In African nations with complex topographies, alternative rainfall estimation methods such as satellites are crucial. This study aimed at predicting the spatial and temporal distribution of rainfall in the lake Tana sub-basin from 1990 to 2020. A satellite-based rainfall estimate of Climate Hazards Group Infrared Precipitation with Station (CHIRPS) data was used with the same spanning period. The validation process employs point-to-pixel analysis, comparing CHIRPS estimates with observed data at specific gauge stations. The findings showed that CHIRPS had well estimated rainfall incidence in the highland areas and significantly overestimated it in the lowland areas. The Mann–Kendall trends for January, June, and August indicate decreasing trends, while the winter and spring seasons show notable declines. Regression analysis reveals a non-significant decrease in annual rainfall with the highest in the summer and relatively dry winters. In addition, the coefficient of variation value of 26.37% suggests a moderate level of variability around the mean annual rainfall. In conclusion, the CHIRPS satellite exhibited varied performance across the Tana sub-basin, with site-specific discrepancies and notable inaccuracies at certain stations. The study underscores the importance of considering local factors and topography in satellite-based rainfall assessments, providing valuable insights for agricultural planning in the region.

  • The variability of rainfall was investigated using the standardized anomaly index, coefficient of variation, and precipitation concentration index.

  • The percentage of daily rainfall events with high intensity was overestimated while the number of daily rainfall events with light precipitation was underestimated by Climate Hazards Group Infrared Precipitation with Station data.

Rainfall is the most important climate input for agriculture, providing the water resources that sustain crops and livestock (Namara et al. 2010; Xu et al. 2022; Chen et al. 2024). Accurate estimation of rainfall amount and variability is critical for agricultural planning and food security (Kogo et al. 2021; Wang et al. 2022; Zhou et al. 2023). Traditionally, rainfall has been measured by ground-based rain gauge networks (Cimini et al. 2013; He et al. 2021; Guan et al. 2023). However, rain gauges provide only point data and are sparsely distributed in many parts of the world, hampering accurate spatial rainfall estimation (Hasan et al. 2016; Li et al. 2020; Dong et al. 2023). In recent decades, satellite-based rainfall products have emerged as an alternative to augment ground observations and provide better spatial coverage (Brunetti et al. 2021; M. Wu et al. 2022; X. Wu et al. 2022; Dong et al. 2023). One such product is the Climate Hazards Group Infrared Precipitation with Station (CHIRPS) data, which combines thermal infrared (TIR) satellite imagery with station data to create high-resolution rainfall time series over land (Shen et al. 2020; Cheng et al. 2023; Yuan et al. 2023). CHIRPS was selected for this study because it has been widely validated across Africa and shown to perform reasonably well compared with ground observations in capturing rainfall variability and trends (Ayehu et al. 2018; Dinku et al. 2018; Esayas et al. 2019).

Several studies have validated the performance of CHIRPS rainfall estimates across Africa. Dinku et al. (2018) compared CHIRPS with ground observations across all of mainland Africa and found a good correlation (r = 0.70) on average, with higher performance in tropical regions compared with arid regions. However, CHIRPS had a dry bias and underestimated rainfall variability. Toté et al. (2015) and Zhou et al. (2022a) evaluated CHIRPS in three climatologically diverse regions in Africa and found it represented the various rainfall regimes well, with correlation coefficients ranging from 0.51 to 0.81 against stations. However, CHIRPS again showed limitations in capturing heavy rainfall events. In Ethiopia specifically, Dinku et al. (2018) found CHIRPS had a high correlation (r = 0.81) with ground data, but underestimated rainfall, especially in the Ethiopian highlands. By contrast, Worqlul et al. (2018) found CHIRPS overestimated annual rainfall in the Upper Blue Nile basin in Ethiopia compared with stations. Both studies noted challenges with properly capturing topography-induced rainfall in the mountainous regions of Ethiopia. Regarding the Tana sub-basin in particular, no studies were found that specifically validated CHIRPS performance in this region.

This study focuses on the Tana sub-basin in the Ethiopian highlands. The region is important for agriculture, but suffers from lack of ground rainfall data. However, no studies have specifically evaluated the capability of CHIRPS to estimate rainfall in the study area. Therefore, there remain knowledge gaps regarding the accuracy of CHIRPS in this region and its potential to augment sparse ground observations for agricultural planning. The objective of this study is to evaluate the capability of CHIRPS satellite rainfall estimates to accurately estimate and predict rainfall variability, compared with ground-based measurements.

Description of the study area

The lake Tana sub-basin, situated in the north-west highlands of Ethiopia, serves as the primary source of the Blue Nile River. Spanning an extensive area of approximately 15,000 km2, inclusive of the lake area covering 3,156 km2, the basin exhibits a diverse topography with an altitude of 1,786 m above mean sea level. It boasts dimensions of 200 km from north to south and 165 km from east to west, lying between 10° 0 0 and 13° 0 0 N latitudes and 36° 0 0 and 39° 0 0 E longitudes (Figure 1). In the lake Tana sub-basin, the intricate interplay between rainfall patterns, topography, and agricultural practices significantly influences agricultural productivity (Tigabu et al. 2019). The region, being a pivotal source of the Blue Nile River, sustains a substantial portion of Ethiopia's agricultural output. The excess agricultural production attributable to the Tana basin is a testament to its vital role in supporting the country's food security (Dile et al. 2016). The highlands and lowlands within the basin exhibit diverse agro-ecological conditions, allowing for the cultivation of various crops. Rain-fed agriculture is prevalent in the region, emphasizing the reliance on precipitation for crop growth. Understanding the spatial and temporal variability of rainfall in the lake Tana sub-basin becomes imperative in assessing its impact on agricultural yields. Furthermore, the distinction between highlands and lowlands is critical, as these areas may experience different precipitation patterns, influencing the types of crops cultivated and the overall agricultural landscape.
Figure 1

Locational map of the study area: (a) Ethiopian regional states, (b) Amhara Regional State and (c) lake Tana sub-basin.

Figure 1

Locational map of the study area: (a) Ethiopian regional states, (b) Amhara Regional State and (c) lake Tana sub-basin.

Close modal

Methods

Data source

Gauge station rainfall data

The study utilized rainfall data from ten meteorological stations covering the period from 1990 to 2020, sourced from the Ethiopian Meteorological Institute (EMI), namely Delgi, Gorgora, Enfranz, Maksegnit, Addis Zemen, Woreta, Dekestifanos, Zegie, Merawi, and Dangila. They were selected as reference points to assess the precision of the CHIRPS rainfall product across the study region. By employing the rain gauge measurements from these stations, the research aimed to evaluate how accurately the CHIRPS rainfall data reflected the actual observed rainfall patterns in the specified area. The chosen time period spans three decades, allowing for the analysis of long-term trends and variability in rainfall patterns. This duration is significant for capturing potential climate changes or shifts in precipitation that may occur over extended periods.

CHIRPS satellite rainfall estimate

The study employed the CHIRPS satellite rainfall estimate as a key data source. CHIRPS is a comprehensive database developed collaboratively by the US Geological Survey (USGS) and the University of California, Santa Barbara (UCSB). CHIRPS offers a 0.05° resolution gridded rainfall time series on a daily, monthly, and annual basis at a global scale. To assess the predictive accuracy of the satellite rainfall estimate over the study area, monthly and annual CHIRPS data for the past 30 years was downloaded from https://earlywarning.usgs.gov/fews. This extensive temporal coverage is valuable for understanding changes in precipitation patterns over the past three decades.

Methods of data processing and analysis

Validation techniques

The primary focus of the validation work was to assess the performance of CHIRPS in predicting rainfall for data-scarce areas within the lake Tana sub-basin, aiming to enable its utilization in spatial and temporal analyses of rainfall variability within the basin. Conducted over the 1990–2020 period, which featured reliable ground-based observations from the EMI, the validation compared CHIRPS rainfall products with observed data. Given the sparse distribution of observation stations in the lake Tana sub-basin, traditional interpolation of ground-based gauge measurements to a gridded dataset was deemed impractical due to significant uncertainties. Consequently, the evaluation adopted a point-to-pixel analysis, overlooking station locations within pixels and anticipating the introduction of random errors. This approach acknowledged uncertainties arising from rain gauge measurements and discrepancies in the spatial alignment of reference gauges and pixel centers. The assessment differentiated between lowland and highland regions, recognizing potential variations in the performance of CHIRPS estimates based on distinct rainfall characteristics.

To evaluate CHIRPS' ability to detect rainfall events on a daily timescale, basic categorical validation statistics were employed, derived from a contingency table (Table 1) encompassing four event combinations: hits (A), false alarms (B), misses (C), and correct negatives (D). The assessment utilized a rainfall threshold of 1 mm to determine rain or no rain events, a value commonly used in similar regional studies (Ayehu et al. 2018; Dinku et al. 2018; Liu et al. 2023) and representing the smallest nonzero recorded value at gauging stations. The categorical validation metrics included the probability of detection (POD), false alarm ratio (FAR), frequency bias index (FBI), and Heidke skill score (HSS). POD gauges the satellite product's skill in detecting rainfall occurrences, ranging from 0 to a perfect score of 1. FAR assesses false detections, with a perfect score of 0. FBI compares the satellite product's rainfall day detection frequency with gauge measurements, indicating underestimation (<1) or overestimation (>1) by CHIRPS. HSS measures estimation accuracy while considering random chance, ranging from minus infinity to 1, where HSS < 0 signifies random chance superiority, HSS = 0 denotes no skill, and HSS = 1 indicates perfect precipitation estimation by the satellite product (Table 2).

