Characterizing the performances of different observational precipitation products and their uncertainties over Africa

Validation of observed gridded precipitation datasets sourced from satellites or reanalysis over Africa remains a challenge due to the dearth of in-situ products that can act as a true estimate. To address this gap, this study compares the performance of different precipitation products (gauge, reanalysis, and satellite-based) sourced from the Frequent Rainfall Observations on GridS (FROGS) database over Africa. Satellite products are classified as corrected (incorporating gauge observations into their algorithms) or uncorrected, which implies that temporal variations depend entirely on the satellite. The main aim is to identify regions where precipitation products depict minimal uncertainties, supporting the use of the datasets in understanding precipitation variability in the specific regions. This is achieved by applying the triple collocation approach, which takes advantage of three collocated datasets of the same variable to derive the mean square error without requiring knowledge of the true value. The results show that light precipitation (1–5 mm d−1) was prevalent in most regions of Africa during the study duration (2001–2016). Estimating the spatial distribution of daily precipitation greater than the 90th percentiles suggests that extreme precipitation is mainly detected over the Central Africa region and coastal regions of West Africa, where the majority of uncorrected satellite products show consistent performance. The satellite product CMORPH_V1_RAW shows higher estimates of 90th percentile precipitation among the uncorrected satellite products. The ability of precipitation products to detect rainy or non-rainy days shows that corrected satellite products depict notable agreement for probability of detection and false alarm ratio over most regions of Africa. Overall, better performance is demonstrated by the IMERG products, ARCv2, CHIRPSv2 and PERSIANN_CDRv1r1 (corrected), and GPCC, CPC_v1.0 and REGEN_ALL (gauge) during the study period. Among the reanalysis products, ERA5 datasets shows good performance in estimating daily precipitation over Africa. The optimal maps that show the classification of products in regions where they depict reliable performance can be recommended for various usage by different stakeholders.

