Evaluation of CHIRPS Data for Hydrological Modeling in Data-Scarce West African Basins and Deep Learning Inflow Prediction

547 Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 p ISSN: 2635-3342; e ISSN: 2635-3350 Original Research Article Evaluation of CHIRPS Data for Hydrological Modeling in Data-Scarce West African Basins and Deep Learning Inflow Prediction *1Sule, J., 2Momodu, O.A. and 1Ibrahim, A.O. 1Department of Civil Engineering, Edo State University, Iyamho, Edo State, Nigeria. 2Department of Estate and Works, Auchi Polytechnics, Edo State, Nigeria. *[email protected] http://doi.org/10.5281/zenodo.18061884 ARTICLE INFORMATION ABSTRACT Article history: Received 24 Oct. 2025 Revised 13 Nov. 2025 Accepted 14 Nov. 2025 Available online 30 Dec. 2025 West Africa faces critical challenges in water and energy development due to limited ground-based hydrological data measurements. This study validates satellite data downloaded from Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) and develops a seasonal Long Short-Term Memory (LSTM) model for predicting dam inflow at Doma Dam, Nigeria. 37 years of CHIRPS dataset (1986-2023), integrated with the climate indices El Niño Southern Oscillation (ENSO) and the Atlantic Multidecadal Oscillation (AMO), was validated against 20 years (1996-2015) ground-based records, achieving a strong correlation (r = 0.774, p < 0.001) with acceptable performance metrics: Mean Absolute Error (MAE) = 49.25 mm, Root Mean Square Error (RMSE) = 71.85 mm, and Nash-Sutcliffe Efficiency (NSE) = 0.537. The LSTM model demonstrated robust forecasting capability with R² = 0.8096, RMSE = 19.54 m³/s, and correlation = 0.9231 on test data, explaining over 75% of inflow variance while maintaining consistent performance across training and validation sets. Future predictions (2024-2030) successfully captured seasonal patterns, with wet season averaging 37.03 m³/s and dry season 10.29 m³/s. This validated CHIRPS-LSTM framework offers practical applications for hydrological forecasting in data-limited West African regions, supporting dam operations, hydropower development, water resource planning, and renewable energy development. The methodology is transferable across the Sudano-Sahelian region, demonstrating satellite data's potential for evidence-based decision-making where ground-based observations are scarce. © 2025 RJEES. All rights reserved. Keywords: Hydrological modeling Data-scare basins CHIRPS validation LSTM neural network Doma dam 1. INTRODUCTION The growing demand for renewable energy sources and the abundance of non-hydropower generating infrastructures present a significant opportunity for sustainable energy development in developing 548 J. Sule et al. / Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 countries. Water and energy are the most important natural resources for life on Earth. The fundamental requirements of the contemporary world. It is no surprise that water and light were the first things mentioned in God's creation tale (Tapaciok, 2024). Their contributions to the comfort of human life are unquestionable. The way a country manages its energy and water resources has the most impact on its social well-being, economic growth, and political stability. The scarcity of these resources presents a significant challenge to the government of every nation. With more than 200 million citizens, Nigeria, the most populated country in Africa, faces serious problems with electricity and water. The nation has more than 240 current dams, many of these are underutilized or built without energy generation in mind. Most of which were built for irrigation and flood control in the 1970s and 1990s (Youdeowei et al., 2019). Retrofitting these dams provides an opportunity to boost energy output without new constructions. While maintaining established water resource functions, retrofitting existing non-power dams with hydropower generation capacity provides a sustainable path for the development of renewable energy (Ilmi, 2025). Dam operation and management are becoming increasingly important for maintaining functional civilizations and economies around the world because failure in a single dam infrastructure has the ability to devastate a whole community (Aquilah et al., 2024). Dam project management requires a