Table 1

Contingency table for comparing CHIRPS satellite-based rainfall estimates and rain gauge observations (daily rainfall threshold used is 1 mm)

CHIRPS ≥ 1 mmCHIRPS < 1 mm
Gauge ≥ 1 mm 
Gauge < 1 mm 
CHIRPS ≥ 1 mmCHIRPS < 1 mm
Gauge ≥ 1 mm 
Gauge < 1 mm 
Table 2

Statistical formulas for rainfall detection capabilities of CHIRPS based on categorical metrics

StatisticsEquationRangeBest value
POD  0 to 1 
FAR  0 to 1 
FBI  0 to ∞ 
HSS  −∞ to 1 
StatisticsEquationRangeBest value
POD  0 to 1 
FAR  0 to 1 
FBI  0 to ∞ 
HSS  −∞ to 1 

In assessing the accuracy of the CHIRPS rainfall product in quantifying precipitation, the authors employed four widely recognized error metrics: mean error (ME), mean absolute error (MAE), the Nash–Sutcliffe efficiency (NSE) coefficient, and bias, examining both daily and monthly timeframes (Table 3). Correlation coefficients were omitted from the evaluation due to their sensitivity to extreme values (outliers) and insensitivity to additive and proportional differences between satellite products and gauge measurements, particularly in daily estimations. ME and MAE, with dimensions in mm, provide insights into the average magnitude of estimation errors, aiming for a perfect score of 0. The root mean squared error was replaced by MAE to mitigate the impact of extremely high rainfall values or outliers. NSE, a dimensionless metric ranging from −∞ to 1, gauges the relative magnitude of residuals compared with observed precipitation values, with higher values indicating better agreement between satellite estimates and gauges. Negative values signify the reference mean as a superior estimate, while 0 implies the reference mean equals the satellite estimates. Bias, ranging from 0 to 1, measures the correspondence between estimated and gauge mean values, with 1 denoting a perfect score. Instances of bias > 1 and positive ME values suggest overestimation.

Table 3

Statistical measures of performance used for analysis based on continuous metrics

StatisticsEquationRangeBest value
ME  −∞ to ∞ 
MAE  0 to ∞ 
NSE  −∞ to 1 
Bias  0 to ∞ 
StatisticsEquationRangeBest value
ME  −∞ to ∞ 
MAE  0 to ∞ 
NSE  −∞ to 1 
Bias  0 to ∞ 

Assessment of spatial–temporal variability and trends of rainfall

Following the validation of the CHIRPS product within the lake Tana sub-basin, the study proceeds to evaluate the spatial and temporal variations in rainfall using the annual and monthly CHIRPS rainfall datasets over a 30-year time period. The analysis was conducted by utilizing ArcMap 10.8.1 and R software, allowing for a comprehensive examination of how rainfall patterns evolved over the specified period. This paper delves into both temporal and spatial aspects, exploring the variability and trends in rainfall within the basin. This helps to provide a nuanced understanding of how precipitation changed across different locations within the lake Tana sub-basin over the three-decade timeframe.

To investigate the long-term evaluation of climatic characteristics through time, several experiments are available. The standardized rainfall anomaly index (SRAI), which examines the frequency and intensity of precipitation, was used to measure the changes in observed climatic trends and variability across time. The Mann–Kendall (MK) trend test was used to detect climate trends in time series data, the coefficient of variation (CV) was used to estimate the seasonal and annual variation of the rainfall patterns, and the standardized precipitation index (SPI) was used to quantify the rainfall deficit for the observed time scale in the study area (Asfaw et al. 2018; Shang et al. 2023).

The standardized anomaly index was used to determine both the negative and positive anomalies of variations in rainfall in a particular area. By identifying the dry and wet years of the recording era, it was easier to pinpoint the drought period (Du & Wang 2014; Gleixner et al. 2017). This index was determined using the following equation (Equation (1)):
formula
(1)
where Z is standardized rainfall anomaly, Xi is the annual rainfall for the historical record, is the mean of the annual rainfall, and S is the standard deviation (SD) of the annual rainfall for the historical observation of the time series.
The dispersion of data points in a data series around the mean is statistically measured by the CV. Even when the means of two data series are quite different from one another, the CV, which measures the SD in relation to the mean, may be used to compare the degree of variation between the two datasets. A CV was employed to detect the seasonal and annual rainfall variability for the observation period (Ahmad et al. 2018). The formula below was used to calculate the coefficients of variation of rainfall variability over the research region (Equation (2)):
formula
(2)
where CV is the coefficient of variation, is the SD, and is the mean precipitation of the recording period.
The precipitation concentration index (PCI) is an effective predictor of the distribution of precipitation across time. Similar to CI, PCI is typically applied to analyze seasonal fluctuations in precipitation to look into the heterogeneity of monthly rainfall data (Gocic et al. 2016). In order to assess how rainfall is distributed throughout the research area on a monthly, seasonal, and yearly basis, PCI was employed (Equation (3)):
formula
(3)
where is the monthly precipitation and is the precipitation concentration index.
The MK trend test (sometimes called the MK test) is used to analyze data collected over time for consistently increasing or decreasing trends (monotonic) in Y values. It is a non-parametric test, which means it works for all distributions, but the data should have no serial correlation. If the data do follow a normal distribution, it can be better to run simple linear regression instead. The test can be used to find trends in as few as four samples. However, with only a few data points, the test has a high probability of not finding a trend when one would be present if more points were provided. The more data points we have, the more likely it is that the test is going to find a true trend.
formula
(4)
formula
(5)
formula
(6)
formula
(7)
where q represents a total number of tied groups. A set of the same values in a dataset is referred to as a tied group. Each tied group is denoted by tp. The positive values of Z indicate upward (increasing) trends in time series, and the negative values show downward (decreasing) trends. Trends are then tested against some critical values (Z1α) to show that either they are statistically significant or not. To quantify the magnitude of detected trends, a frequently used non-parametric method, the Sen's slope method (Ahmad et al. 2014; Zhou et al. 2022b), was applied in the present study. This method is robust against outliers in a time series. Sen's slope (SS) is calculated as below:
formula
(8)
where Xi is the value of data at time step i and Xj at time step j. Before application of the MK test, the time series must be free of serial correlation, which can mislead the actual result of the trends.

CHIRPS rainfall detection potential

Statistical indicators such as POD, FAR, FBI, and HSS were calculated for the years 1990–2020 across the Tana sub-basin to evaluate the performance of the CHIRPS satellite rainfall estimate. Ten meteorological stations in the Tana sub-basin were quantified for the pixel-to-pixel comparison of gauge and satellite rainfall (Table 4). The POD, which runs from 0 to a flawless 1, assesses how well the satellite product can identify when it rains. With a POD value score of 1, the CHIRPS satellite can accurately predict the amount of rainfall over the Dangila station. This demonstrates that the rainfall measurement at this station was entirely consistent with the rainfall estimate from the CHIRPS satellite. However, compared with rain gauge measurements of rainfall, the CHIRPS satellite's estimate of rainfall above the Dekestifanos station was 15% off. A miss estimation of rainfall with less than 10% from gauge station measurements was found in CHIRPS satellite rainfall estimates across the remaining stations (Table 4). The FAR statistic, which rates false detections from 0 to 1 with a perfect score of 0, evaluates the effectiveness of the CHIRPS satellite rainfall estimate. The findings showed that CHIRPS satellite rainfall estimates over the Addis Zemen station were inaccurate by 25%, according to FAR values. This demonstrates that the amount of rainfall determined by the CHIRPS satellite has probably detected a false alert with a 25% variation at this site. As a result, it was noted that the CHIRPS satellite's performance over the Addis Zemen station was much worse as compared with that of the other similar stations. With FAR values of 5% and 4%, respectively, the CHIRPS satellite conducts rainfall estimation quite effectively over the Woreta and Gorgora stations. Over the study area of the Tana sub-basin, CHIRPS satellite estimates rainfall over the Zegie and Delgi stations with an equivalent FAR of 9%. A FAR of 11% was experienced by CHIRPS rainfall estimations across the Merawi and Enfranz stations at the same time. This is related to the stations' similar patterns in agro-climatic conditions. This indicates that the CHIRPS satellite estimate false alarm rates are similar at stations in roughly similar agro-climatic zones.