Considering sub-regions in Africa, for example, Ageet et al [22] employed in-situ daily precipitation datasets sourced from 36 stations across East Africa (EA) and used them to evaluate four satellite products that have been corrected with gaugebased observations.The study reported that over EA, datasets sourced from the Integrated Multisatellite Retrieval for Global Precipitation Measurement (IMERG-NRT) are suitable for estimating extreme events.Despite the availability of many satellite products, only a few were considered.Still over the EA region, Monsieurs et al [23] assessed the performance of a single satellite data sourced from the Tropical Rainfall Measuring Mission Multisatellite Precipitation Analysis (TMPA).The study revealed that complex topography and high precipitation intensities limit the performance of this product, thus leading to underestimation over the region.Nevertheless, the study noted that area-averaged TMPA precipitation estimates are more suitable for regional hazard assessment compared to gauge-based estimates that are scarcely distributed.These findings agreed with a similar study that assessed the performance of TMPA along with products such as the Climate Hazards Infrared Precipitation with Station (CHIRPS), African Rainfall Climatology version 2 (ARC2), and the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks Climate Data Records (PERSIANN-CDR) [24].However, the aforementioned study was limited to a few gauges against the satellite datasets that have varying horizontal grid resolution.In Algeria, Babaousmail et al [25] reported better performance of CHIRPS products and unsatisfactory performance of the Climate Prediction Center (CPC) Morphing Technique (CMORPH) datasets.The conclusion of these studies is that most precipitation products are characterized by various strengths and weaknesses.
Similar observations were reported in other regions of the globe.For instance, Bai and Liu [26], despite evaluating various satellite products and employing robust statistical approaches, limited the study to a small catchment along the Tibetan Plateau that is critical for Asian river sources.Over the Peruvian Andes region, Condom et al [27] employed TMM 3B43 to examine the spatial and temporal behaviour of precipitation.Findings of this study proposed the need to correct the precipitation products due to the discrepancies arising from their algorithms.Like many other studies, this one was limited to a single satellite product.
Clearly, these studies suggested the need to evaluate the satellite products and correct with gauges due to their varying performance.However, the dearth coverage and limited ground-based observations in some regions (especially in Africa) have resulted in some existing studies utilizing the few available synoptic datasets for validation of precipitation products, thus leading to some uncertainties [28][29][30][31].Overall, recent studies have shown that inter-product differences are generally larger among reanalysis products compared to in-situ products [14,15,32].More recently, Alexander et al [33] and Bador et al [34] have agreed to an existing and widely accepted view that satellite and in-situ precipitation products are skilful, which could be verified over Africa with this study.
Precipitation distribution over Africa is influenced by numerous factors (locally and regionally), which make it a highly heterogeneous climate [35][36][37].The factors range from complex orographic features (e.g.Mount Kilimanjaro ∼5800 m), large expanses of desert lands (Kalahari Desert in Southern Africa), complex equatorial forests in Central Africa (CAF), and large water bodies such as Lake Victoria in East Africa [38].The combination of the aforementioned processes significantly influences the distribution of precipitation across the continent, thus contributing to many uncertainties in the daily variation [38].Characterizing the performance of different precipitation products and their uncertainties is important.In fact, the concerted efforts by the International Precipitation Working Group to address this gap have led to the inter-comparison of precipitation products across regions of Australia, Europe, Japan and South America, among others, where validation projects have been undertaken.However, over sub-Saharan Africa, it is only in South Africa where a similar validation project has been conducted for satellite-surface inter-comparisons products [39].
Meanwhile, Africa faces the serious challenge of limited observed data, hindering the validation process of global models that are useful for climate projections [40].In an effort to overcome such setbacks, emerging studies prefer evaluating gridded satellite and reanalysis products against themselves without necessarily using reference datasets [41][42][43].On the other hand, other studies have employed a recent statistical approach referred to as 'triple collocation' to estimate the unknown errors of varying geophysical variables such as ocean wind speed and wave height that are mutually independent without treating one product as a perfectly observed 'truth' [44][45][46].The robustness of the triple collocation approach is its ability to evaluate the products over the entire study region so that data quality for the different climatic zones can be properly understood.Moreover, it takes advantage of three collocated products of the same variables and estimates the mean square error of each without requiring knowledge of the true value.Massari et al [47] used a recently employed precipitation product, SM2RAIN, built from soil moisture data, to validate satellite and reanalysis precipitation products over global land.The study recommended using triple collocation to identify the best precipitation products for hydrologic models over sparsely gauged regions and provide the benchmark for optimal integration among varying precipitation products.Few studies have explored the use of triple collocation as an alternative technique for evaluating precipitation products across the global land [48][49][50].
In a bid to address the lack of reliable groundbased observations, the study employs a triple collocation approach to compare the performance of different precipitation products over diverse climate regions in Africa.This will aid in identifying regions where the products depict minimal uncertainties and thus can be used in building optimal maps to guide scientific and operational applications.To this end, the study will address the following key scientific questions: (i) how do uncorrected or corrected satellite and reanalysis precipitation datasets simulate daily precipitation events over diverse climate zones of Africa?(ii) How do the precipitation products detect rainy or non-rainy days over the continent?(iii) Which datasets perform best over each of the diverse climates of Africa?The rest of this paper is organized as follows: section 2 describes the data and methods utilized for the intercomparison process.Section 3 presents the results and discussion, while the conclusion and recommendation of the study are presented in section 4.

Data and methods
This study employed different datasets sourced from the Frequent Rainfall Observations on GridS (FROGS; [51]) platform.Tables S1-4 highlight the different categories of precipitation products.Various statistical techniques are applied to examine the strengths and weaknesses of various precipitation products (reanalysis, satellite and gauge-based) in characterizing observed precipitation patterns.The aim is to identify similar features and corresponding differences displayed by the precipitation products, rather than the common practice where validation is conducted using the 'true' dataset as a reference.The first approach assesses the capabilities of datasets to estimate the frequency of daily precipitation occurrences during the study duration (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016).Frequency occurrences are divided into light (1-5 mm d −1 ), moderate (5-20 mm d −1 ) and heavy (⩾20 mm d −1 ) precipitation events.Secondly, categorical and descriptive statistics are used to determine the detection capabilities and performance of each product [52].Probability of detection (POD; equation (S1)) and false alarm ratio (FAR; equation (S2)) are used to distinguish rainy and non-rainy days with the Global Precipitation Climatology Centre Full Daily Data version 2020 (GPCC_FDD_v2020) datasets as the sample reference.A precipitation threshold of 1 mm day −1 is used to categorize rainy and non-rainy days, following the Expert Team on Sector-Specific Climate Indices recommendations.
On the other hand, descriptive statistics used to quantify the statistical parameters include mean bias (B), correlation of determination (R 2 ), and root mean square error (RMSE).RMSE identifies the amplitude of errors between the products.Detailed equations are presented in table S5.The descriptive statistics are presented in a box plot diagram with the box and whisker representing the median, and minimum and maximum of 25% and 75% quantiles for each statistical value.Finally, the triple collocation technique is utilized to build optimal maps for the precipitation products for each region where they demonstrate consistent, satisfactory performance.This technique estimates the amplitude variation of unknown true values and thus provides an appealing alternative for the assessment of precipitation products in regions lacking observation.Triple collocation considers the assumptions of 'no absolute correlation' among the products and linearity between the products when evaluating the precipitation from varying sources.In this study, triplets comprising a combination of either (1) gauge-based products + reanalysis to evaluate uncorrected satellite products or (2) best-performing corrected satellites from first-tier analysis + gaugebased products to estimate the performance of reanalysis are used to build a triplet to estimate correlation and covariance in the products.

