposed

to calibrations being performed for each product. However, model efficiencies were observed to be

significantly lower for non-corrected precipitation products (CFSR, CMORPH RAW, TRMM RT)

compared to calibrations for each product individually. It can therefore be argued that an initial

calibration using a dataset that has already been used to bias-correct the evaluated datasets

introduces further bias into the analysis by discriminating non-corrected products. Therefore, the

initial calibration was rejected for this study.

Fig. 6: Initial calibration results compared to calibrations for each product in the Lawra basin

The results of the hydrological evaluation are diverse, as is visible in the NSE score diagrams (Figure

7). Trends between the products and differences between the basins are evident. For the southern

Prestea basin, none of the products yielded good simulations, the only satisfactorily performing

being PERSIANN-CDR (NSE 0.52) and TMPA RT (NSE 0.50). All other products scored unsatisfactorily,

with the worst scoring being CMORPH RAW (NSE -0.53) and CHIRPS (NSE 0.26). For the validation

phase, no product performed satisfactorily.

In the northernmost and largest Lawra basin, the overall best calibrations were achieved. Very good

NSE values of between 0.80 and 0.84 were attained by CHIRPS, CMORPH CRT, GPCC, RFE 2.0,

TAMSAT and TMPA. TMPA RT and CFSR achieved only satisfactory NSE values and CMORPH RAW

was identified as performing poorly. For the validation phase, very good simulations were achieved

by all products except CFSR and CMORPH RAW, which did not manage to simulate the streamflow

accurately. For the Aval Sani basin only CMORPH CRT achieved a very good simulation (NSE = 0.79).

CHIRPS, PERSIANN CDR and TAMSAT produced good simulations (NSE between 0.70 and 0.74), while

CMORPH RAW, GPCC, RFE 2.0 and TMPA only performed satisfactory and CFSR as well as TMPA RT

unsatisfactory. During the validation phase, very good results were observed for CHIRPS and

18
CMORPH CRT, while CMORPH RAW, GPCC, TMPA and TMPA RT produced good results. For the

Kaboua basin calibration, CHIRPS, CMORPH CRT, RFE 2.0 and TAMSAT produced good simulations

with NSE values between 0.71 and 0.73. CMORPH RAW, PERSIANN-CDR and TMPA performed

satisfactorily, while CFSR, GPCC and TMPA RT calibrated unsatisfactorily. Efficiencies are better for

the validation period with all products except CFSR performing good or very good simulations. For

the calibration of the Atchérigbé basin, good simulations were produced by CHIRPS, TMPA and TMPA

RT (NSE 0.65-0.72). While CFSR and CMORPH RAW performed unsatisfactory, all other products

performed satisfactory calibrations. For the validation phase, results differ. Here, TMPA produced a

very good (NSE = 0.79) and CMORPH CRT, GPCC and PERSIANN CDR good simulations, while CMORPH

RAW, RFE 2.0 and TAMSAT only produced satisfactory results. CFSR, CHIRPS and TMPA RT performed

unsatisfactorily. For the calibration of the Gbasse basin, TAMSAT produced a very good simulation

(NSE 0.76), while CHIRPS, CMORPH CRT, GPCC, PERSIANN CDR, RFE 2.0 and TMPA RT produced good

simulations. CMORPH RAW and TMPA performed satisfactorily and CFSR delivered an unsatisfactory

result. During the validation, almost all products performed unsatisfactorily with sometimes highly

negative NSE values (-7.89 on average), with the exception of CMORPH RAW and TMPA RT, which

produced satisfactory simulations. Considering the performances for all basins during the calibration

and validation it becomes apparent that CMORPH CRT performed best, followed by CHIRPS, TAMSAT

and TMPA. PERSIANN CDR and RFE 2.0 performed averagely while GPCC, TMPA RT, CMORPH RAW

and CFSR performed poorly.

Figure 7: NSE score for model calibration and validation period. PR: Prestea, LA: Lawra, AS: Aval Sani,

KA: Kaboua, AT: Atchérigbé, GB: Gbasse

Interestingly, discharges simulated by CMORPH CRT, TMPA, PERSIANN CDR and TMPA RT show

similar correlations and standard deviations to observed data in the model results. This might be due

to the same resolution of is 0.25° for each product on the one hand and the input data on the other,

as some of these datasets share input data from the same satellites. Where products are bias-

19
corrected, the limited number of available gauge products makes it likely that the same data is used

for multiple products.

The evaluation of the cumulative discharges and biases for the calibration periods are depicted in

Figures 8 and 9.The cumulative discharges immediately reveal inconsistencies in some of the models.