Table 4

Performance of the CHIRPS satellite rainfall estimate over the Tana sub-basin meteorological stations

StationStatistics
PODFARFBIHSS
Dangila 1.00 0.19 1.24 0.40 
Addis Zemen 0.95 0.25 1.27 0.28 
Dekestifanos 0.85 0.10 0.95 0.62 
Enfranz 0.93 0.11 1.05 0.73 
Maksegnit 0.90 0.10 1.00 0.41 
Merawi 0.89 0.11 1.10 0.35 
Woreta 0.95 0.05 0.85 0.39 
Zegie 0.91 0.09 1.05 0.36 
Delgi 0.91 0.09 1.02 0.39 
Gorgora 0.96 0.04 0.86 0.40 
StationStatistics
PODFARFBIHSS
Dangila 1.00 0.19 1.24 0.40 
Addis Zemen 0.95 0.25 1.27 0.28 
Dekestifanos 0.85 0.10 0.95 0.62 
Enfranz 0.93 0.11 1.05 0.73 
Maksegnit 0.90 0.10 1.00 0.41 
Merawi 0.89 0.11 1.10 0.35 
Woreta 0.95 0.05 0.85 0.39 
Zegie 0.91 0.09 1.05 0.36 
Delgi 0.91 0.09 1.02 0.39 
Gorgora 0.96 0.04 0.86 0.40 

The FBI statistic, which ranges from 0 to infinity with a perfect score of 1, compares the rainfall detection frequency of the satellite product with that of the gauge measurements and is another statistical method used to evaluate the performance of the CHIRPS satellite. An FBI of less than 1 suggests that CHIRPS has underestimated the number of rainy days, whereas an FBI of more than 1 indicates the opposite. Based on this stance, the FBI gave a perfect score of 1 for the CHIRPS satellite rainfall estimate's detection capabilities over the Maksegnit station. This indicates that gauge measurements and the frequency of rainfall detection by the CHIRPS satellite were in agreement. The CHIRPS satellite has estimated rainfall day frequencies of 1.24 and 1.27, respectively, for the Dangila and Addis Zemen stations on a wide scale (Table 4). With FBI values of 0.85, 0.86, and 0.95, respectively, the CHIRPS satellite underestimated the frequency of rainy days over the stations in Woreta, Gorgora, and Dekestifanos.

While accounting for matches owing to random chances (after rain events discovered by chance have been excluded), the HSS statistic was also calculated to determine the accuracy of the estimates. Its range is from minus infinity to 1. HSS 0 denotes that chance is better than the satellite product, HSS = 0 denotes that the product has no skill and HSS = 1 denotes that the satellite estimate product has accurately predicted the amount of precipitation. This claim is supported by the fact that the CHIRPS satellite has the ability to predict rainfall over the Enfranz station with HSS values of 0.73, or the satellite has the opportunity to estimate rainfall at this station at a random percentage of 73%.

Accuracy of the CHIRPS satellite rainfall estimate over the Tana sub-basin

In comparison with rainfall recorded at the ten gauge stations, the accuracy of CHIRPS satellite rainfall estimates across the Tana sub-basin was assessed. At each gauge station over the Tana sub-basin, statistical metrics including ME, MAE, NSE coefficient, and bias error were used to assess the accuracy of CHIRPS satellite rainfall estimates. The statistical metric of MAE was used to assess the variation of the CHIRPS satellite rainfall estimate from gauge station measurements. To examine the fluctuation of the CHIRPS satellite rainfall estimate over each individual site, gauge-specific validation was carried out, and for each of the validation sites, MAE was calculated. As a result, the rainfall estimates from the CHIRPS satellite varied over each validation site.

The MAE value for each of the ten validation locations across the research region is displayed in Table 5. The best value for MAE between two measurements is zero, and its range is zero to positive infinity. The variance of the CHIRPS satellite rainfall estimate over each validation site increases with increasing MAE from zero. With MAE values of 0.95 and 0.85, respectively, the CHIRPS satellite performs better than other gauge station measurements in its estimation of rainfall over the Woreta and Zegie gauge stations. The CHIRPS satellite rainfall estimate across four places differs significantly from the ground measurements of rainfall obtained from gauge stations. Therefore, CHIRPS satellite rainfall estimates over the stations at Dangila, Maksegnit, Delgi, and Gorgora had high MAE values of 1.43, 1.47, 1.57, and 1.45 mm, respectively, compared with ground measurements.

Table 5

Statistical validation results of the CHIRPS satellite rainfall estimate

StationStatistics
MEMAENSEBIAS
Dangila 0.52 1.43 −2.54 1.17 
Addis Zemen 1.00 1.30 −3.06 1.23 
Dekestifanos 0.80 1.20 −1.57 1.20 
Enfranz 0.68 1.29 −4.39 1.40 
Maksegnit 0.87 1.47 −5.21 1.46 
Merawi 0.35 1.16 −0.98 1.20 
Woreta 0.70 0.95 −1.33 1.13 
Zegie 0.35 0.85 −0.26 1.13 
Delgi 1.11 1.57 −4.73 1.42 
Gorgora 1.30 1.45 −5.15 1.41 
StationStatistics
MEMAENSEBIAS
Dangila 0.52 1.43 −2.54 1.17 
Addis Zemen 1.00 1.30 −3.06 1.23 
Dekestifanos 0.80 1.20 −1.57 1.20 
Enfranz 0.68 1.29 −4.39 1.40 
Maksegnit 0.87 1.47 −5.21 1.46 
Merawi 0.35 1.16 −0.98 1.20 
Woreta 0.70 0.95 −1.33 1.13 
Zegie 0.35 0.85 −0.26 1.13 
Delgi 1.11 1.57 −4.73 1.42 
Gorgora 1.30 1.45 −5.15 1.41 

The NSE is a metric used to assess the performance of rainfall models, with a value of 1 indicating a perfect match between the modeled and observed data. As per the analysis in Table 5, the Merawi and Zegie stations exhibit NSE values greater than −1, specifically −0.98 and −0.26, respectively. This discrepancy can be attributed due to several factors. In the case of the Merawi and Zegie stations, the CHIRPS satellite rainfall estimates might align more closely with the observed data, resulting in lower error variances and higher efficiency scores. This can be attributed to specific local conditions, topography, and microclimates unique to these stations, making the CHIRPS satellite estimates relatively more accurate (Ritter & Muñoz-Carpena 2013; Seong et al. 2022). Conversely, for the Maksegnit and Gorgora stations, the NSE values are notably below 1, indicating poor model performance. The unfavorable performance at these stations could be influenced by diverse local climatic patterns, geographical features, or factors affecting rainfall distribution that are not adequately captured by the CHIRPS satellite model.

The overall unreliability of the CHIRPS satellite rainfall estimation across the Tana sub-basin, as indicated by NSE values, underscores the need for considering site-specific characteristics and potential limitations of the satellite model in capturing complex local variations. The favorable ME of 0.35 for the Merawi and Zegie stations suggests that the CHIRPS satellite tends to slightly overestimate rainfall at these locations, possibly due to specific regional dynamics not fully accounted for in the model. In contrast, the inconsistent performance observed in the Maksegnit and Gorgora stations can be attributed to their distinct climatic conditions, leading to significant discrepancies between satellite estimates and ground measurements. This nuanced understanding provides a scientifically grounded basis for the observed statistical outcomes, emphasizing the importance of considering local context and environmental factors in satellite-based rainfall assessments.

A systematic inaccuracy known as bias results in an estimate of an effect or association that is erroneous. To calculate the bias difference between gauge station measurements of rainfall and the CHIRPS satellite rainfall estimate, the bias error of each validation site was also assessed. All of the validation sites were determined to have positive bias errors. This demonstrates that the rainfall measurement acquired from the gauge station has a lower value than the rainfall estimates from the CHIRPS satellite. The overestimation of rainfall events by the CHIRPS satellite over the Tana sub-basin was blamed for the higher values of the CHIRPS satellite rainfall estimate at each validation site. To ascertain the discrepancy between the gauge station measurement and the rainfall estimate provided by the CHIRPS satellite, the ME of rainfall measurement was also assessed. The discrepancy between the ground observational rainfall estimates and the CHIRPS satellite estimate represents equivalent measurement values, according to an ME value that is near zero, while certain validation sites in the Tana sub-basin showed positive and negative ME values that were overestimated and underestimated by the CHIRPS satellite, Because of this, ME values found in almost every location showed that the ground observation and the CHIRPS satellite estimate of rainfall are very similar. With ME values of 0.35, which are substantially closer to zero than the other station measurements, the CHIRPS satellite specifically delivers better rainfall estimation over the Merawi and Zegie stations.

Cumulative distribution function of CHIRPS rainfall estimates and gauge measurements over the Tana sub-basin

The cumulative distribution functions (CDFs) were created using the daily rainfall data for each of the ten rainfall gauge stations in the research area in order to determine the probabilities of the daily rainfall volumes. However, four target rainfall stations were chosen from the total of ten rainfall stations in order to describe the results of CDFs and their related P–P plots due to the vast quantity of data gathered, in an effort to better comprehend the approach utilized. The CHIRPS satellite rainfall estimate's cumulative probability over a given time period is calculated using the CDF. The probability that a chance observation of a CHIRPS satellite rainfall estimate will be less than or equal to gauge station rainfall observations can be calculated using CDF values. The likelihood that a CHIRPS satellite rainfall estimate is greater than actual rainfall measurement data received from gauge stations, or between two values, is also calculated using CDF information.

The distribution of CHIRPS daily rainfall occurrences at the Addis Zemen station follows the gauge measurement trend, as seen in Figure 2. The CDF gamma P–P map, however, shows that the rainfall event captured by CHIRPS was underestimated compared with what was actually observed at this particular station. On the other hand, the daily rainfall events over the Maksegnit gauge station were overstated by the CHIRPS satellite (Figure 3). The satellite estimate does not match the measurements of rainfall events made at this gauge station. When compared with the Addis Zemen station, which completely matched gauge station data with the exception of underestimating the rainfall event, the CHIRPS satellite's performance in estimating the rainfall event at this particular site is quite subpar. The CDF distribution of CHIRPS rainfall events over the Dekestifanos station of the Tana sub-basin is described in Figure 4.
Figure 2

CDF distribution of CHIRPS rainfall events over Addis Zemen station of the Tana sub-basin.