Daily intercomparisons of precipitation estimates
Figure 1 shows the frequency (%) of light (1-5 mm d −1 ), moderate (5-20 mm mm d −1 ), and heavy (⩾20 mm d −1 ) precipitation events over the nine climate regions of Africa during 2001-2016.The results indicate that uncertainties in light precipitation are most prevalent over most regions in Africa.West Africa (WAF), Madagascar (MDG), and CAF regions record higher frequencies of light precipitation (65%-95%), while the Sahara (SAH) shows lower frequency estimates at 35% (figure 1).WAF and CAF show the occurrence of moderate precipitation, with most products estimating frequencies of 10%-45%, except for JRA-55 reanalysis products over WAF and CFSR products over CAF that record higher frequency estimates.Over MDG, a large spread is noted for the frequency of light precipitation (60%-80%), while moderate precipitation shows a lower frequency of occurrence with a maximum value of 45%.Few datasets, such as MERRA2, PERSIANN_CCS_CDR and REGEN_LONG v1 gauge products, detect varying precipitation frequencies over the SAH region.Generally, most of the precipitation datasets are unable to capture precipitation greater than 20 mm in most regions of Africa, except for the PERSIANN_CCS_CDR product.These results provide an important insight into how the products perform across various regions with varying topographical features.Taking the case of the detection of heavy precipitation events among the products, the results show that gauge products sourced from GPCC_FDD for both versions 2018 and 2020 and REGEN_LONG and REGEN_ALL show similar performance across most regions.As for uncorrected satellite products, GSMAP_NRT and PERSIAN_CCS_CDR show notable detection capability for heavy precipitation over most regions, especially WAF and CAF.Finally, corrected satellite products such as CHIRPS, GSMAP_gauge_NRT_V6 and IMERG_V06_EU/FC/FU/LU demonstrate a high capability of detecting heavy precipitation over most regions in Africa.A remarkable performance of IMERG compared to other corrected satellite products has been noted over other regions, owing to the improvement in its algorithms [51].The GPM IMERG products feature the first spaceborne Ku/Kaband dual-frequency precipitation radar, in addition to the multi-channel GMI, which enables it to detect light precipitation over diverse regions [51].The superior performance of the IMERG V06 products has been noted over typical agricultural regions in the Huanghuaihai Plain of China [52].Remarkably, only MERRA2 from the reanalysis products captured the heavy precipitation events of ⩾20 mm d −1 during the study period over all climate regions in Africa (figure 1).In a similar study over Africa, it was noted that most reanalysis products depict higher numbers of light and moderate precipitation events compared to satellite and gauge-based observations, while they tend to show the least number of heavy precipitation events [42].The study attributed the performance of reanalysis products to their convection parameterization schemes, which tend to show large values for light and moderate precipitation events.
An analysis of the daily precipitation 90th percentile during 2001-2016 is conducted to further explore the performance of precipitation products in estimating extreme precipitation events.Figure 2 displays the precipitation 90th percentile, representing extreme values for the specific locations.CAF exhibits the highest values of the 90th percentile.The majority of uncorrected satellite products show clear performance in depicting the extreme values of precipitation events over the coastal regions of WAF.For instance, CMORPH_V1_RAW shows high extreme event estimates, while PERSIANN_CCS_CDR shows low detection of extreme precipitation events among the uncorrected products over the CAF and WAF regions (figure 2).For the corrected satellite product, IMERG_V06_EU shows higher estimates of 90th percentile precipitation among the corrected products, while ARC2 shows lower values.Notably, only GPCC_FDD_v2018 and GPCC_FDD_v2020 among the observations show similar performance in detecting extreme precipitation events over CAF and southern parts of WAF.For the reanalysis, CFSR and MERRA2 demonstrate satisfactory skills in detecting higher precipitation values, especially over Northern Eastern Africa (NEAF), MDG and WAF, which is not shown among other products.For instance, ERA5 shows low estimates of 90th percentile precipitation despite the improvement from ERAi, consistent with related studies [53,54].Overall, most products show large differences in the estimation of extreme precipitation events over the wet humid region of CAF, while consistent performance is noted over the arid zones of SAH, Mediterranean (MED), and Southern Africa (SAF).
In a similar study over the global land, Sun et al [15] reported that satellite and gauge products demonstrated higher extreme precipitation over Africa, southern Asia, and South America than the reanalysis products, except for MERRA.Interestingly, our findings show that CFSR records higher extreme event estimates compared to all other reanalysis products, including MERRA products (figure 2).The ability of CFSR to reproduce regional patterns of extreme precipitation events could be attributed to the higher horizontal grid resolution and assimilation of coupled ocean-atmosphere in the models [55].Sun et al [15] further noted that the low extreme precipitation event estimates that characterize other reanalysis products, such as ERA, agree with the results of the present study over Africa.
The ability of precipitation products to estimate rainy days (1 mm d −1 ) and non-rainy days is tested based on the categorical statistics of POD and FAR.The spatial distribution of POD and FAR is presented in figures 3 and 4 during 2001-2016 for different precipitation products, with the GPCC_FDD_v2020 as the sample reference dataset.Compared with other gauge-based products, GPCC_FDD_v2020 has the highest number of station data points (i.e.84 800) incorporated into their algorithm to develop the spatially gridded products [3].For instance, CPC unified gauge-based analysis for global precipitation has ∼30 000 stations spatially distributed over the global land, while REGEN-ALL and REGEN-LONG have about 50 530 stations, mostly obtained from the GPCC and GHCN-Daily archives [2].Despite the large base of station data incorporated through robust interpolation schemes, Africa still accounts for the lowest number of station data used.Despite the low number of stations used in most gauge gridded products, GPCC_FDD_v2020 is used as a benchmark in other intercomparison analyses to compare the limitations and strengths of other precipitation products across the continent [e.g.19,42,43].
The thresholds of categorical statistics are defined for each product, where a value of zero denotes 'no rain' whereas 1 mm day −1 is the threshold to define a rainy day.Generally, a good agreement is noted in the corrected satellite products for POD and FAR over most regions of Africa (figure 3), with CHIRPS indicating 'perfect' FAR over the eastern parts of the SAH region (figure 4).It is worth noting that the SAH region has the lowest gauge density, so the POD scores over this region remain uncertain.The results show that gauge observations depict similar patterns with the reference product in POD (0.7-0.9), except for REGEN_LONG_V1, which shows low POD values (0.2-0.4) and high amplitudes (± 0.84) over EA and Eastern Southern Africa (ESAF) (figure 3).Conversely, only CPC_v1.0 exhibits a consistent FAR value of 0.2 over CAF and some parts of Southern Eastern Africa (SEAF) (figure 4).The results show weak POD in JRA-55 (0.2 ± 0.41) among the reanalysis products except over Western Southern Africa (WSAF), which depicts robust POD of 0.7 (figure 3).However, the insufficient FAR of >0.7 is evident in reanalysis products, except for JRA-55, which shows an FAR of 0.1 over the MED region (figure 4).Interestingly, ERAi shows strong POD over CAF and the coastal belt of WAF with values ∼0.9 while failing to detect FAR over similar regions (figure 3).Among the uncorrected satellite products, PERSIANN_CCS_CDR demonstrates strong POD and FAR values over most climatic regions (figures 3  and 4).Overall, the analysis suggests that a reliable score of POD or FAR is unattainable, irrespective of the source of the product.In agreement with other studies [e.g.[56][57][58], the results demonstrate that the performance of precipitation products is very much dependent on the techniques and algorithms used to build them.
Lastly, the uncertainties in the daily precipitation products is determined using the quantitative metrics of MB, RMSE and R 2 .Figure 5 shows the results for R 2 for each climatic zone, while RMSE and bias are shown in figures S1 and S2.The box and whisker represent the median, and minimum and maximum of 25% and 75% quantiles for each statistical value.Similar to the categorical metrics, GPCC_FDD_v2020 is used as a reference to ascertain how the other products represent the spread of biases against each other, their amplitudes and correlations.Considering all sub-regions in Africa, notable agreement with the reference dataset is depicted by REGEN_ALL, CPC_v1.0,ARCv2, and IMERG_V06_EU (figure 5).Conversely, MERRA2, CFSR, JRA_55, and ERAi exhibit the least similarity with GPCC_FDDv2020 products across all regions.It is interesting to note that NEAF and SAH depict the least R 2 with the reference product.In contrast, the SAF region generally shows better agreement with the reference product.This feature could be attributed to the low number of stations that are incorporated into the GPCC algorithms [3].A similar situation has been reported by Dinku et al [29], where most regions in Africa have reported a sharp decline in the number of recorded station datasets sent to the Global Telecommunication System for archive.Several challenges attributed to the discontinuity include the lack of temporal coverage and the dearth of spatial coverage across the continent.
The results for RMSE and MB as presented in figures S1 and S2 demonstrate consistent performance for REGEN_ALL, CPC_v1.0, and IMERG_V06_EU with low RMSE and MB, respectively.On the contrary, MERRA2, CFSR, and IMERG_V06_LU show high amplitude and a large bias for daily precipitation.Most products depict the least RMSE and MB over the MED region compared to other zones.CHIRPSv2.0exhibits robust performance over NEAF and CAF as compared to other regions.The best performance demonstrated by satellite-based products was equally noted by Zhang et al [59], who reported CHIRPSv2.0as the  correlation and relatively high RMSE over the continent.Overall, the products show acceptable performance over Africa, and their accuracy depends on the product sources or combination of techniques.Moreover, the skill score of products varies depending on the timescale (e.g.correlation increases as the time step increases).It is worth noting that while the scope of this study is limited to the analysis of precipitation products in estimating the mean daily precipitation characteristics, recent studies over Africa have noted a large spread in precipitation products, representing extreme events over Africa [42,[60][61][62].