Both CMORPH CRT and RAW models underestimate discharges during certain periods in the Prestea

basin but overestimate in mid-2003 in Atchérigbé. During certain points in time CFSR underestimates

streamflow in Lawra, Aval Sani, Kaboua, Atchérigbé and Gbasse and TMPA RT underestimates

periods in Aval Sani. Interestingly, all models underestimate streamflow in Lawra. Biases for the

calibration period are generally low at below 15 %. Only in the Lawra basin we observed an

underestimation of streamflow by every model, the strongest bias being exhibited by CFSR and

TMPA RT (both > 30 %). The only product that performed well in this regard is CMORPH RAW (< 15

%). While RFE 2.0 and TMPA overestimated in the Aval Sani basin, CHIRPS overestimated in Prestea.

For the validation phase, higher biases were exhibited by all models. Especially in the Gbasse basin,

all models highly overestimated discharge by over 30 % with the exception of CMORPH CRT and

TMPA RT. Best performances were achieved in the Lawra, Aval Sani and Atchérigbé basins with only

two models each being highly biased (CFSR in Lawra and Aval Sani, TMPA RT in Lawra and Atchérigbé,

PERSIANN CDR in Aval Sani and CHIRPS in Atchérigbé). While most models exhibited less than 30 %

bias in Prestea and Kaboua, high biases were observed for CMORPH RAW, GPCC, TAMSAT and TMPA

in Prestea and CFSR, GPCC, PERSIANN CDR and TAMSAT in Kaboua. For both the calibration and

validation phase, the models exhibiting the least bias are CMORPH CRT, TMPA, CMORPH RAW,

TAMSAT and TMPA RT.

Fig. 8: Cumulative yearly discharges for the calibration period in mm

Fig. 9: Percent bias for model calibration and validation period. PR: Prestea, LA: Lawra, AN: Aval Sani,

KA: Kaboua, AT: Atchérigbé, GB: Gbasse

20
When we calculated the scores for each product based on NSE and PBIAS during the calibration and

validation period as shown in Figure 10, best results were achieved by the CMORPH CRT product,

followed by TAMSAT, TMPA and CHIRPS. RFE 2.0, PERSIANN CDR, TMPA RT and GPCC which scored

well to average and CMORPH RAW as well as TMPA RT which scored unsatisfactory. During

calibration, best performances were reached in the Gbasse, Kaboua and Atchérigbé basins, followed

by Aval Sani, Presta and Lawra. Interestingly, if NSE and PBIAS values were weighted equally, no

product performed unsatisfactory during the calibration. It follows that products with unsatisfactory

NSE values show low biases and vice-versa. For the validation period, results differ. Here the worst

scores are exhibited for the Gbasse and Prestea basins. This may be due to the short validation

periods and missing discharge data. Nevertheless, the CMORPH CRT product scored well in Prestea

and satisfactory in Gbasse. The best validations were achieved for Lawra, Kaboua, Aval Sani and

Atchérigbé with CMORPH CRT being the only product consistently scoring very well, followed by

TMPA with only good or very good scores. CFSR performed worst with only satisfactory scores in

Prestea and Atchérigbé and unsatisfactory performances in all other basins.

Similar results were achieved by Thiemig et al. (2013), with RFE 2.0 and TMPA performing best over

the Volta basin. While the lower performance of their PERSIANN product might be due to the fact

that the PERSIANN CDR product was created using different input data, it is interesting to note that

Thiemig et al. observed high biases for the CMORPH product which we were also able to reproduce

with the RAW product. Gosset et al. (2013) also produced comparable results with RFE 2.0 and TMPA

scoring best in Benin and Niger with good model performances and low biases, followed by CMORPH

RAW which was biased and performed lower over the Volta catchment. Dile and Srinivasan (2014)

discovered high over- and underestimations of CFSR for several subbasins of the Blue Nile Basin.

Interestingly, the two products which rely largely on ground data, CFSR and GPCC, produce among

the worst results. This may be due to the scarcity of rain gauges in West Africa and the subsequent

large scale interpolation of the data. Since the resolution of these two products is the coarsest, this

may also influence the model, e.g. in regions where multiple subbasins fall within the same pixel.

21
Products relying on a combination of infrared and microwave satellite instruments as well as rain

gauges for bias-correction, such as CMORPH CRT, TMPA, CHIRPS, TAMSAT (uses only infrared and

rain gauge data), RFE 2.0 and PERSIANN CDR score the highest. However, neither satellite only- nor

ground only products, such as TMPA RT and CMORPH RAW (both using only infrared and microwave

satellite instruments), as well as GPCC (ground-only), are able to produce results of similar quality.

The CFSR reanalysis product scored worst of all considered products. While the analyzed products

feature different spatial resolutions, we were unable to remark any impact on the products’

performance. Although CFSR and GPCC have the largest pixel sizes and worst results, especially if

considering the rain gauge density in West Africa we conclude that the lack of data and subsequent

large interpolation will have a greater influence than the size of the pixel. Overall the results show

that especially in the West African region, which is characterized by few reporting rain gauges but

also a relatively flat topography, combined satellite-gauge precipitation products prove

advantageous over single-source products.