Figure 2

CDF distribution of CHIRPS rainfall events over Addis Zemen station of the Tana sub-basin.

Close modal
Figure 3

CDF distribution of CHIRPS rainfall events over Maksegnit station of the Tana sub-basin.

Figure 3

CDF distribution of CHIRPS rainfall events over Maksegnit station of the Tana sub-basin.

Close modal
Figure 4

CDF distribution of CHIRPS rainfall events over Dekestifanos station of the Tana sub-basin.

Figure 4

CDF distribution of CHIRPS rainfall events over Dekestifanos station of the Tana sub-basin.

Close modal

Spatiotemporal variability of rainfall over the Tana sub-basin

Spatial variability of rainfall

In terms of mean monthly and seasonal rainfall patterns, the spatial variability of the Tana sub-basin's CHIRPS satellite rainfall estimate was investigated. Figure 5 shows that over the past 30 years of study, there have been variations in the CHIRPS rainfall observation patterns across each month, each season, and even each year. The Tana sub-basin experiences extremely little rainfall overall when comparing equivalent months, with January, February, and March receiving the least amount of rain. Figure 5 clearly illustrates this, by the bright green hue representing the maximum quantity of rainfall and the red color representing the least amount. The Tana sub-basin experiences its heaviest rainfall during the month of June.
Figure 5

Spatial variability of rainfall across each month over the lake Tana sub-basin (1990–2020).

Figure 5

Spatial variability of rainfall across each month over the lake Tana sub-basin (1990–2020).

Close modal

Examining seasonal variations in rainfall patterns is pivotal for understanding agricultural production dynamics in the Tana sub-basin, celebrated for its substantial agricultural productivity (Fazzini et al. 2015; Behailu et al. 2021). Following the Ethiopian agro-climatology classification, which recognizes winter, autumn, summer, and spring, each spanning three specific months, this study delves into the unique characteristics of these seasons. The winter season, encompassing January, February, and March, is identified as the driest among the three main seasons (Behailu et al. 2021). Autumn, comprising April, May, and June, aligns with the onset of the rainy season and is a key cropping period. The summer months of July, August, and September are anticipated to receive the highest rainfall, crucial for the primary cropping season, during which over 80% of agricultural crop production depends on rainfall (Gebremichael et al. 2014; Wakjira et al. 2021). In contrast, the months of October, November, and December, categorized as spring following the summer season, experience less precipitation compared with the summer season. The northern half of the study region encounters minimal winter rainfall, except on the southern edge of the basin. Conversely, the southern part experiences peak rainfall during the autumn season. Surprisingly, the northern region receives the least rainfall in autumn, potentially linked to the basin's topographic features. Higher elevations correlate with increased rainfall, with the southernmost region of the basin exhibiting both elevated terrain and the highest rainfall levels (Aher et al. 2019).

On the other hand, the entire study area received a significant amount of rain during the summer. For the past 30 years of the study period, the southern and western edges of the study region have received the most rainfall. In each of the four distinct seasons, rainfall was lowest in the northern part of the study region. The amount of rainfall experienced in the spring season shares a pattern with that of the autumn season, with the southern and northern edges experiencing the highest and lowest amounts of precipitation, respectively. This demonstrates that over the previous 30 years of the study period, the agricultural production of the Tana sub-basin has followed a similar trend in the spring and autumn seasons (Figure 6).
Figure 6

Seasonal rainfall variability of the CHIRPS satellite rainfall estimate over the Tana sub-basin: (a) winter season, (b) autumn season, (c) summer season, and (d) spring season.

Figure 6

Seasonal rainfall variability of the CHIRPS satellite rainfall estimate over the Tana sub-basin: (a) winter season, (b) autumn season, (c) summer season, and (d) spring season.

Close modal
It is crucial to define the spatial variability of rainfall over the Tana sub-basin in addition to looking at the independent seasonal rainfall distribution over the study area. The southern border experienced the most rainfall, with minimum totals of 1,500 mm and maximum totals of 1,900 mm (Figure 7). From the southern edge to the center of the basin, the amount of rainfall measured at and close to the Dangila and Zegie stations follows a similar pattern, ranging from a minimum of 1,400 mm to a maximum of 1,570 mm. The center portion of the research area, which is close to the Dekestifanos, Woreta, and Addis Zemen stations, likewise exhibits a similar distribution pattern, ranging from a minimum of 1,300 mm to a high of 1,500 mm. In comparison with the other portions of the study area, the northern portion of the region, which is home to the gauge stations at Delgi, Gorgora, Enfranz, and Maksegnit, receives the least amount of rainfall each year.
Figure 7

Annual rainfall variability of the CHIRPS satellite rainfall estimate over the past 30 years in the case of the lake Tana sub-basin.

Figure 7

Annual rainfall variability of the CHIRPS satellite rainfall estimate over the past 30 years in the case of the lake Tana sub-basin.

Close modal

Exploring the relationship between CHIRPS rainfall and elevation over the Tana sub-basin

Precipitation is highly influenced by elevation, particularly in mountainous areas. It is heavier on a mountain's windward side (Pepin et al. 2022). On the windward side of the mountain, as the air parcel rises due to the higher elevation, it cools, condenses, and rains (Houze 2014). To describe the distribution of rainfall across the Tana sub-basin, it is crucial to investigate the relationship between rainfall and elevation (Figure 8). Woreta, Addis Zemen, and Dekestifanos are all located in the study area's middle section, in lowland regions with low elevation values. In comparison with Dangila, where high elevations are predominantly associated with the maximum amount of rainfall, it is claimed that the dispersion of rainfall over these particular sites is rather modest. The association between elevation and rainfall distribution over the Tana sub-basin is shown by the scatter plot in Figure 9. The research area's topography has a big impact on how much rainfall is available. As a result, with coefficient of determination values of R2 = 0.816, the elevation of the Tana sub-basin determines the occurrence of rainfall events. The SD of the CHIRPS satellite rainfall estimate over the Tana sub basin varies from 10.23 to 13.07 mm (Figure 10).
Figure 10

SD of the CHIRPS satellite rainfall estimate over the Tana sub-basin (1990–2020).

Figure 10

SD of the CHIRPS satellite rainfall estimate over the Tana sub-basin (1990–2020).

Close modal
Figure 9

Scatter plot depicting the relationship between the CHIRPS satellite rainfall estimate and elevation over the study region.

Figure 9

Scatter plot depicting the relationship between the CHIRPS satellite rainfall estimate and elevation over the study region.

Close modal
Figure 8

Comparison of the (a) CHIRPS satellite rainfall estimate and (b) elevation over the Tana sub-basin.

Figure 8

Comparison of the (a) CHIRPS satellite rainfall estimate and (b) elevation over the Tana sub-basin.

Close modal

Rainfall trend analysis

Rainfall summary statistic

Figure 11 displays the rainfall and Table 6 displays the computed summary statistics of rainfall, including mean, SD, rainfall distribution, and CV over the study area from 1990 to 2020. The annual mean rainfall was 1,088.14 mm, with an SD of 286.971 mm. July (325.29 mm) and August (334.98 mm) had the highest maximum monthly rainfall, accounting for 29.8% and 30.7% of total rainfall, respectively. The least quantity of rainfall for the study period was observed in January and November, with 7.58 and 9.54 mm, respectively. The rainfall CV was calculated using the mean and SD for the last 30 years of data. The study area's mean annual rainfall has varied during the last 30 years, with a CV of 26.37%.
Figure 11

Monthly rainfall distribution of the study region (1990–2020).

Figure 11

Monthly rainfall distribution of the study region (1990–2020).

Close modal

The CV in the analysis of mean annual rainfall over the last 30 years holds significance as it reflects the relative variability or stability in precipitation patterns within the study area during this period. A CV value of 26.37% suggests a moderate level of variability around the mean annual rainfall. This information is crucial for understanding the consistency or fluctuations in precipitation, which is pivotal for various applications, including agriculture, water resource management, and climate studies (Ogunrinde et al. 2019; Zhu et al. 2022). Moderate variance was reported in the months of February and December, with CV values of 167.87% and 222.89%, respectively. The observed highest variances in rainfall, particularly in February and December, can be attributed to several factors influencing precipitation patterns in the study area. In February, factors such as atmospheric circulation patterns, local topography, and climate anomalies can contribute to the pronounced variability. December, being the beginning of the dry season, experiences irregularities such as sporadic rainfall events and the influence of climate phenomena such as El Niño or La Niña, which can significantly impact precipitation patterns (Supari et al. 2018).

Mann–Kendall test

The MK trend test is a popular approach for detecting rainfall trends in time series data. The MK test is used to find monotonic (increasing or declining) patterns in yearly and seasonal rainfall parameter bases. The MK trend test and Sen's estimator were used to explore the existence of long-term changes in rainfall indices. The MK test detects annual and seasonal trend changes with decreased sensitivity to climate outliers. However, if autocorrelation occurs in the time series data, the MK test result may contain some mistake. To address this issue, the pre-whitening process was used, and there was no significant amount of serial autocorrelation at all lags and with no modifications. The MK test from the Z value and trend from Sen's slope (s) calculation were computed based on monthly and annual rainfall data from 1990 to 2020 in the Tana sub-basin, and the results are shown in Table 6.