Triple collocation analysis for optimal maps
The study employs the triple collocation approach to estimate the relative error in precipitation products by computing their RMSE at each grid point.In the analysis of uncorrected satellite products, the gauge-based products (GPCC_FDD_v2020) and reanalysis (ERA5) are kept constant as inputs and all satellite data are interchanged, whereas, for reanalysis ranking, we keep GPCC_FDD_v2020 and PERSIANN-CCS_CDR products and alternate all the reanalysis data.Consequently, low RMSE denotes robust performance, while larger errors reflect strong disagreement among the precipitation products.Figure 6 shows the optimal maps that illustrate the RMSE values of satellite and reanalysis products based on triple collocation analysis.It is apparent that regions that experience less precipitation depict minimal RMSE, while regions that experience more precipitation in most seasons display large RMSE values.For instance, most products illustrate low RMSE (<5) over SAH, MED, and WSAF, whereas high RMSE (>15) is recorded over WAF, CAF, and MDG (figure 6).Comparative analysis among the shows relatively low RMSE among the corrected satellite products, followed by uncorrected satellites, and lastly, the reanalysis data.In addition, reanalysis products also exhibit large RMSEs over SAH, especially for the CFSR datasets.Overall, the optimal maps can provide a guide for end users of the products and for various purposes based on the classification of regions where they show varying performance.For instance, products that consistently demonstrate high RMSE over different regions, such as CFSR and MERRA2, can be useful for flood monitoring since a good estimation of flood events relies on precipitation products that represent high-intensity precipitation occurrences well.Thus, any products that underestimate the occurrence and amount should be avoided [21].Meanwhile, IMERG_V06_FC and GSMAP_NRT_V6.0could be useful for drought monitoring due to their low RMSE based on the triple collocation analysis (figure 6).A good illustration of the low-intensity precipitation events is more important as compared to the products that depict higher values.