complex interplay of various factors, including demand volatility. The main difficulty in retrofitting dams is still striking a balance between the generation of electricity (hydropower) and other goals, including ecosystem preservation, flood control, irrigation, and water supply. For instance, flood management may recommend a significantly lower water elevation to lessen the risk of overtopping, while a hydropower target would mandate that water elevation be maintained close to maximum storage to boost power generation reliability. It is necessary to update operating standards and modify water release policies to reflect changes in energy and water demand in order to guarantee the long-term operation of retrofitting dams. It is essential to have access to trustworthy rainfall data, both historical and near-real time, in order to retrofit an existing dam to generate hydropower. For assessing climatic patterns and tracking above or below average inflow to the dam, long-term historical data (daily or monthly) can be helpful. Also, it provides information necessary to understand the amount of water retained in the dam structure. Sun et al., (2018) stated that accurate precipitation data forms the foundation of hydrological modeling, water resource management and sustainable energy planning. However, Sylla et al., (2013) stated that West African countries face significant challenges in maintaining dense, reliable ground-based rainfall monitoring networks due to limited resources, vast geographic areas, and harsh environmental conditions. In many developing regions, including parts of Africa and Asia, ground-based observation networks suffer from irregular maintenance, missing data, limited spatial coverage, and short historical records (Arregoces et al., 2023). These inconsistencies create gaps in critical datasets such as rainfall, temperature, evaporation, streamflow, and reservoir inflow variables essential for hydrological modeling, flood forecasting, and dam operation analysis. The River Basin Authority, which encompasses the study area, exemplifies these challenges with only sporadic gauge measurements despite supporting critical agricultural and municipal water supplies for over 10 million people (Abah and Petja, 2016). This data scarcity constrains hydrological modeling efforts, limiting the development of robust forecasting systems necessary for evidence-based water resources planning. To address these challenges, satellite-based remote sensing data has become a vital alternative and complement to ground-based observations (Ibrahim et al., 2024). Modern satellites provide continuous, highresolution, and spatially extensive measurements of key hydrological and meteorological parameters, including precipitation (e.g., Global Precipitation Measurement (GPM), Tropical Rainfall Measuring Mission (TRMM)), soil moisture (e.g., Soil Moisture Active Passive (SMAP)), evapotranspiration (e.g., Moderate Resolution Imaging Spectroradiometer (MODIS)), and surface water levels (e.g., altimetry missions like Surface Water and Ocean Topography (SWOT)). A game-changing tool for overcoming data shortages in underdeveloped nations is satellite-based precipitation estimation (Joyce et al., 2004; Huffman et al., 2007). Among available products, the CHIRPS developed in 1981 by Geological Survey (USGS) and University of California, Santa Barbara 549 J. Sule et al. / Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 (UCSB), has gained prominence for hydrological applications due to its: (1) high spatial resolution (0.05° × 0.05°, approximately 5.5 km at the equator), (2) long temporal coverage (1981-present), (3) quasi-global coverage (50°S-50°N), (4) incorporation of ground-based station data through blending algorithms, and (5) specific design optimization for agricultural and drought monitoring applications (Funk et al., 2015). Using advanced blending techniques, CHIRPS combines data from several sources, including atmospheric model rainfall fields (Climate Forecast System, CFSv2), Cold Cloud Duration (CCD) observations from thermal infrared satellites, and in-situ precipitation observations (Funk et al., 2015). In theory, predictions from this multi-source technique are more accurate than those from only satellites, especially in areas with complicated