Table 6

Summary statistical and monthly MK test results for Tana sub-basin (1990–2020)

MonthMean (mm)SD (mm)CV (%)Sen's (s)MK test (P-value)MK test (Z)Trend
Jan 7.589 30.886 407.008 −84 0.019* −1.293 
Feb 12.09 20.295 167.872 −39 0.648 −0.456 
March 36.528 43.54 119.197 −313 0.0004*** −3.523 
April 44.459 30.529 68.667 −133 0.137 −1.486 
May 58.578 40.586 69.286 111 0.215 1.238 
June 100.46 54.694 54.444 −213 0.017** −2.389 
July 325.293 122.713 37.724 2.9 0.753 0.314 
Aug 334.986 93.691 27.969 −1.7 0.087** −1.708 
Sep 114.025 50.833 44.58 −86 0.338 −0.956 
Oct 34.563 39.283 113.658 3.3 0.717 0.361 
Nov 9.546 15.068 157.837 112 0.017* 1.361 
Dec 10.028 22.351 222.899 −79 0.297 −1.041 
Annual 1,088.147 286.971 26.372 −133 0.138 −1.482 
MonthMean (mm)SD (mm)CV (%)Sen's (s)MK test (P-value)MK test (Z)Trend
Jan 7.589 30.886 407.008 −84 0.019* −1.293 
Feb 12.09 20.295 167.872 −39 0.648 −0.456 
March 36.528 43.54 119.197 −313 0.0004*** −3.523 
April 44.459 30.529 68.667 −133 0.137 −1.486 
May 58.578 40.586 69.286 111 0.215 1.238 
June 100.46 54.694 54.444 −213 0.017** −2.389 
July 325.293 122.713 37.724 2.9 0.753 0.314 
Aug 334.986 93.691 27.969 −1.7 0.087** −1.708 
Sep 114.025 50.833 44.58 −86 0.338 −0.956 
Oct 34.563 39.283 113.658 3.3 0.717 0.361 
Nov 9.546 15.068 157.837 112 0.017* 1.361 
Dec 10.028 22.351 222.899 −79 0.297 −1.041 
Annual 1,088.147 286.971 26.372 −133 0.138 −1.482 

Note: *, **, *** significant at 0.1, 0.05, 0.01 levels, respectively; I = increasing trend, D = decreasing trend.

The MK trend test revealed a statistically significant decreasing trend for the months of January, March, June, and August at 10%, 1%, 5%, and 5% levels of significance, respectively (Table 6), while there was a considerable upward trend in November. In October and July, there was no significant increase in rainfall, whereas there was no significant decrease in February, April, September, and December. In terms of rainfall volumes, the monthly share of rainfall showed a 61.54% negative trend and a 39.33% positive trend. Winter and spring, on the other hand, showed a decreasing tendency at a 5% and 1% level of significance, respectively. The geographical trend analysis for the study region likewise reveals a consistent pattern with the area-averaged trend analysis, which is a lowering tendency of the winter and spring rainfall in the region. Annual rainfall in summer and autumn has decreased in a non-significant way.

Seasonal rainfall variability is illustrated in Tables 7and 8 for the winter, autumn, summer, and spring seasons. The seasonal rainfall variability trend for the winter, autumn, summer, and spring seasons was −0.431, −0.676, −1.123, and 0.043 mm/year, respectively. The negative and positive numbers indicate, respectively, a declining and increasing trend of rainfall distribution over the measured period. The autumn rainfall pattern shows a decreasing trend but is not statistically significant with p-value 0.1567, which is greater than the significance level (0.05). The autumn season's rainfall is critical for farmers since it impacts their preparation and planting operations in the studied area.

Table 7

Seasonal MK test results for Tana sub-basin (1990–2020)

SeasonsMean (mm)SD (mm)CV (%)Sen's (s)MK test (P-value)MK test (Z)
Winter 9.902 2.253 22.754 −55 0.5521 −0.0059*** 
Autumn 46.522 11.169 24.008 −127 0.1567 −1.4164 
Summer 253.580 132.694 52.328 −106 0.238 −1.1797 
Spring 52.711 54.553 103.493 −3 0.982 −0.0224* 
SeasonsMean (mm)SD (mm)CV (%)Sen's (s)MK test (P-value)MK test (Z)
Winter 9.902 2.253 22.754 −55 0.5521 −0.0059*** 
Autumn 46.522 11.169 24.008 −127 0.1567 −1.4164 
Summer 253.580 132.694 52.328 −106 0.238 −1.1797 
Spring 52.711 54.553 103.493 −3 0.982 −0.0224* 

Note: *, *** significant at 0.1, 0.01 levels, respectively.

The regression model was used to compute the annual and seasonal patterns of rainfall analysis, and the findings are described in Table 8. The annual and seasonal CV computations show substantial seasonal rainfall variation at the p-value of 0.05 level of significance. Annual rainfall variations, on the other hand, revealed a non-significant drop in trend, with the average rate of change computed as −0.547 mm/year. Summer has had the most rainfall during the previous 30 years, with a mean annual rainfall of 253.58 mm. In most parts of Ethiopia, the summer season receives the most rainfall compared with the other three seasons. Winter, on the other hand, was the dry season across much of Ethiopia, including the study region, with the least quantity of rainfall compared with the other seasons. With 46.5 and 52.71 mm, respectively, the autumn and spring seasons received the third and second largest amounts of rainfall after the summer seasons. The amount of rainfall seen in the summer season differed from the norm by 132.6 mm, representing a 52.3% deviation over the last 30 years of observation. Furthermore, the CV value of the spring season was exceptionally high, accounting for 103.49% of the overall CV value of the four seasons (Figure 12). This demonstrates that the rainfall variability trend was extremely high in this season across the study area.
Table 8

Linear regression results of seasonal rainfall over Tana sub-basin (1990–2020)

SeasonsMean (mm)SD (mm)CV (%)R2MK test (P-value)Change in rainfall (mm/year)
Winter 9.902 2.253 22.754 0.088 0.5521 −0.431 
Autumn 46.522 11.169 24.008 0.083 0.1567 −0.676 
Summer 253.580 132.694 52.328 0.037 0.238 −1.123 
Spring 52.711 54.553 103.493 0.005 0.982 0.043 
SeasonsMean (mm)SD (mm)CV (%)R2MK test (P-value)Change in rainfall (mm/year)
Winter 9.902 2.253 22.754 0.088 0.5521 −0.431 
Autumn 46.522 11.169 24.008 0.083 0.1567 −0.676 
Summer 253.580 132.694 52.328 0.037 0.238 −1.123 
Spring 52.711 54.553 103.493 0.005 0.982 0.043 
Figure 12

Seasonal rainfall variability trend over the study region (1990–2020).

Figure 12

Seasonal rainfall variability trend over the study region (1990–2020).

Close modal
There was relatively little rainfall in the study area in 2002, 2003, and 2004, with only 585 mm of yearly rainfall. In contrast, unusually high amounts of rainfall were seen in 2006 and 2020, accounting for 1,534 and 1,472 mm of annual rainfall, respectively. With an R2 value of 0.075, the coefficient of determination value between annual rainfall and observed years can be said to be quite low. This demonstrates that the amount of annual rainfall is not considerably governed by the passage of time but rather by other environmental and topographic characteristics. Overall yearly rainfall statistics over the previous 30 years show a decreasing tendency (Figure 13). The amount of yearly rainfall in the study region was reduced by −6.56 mm/year. The decreasing trends in yearly rainfall over the study region could be attributed to environmental, economic, and political factors. The excessive rainfall pattern, particularly the late arrival of the autumn season, supported the existing rainfall variability across the study area.
Figure 13

Annual rainfall trends over the study area (1990–2020).

Figure 13

Annual rainfall trends over the study area (1990–2020).

Close modal

Rainfall anomaly index

Van Rooy (1965) developed the rainfall anomaly index (RAI), which uses the rank of precipitation readings to calculate positive and negative precipitation anomalies. The RAI was originally calculated for the 30-year period 1990–2020 across the Tana sub-basin area. Figure 14 depicts the largest rainfall anomalies for the period of 30 years (1990–2020). The findings show that moderately dry rainfall anomalies account for the largest proportion of the frequency of years in the study area. It can be observed that a moderately dry anomaly occurred for eight consecutive years, namely 2000, 2001, 2002, 2003, 2004, 2005, 2006, and 2011. A near-normal anomaly index was observed in the years 1990, 1991, 1992, 1996, 1997, 2007, and 2009 with RAI values of −0.31, −0.37, −0.33, 0.47, −0.24, −0.17, and −0.20, respectively.
Figure 14

RAI calculated for the 30-year study period in the Tana sub-basin.

Figure 14

RAI calculated for the 30-year study period in the Tana sub-basin.