Fig. 10: Calibration and validation score calculated based on performance of NSE and PBIAS criteria.

PR: Prestea, LA: Lawra, AS: Aval Sani, KA: Kaboua, AT: Atchérigbé, GB: Gbasse

Uncertainties may be introduced into an analysis by the choice of the hydrological model and

calibrating that model for discharge. Especially parameters governing infiltration and evaporation

may dampen the effects of over- or underestimated precipitation. It is therefore important to select

realistic ranges for these parameters before the calibration. These parameters can only influence

results in streamflow sums but not in daily distribution. This can explain the fact that CFSR, CMORPH

RAW and TMPA RT show low biases in streamflow although the products either over- or

underestimate precipitation. When analyzing the NSE however it can be seen that the products

deliver weaker results. The model may also be influenced by the potential evapotranspiration inputs,

where CFSR estimates were used, but since the same product was used for every model, it should

not influence the comparison between the precipitation datasets. The length of the time series also

has a certain influence on the performance of the satellite estimates using the hydrological
22
evaluation approach, especially if it is too short for proper calibration and validation of the model.

Thiemig et al. (2013) emphasized that longer time periods would obtain results that are more robust,

as they only analyzed data over five years. This could be of concern for some of the basins included in

this study, e.g. Gbasse, as the time series records are relatively short to sufficiently calibrate and

validate the HBV model.

4 Conclusion

The evaluation of remotely sensed and reanalysed precipitation datasets is invaluable in

undergauged regions. Especially if considering the challenges for Africa that arise due to climate

change and intensifying rainfall variabilities, these datasets can provide valuable up to date

information beneficial to decision makers. In search of a method to validate precipitation products in

the notoriously undergauged West African subcontinent, ten datasets were evaluated, including

satellite estimates, reanalysis data and one gauge product. We furthermore placed emphasis on

comparing real time, post processed, global and regional products in six basins of different location

and size as well as using two sampling strategies. Also, it was evaluated whether results change

significantly if the HBV model is initially calibrated or calibrated for each rainfall product separately.

The calibration of the model for each product seems the most sensible approach, since the possible

introduction of bias during an initial calibration influences the results on the expense of unrealistic

evapotranspiration values. Results showed that satellite products which use MW, IR and RG data

generally produce good to very good simulations. Best calibrations in terms of NSE were achieved by

CMORPH CRT, PERSIANN CDR, TAMSAT, CHIRPS and TMPA, all of which show significant potential as

an alternative to RG data. RFE 2.0, GPCC and TMPA RT performed satisfactorily, while CMORPH RAW

and CFSR deliver the least robust simulations. Interestingly, CFSR predicted very low precipitation

rates in the northern basins and high rates in the south. This was not observed by any other product

and leads to questioning the intrinsic data quality of the set for this region. The results show that the

23
best results can be achieved using bias-corrected satellite products, while satellite only or gauge only

products deliver less robust simulations. Many hydrological evaluations of satellite and reanalysis

precipitation data have employed the more complex, physically based SWAT model (Dile and

Srinivasan, 2014; Tobin and Bennett, 2014). In this study, the easy to learn conceptual HBV light

model has proven to be able to generate robust results. Nevertheless, uncertainties remain due to

possible inaccuracies in the discharge data used and uncertainties in the model structure. It can be

expected that, with longer and more complete discharge time series, calibration and validation

results can be improved.

Acknowledgements

This study is part of the COAST project (Studying changes of sea level and water storage for coastal

regions in West-Africa using satellite and terrestrial data sets) of the University of Bonn, supported

by the Deutsche Forschungsgemeinschaft under Grant No. DI 443/6-1.

We are grateful to Dr. Christophe Peugeot and the AMMA-Catch project as well as to the Global

Runoff Data Centre in Koblenz, Germany for providing discharge data.

The AMMA-CATCH regional observing system was set up thanks to an incentive funding of the

French Ministry of Research that allowed pooling together various preexisting small scale

observing setups. The continuity and long term perenity of the measurements are made possible by

an undisrupted IRD funding since 1990 and by a continuous CNRS-INSU funding since 2005.

The Precipitation – PERSIANN CDR used in this study was acquired from NOAA's National Centers for

Environmental Information (http://www.ncdc.noaa.gov). This CDR was originally developed by

Soroosh Sorooshian and colleagues for NOAA's CDR Program.