topography or West African convective precipitation regimes. Numerous researches have been carried out to evaluate CHIRPS's performance; however, as far as we are aware, no validation of the CHIRPS data studies in Nigeria has yet been done despite the fact that CHIRPS has been widely used globally for a number of applications, such as climate trend analysis (Ayehu et al., 2018), hydrological modeling (Duan et al., 2016), drought assessment (Toté et al., 2015), and agricultural monitoring (Katsanos et al., 2016). Previous West African validations have shown strong performance in Ethiopia (Dinku et al., 2018; NSE = 0.65-0.75), reasonable accuracy in Ghana (Dembélé and Zwart, 2016; r = 0.70-0.85), and lower performance in semi-arid Sahel regions (Dembélé and Zwart, 2016; NSE = 0.30-0.50). The need for basin-specific validation prior to practical deployment is highlighted by these spatially disparate results. Also, conventional hydrological forecasting uses statistical models (Autoregressive Integrated Moving Average (ARIMA), regression) that assume linear correlations or physically-based models (Soil and Water Assessment Tool (SWAT), Hydrologic Engineering Center - Hydrologic Modeling System (HEC-HMS)) that require large amounts of calibration data (Arnold et al., 1998; Xu, 1999). Strong substitutes for capturing intricate non-linear temporal correlations in hydrological systems are provided by recent developments in deep learning, especially LSTM neural networks (Kratzert et al., 2018; Hu et al., 2018). There are still a number of important research gaps in West African water studies, despite increasing progress in hydrological modeling. There is currently little thorough validation of CHIRPS data across Nigeria's diverse climate zones, especially for operational timescales that are important for reservoir management. The majority of research compares satellite precipitation with ground gauges, but they rarely examine how it affects streamflow modeling, which is the real indicator of operational utility in dam retrofitting decisions. Furthermore, there is still a lack of research on the use of deep learning models like LSTM in the highly seasonal, monsoon-driven basins of the region, and these models seldom ever incorporate important climatic indices like ENSO, WAM, and AMO. Last but not least, there is still a significant obstacle in moving from research-based models to operational forecasting frameworks appropriate for institutions with limited resources, which call for advancements in precision, effectiveness, and interpretability. The main objectives of this study are to validate CHIRPS satellite precipitation data for hydrological modeling applications in the Doma Dam, Nasarawa State, Nigeria, and develop a seasonal ensemble LSTM framework for operational inflow forecasting which will enable us substantiate the amount of water that can be retained in the dam for hydropower retrofitting decisions. The degree of accuracy of CHIRPS data and its possible application in different fields are determined by this validation. The results of this investigation should demonstrate a strong and trustworthy link between rainfall data from CHIRPS satellites. This research makes a number of new methodological, practical, and scientific contributions. In terms of methodology, it offers the first thorough CHIRPS validation for dam basins in Nigeria, presents a seasonal ensemble LSTM architecture that specifically accounts for wet-dry season transitions, and creates an integration framework that combines climate indices, satellite data, and ground observations with end to end validation for inflow prediction. In practice, it creates a verified 37-year hydrological dataset (1986–2023) for the data-poor Doma dam in Nigeria and suggests an operational forecasting framework appropriate for organizations with limited resources, offering a tool to assist in decisionmaking regarding dam retrofitting, reservoir management, and water distribution. Scientifically, it 550 J. Sule et al. / Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 measures the accuracy of CHIRPS across seasons and rainfall intensities, shows how deep learning techniques outperform conventional models in West Africa, emphasizes the value of climate indices in enhancing monthly forecasts, and sheds light on the seasonal modeling requirements in basins driven by the monsoon. The paper is structured as follows: Section 2 outlines the study area and methodology, Section 3 reports and discuss the results, Section 4 concludes with key findings and recommendations. 