Close modal
A slightly wet anomaly index occurred with equal frequency in seven years, namely 1994, 1995, 1999, 2010, 2013, 2014, and 2016, with RAI class values ranging from 0.5 to 0.9 over the past 30 years. A moderately wet anomaly was also observed in the study region over the years 1993, 1998, and 2006, with RAI values of 1.13, 1.09, and 1.55, respectively. Slightly dry anomalies were observed in the years 2012, 2015, and 2017 over the Tana sub-basin. Based on the RAI analysis results, the study area is characterized as having a moderately dry condition with an eight-year frequency over the past 30 years, as indicated in Table 9 and Figure 15. In addition, the monthly rainfall anomaly index calculated for the study area from 1990 to 2020 is shown in Figure 16.
Table 9

Estimated RAI with their respective frequency of years over the past 30 years (1990–2020)

RAI class descriptionFrequencyYears
Near-normal 1990, 1991, 1992, 1996, 1997, 2007 and 2009 
Normal 2008 and 2020 
Moderately wet 1993, 1998 and 2006 
Moderately dry 2000, 2001, 2002, 2003, 2004, 2005, 2006 and 2011 
Slightly dry 2012, 2015 and 2017 
Slightly wet 1994, 1995, 1999, 2010, 2013, 2014 and 2016 
RAI class descriptionFrequencyYears
Near-normal 1990, 1991, 1992, 1996, 1997, 2007 and 2009 
Normal 2008 and 2020 
Moderately wet 1993, 1998 and 2006 
Moderately dry 2000, 2001, 2002, 2003, 2004, 2005, 2006 and 2011 
Slightly dry 2012, 2015 and 2017 
Slightly wet 1994, 1995, 1999, 2010, 2013, 2014 and 2016 
Figure 15

Frequency of RAI distribution with index class description in the study region (1990–2020).

Figure 15

Frequency of RAI distribution with index class description in the study region (1990–2020).

Close modal
Figure 16

Monthly rainfall anomaly index calculated for the study area (1990–2020).

Figure 16

Monthly rainfall anomaly index calculated for the study area (1990–2020).

Close modal

Precipitation concentration index

The study area PCI was also generated to assess rainfall variability across the study period (1990–2020). The study period was divided into three classes to allow for a clearer evaluation of the PCI trend over the study region. The first designation was an almost ten-year study period examination spanning the years 1990 to 2000. The second group was a ten-year study period description beginning in 2000 and ending in 2010. The final designations were a ten-year study period description beginning in 2010 and ending in 2020. As shown in Figure 17, the PCI values showed an up-and-down trend for ten consecutive years from 1990 to 2000. It dropped from 1990 to 1997 with PCI values of 21.31–15.19 within seven consecutive years.

Starting from 1997 to 2000, it then increased from a PCI value of 15.19 to 17.23. Thus, the PCI followed decreasing and increasing trends for the ten consecutive years categorized under the first designation. Starting from 2000, the PCI values increased from 17.23 to 26.01 within ten consecutive years (2000–2010). However, the PCI still followed an up–and-down trend within that ten-year designation (2000–2010). It significantly dropped from 17.23 to 16.29 from the year of 2000 to 2006 for six consecutive years. Following that, it again rose from 16.29 to 27.75 from 2006 to 2009 and then declined to 26.01 in 2010. The third designation period (2010–2020) following the same PCI trends as those of the first and second designation periods. As indicated in Figure 17, a high incremental trend was observed between 2011 and 2012 with PCI values of 19.98–30.85 and this was the highest incremental trend over a year (2011–2012). The third designation period (2010–2020) showed the highest declining trend of PCI values over the past 30 years in the Tana sub-basin with 9.89 PCI difference from 2012 to 2020. The higher the PCI values of rainfall, the greater the annual rainfall fluctuations, which may influence the vegetation condition.
Figure 17

PCI for annual rainfall series (1990–2020).

Figure 17

PCI for annual rainfall series (1990–2020).

Close modal

Meanwhile, the lower the PCI values of the rainfall pattern, the lower the annual rainfall fluctuations will be. This argument is supported by the fact that rainfall fluctuation has steadily decreased from 1990 to 2020, reflecting the lowering tendencies of yearly rainfall fluctuation over 30 years. This rainfall fluctuation significantly affected the vegetation condition in the study region and this situation was confirmed with field observations from the ground. The rainfall distribution pattern was somewhat lowered with an up-and-down trend from 1990 to 2020, which was mainly linked to the state of vegetation condition. In summary, the PCI values of the research region demonstrated that the rainfall pattern distribution was aberrant. The annual rainfall fluctuation in the study area can be said to be particularly high throughout the past 30 years of investigation.

The robust capability of satellite rainfall estimates to detect precipitation is evident through a synergistic combination of high POD, FBI, and HSS, along with a low FAR. In the Tana sub-basin, the CHIRPS rainfall product tends to overestimate the frequency of rainy days at the Dangila station (POD = 1.00) and underestimate the frequency at the Addis Zemen gauge station (POD = 0.95, FAR = 0.25, and FBI = 1.27). Our investigation demonstrates the commendable performance of the CHIRPS product, especially when compared with a previous study in the surrounding area (Fenta et al. 2018), where POD ranged between 0.54 and 0.6 and FBI ranged between 0.70 and 0.87.

CHIRPS utilizes a fixed TIR cloud coldness duration threshold of 235 K, without temporal or spatial fluctuations. This fixed threshold may contribute to the slight underestimation of rainy days in highland regions, where warm-rain mechanisms related to the warm orographic rain process dominate, rendering the threshold relatively cold. Conversely, overestimation of rainy days in lowland regions may be attributed to sub-cloud evaporation, where the dry atmosphere beneath the clouds evaporates before reaching the surface, leading to an overestimation of rainfall (Dinku et al. 2018; Areerachakul et al. 2022).

In the study region spanning from 1990 to 2020, the annual rainfall exhibited a smaller CV (26.37%) compared with the interannual variability in rainfall. Likewise, there was reduced variability in rainfall amounts during the months of July, August, and September when contrasted with other months. This concurs with recent scholarly investigations highlighting the notable diversity and unpredictability in Ethiopia's seasonal and annual climate patterns (Gebremichael et al. 2014; Gizaw & Gan 2017; Mesfin et al. 2020). Standard rainfall anomalies, reflecting both positive and negative deviations, signify the interannual variability in rainfall over the observed time series. The year 2001 recorded the highest positive anomaly (+2.74), while 2015 marked the most substantial negative anomaly (−3.62). The adverse anomaly in 2015 was predominantly attributed to El Niño, causing detrimental impacts on the primary means of subsistence for rural communities in various regions of Ethiopia (Haile et al. 2021).

Recent studies underscore that Ethiopia has undergone approximately 12 significant historical droughts, influencing the country's economic development (Liou & Mulualem 2019; Kassaye et al. 2021). The observed negative influence of Ethiopian drought events on the livelihoods of millions of people aligns with the findings in our study, particularly in the Tana sub-basin region, from 1990 to 2020. The documented severe droughts in Ethiopia over the past three decades, connected to El Niño–Southern Oscillation occurrences (Zeleke et al. 2017; M. Wu et al. 2022; X. Wu et al. 2022), resonate with our examination of annual rainfall anomalies, revealing significant negative anomalies in 2015, a year associated with El Niño. Our study identifies drought years coinciding with El Niño episodes, including 2015, which had the largest unfavorable anomaly. This concurs with recent studies (Gleixner et al. 2017) that emphasize the effects of El Niño events on rainfall distribution in Ethiopia.

This study focuses on a comprehensive evaluation of CHIRPS satellite rainfall estimates in the Tana sub-basin by employing statistical indicators such as POD, FAR, FBI, and HSS. The analysis of CHIRPS estimates against gauge station measurements revealed fluctuations in ME, MAE, NSE, and bias error at different stations. This study emphasized the importance of understanding monthly and seasonal rainfall patterns, including the impact of elevation on precipitation distribution. Trend analyses via the MK test uncovered significant decreasing trends in specific months and seasons. RAI and PCI assessments indicated distinct patterns of moderately dry conditions and changing trends in rainfall concentration over the past three decades. Overall, the study provides crucial insights into the performance and implications of CHIRPS estimates, spatial–temporal rainfall dynamics, and long-term trends in the Tana sub-basin, influencing various sectors such as agriculture, water resource management, and climate studies.

The authors acknowledge Department of Earth Science, College of Natural and Computational Science, Wollega University Nekemte Campus; Department of Geography and Environmental Studies, College of Social Science and Humanities, Arsi University; Department of Forestry, College of Natural Resource and Agricultural Economics, Mettu University Bedele Campus, Bedele, Ethiopia; Department of Geography and Environmental Studies, Raya University, College of Natural Resource and Agricultural Economics; Department of Forestry and Jimma University College of Agriculture and Veterinary Medicine for the existing facilities to conduct this research.

S.S.D. participated in research design, data collection, and data analysis and draft manuscript. Z.R.R. and M.B.M. participated in methodology. K.T.D., H.H.H., and D.O.G. worked on literature and edited the manuscript to the journal style. All authors read and approved the final manuscript.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.