24
Literature

Adeyewa, Z.D., Nakamura, K., 2003. Validation of TRMM Radar Rainfall Data over Major Climatic

Regions in Africa. J. Appl. Meteorol. 42, 331–347. doi:10.1175/1520-

0450(2003)042<0331:VOTRRD>2.0.CO;2

Adjei, K.A., Ren, L., Appiah-Adjei, E.K., Kankam-Yeboah, K., Agyapong, A.A., 2012. Validation of TRMM

Data in the Black Volta Basin of Ghana. J. Hydrol. Eng. 17, 647–654. doi:10.1061/(ASCE)HE.1943-

5584.0000487

Aghakouchak, A., Habib, E., 2010. Application of a Conceptual Hydrologic Model in Teaching

Hydrologic Processes. Int. J. Eng. Educ. 26, 1–11. doi:0949-149X/91

Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration: Guidelines for

computing crop requirements. Irrig. Drain. Pap. No. 56, FAO 300. doi:10.1016/j.eja.2010.12.001

Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologic modeling and

assessment Part 1: model development. J. Am. Water Resour. Assoc. 34, 73–89.

doi:10.1111/j.1752-1688.1998.tb05961.x

Ashouri, H., Hsu, K.L., Sorooshian, S., Braithwaite, D.K., Knapp, K.R., Cecil, L.D., Nelson, B.R., Prat,

O.P., 2015. PERSIANN-CDR: Daily precipitation climate data record from multisatellite

observations for hydrological and climate studies. Bull. Am. Meteorol. Soc. 96, 69–83.

doi:10.1175/BAMS-D-13-00068.1

Awange, J.L., Ferreira, V.G., Forootan, E., Khandu, Andam-Akorful, S. a., Agutu, N.O., He, X.F., 2015.

Uncertainties in remotely sensed precipitation data over Africa. Int. J. Climatol.

doi:10.1002/joc.4346

Behrangi, A., Khakbaz, B., Jaw, T.C., AghaKouchak, A., Hsu, K., Sorooshian, S., 2011. Hydrologic

evaluation of satellite precipitation products over a mid-size basin. J. Hydrol. 397, 225–237.

doi:10.1016/j.jhydrol.2010.11.043

25
Bitew, M.M., Gebremichael, M., 2011. Assessment of satellite rainfall products for streamflow

simulation in medium watersheds of the Ethiopian highlands. Hydrol. Earth Syst. Sci. 15, 1147–

1155. doi:10.5194/hess-15-1147-2011

Bitew, M.M., Gebremichael, M., Gebremichael, L.T., Bayissa, Y.A., 2012. Evaluation of High-

Resolution Satellite Rainfall Products through Streamflow Simulation in a Hydrological Modeling

of a Small Mountainous Watershed in Ethiopia. J. Hydrometeorol. 13, 338–350.

doi:10.1175/2011JHM1292.1

Blacutt, L.A., Herdies, D.L., de Gonçalves, L.G.G., Vila, D.A., Andrade, M., 2015. Precipitation

Comparison for the CFSR, MERRA, TRMM3B42 and Combined Scheme datasets in Bolivia.

Atmos. Res. 163, 117–131. doi:10.1016/j.atmosres.2015.02.002

Bodian, A., Dezetter, A., Deme, A., Diop, L., 2016. Hydrological Evaluation of TRMM Rainfall over the

Upper Senegal River Basin. Hydrology 3, 1–18. doi:10.3390/hydrology3020015

Cohen Liechti, T., Matos, J.P., Boillat, J.L., Schleiss, A.J., 2012. Comparison and evaluation of satellite

derived precipitation products for hydrological modeling of the Zambezi River Basin. Hydrol.

Earth Syst. Sci. 16, 489–500. doi:10.5194/hess-16-489-2012

Cornelissen, T., Diekkrüger, B., Giertz, S., 2013. A comparison of hydrological models for assessing the

impact of land use and climate change on discharge in a tropical catchment. J. Hydrol. 498, 221–

236. doi:10.1016/j.jhydrol.2013.06.016

Dile, Y.T., Srinivasan, R., 2014. Evaluation of CFSR climate data for hydrologic prediction in data-

scarce watersheds: an application in the Blue Nile River Basin. JAWRA J. Am. Water Resour.

Assoc. 77845, 1–16. doi:10.1111/jawr.12182

Fink, A.H., Christoph, M., Born, K., Bruecher, T., Piecha, K., Pohle, S., Schulz, O., Ermert, V., 2010.

Climate, in: Speth, P., Christoph, M., Diekkrüger, B. (Eds.), Impacts of Global Change on the

Hydrological Cycle in West and Northwest Africa. Springer, Heidelberg, pp. 54–58.

26
Fujihara, Y., Yamamoto, Y., Tsujimoto, Y., Sakagami, J., 2014. Discharge Simulation in a Data-Scarce

Basin Using Reanalysis and Global Precipitation Data : A Case Study of the White Volta Basin. J.

Water Resour. Prot. 1316–1325.

