Evaluating the Performance of Multiple Precipitation Datasets over the Transboundary Ili River Basin Between China and Kazakhstan

Abstract

The Ili River Basin is characterized by complex topography and diverse climatic zones with limited in situ observations. This study evaluates the performance of six widely used precipitation datasets, CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data), ERA5_Land (European Centre for Medium-Range Weather Forecasts—ECMWF Reanalysis 5_Land), GPCC (Global Precipitation Climatology Centre), IMERG (Integrated Multi-satellite Retrievals for GPM), PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks), and TerraClimate, against ground-based data from 2001 to 2023. The evaluation is conducted across multiple spatial scales and temporal resolutions. At the basin scale, most datasets exhibit strong correlations with in situ observations across all temporal scales (r > 0.7), except for PERSIANN, which demonstrates a relatively weaker performance during summer and winter (r < 0.6). All datasets except ERA5_ Land show low annual and monthly bias (<5%), although larger errors are observed during summer, particularly for IMERG and PERSIANN. Dataset performance generally declines with increasing elevation. Basin-wide gridded evaluations reveal distinct spatial variations across all elevation zones, with CHIRPS showing the strongest ability to capture orographic precipitation gradients throughout the basin. All datasets correctly identified 2008 as a drought year and 2016 as a wet year, even though the magnitude and spatial resolution of the anomalies varied among them. These findings highlight the importance of selecting precipitation datasets that are suited to the complex topographic and climatic characteristics of transboundary basins. Our study provides valuable insights for improving hydrological modeling and can be used for water sustainability and flood–drought mitigation support activities in the Ili River Basin.

1. Introduction

Water sustainability is a critical issue related to national security and for planning development strategies regionally and nationally. The sustainability of water resources and water availability are complicated issues in Central Asia. Being located downstream of the most river basins of Central Asia, Kazakhstan depends greatly on neighboring countries for transboundary river basin water movements. Transboundary water movements are challenging for negotiations and regulations. One such river basin is the Ili River, an endorheic basin of Central Asia [1], and a primary water resource that has influenced regional development in the Ili region of China and the Almaty region of Kazakhstan. The Ili River Basin has issues with water conservation during the summer drought seasons as well as flood emergencies. The Ili River is fed by transmountain streams, is the largest tributary to Lake Balkhash, one of the largest lakes in Central Asia, and is characterized by complex topography and diverse climatic conditions that challenge the accuracy of conventional and satellite-based precipitation datasets. Given the scarcity and uneven spatial distribution of ground-based observations, the evaluation of multiple precipitation datasets can be helpful to support water sustainability as well as flood and drought mitigation activities for communities within the Ili River Basin. This study aims to systematically evaluate the accuracy and applicability of multiple gridded precipitation datasets over the Ili transboundary River Basin using a suite of statistical metrics (e.g., correlation, bias, normalized root mean square error) in comparison with ground-based observations. The results could provide insights into the relative strengths and limitations of each dataset, informing us of their suitability for regional-scale hydrological modeling, flood–drought monitoring and mitigation, and sustainable water use program development for communities within the Ili River Basin.

Accurate precipitation data are essential for effective water resource management, hydrological modeling, and climate change impact assessments, particularly in transboundary river basins where shared resources necessitate coordinated planning and monitoring [2]. CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data) has demonstrated reasonable accuracy estimating precipitation in arid and semi-arid regions and is often used because of its long temporal coverage and consistency [3]. Studies have shown CHIRPS effectively captures seasonal and interannual precipitation variability in including Central Asia, although biases may occur in areas with complex terrain or sparse station data [4,5,6]. ERA5_Land (ECMWF Reanalysis 5) precipitation data are derived from model outputs constrained by observations, offering consistency and global coverage [7]. However, ERA5_Land has been shown to overestimate precipitation in arid regions and underestimate extreme rainfall events, due to its reliance on parameterized convective processes [8,9]. Due to its reliance on ground-based data, GPCC (Global Precipitation Climatology Centre) accurately estimates precipitation in areas with dense station networks but may lack accuracy in data-sparse or mountainous regions [10,11]. GPCC is commonly used as a benchmark in performance validation studies [12]. IMERG (Integrated Multi-satellite Retrievals for Global Precipitation Measurement) can be used for real-time monitoring and short-term hydrological modeling [13]. However, its accuracy can be limited by surface characteristics such as snow cover or orographic effects, especially in mountainous basins [14,15,16]. PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks) by [17] can be used for near-real-time applications, but several studies have reported its tendency to underestimate high-intensity precipitation events and overestimate light rainfall, particularly in tropical and mountainous regions [18,19,20]. TerraClimate offers good spatial coverage and incorporates both station and model-derived data, but TerraClimate’s precipitation estimates may be less responsive to sub-monthly variability and often rely heavily on downscaling procedures, which can introduce biases in mountainous terrain [21,22]. Overall, these datasets show varying degrees of accuracy depending on regional topography, climate, and the density of observation networks. Comparative assessments suggest that combining multiple datasets or applying region-specific bias correction methods can substantially enhance the reliability of precipitation estimates for hydrological and climate applications [23,24].