Aher
S.
,
Shinde
S.
,
Gawali
P.
,
Deshmukh
P.
&
Venkata
L. B.
2019
Spatio-temporal analysis and estimation of rainfall variability in and around upper Godavari River basin, India
.
Arabian Journal of Geosciences
12
(
22
),
682
.
doi:10.1007/s12517-019-4869-z
.
Ahmad
W.
,
Fatima
A.
,
Awan
U. K.
&
Anwar
A.
2014
Analysis of long term meteorological trends in the middle and lower Indus Basin of Pakistan – a non-parametric statistical approach
.
Global and Planetary Change
122
,
282
291
.
doi:10.1016/j.gloplacha.2014.09.007
.
Ahmad
I.
,
Zhang
F.
,
Tayyab
M.
,
Anjum
M. N.
,
Zaman
M.
,
Liu
J.
,
Farid
H. U.
&
Saddique
Q.
2018
Spatiotemporal analysis of precipitation variability in annual, seasonal and extreme values over upper Indus River basin
.
Atmospheric Research
213
,
346
360
.
doi:10.1016/j.atmosres.2018.06.019
.
Areerachakul
N.
,
Prongnuch
S.
,
Longsomboon
P.
&
Kandasamy
J.
2022
Quantitative precipitation estimation (QPE) rainfall from meteorology radar over Chi Basin
.
Hydrology
9
(
10
),
178
.
doi:10.3390/hydrology9100178
.
Asfaw
A.
,
Simane
B.
,
Hassen
A.
&
Bantider
A.
2018
Variability and time series trend analysis of rainfall and temperature in northcentral Ethiopia: a case study in Woleka sub-basin
.
Weather and Climate Extremes
19
,
29
41
.
doi:10.1016/j.wace.2017.12.002
.
Ayehu
G. T.
,
Tadesse
T.
,
Gessesse
B.
&
Dinku
T.
2018
Validation of new satellite rainfall products over the Upper Blue Nile Basin, Ethiopia
.
Atmospheric Measurement Techniques
11
(
4
),
1921
1936
.
doi:10.5194/amt-11-1921-2018
.
Behailu
G.
,
Ayal
D. Y.
,
Zeleke
T. T.
,
Ture
K.
&
Bantider
A.
2021
Comparative analysis of meteorological records of climate variability and farmers’ perceptions in Sekota Woreda, Ethiopia
.
Climate Services
23
,
100239
.
doi:10.1016/j.cliser.2021.100239
.
Brunetti
M. T.
,
Melillo
M.
,
Gariano
S. L.
,
Ciabatta
L.
,
Brocca
L.
,
Amarnath
G.
&
Peruccacci
S.
2021
Satellite rainfall products outperform ground observations for landslide prediction in India
.
Hydrology and Earth System Sciences
25
(
6
),
3267
3279
.
doi:10.5194/hess-25-3267-2021
.
Chen
H.
,
Ye
Q.
,
Wang
X.
,
Sheng
J.
,
Yu
X.
,
Zhao
S.
,
Zou
X.
,
Zhang
W.
&
Xue
G.
2024
Applying sludge hydrolysate as a carbon source for biological denitrification after composition optimization via red soil filtration
.
Water Research
249
,
120909
.
doi:10.1016/j.watres.2023.120909
.
Cheng
Y.
,
Lan
S.
,
Fan
X.
,
Tjahjadi
T.
,
Jin
S.
&
Cao
L.
2023
A dual-branch weakly supervised learning based network for accurate mapping of woody vegetation from remote sensing images
.
International Journal of Applied Earth Observation and Geoinformation
124
,
103499
.
doi:10.1016/j.jag.2023.103499
.
Cimini
D.
,
Romano
F.
,
Ricciardelli
E.
,
Di Paola
F.
,
Viggiano
M.
,
Marzano
F. S.
,
Colaiuda
V.
,
Picciotti
E.
,
Vulpiani
G.
&
Cuomo
V.
2013
Validation of satellite OPEMW precipitation product with ground-based weather radar and rain gauge networks
.
Atmospheric Measurement Techniques
6
(
11
),
3181
3196
.
doi:10.5194/amt-6-3181-2013
.
Dile
Y. T.
,
Karlberg
L.
,
Daggupati
P.
,
Srinivasan
R.
,
Wiberg
D.
&
Rockström
J.
2016
Assessing the implications of water harvesting intensification on upstream–downstream ecosystem services: a case study in the Lake Tana basin
.
Science of the Total Environment
542
,
22
35
.
doi:10.1016/j.scitotenv.2015.10.065
.
Dinku
T.
,
Funk
C.
,
Peterson
P.
,
Maidment
R.
,
Tadesse
T.
,
Gadain
H.
&
Ceccato
P.
2018
Validation of the CHIRPS satellite rainfall estimates over eastern Africa
.
Quarterly Journal of the Royal Meteorological Society
144
(
S1
),
292
312
.
doi:10.1002/qj.3244
.
Dong
W.
,
Zhao
J.
,
Qu
J.
,
Xiao
S.
,
Li
N.
,
Hou
S.
&
Li
Y.
2023
Abundance matrix correlation analysis network based on hierarchical multihead self-cross-hybrid attention for hyperspectral change detection
.
IEEE Transactions on Geoscience and Remote Sensing
61
,
5501513
.
doi:10.1109/TGRS.2023.3235401
.
Du
W.
&
Wang
G.
2014
Fully probabilistic seismic displacement analysis of spatially distributed slopes using spatially correlated vector intensity measures
.
Earthquake Engineering & Structural Dynamics
43
(
5
),
661
679
.
doi:10.1002/eqe.2365
.
Esayas
B.
,
Simane
B.
,
Teferi
E.
,
Ongoma
V.
&
Tefera
N.
2019
Climate variability and farmers’ perception in Southern Ethiopia
.
Advances in Meteorology
2019
,
7341465
.
doi:10.1155/2019/7341465
.
Fazzini
M.
,
Bisci
C.
&
Billi
P.
2015
The climate of Ethiopia
. In:
Landscapes and Landforms of Ethiopia (Billi, P., ed.)
,
Springer
,
Dordrecht, The Netherlands
, pp.
65
87
.
doi:10.1007/978-94-017-8026-1_3
.
Fenta
A. A.
,
Yasuda
H.
,
Shimizu
K.
,
Ibaraki
Y.
,
Haregeweyn
N.
,
Kawai
T.
,
Belay
A. S.
,
Sultan
D.
&
Ebabu
K.
2018
Evaluation of satellite rainfall estimates over the Lake Tana basin at the source region of the Blue Nile River
.
Atmospheric Research
212
,
43
53
.
doi:10.1016/j.atmosres.2018.05.009
.
Gebremichael
A.
,
Quraishi
S.
&
Mamo
G.
2014
Analysis of seasonal rainfall variability for agricultural water resource management in Southern Region, Ethiopia
.
Journal of Natural Sciences Research
4
(
11
),
56
79
.
Gizaw
M. S.
&
Gan
T. Y.
2017
Impact of climate change and El Niño episodes on droughts in sub-Saharan Africa
.
Climate Dynamics
49
(
1–2
),
665
682
.
doi:10.1007/s00382-016-3366-2
.
Gleixner
S.
,
Keenlyside
N.
,
Viste
E.
&
Korecha
D.
2017
The El Niño effect on Ethiopian summer rainfall
.
Climate Dynamics
49
(
5–6
),
1865
1883
.
doi:10.1007/s00382-016-3421-z
.
Gocic
M.
,
Shamshirband
S.
,
Razak
Z.
,
Petković
D.
,
Chintalapati
S.
&
Trajkovic
S.
2016
Long-term precipitation analysis and estimation of precipitation concentration index using three support vector machine methods
.
Advances in Meteorology
2016
,
7912357
.
doi:10.1155/2016/7912357
.
Guan
H.
,
Huang
J.
,
Li
L.
,
Li
X.
,
Miao
S.
,
Su
W.
,
Ma
Y.
,
Niu
Q.
&
Huang
H.
2023
Improved Gaussian mixture model to map the flooded crops of VV and VH polarization data
.
Remote Sensing of Environment
295
,
113714
.
doi:10.1016/j.rse.2023.113714
.
Haile
B. T.
,
Zeleke
T. T.
,
Beketie
K. T.
,
Ayal
D. Y.
&
Feyisa
G. L.
2021
Analysis of El Niño Southern Oscillation and its impact on rainfall distribution and productivity of selected cereal crops in Kembata Alaba Tembaro zone
.
Climate Services
23
,
100254
.
doi:10.1016/j.cliser.2021.100254
.
Hasan
M. M.
,
Sharma
A.
,
Mariethoz
G.
,
Johnson
F.
&
Seed
A.
2016
Improving radar rainfall estimation by merging point rainfall measurements within a model combination framework
.
Advances in Water Resources
97
,
205
218
.
doi:10.1016/j.advwatres.2016.09.011
.
He
M.
,
Dong
J.
,
Jin
Z.
,
Liu
C.
,
Xiao
J.
,
Zhang
F.
,
Sun
H.
,
Zhao
Z.
,
Gou
L.
,
Liu
W.
,
Luo
C.
,
Song
Y.
,
Ma
L.
&
Deng
L.
2021
Pedogenic processes in loess-paleosol sediments: clues from Li isotopes of leachate in Luochuan loess
.
Geochimica et Cosmochimica Acta
299
,
151
162
.