Fuka, D.R., Walter, M.T., Macalister, C., Degaetano, A.T., Steenhuis, T.S., Easton, Z.M., 2013. Using

the Climate Forecast System Reanalysis as weather input data for watershed models. Hydrol.

Process. doi:10.1002/hyp.10073

Funk, C., Peterson, P., Landsfeld, M., Pedreros, D., Verdin, J., Shukla, S., Husak, G., Rowland, J.,

Harrison, L., Hoell, A., Michaelsen, J., 2015. The climate hazards infrared precipitation with

stations—a new environmental record for monitoring extremes. Sci. Data 2, 150066.

doi:10.1038/sdata.2015.66

Gosset, M., Viarre, J., Quantin, G., Alcoba, M., 2013. Evaluation of several rainfall products used for

hydrological applications over West Africa using two high-resolution gauge networks. Q. J. R.

Meteorol. Soc. 139, 923–940. doi:10.1002/qj.2130

Gourley, J.J., Hong, Y., Flamig, Z.L., Wang, J., Vergara, H., Anagnostou, E.N., 2011. Hydrologic

Evaluation of Rainfall Estimates from Radar, Satellite, Gauge, and Combinations on Ft. Cobb

Basin, Oklahoma. J. Hydrometeorol. 12, 973–988. doi:10.1175/2011JHM1287.1

Grimes, D.I.F., Pardo-Igúzquiza, E., Bonifacio, R., 1999. Optimal areal rainfall estimation using

raingauges and satellite data. J. Hydrol. 222, 93–108. doi:10.1016/S0022-1694(99)00092-X

Hsu, K., Gao, X., Sorooshian, S., Gupta, H. V., 1997. Precipitation Estimation from Remotely Sensed

Information Using Artificial Neural Networks. J. Appl. Meteorol. 36, 1176–1190.

doi:10.1175/1520-0450(1997)036<1176:PEFRSI>2.0.CO;2

Huffman, G.J., Bolvin, D.T., Nelkin, E.J., Wolff, D.B., Adler, R.F., Gu, G., Hong, Y., Bowman, K.P.,

Stocker, E.F., 2007. The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-Global,

Multiyear, Combined-Sensor Precipitation Estimates at Fine Scales. J. Hydrometeorol. 8, 38–55.

27
 doi:10.1175/JHM560.1

Hughes, D.A., 2006. Comparison of satellite rainfall data with observations from gauging station

networks. J. Hydrol. 327, 399–410. doi:10.1016/j.jhydrol.2005.11.041

Jobard, I., Chopin, F., Berges, J.C., Roca, R., 2011. An intercomparison of 10-day satellite precipitation

products during West African monsoon. Int. J. Remote Sens. 32, 2353–2376.

doi:10.1080/01431161003698286

Joyce, R.J., Janowiak, J.E., Arkin, P.A., Xie, P., 2004. CMORPH: A Method that Produces Global

Precipitation Estimates from Passive Microwave and Infrared Data at High Spatial and Temporal

Resolution. J. Hydrometeorol. 5, 487–503. doi:10.1175/1525-

7541(2004)005<0487:CAMTPG>2.0.CO;2

Katsanos, D., Retalis, A., Michaelides, S., 2016. Validation of a high-resolution precipitation database

(CHIRPS) over Cyprus for a 30-year period. Atmos. Res. 169, 459–464.

doi:10.1016/j.atmosres.2015.05.015

Koutsouris, A.J., Chen, D., Lyon, S.W., 2016. Comparing global precipitation data sets in eastern

Africa: A case study of Kilombero Valley, Tanzania. Int. J. Climatol. 36, 2000–2014.

doi:10.1002/joc.4476

Li, D., Ding, X., Wu, J., 2015. Simulating the regional water balance through hydrological model based

on TRMM satellite rainfall data. Hydrol. Earth Syst. Sci. Discuss. 12, 2497–2525.

doi:10.5194/hessd-12-2497-2015

Maidment, R.I., Grimes, D., Allan, R.P., Tarnavsky, E., Stringer, M., Hewison, T., Roebeling, R., Black,

E., 2014. The 30 year TAMSAT African Rainfall Climatology And Time series (TARCAT) data set. J.

Geophys. Res. Atmos. 119, 10619–10644. doi:10.1002/2014JD021927.Received

Maidment, R.I., Grimes, D.I.F., Allan, R.P., Greatrex, H., Rojas, O., Leo, O., 2013. Evaluation of

satellite-based and model re-analysis rainfall estimates for Uganda. Meteorol. Appl. 20, 308–

28
 317. doi:10.1002/met.1283

Moazami, S., Golian, S., Kavianpour, M.R., Hong, Y., 2013. Comparison of PERSIANN and V7 TRMM

Multi-satellite Precipitation Analysis (TMPA) products with rain gauge data over Iran. Int. J.