Conclusion
The present study employs multiple statistical techniques, including the triple collocation approach, to estimate the performance of multiple sources of precipitation products over diverse climate regions in Africa and identify the regions where the products depict minimal uncertainties to build an optimal map for climate users.The study uses GPCC_FDD_v2020 as a benchmark in intercomparison analysis to compare the limitations and strengths of other precipitation products across the continent.The first-tier analysis examines their capability to satisfactorily estimate spatiotemporal patterns of precipitation at daily scales over diverse climate regions.Ultimately, the triple collocation approach is utilized to build the optimal maps for the datasets for each region where they demonstrate consistent, satisfactory performance.The findings in the present study are itemized as follows: (a) Examination of the capability of precipitation products to detect the occurrence of daily PRE intensity at different frequencies shows that the uncertainties in light precipitation are most prevalent over most regions in Africa during the study duration.Overall, detection of heavy precipitation events is demonstrated by GPCC_FDD and REGEN_ALL among the gauge products GSMAP_NRT_v6.0and PERSIAN_CCS_CDR (uncorrected satellite products), and CHIRPS, GSMAP_gauge_NRT_V6, and all IMERG for corrected satellite products.(b) Further analysis of quantifying the performance of precipitation products in estimating the occurrence of extreme precipitation events based on the daily precipitation 90th percentile suggests that extreme precipitation is mainly detected over the CAF region and some coastal regions of WAF, where the majority of uncorrected satellite products show good agreement.Concurrently, CMORPH_V1_RAW shows higher estimates of 90th percentile precipitation over CAF and southern parts of WAF.
For the reanalysis, CFSR and MERRA2 demonstrate satisfactory skills for detecting higher precipitation values, especially over NEAF, MDG and WAF, which is not shown among other products.These products can thus be utilized for flood monitoring and extreme event quantification.(c) The detection of rainy days and non-rainy days using a threshold of 1 mm d −1 based on categorical statistics suggests that a satisfactory score for POD or FAR is unattainable irrespective of product type.However, some products, such as REGEN_ALL or PERSIANN-CDR, could be useful for flood monitoring, while products that demonstrate low FAR, such as CHIRPS or CPC, could be applied for drought monitoring.(d) A descriptive statistical estimation of daily precipitation uncertainties determined based on quantitative metrics of bias, RMSE and R 2 shows that consistent satisfactory performance is demonstrated by the uncorrected satellite products from IMERG_V06_EU.The corrected satellite products include ARCv2 and CHIRPSv2.Lastly, gauge-based datasets of GPCC, CPC_v1.0, and REGEN_ALL demonstrate robust performance during the study period.The products that depict low RMSE/Bias and high R 2 could be recommended for agriculture and crop modelling over various climate regions.(e) As a final step to identify a precipitation product that robustly and consistently estimates the spatiotemporal characteristics of precipitation over diverse climate regions and temporal scales, the reanalysis product of ERA5 shows good performance in estimating daily precipitation over Africa.Corrected satellites sourced from IMERG show consistent performance, while uncorrected satellites from PERSIANN_CCS_CDR demonstrate notable performance over most regions, except for SEAF.The optimal maps that show the classification of products in regions where they demonstrate reliable performance can be recommended for various stakeholders.