2. MATERIALS AND METHODS 2.1. Study Area Doma Dam (8°24'N, 8°18'E) is an earth-fill dam located in Nasarawa State, within the Lower Benue River Basin (Figure 1), North-central Nigeria. It was commissioned in 1986 to support dry-season irrigation for approximately 2,500 hectares of farmland serving over 15,000 farmers (Abah and Petja, 2016). The dam has a maximum capacity of 93 million cubic meters and an active storage of 65 million cubic meters, with its 1,850 square kilometer catchment area experiencing a tropical continental climate characterized by distinct wet and dry seasons, where 85-90% of annual rainfall occurs between May and October. The structure stands 18.5 meters high with a 1,200-meter crest length and currently serves multiple purposes including irrigation, domestic water supply for about 50,000 residents, flood control, and environmental flow maintenance. Although the dam does not currently generate hydropower, feasibility studies suggest potential for developing a 20-40 MW run-of-river hydropower facility with minimal modifications to the existing structure (Ilmi, A. 2025). Determining the optimal hydropower generation capacity depends on reliable longterm inflow forecasting that accounts for climate variability, which is the essential focus of this study. Figure 1: Study area map showing Doma dam location 2.2. Methods 2.2.1. Collection of hydrological data for Doma Dam from LBRBDA 20 years (1996 – 2015) monthly precipitation, water level (inflow), temperature (minimum and maximum) and evaporation data were collected. 2.2.2. Download hydrological data for Doma Dam from CHIRPS Considering the month of commissioning and effective operation of Doma dam which is March 1986 (Yusuf and Akashe, 2014), R programming script was written with the key features for Doma dam; accurate location data, regional weather network, middle belt climate characteristics and the West African Climate Indices like ENSO, AMO, Atlantic Nino Index (crucial for West African rainfall), West African Monsoon Index, Sahel Precipitation index and Indian Ocean Dipole to download 37 years (1986 – 2023) monthly robust dataset for Doma dam from Digital Earth Africa (DE Africa) which provides free and open access to CHIRPS on monthly and daily basic over Africa. The required packages for this operation in R are: lubridata, dplyr and zoo. Figure 2 gives the key elements to ensure reproducibility. 551 J. Sule et al. / Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 Figure 2: Key reproducibility elements 2.2.3. CHIRPS data validation and suitability assessment The downloaded CHIRPS data was temporally filtered to match the available ground-based data collected from the River Basin authority that manages the dam for the period of January 1996 to December 2015. The date strings of the CHIRPS dataset were converted to proper date format, and the column names were standardised before being interpolated for any missing values in the data. Also, the date format for the ground-based dataset was corrected. The original format (M/D/YYYY) used the day field to represent months (1/1/1996 = January 1996, 1/2/1996 = February 1996, etc.). The data were converted into the time series objects with a complete monthly frequency of 240 observations and validated. To evaluate the linear relationship between CHIRPS and ground-based rainfall data, Pearson product-moment correlation coefficients were computed. The correlation strength was evaluated in view of Schober et al. (2018): Strong correlation: r > 0.7, Moderate correlation: 0.5 < r ≤ 0.7 and Weak correlation: r ≤ 0.5. The CHIRPS data performance metrics were evaluated using; Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Bias and Nash-Sutcliffe Efficiency (NSE). 