Remote Sens. 34, 8156–8171. doi:10.1080/01431161.2013.833360

Moriasi, D.N., Arnold, J.G., Liew, M.W. Van, Bingner, R.L., Harmel, R.D., Veith, T.L., 2007. Model

evaluation guidelines for systematic quantification of accuracy in watershed simulations. Trans.

ASABE 2007 50, 885–900.

Nicholson, S.E., Some, B., McCollum, J., Nelkin, E., Klotter, D., Berte, Y., Diallo, B.M., Gaye, I.,

Kpabeba, G., Ndiaye, O., Noukpozounkou, J.N., Tanu, M.M., Thiam, A., Toure, A.A., Traore, A.K.,

2003. Validation of TRMM and Other Rainfall Estimates with a High-Density Gauge Dataset for

West Africa. Part I: Validation of GPCC Rainfall Product and Pre-TRMM Satellite and Blended

Products. J. Appl. Meteorol. 42, 1355–1368. doi:10.1175/1520-

0450(2003)042<1355:VOTAOR>2.0.CO;2

Nikulin, G., Jones, C., Giorgi, F., Asrar, G., Büchner, M., Cerezo-Mota, R., Christensen, O.B., Déqué, M.,

Fernandez, J., Hänsler, A., van Meijgaard, E., Samuelsson, P., Sylla, M.B., Sushama, L., 2012.

Precipitation climatology in an ensemble of CORDEX-Africa regional climate simulations. J. Clim.

25, 6057–6078. doi:10.1175/JCLI-D-11-00375.1

Novella, N.S., Thiaw, W.M., 2013. African rainfall climatology version 2 for famine early warning

systems. J. Appl. Meteorol. Climatol. 52, 588–606. doi:10.1175/JAMC-D-11-0238.1

Pfeifroth, U., Trentmann, J., Fink, A.H., Ahrens, B., 2016. Evaluating satellite-based diurnal cycles of

precipitation in the African tropics. J. Appl. Meteorol. Climatol. 23–39. doi:10.1175/JAMC-D-15-

0065.1

Radchenko, I., Breuer, L., Forkutsa, I., Frede, H.-G., 2014. Simulating Water Resource Availability

under Data Scarcity—A Case Study for the Ferghana Valley (Central Asia). Water 6, 3270–3299.

29
 doi:10.3390/w6113270

Ramarohetra, J., Sultan, B., Baron, C., Gaiser, T., Gosset, M., 2013. How satellite rainfall estimate

errors may impact rainfed cereal yield simulation in West Africa. Agric. For. Meteorol. 180, 118–

131. doi:10.1016/j.agrformet.2013.05.010

Rientjes, T.H.M., Muthuwatta, L.P., Bos, M.G., Booij, M.J., Bhatti, H.A., 2013. Multi-variable

calibration of a semi-distributed hydrological model using streamflow data and satellite-based

evapotranspiration. J. Hydrol. 505, 276–290. doi:10.1016/j.jhydrol.2013.10.006

Rusli, S.R., Yudianto, D., Liu, J., 2015. Effects of temporal variability on HBV model calibration. Water

Sci. Eng. 8, 291–300. doi:10.1016/j.wse.2015.12.002

Saha, S., Moorthi, S., Pan, H.L., Wu, X., Wang, J., Nadiga, S., Tripp, P., Kistler, R., Woollen, J.,

Behringer, D., Liu, H., Stokes, D., Grumbine, R., Gayno, G., Wang, J., Hou, Y.T., Chuang, H.Y.,

Juang, H.M.H., Sela, J., Iredell, M., Treadon, R., Kleist, D., Van Delst, P., Keyser, D., Derber, J., Ek,

M., Meng, J., Wei, H., Yang, R., Lord, S., Van Den Dool, H., Kumar, A., Wang, W., Long, C.,

Chelliah, M., Xue, Y., Huang, B., Schemm, J.K., Ebisuzaki, W., Lin, R., Xie, P., Chen, M., Zhou, S.,

Higgins, W., Zou, C.Z., Liu, Q., Chen, Y., Han, Y., Cucurull, L., Reynolds, R.W., Rutledge, G.,

Goldberg, M., 2010. The NCEP climate forecast system reanalysis. Bull. Am. Meteorol. Soc. 91,

1015–1057. doi:10.1175/2010BAMS3001.1

Schamm, K., Ziese, M., Raykova, K., Becker, A., Finger, P., Meyer-Christoffer, A., Schneider, U., 2015.

GPCC Full Data Daily Version 1.0 at 1.0°: Daily Land-Surface Precipitation from Rain-Gauges built

on GTS-based and Historic Data [WWW Document]. URL

ftp://ftp.dwd.de/pub/data/gpcc/html/fulldata-daily_v1_doi_download.html (accessed 6.7.16).