Figure 1 .
Figure 1.Frequency of light (L), moderate (M), and heavy (H) precipitation events from 31 datasets over the nine climate zones of Africa during 2001-2016.The frequencies are calculated with respect to rainy days only.The different data categories are: circle for the corrected satellites, star for the gauges, rectangle for reanalysis data, and triangle for uncorrected satellites.

Figure 2 .
Figure 2. Spatial variation of daily 90th percentile during 2001-2016 for different precipitation products over Africa and its sub-regions.The color of the figure title represents the data type, with black, blue, red and green representing the corrected satellite, gauge products, reanalysis and uncorrected satellite data, respectively.

3 .
Spatial distribution of probability of detection (POD) over Africa and its sub-regions during 2001-2016 for different precipitation products, with GPCC_FDDv2020 as the reference dataset.The color of the figure title represents the data type, with black, blue, red and green representing the corrected satellite, gauge products, reanalysis and uncorrected satellite data, respectively.

Figure 4 .
Figure 4. Spatial distribution of false alarm ratio (FAR) over Africa and its sub-regions during 2001-2016 for precipitation products, with GPCC_FDD_v2020 as the reference dataset.

Figure 5 .
Figure 5. Precipitation coefficient of determination averaged over Africa during 2001-2016.The coefficient of determination is calculated between and the remaining 30 precipitation estimates.

Figure 6 .
Figure 6.Triple collocation analysis RMSE results averaged over the nine zones of Africa during 2001-2016.The dashed black lines in the histogram demarcate the different types of datasets.