𝑀𝐴𝐸=(1 𝑛)∑(𝑃𝑖− 𝑂𝑖) (1) 𝑅𝑀𝑆𝐸= √[(1 𝑛)∑(𝑃𝑖− 𝑂𝑖)2] (2) 𝐵𝑖𝑎𝑠=(1 𝑛)∑(𝑃𝑖− 𝑂𝑖) (3) 𝑁𝑆𝐸=1−[∑(𝑂𝑖− 𝑃𝑖)2] [∑(𝑂𝑖− 𝑂 )2] (4) Where Pi = CHIRPS rainfall, Oi = observed rainfall and 𝑂  = mean observed rainfall and n = number of observations. Plotting the time series helped identify the trend. The stationarity assessment in this study was conducted using the Augmented Dicky–Fuller (ADF) test. When the ADF test is used, it means that the null hypothesis which holds that the hydrological time series is not stationary will be disproved. The Pearson correlation significance tests (α = 0.05) and 95% confidence intervals for correlation coefficients were used to assess 552 J. Sule et al. / Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 the statistical significance. The analysis was conducted using R statistical software (version 4.5.1) with the following packages: i. dplyr for data manipulation ii. lubridate for date handling iii. ggplot2 for visualization iv. corrplot for correlation matrices v. tseries for time series analysis vi. forecast for decomposition analysis The suitability of CHIRPS dataset was evaluated based on: i. Correlation coefficient threshold (r > 0.7 for strong relationship) ii. Nash-Sutcliffe Efficiency (NSE > 0.5 for satisfactory performance) iii. Statistical significance of correlation (p < 0.05) iv. Temporal coverage completeness and v. Bias magnitude and systematic patterns 2.2.4. Development of LSTM Model for inflow prediction The R program to develop the LSTM model was implemented in Python, reticulated through R using TensorFlow/Keras with the following configuration: The model architecture is show on Figure 3. Figure 3: The LSTM model architecture The dropout rate of 0.2 was adopted in accordance with the recommendation of Gal and Ghahramani (2016) who stated that 20% dropout prevents overfitting without excessive regularization. 553 J. Sule et al. / Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 a. Model Training and Hyperparameters i. epochs = 100, batch size = 16 ii. data split in accordance with Ng (2018) = 70:15:15 iii. n_steps = 10 (the model considers the last 10 days to predict the next inflow) b. The execution steps in R involve the following i. Load the following required libraries: lubridate, ggplot2, gridExtra, dplyr and reticulate. ii. Set Python environment; Sys.setenv(RETICULATE_PYTHON = "C:/Users/MY PC/AppData/Local/Programs/Python/Python311/python.exe") iii. Load the dataset into R, and standardized the column names, preprocessing the data with wet/dry season focus and split data into train/validation/test sets (70-15-15 split) iv. Build enhanced LSTM model for seasonal patterns and create the model. v. Evaluate the model, generate future predictions to 2030 with seasonal focus and create individual plot and compare the performance metrics. c. Model Evaluation The performance model were assessed using; R² Score: Coefficient of determination, Mean Absolute Percentage Error (MAPE) and Correlation Coefficient: Pearson correlation between actual and predicted values. d. Seasonal Analysis Framework Two season adopted for this research i. Wet Season: May to October (months 5-10) ii. Dry Season: November to April (months 11,12,1,2,3,4) e. Future prediction Future predictions (2024-2030) were generated using: i. Rolling Prediction: Using the last 12 months as seed ii. Seasonal Feature Generation: Based on historical seasonal statistics iii. Climate Index Projection: Using seasonal averages with controlled variation Figure 4 gives the workflow for the LSTM model. 3. RESULTS AND DISCUSSION 3.1. Ground-based Data Collected for Doma Dam from the Basin Authority Statistical summary of 20 years of monthly water level, rainfall, evaporation, minimum and maximum temperatures observed data for Doma dam collected from the basin authority that manages the dam is presented on Table 1. The World Meteorological Organization (WMO) requires a minimum of 30 years of continuous data for credible climatological research (Hasan et al, 2025). The available recorded dataset does not meet this standard, limiting its usefulness for dam retrofitting decisions. Without three decades of observations, the dataset fails to capture complete climate cycles, long-term trends, and extreme weather events. The quality of hydrological and meteorological data collection is seriously compromised in many developing regions of Africa and Asia due to ground-based observation networks' extensive data gaps, irregular maintenance, limited spatial coverage, and shortened historical records. This River Basin serves as an example of these limitations, as it provides vital water resources to more than 10 million people yet only sometimes has gauge measurements available. This prevents thorough hydrological modeling and evidence-based planning initiatives. 