Schneider, U., Ziese, M., Becker, A., Meyer-Christoffer, A., Finger, P., 2015. Global Precipitation

Analysis Products of the GPCC [WWW Document]. Glob. Precip. Climatol. Cent. URL

ftp://ftp.dwd.de/pub/data/gpcc/PDF/GPCC_intro_products_2008.pdf (accessed 1.1.16).

30
Schuol, J., Abbaspour, K.C., Yang, H., Srinivasan, R., Zehnder, A.J.B., 2008. Modeling blue and green

water availability in Africa. Water Resour. Res. 44, 1–18. doi:10.1029/2007WR006609

Seibert, J., 2000. Multi-criteria calibration of a conceptual runoff model using a genetic algorithm.

Hydrol. Earth Syst. Sci. 4, 215–224. doi:10.5194/hess-4-215-2000

Seibert, J., 1999. Regionalisation of parameters for a conceptual rainfall-runoff model. Agric. For.

Meteorol. 98–99, 279–293. doi:10.1016/S0168-1923(99)00105-7

Seibert, J., 1997. Estimation of Parameter Uncertainty in the HBV Model. Nord. Hydrol. 28, 247–262.

doi:10.2166/nh.1997.015

Seibert, J., Beven, K.J., 2009. Gauging the ungauged basin: how many discharge measurements are

needed? doi:10.5194/hessd-6-2275-2009

Seibert, J., Vis, M.J.P., 2012. Teaching hydrological modeling with a user-friendly catchment-runoff-

model software package. Hydrol. Earth Syst. Sci. 16, 3315–3325. doi:10.5194/hess-16-3315-

2012

Singh, V.P., 1995. Computer models of watershed hydrology. Water Resources Publications, Baton

Rouge.

Tarnavsky, E., Grimes, D., Maidment, R., Black, E., Allan, R.P., Stringer, M., Chadwick, R., Kayitakire, F.,

2014. Extension of the TAMSAT satellite-based rainfall monitoring over Africa and from 1983 to

present. J. Appl. Meteorol. Climatol. 53, 2805–2822. doi:10.1175/JAMC-D-14-0016.1

The NOAA Climate Prediction Center, 2002. African Rainfall Estimation Algorithm Version 2.0.

Thiemig, V., Rojas, R., Zambrano-Bigiarini, M., De Roo, A., 2013. Hydrological evaluation of satellite-

based rainfall estimates over the Volta and Baro-Akobo Basin. J. Hydrol. 499, 324–338.

doi:10.1016/j.jhydrol.2013.07.012

Thiemig, V., Rojas, R., Zambrano-Bigiarini, M., Levizzani, V., De Roo, A., 2012. Validation of Satellite-

31
 Based Precipitation Products Over Sparsely-Gauged African River Basins. J. Hydrometeorol.

120713072538009. doi:10.1175/JHM-D-12-032.1

Tobin, K.J., Bennett, M.E., 2014. Satellite precipitation products and hydrologic applications. Water

Int. 1–21. doi:10.1080/02508060.2013.870423

Tobin, K.J., Bennett, M.E., 2009. Using SWAT to model streamflow in two river basins with ground

and satellite precipitation data. J. Am. Water Resour. Assoc. 45, 253–271. doi:10.1111/j.1752-

1688.2008.00276.x

Uhlenbrook, S., Seibert, J., Leibundgut, C., Rodhe, A., 1999. Prediction uncertainty of conceptual

rainfall-runoff models caused by problems in identifying model parameters and structure.

Hydrol. Sci. J. 44, 779–797. doi:10.1080/02626669909492273

Winchell, M., Srinivasan, R., Di Luzio, M., Arnold, J., 2010. ArcSWAT Interface for SWAT 2009,

Database. USDA, Temple.

Worqlul, A.W., Maathuis, B., Adem, A.A., Demissie, S.S., Langan, S., Steenhuis, T.S., 2014. Comparison

of TRMM, MPEG and CFSR rainfall estimation with the ground observed data for the Lake Tana

Basin, Ethiopia. Hydrol. Earth Syst. Sci. Discuss. 11, 8013–8038. doi:10.5194/hessd-11-8013-

2014

32
Figure 1: Research area. Pixels of precipitation products including the subbasin centroids
were used for data extraction. Highlighted GPCC pixels have station data for at least 80 % of
days of the period 2003-2013

Figure 2: HBV model structure (Seibert 2000)

Figure 3: Overview of GPCC station data in percent of pixels with reporting station for the
research area of 5°N-15°N and 6°W-4°E (see Figure 1)

Figure 4: Average yearly precipitation (2003-2013) for all products and basins in mm/y

Figure 5: Average monthly precipitation (2003-2013) for all products and basins in mm/y

Figure 6: Initial calibration results compared to calibrations for each product in the Lawra
basin