doi:10.1016/j.gca.2021.02.021
.
Houze
R. A.
2014
Clouds and precipitation associated with hills and mountains
. In:
Cloud Dynamics (Houze, R. A., ed.)
.
Academic Press
,
Oxford, UK
, pp.
369
402
.
doi:10.1016/B978-0-12-374266-7.00012-3
.
Kassaye
A. Y.
,
Shao
G.
,
Wang
X.
&
Wu
S.
2021
Quantification of drought severity change in Ethiopia during 1952–2017
.
Environment, Development and Sustainability
23
(
4
),
5096
5121
.
doi:10.1007/s10668-020-00805-y
.
Kogo
B. K.
,
Kumar
L.
&
Koech
R.
2021
Climate change and variability in Kenya: a review of impacts on agriculture and food security
.
Environment, Development and Sustainability
23
(
1
),
23
43
.
doi:10.1007/s10668-020-00589-1
.
Li
J.
,
Wang
Z.
,
Wu
X.
,
Xu
C.
,
Guo
S.
&
Chen
X.
2020
Toward monitoring short-term droughts using a novel daily scale, standardized antecedent precipitation evapotranspiration index
.
Journal of Hydrometeorology
21
(
5
),
891
908
.
doi:10.1175/JHM-D-19-0298.1
.
Liou
Y. A.
&
Mulualem
G. M.
2019
Spatio-temporal assessment of drought in Ethiopia and the impact of recent intense droughts
.
Remote Sensing
11
(
15
),
1828
.
doi:10.3390/rs11151828
.
Liu
Q. Y.
,
Li
D. Q.
,
Tang
X. S.
&
Du
W.
2023
Predictive models for seismic source parameters based on machine learning and general orthogonal regression approaches
.
Bulletin of the Seismological Society of America
113
(
6
),
2363
2376
.
doi:10.1785/0120230069
.
Mesfin
D.
,
Simane
B.
,
Belay
A.
,
Recha
J. W.
&
Schmiedel
U.
2020
Assessing the adaptive capacity of households to climate change in the Central Rift Valley of Ethiopia
.
Climate
8
(
10
),
106
.
doi:10.3390/cli8100106
.
Namara
R. E.
,
Hanjra
M. A.
,
Castillo
G. E.
,
Ravnborg
H. M.
,
Smith
L.
&
Van Koppen
B.
2010
Agricultural water management and poverty linkages
.
Agricultural Water Management
97
(
4
),
520
527
.
doi:10.1016/j.agwat.2009.05.007
.
Ogunrinde
A. T.
,
Oguntunde
P. G.
,
Akinwumiju
A. S.
&
Fasinmirin
J. T.
2019
Analysis of recent changes in rainfall and drought indices in Nigeria, 1981–2015
.
Hydrological Sciences Journal
64
(
14
),
1755
1768
.
doi:10.1080/02626667.2019.1673396
.
Pepin
N. C.
,
Arnone
E.
,
Gobiet
A.
,
Haslinger
K.
,
Kotlarski
S.
,
Notarnicola
C.
,
Palazzi
E.
,
Seibert
P.
,
Serafin
S.
,
Schöner
W.
,
Terzago
S.
,
Thornton
J. M.
,
Vuille
M.
&
Adler
C.
2022
Climate changes and their elevational patterns in the mountains of the world
.
Reviews of Geophysics
60
(
1
),
e2020RG000730
.
doi:10.1029/2020RG000730
.
Ritter
A.
&
Muñoz-Carpena
R.
2013
Performance evaluation of hydrological models: statistical significance for reducing subjectivity in goodness-of-fit assessments
.
Journal of Hydrology
480
,
33
45
.
doi:10.1016/j.jhydrol.2012.12.004
.
Shang
K.
,
Xu
L.
,
Liu
X.
,
Yin
Z.
,
Liu
Z.
,
Li
X.
,
Yin
L.
&
Zheng
W.
2023
Study of urban heat island effect in Hangzhou metropolitan area based on SW-TES algorithm and image dichotomous model
.
SAGE Open
13
(
4
).
doi:10.1177/21582440231208851
.
Shen
Z.
,
Yong
B.
,
Gourley
J. J.
,
Qi
W.
,
Lu
D.
,
Liu
J.
,
Ren
L.
,
Hong
Y.
&
Zhang
J.
2020
Recent global performance of the Climate Hazards group Infrared Precipitation (CHIRP) with Stations (CHIRPS)
.
Journal of Hydrology
591
,
125284
.
doi:10.1016/j.jhydrol.2020.125284
.
Supari
Tangang, F.
,
Salimun
E.
,
Aldrian
E.
,
Sopaheluwakan
A.
&
Juneng
L.
2018
ENSO modulation of seasonal rainfall and extremes in Indonesia
.
Climate Dynamics
51
(
7–8
),
2559
2580
.
doi:10.1007/s00382-017-4028-8
.
Tigabu
T. B.
,
Wagner
P. D.
,
Hörmann
G.
&
Fohrer
N.
2019
Modeling the impact of agricultural crops on the spatial and seasonal variability of water balance components in the Lake Tana basin, Ethiopia
.
Hydrology Research
50
(
5
),
1376
1396
.
doi:10.2166/nh.2019.170
.
Toté
C.
,
Patricio
D.
,
Boogaard
H.
,
Van der Wijngaart
R.
,
Tarnavsky
E.
&
Funk
C.
2015
Evaluation of satellite rainfall estimates for drought and flood monitoring in Mozambique
.
Remote Sensing
7
(
2
),
1758
1776
.
doi:10.3390/rs70201758
.
Van Rooy
M. P.
1965
A rainfall anomaly index independent of time and space
.
Notos
14
,
43
48
.
Wakjira
M. T.
,
Peleg
N.
,
Anghileri
D.
,
Molnar
D.
,
Alamirew
T.
,
Six
J.
&
Molnar
P.
2021
Rainfall seasonality and timing: implications for cereal crop production in Ethiopia
.
Agricultural and Forest Meteorology
310
,
108633
.
doi:10.1016/j.agrformet.2021.108633
.
Worqlul
A. W.
,
Ayana
E. K.
,
Maathuis
B. H. P.
,
MacAlister
C.
,
Philpot
W. D.
,
Osorio Leyton
J. M.
&
Steenhuis
T. S.
2018
Performance of bias corrected MPEG rainfall estimate for rainfall-runoff simulation in the upper Blue Nile Basin, Ethiopia
.
Journal of Hydrology
556
,
1182
1191
.
doi:10.1016/j.jhydrol.2017.01.058
.
Wu
M.
,
Ba
Z.
&
Liang
J.
2022
A procedure for 3D simulation of seismic wave propagation considering source-path-site effects: theory, verification and application
.
Earthquake Engineering & Structural Dynamics
51
(
12
),
2925
2955
.
doi:10.1002/eqe.3708
.
Wu
X.
,
Guo
S.
,
Qian
S.
,
Wang
Z.
,
Lai
C.
,
Li
J.
&
Liu
P.
2022
Long-range precipitation forecast based on multipole and preceding fluctuations of sea surface temperature
.
International Journal of Climatology
42
(
15
),
8024
8039
.
doi:10.1002/joc.7690
.
Xu
Z.
,
Li
X.
,
Li
J.
,
Xue
Y.
,
Jiang
S.
,
Liu
L.
,
Luo
Q.
,
Wu
K.
,
Zhang
N.
,
Feng
Y.
,
Shao
M.
,
Jia
K.
&
Sun
Q.
2022
Characteristics of source rocks and genetic origins of natural gas in deep formations, Gudian Depression, Songliao Basin, NE China
.
ACS Earth and Space Chemistry
6
(
7
),
1750
1771
.
doi:10.1021/acsearthspacechem.2c00065
.
Yuan
J.
,
Wang
T. J.
,
Chen
J.
&
Huang
J. A.
2023
Microscopic mechanism study of the creep properties of soil based on the energy scale method
.
Frontiers in Materials
10
,
1137728
.
doi:10.3389/fmats.2023.1137728
.
Zeleke
T. T.
,
Giorgi
F.
,
Diro
G. T.
&
Zaitchik
B. F.
2017
Trend and periodicity of drought over Ethiopia
.
International Journal of Climatology
37
(
13
),
4733
4748
.
doi:10.1002/joc.5122
.
Zhou
G.
,
Wang
Z.
&
Li
Q.
2022a
Spatial negative co-location pattern directional mining algorithm with join-based prevalence
.
Remote Sensing
14
(
9
),
2103
.
doi:10.3390/rs14092103
.
Zhou
G.
,
Zhou
X.
,
Chen
J.
,
Jia
G.
&
Zhu
Q.
2022b
LiDAR echo Gaussian decomposition algorithm for FPGA implementation
.
Sensors
22
(
12
),
4628
.
doi:10.3390/s22124628
.
Zhou
G.
,
Zhang
H.
,
Xu
C.
,
Zhou
X.
,
Liu
Z.
,
Zhao
D.
,
Lin
J.
&
Wu
G.
2023
A real-time data acquisition system for single-band bathymetric LiDAR
.
IEEE Transactions on Geoscience and Remote Sensing
61
,
5702721
.
doi:10.1109/TGRS.2023.3282624
.
Zhu
W.
,
Chen
J.
,
Sun
Q.
,
Li
Z.
,
Tan
W.
&
Wei
Y.
2022
Reconstructing of high-spatial-resolution three-dimensional electron density by ingesting SAR-derived VTEC into IRI model
.
IEEE Geoscience and Remote Sensing Letters
19
,
4508305
.
doi:10.1109/LGRS.2022.3178242
.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/).