554 J. Sule et al. / Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 Thus, satellite-based remote sensing has emerged as a crucial option, providing high-resolution, continuous, and spatially extensive monitoring of critical parameters such as surface water levels, evapotranspiration, precipitation, and soil moisture. These satellite datasets provide the significant data quantities needed to create trustworthy correlations between input factors and output predictions for dam retrofitting applications, which in turn allows the creation of strong predictive models such as CHIRPS-LSTM. Figure 4: LSTM neural networks workflow Table 1: Summary of the recorded dataset Variable Minimum 1st Quartile Median Mean 3rd Quartile Maximum Water level (m) 0.05 0.50 0.60 0.58 0.73 1.00 Rainfall (mm) 0.0 0.0 90.0 106.3 197.5 298.0 Evaporation (mm) 58.0 83.0 100.0 102.7 120.0 149.0 Minimum Temperature (°C) 21.4 23.6 24.5 24.5 25.3 27.4 Maximum Temperature (°C) 23.1 32.0 33.4 33.3 34.7 36.5 3.2. Hydrological Data Downloaded for Doma Dam from CHIRPS A significant quantity of data is required for hydrological analyses for retrofitting a non-power-generating dam in order to create solid correlations between input variables and output forecasts. Kurnianingsih et al. (2025) suggested using satellite data as an option in places with insufficient data. After running the R script which contained the exact dam coordinates and parameters, 37 years (1981 – 2023) real CHIRPS monthly observations data set with overview shows 454 rows by 23 columns was downloaded from the Digital Earth Africa CHIRPS RainfallRegistry of Open Data on AWS. The variables columns title are: date, year, month, rainfall (mm), Lafia, Makurdi, Kaduna, Jos, Abuja, WAM Index, ENSO ONI, AMO Index, Weighted rainfall, Inflow (m3/s), rainfall (mA3), rainfall (MA6), season sin, season cos, ENSO Rainfall, WAM Rainfall, Inflow Lag1, Inflow Lag2, Rainfall Lag1. The script-generated files were automatically saved in the working directory, as shown in Table 2, and the statistical summary is presented in Table 3. 555 J. Sule et al. / Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 Table 2: Generated files File name Content doma_dam_lstm_sdp_dataset.csv Main dataset (all variables) doma_train_lstm_sdp.csv Training data (70%) doma_validation_lstm_sdp.csv Validation data (15%) doma_test_lstm_sdp.csv Test data (15%) doma_lstm_sdp_summary.txt Documentation Table 3: The statistical summary of the variables Year Rainfall (mm) Lafia Makurdi Kaduna Jos Abuja WAM index ENSO ONI AMO index Minimum 1986 1.000 0.50 0.20 0.10 0.30 0.40 -1.335 -4.724 -0.644 1st Quarter 1995 6.725 6.30 8.80 3.05 4.95 5.25 -0.587 -0.733 -0.132 Median 2005 55.700 47.85 48.45 36.30 36.30 40.85 -0.303 -0.103 0.097 Mean 2005 85.199 84.39 101.02 73.70 79.52 79.62 -0.057 0.000 0.076 3rd Quarter 2014 145.775 120.58 151.70 107.78 118.83 119.65 -0.655 0.672 0.280 Maximum 2023 453.500 776.10 949.10 579.30 609.10 561.30 1.514 4.004 0.654 Weighted rainfall Inflow m3s Rainfall MA3 Rainfall MA6 Season Sin Season Cos ENSO rainfall WAM rainfall Inflow Lag 1 Inflow Lag 2 Rainfall Lag 1 1.000 0.73 1.90 8.80 -1.000 -1.000 -849.49 - 145.670 0.73 0.730 1.00 6.725 1.33 15.88 35.40 -0.767 -0.769 -44.628 -10.562 1.33 1.330 6.80 55.700 2.27 67.80 86.40 0.011 0.001 -0.710 -1.295 2.32 2.685 55.80 85.199 32.90 85.50 85.45 0.003 -0.001 0.5145 41.705 32.97 33.038 85.38 145.775 64.13 150.32 129.40 0.775 0.524 14.0123 104.545 64.22 65.025 145.80 453.500 120 305.40 206.60 1.000 0.999 1110.46 452.140 120.00 120.000 453.50 3.3. Validation of CHIRPS Data and Suitability Assessment The preprocessing and integration of CHIRPS and observed datasets resulted in full temporal alignment. The merged dataset included 240 monthly observations from January 1996 to December 2015. This perfect overlap permitted strong statistical validation with no temporal gaps or missing data periods. The effective data integration addressed initial formatting issues in the observed dataset, where the unusual date structure (M/D/YYYY with day denoting month) necessitated specialized parsing