Figure 7: NSE score for model calibration and validation period. PR: Prestea, LA: Lawra, AS:
Aval Sani, KA: Kaboua, AT: Atchérigbé, GB: Gbasse

Figure 8: Cumulative yearly discharges for the calibration period in mm

Figure 9: Percent bias for model calibration and validation period. PR: Prestea, LA: Lawra, AS:
Aval Sani, KA: Kaboua, AT: Atchérigbé, GB: Gbasse

Figure 10: Calibration and validation score calculated based on performance of NSE and
PBIAS criteria. PR: Prestea, LA: Lawra, AS: Aval Sani, KA: Kaboua, AT: Atchérigbé, GB: Gbasse

33
Table 1: Selected basins

Basin River Country Area Basin Discharge

observations
Prestea (PR) Ankobra Ghana 4,305 km² Ankobra 1998-2006
Lawra (LA) Black Volta Burkina Faso, 106,074 km² Volta 1998-2006
Ghana, Mali
Aval Sani (AS) Ouémé Benin 2,594 km² Ouémé 1999-2011
Kaboua (KA) Okpara Benin, Nigeria 9,714 km² Ouémé 1998-2010
Atchérigbé (AT) Zou Benin, Togo 7,115 km² Ouémé 2001-2010
Gbasse (GB) Sota Benin 8,504 km² Niger 2003-2006

34
Table 2: Selected precipitation products

Product Instrument Spatial Spatial Temp. Provider Reference

Coverage Resolution Coverage
CFSR RA, RG 90°N-90°S 0.3125° 1979-2014 NOAA-NCEP Saha et al.,
2010
CHIRPS MW,IR,RG 50°N-50°S 0.05° 1981-pres. NOAA-CPC Funk et al.,
2015
CMORPH v1 RAW MW,IR 60°N-60°S 0.25° 1998-pres. NOAA-CPC Joyce et al.,
2004
CMORPH v1 CRT MW,IR,RG 60°N-60°S 0.25° 1998--pres. NOAA-CPC Joyce et al.,
2004; Xie et al.
2011
PERSIANN-CDR MW,IR,RG 60°N-60°S 0.25° 1983-pres. NOAA-NCDC Ashouri et al.,
2015
RFE 2.0 MW,IR,RG AFRICA 0.1° 2000-pres. NOAA-CPC The NOAA
Climate
Prediction
Center, 2002
TAMSAT IR,RG AFRICA 0.0375° 1983-pres. University of Maidment et
Reading al., 2014;
Tarnavsky et
al., 2014
TMPA 3B42 v7 MW,IR,RG 50°N-50°S 0.25° 1998-pres. NASA/JAXA Huffman et al.,
2007
TMPA 3B42RT v7 MW,IR 60°N-60°S 0.25° 2000-pres. NASA/JAXA Huffman et al.,
2007
GPCC FDDv1 RG 90°N-90°S 1° 1988-2013 DWD Schamm et al.,
2015

RA: Reanalysis, MW: Microwave imager, IR: Infrared, RG: Rain gauges (used for bias correction), pres: present,
FDDv1: Full Data Daily Version 1

35
Table 3: Parameters and their initial (default) ranges in the gap optimization runs

Parameter Description Unit Minimum Maximum

Soil Moisture Routine
FC Maximum soil moisture mm 100 550
storage
LP Soil moisture threshold for mm 0.3 1
reduction of evaporation
BETA Shape coefficient 1 5
Response Routine
PERC Maximum flow from mm d-1 0 4
upper to lower
groundwater box
UZL Threshold parameter for 0 70
K0 outflow
K0 Recession coefficient d-1 0.1 0.5
K1 Recession coefficient d-1 0.01 0.2
K2 Recession coefficient d-1 0.00005 0.1
Routing Routine
MAXBAS Routing, length of d 1 2.5
weighting function

36
Table 4: Calibration and validation periods

Prestea/Lawra Aval Sani Kaboua Atchérigbé Gbasse

Warm-up 2001 2001-2002 2001 2001-2002 2003
Calibration 2002-2005 2003-2008 2002-2008 2003-2007 2004-2005
Validation 2006 2009-2011 2009-2010 2008-2010 2006

37
Figure2
 Percent of pixels (1°*1°) with at least
Figure3

one reporting station

20
60

40

2003-2013
Percent of days in the period
Figure4
Figure5
Figure6

1.00

0.75

0.50
NSE

0.25

0.00

-4.29 -21.12 -14.87

-0.25

Initial calibration Each Product

Figure7
Figure8
Figure9
Figure10
Highlights:

Hydrological evaluation of precipitation datasets over West Africa using HBV light

Analysis of performance over large and small basins with point source / averaged data

Evaluation of performance in areas with unimodal and bimodal precipitation patterns

Good performance of datasets combining multiple sources

38