techniques. This methodological technique enabled precise temporal correlation between satellite-derived and ground-based observations. Table 4 gives the monthly statistical analysis revealed of the seasonal correlation patterns. The seasonal analysis reveals that CHIRPS performance varies throughout the year: 1. Dry Season (Nov-Mar): CHIRPS overestimates low rainfall amounts, possibly due to detection threshold sensitivities 2. Wet Season (Apr-Oct): CHIRPS generally underestimates high rainfall events, with largest discrepancies during peak rainfall months (May, September, October) 3. Transition Periods: Better agreement during moderate rainfall conditions (July-August) 562 J. Sule et al. / Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 4. CONCLUSION This study addresses the challenge of inconsistent record-keeping in developing countries by validating CHIRPS satellite precipitation data and developing a seasonal LSTM neural network to predict dam inflow at Doma Dam, Nigeria, using 20 years of observed data that revealed significant flow variability. The research successfully integrated CHIRPS satellite-derived precipitation data at 0.05° resolution across five regional stations, providing continuous monthly rainfall estimates that eliminated data gaps in conventional gauge records and enabled comprehensive spatial representation of precipitation patterns across the catchment area. Rigorous validation confirmed CHIRPS data suitability for hydrological modeling, achieving a strong correlation coefficient of 0.774 with observed rainfall and a NSE of 0.537, which exceeds the conventional threshold for operational applications and demonstrates the practical utility of freely available global datasets for local-scale modeling in data-scarce regions. The seasonal LSTM neural network model achieved exceptional predictive performance with R² = 0.8387, RMSE = 17.98 m³/s, MAE = 10.90 m³/s, and overall accuracy of 90.53%, successfully capturing complex non-linear hydrological relationships and seasonal dynamics with projected wet season means of 47.10 m³/s and dry season means of 4.29 m³/s through 2030. This methodological advancement demonstrates significant potential for improving water resources management in data-scarce basins across West Africa by combining satellite remote sensing with advanced machine learning techniques. 5. ACKNOWLEDGMENT The authors appreciate the Executive Director and Staff of the Meteorological Department of Lower Benue River Basin Development Authority for their kind assistance. 6. CONFLICT OF INTEREST There is no conflict of interest associated with this work. REFERENCES Abah, R. C., and Petja, B. M. (2016). Assessment of potential impacts of climate change on agricultural development in the Lower Benue River Basin. Environmental monitoring and assessment, 188(12), 683. Aqilah, S. S., Karmilla K. Z., Tamanna K., Ali Z. and Mat N. (2024). Assessing socio-economic and environmental losses of dam-failure flood risk: a review on sustainable framework. Journal of Sustainability Science and Management, 19(1), pp. 171-195. Arnold, J. G., Srinivasan R., Muttiah R. S. and Williams J. R. (1998). Large area hydrologic modeling and assessment part I: model development 1. JAWRA Journal of the American Water Resources Association, 34(1), pp. 73-89. Arregocés, H. A., Rojano R. and Pérez J. (2023). Validation of the CHIRPS dataset in a coastal region with extensive plains and complex topography. Case Studies in Chemical and Environmental Engineering, 8, 100452. Ayehu, G. T., Tadesse T., Gessesse B. and Dinku T. (2018). Validation of new satellite rainfall products over the Upper Blue Nile Basin, Ethiopia. Atmospheric Measurement Techniques, 11(4), pp. 1921-1936. Dembélé, M. and Zwart, S. J. (2016). Evaluation and comparison of satellite-based rainfall products in Burkina Faso, West Africa. International Journal of Remote Sensing, 37(17), 3995-4014. Dinku, T., Funk C., Peterson P., Maidment R., Tadesse T., Gadain H. and Ceccato P. (2018). Validation of the CHIRPS satellite rainfall estimates over eastern Africa. Quarterly Journal of the Royal Meteorological Society, 144, pp. 292-312. Duan, Y., Chen X., Houthooft R., Schulman J. and Abbeel P. (2016). Benchmarking deep reinforcement learning for continuous control. In International conference on machine learning (pp. 1329-1338). PMLR. Funk, C., Peterson P., Landsfeld M., Pedreros D., Verdin J., Shukla S. ... and Michaelsen J. (2015). The climate hazards infrared precipitation with stations a new environmental record for monitoring extremes. Scientific data, 2(1), pp. 1-21. Gal, Y. and Ghahramani, Z. (2016). A theoretically grounded application of dropout in recurrent neural networks. Advances in neural information processing systems, 29. Hasan, K., Islam M. S., Mitu K. N., Salam M. I. and Aghashahi N. (2025). Evaluating the performance of CHIRPS and CPC precipitation data for streamflow forecasting using multiple linear regression and Long Short-Term Memory Neural Network model. MethodsX, 103443. 563 J. Sule et al. / Nigerian Research Journal of Engineering and Environmental Sciences 10(2) 2025 pp. 547-563 Hu, J., Shen L., Albanie S., Sun G. and Vedaldi A. (2018). Gather-excite: Exploiting feature context in convolutional neural networks. Advances in neural information processing systems, 31. Huffman, G. J., Bolvin D. T., Nelkin E. J., Wolff D. B., Adler R. F., Gu G., ... and Stocker E. F. (2007). The TRMM multisatellite precipitation analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. Journal of hydrometeorology, 8(1), pp. 38-55. Joyce, R. J., Janowiak J. E., Arkin P. A. and Xie P. (2004). A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. Journal of hydrometeorology, 5(3), pp. 487-503. Ibrahim, A. H., Molla D. D. and Lohani T. K. (2024). Performance evaluation of satellite-based rainfall estimates for hydrological modeling over Bilate River Basin, Ethiopia. World Journal of Engineering, 21(1), pp. 1-15. Ilmi, A. (2025). Sustainable Dam Development: Balancing Hydropower, Water Management, and Environmental Conservation in a Changing Climate. Water Management, and Environmental Conservation in a Changing Climate (April 05, 2025). Katsanos, D., Retalis A and Michaelides S. (2016). Validation of a high-resolution precipitation database (CHIRPS) over Cyprus for a 30-year period. Atmospheric Research, 169, pp. 459 - 464. Kratzert, F., Klotz D., Brenner C., Schulz K. and Herrnegger M. (2018). Rainfall–runoff modelling using long shortterm memory (LSTM) networks. Hydrology and Earth System Sciences, 22(11), pp. 6005-6022. Kurnianingsih, O., Hadiani R. R., Setiawan B. and Sobriyah. (2025). Evaluation and Reliability of CHIRPS Satellite Rainfall Products against Station Rainfall Data in the Siwaluh Basin. Engineering Proceedings, 84(1), 94. Le, X. H., Ho H. V., Lee G. and Jung S. (2019). Application of long short-term memory (LSTM) neural network for flood forecasting. Water, 11(7), 1387. Moriasi, D. N., Arnold J. G., Van Liew M. W., Bingner R. L., Harmel R. D and Veith T. L. (2007). Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Transactions of the ASABE, 50(3), pp. 885-900. Ng, A. (2018). Machine learning yearning: Technical strategy for AI engineers, in the era of deep learning. Schober, P.; Boer C. and Schwarte L. A. (2018). Correlation coefficients: appropriate use and interpretation. Anesthesia & analgesia, 126(5), pp. 1763-1768. Sun, Y., Huang X., Kroening D., Sharp J., Hill M. and Ashmore R. (2018). Testing deep neural networks. arXiv preprint arXiv:1803.04792. Sylla, M. B., Diallo I. and Pal J. S. (2013). West African monsoon in state-of-the-science regional climate models. In Climate variability-regional and thematic patterns. IntechOpen. Тарасюк, А. М. (2024). Suggestive linguistic devices for world creation in middle English religious discourse. In The 3rd International scientific and practical conference “Technologies in education in schools and universities”(January 23-26, 2024) Athens, Greece. International Science Group. 2024. 363 p. (p. 251). Toté, C., Patricio D., Boogaard H., Van Der Wijngaart R., Tarnavsky V, and Funk C. (2015). Evaluation of satellite rainfall estimates for drought and flood monitoring in Mozambique. Remote Sensing, 7(2), pp. 1758-1776. Xu, C. Y. (1999). Climate change and hydrologic models: A review of existing gaps and recent research developments. Water Resources Management, 13(5), pp. 369-382. Youdeowei, P. O.; Nwankwoala H. O. and Desai D. D. (2019). Dam structures and types in Nigeria: sustainability and effectiveness. Water conservation and management, 3(1), pp. 20 - 26.