Several studies have examined hydrological variability, water resource management, and climate impacts in the Lake Balkhash Basin and its sub-basins, particularly the Ili River Basin. For instance, Ref. [25] assessed hydroclimatic changes and their impacts on streamflow in the IRB, revealing clear shifts in precipitation and temperature patterns influencing seasonal runoff. Refs. [26,27,28] investigated water governance challenges, emphasizing the institutional and political complexities that affect transboundary water management in the broader Balkhash system. Ref. [29] provided critical insights into the role of snow and glacier melt in sustaining river discharge from the Tianshan Mountains’ headwaters, highlighting the sensitivity of the region’s water supply to cryospheric changes. More recently, Ref. [30] used remote sensing and model-based approaches to quantify water balance components across the transboundary basin, demonstrating spatial disparities in runoff contributions. Additionally, Ref. [31] focused on the verification and refinement of climate data in the Almaty region, supporting more reliable detection of climate change signals critical for regional water and environmental planning. Parallel to these hydrological assessments, several studies have addressed drought and flood monitoring to identify regional risks and support mitigation strategies. For example, Ref. [32] analyzed meteorological and hydrological drought regimes, while Ref. [33] examined flood hazards under climate-induced hydrological changes. Also, Ref. [34] examined the impacts of climate change on agricultural production and crop water requirements through modeling approaches and emphasized the importance of adopting agricultural adaptation measures in response to climate change. Using satellite-based indices, Ref. [35] demonstrated the utility of remote sensing in large-scale drought monitoring across the Lake Balkhash Basin. Ref. [36] investigated the response of Lake Balkhash water levels to drought events, and Ref. [37] introduced advanced flood vulnerability assessments in the IRB using novel statistical metrics. Moreover, Ref. [38] emphasized on the anthropogenic issues related to the flood issues importance of industry engagement in flood water management.

Recent developments in satellite remote sensing and reanalysis have yielded a suite of high-resolution global precipitation datasets, such as CHIRPS, ERA5_Land, GPCC, IMERG, PERSIANN-CDR, and TerraClimate, that are increasingly applied in climate and water studies [3,11,13]. When coupled with in situ observations, these developments can enhance hydrological modeling, mitigation planning for droughts and flooding occurrences and sustainable water management. However, their performance can vary significantly depending on seasonality, topography, and data-scarcity, particularly in complex environments such as the mountainous and semi-arid Ili River Basin. Therefore, evaluating the accuracy and applicability of multiple gridded precipitation datasets over the transboundary Ili River Basin, in comparison with ground-based observations, is crucial. The results will contribute to broader studies aimed at promoting hydrological modeling, hazard monitoring, and sustainable water management at the regional scale.

2. Study Area

The Ili River Basin (hereafter referred to as “the Basin”) is a significant transboundary watershed shared between the southeastern region of Kazakhstan and the Xinjiang Uyghur Autonomous Region of China (Figure 1). The Ili river (1439 km) originates from the Kax, Künes, and Tekes Rivers in the eastern Tianshan Mountains (Figure 1). Covering a total area of approximately 140,000 km² (with around 60% in Kazakhstan and 40% in China), the basin plays a critical role in supporting agricultural productivity, regional ecosystems, and water supply for millions of people across both countries. The river originates from the Tekes and Kunes rivers in the Tianshan Mountains of western China and flows northwestward into Kazakhstan, ultimately discharging into Lake Balkhash, which is one of Central Asia’s largest inland water bodies.

The basin exhibits a diverse range of topographic, climatic, and hydrological conditions. Elevation in the basin ranges from over 3000 m above sea level (m) in the mountainous eastern upstream regions to below 200 m in the lowland floodplains of the western basin (Figure 1). This elevation gradient result in substantial climatic variation from alpine and subalpine climates in the upper reaches to semi-arid and arid environments downstream with variations in isotopic composition of different water sources [39,40].

Land use/land cover (LULC) within the basin varies widely, with forested and alpine zones in the upstream areas, and extensive agricultural zones in the midstream and downstream areas, particularly grassland (Figure 1). Irrigation-intensive agriculture, supported in part by groundwater wells and surface water diversions, exerts considerable pressure on the region’s water resources, especially in China (Figure 1). Due to increasing water demand, climate variability, and transboundary water governance challenges, the IRB has become a focal point for hydrological and environmental research [41,42].

The Basin is characterized by a continental climate, featuring cold winters and hot summers. The seasonal and elevational precipitation gradients (Figure 1 and Figure 2) suggest that spring is the dominant wet season in the lowland and mid-altitude regions, while summer dominates in the upper mountainous areas (Figure 2). The contribution of snowmelt and glacier runoff from the Tianshan Mountains is especially significant during spring and early summer, driving seasonal peaks in river discharge [43,44]. The basin’s complex topography, steep climatic gradients, and spaced distribution of meteorological stations, particularly in remote high-elevation zones, pose substantial challenges for reliable precipitation monitoring. As such, the basin represents a critical testbed for evaluating the performance of satellite- and model-based precipitation datasets under diverse hydroclimatic and topographic conditions.

Figure 1. Study area showing the Ili River Basin between China and Kazakhstan. (a) Digital Elevation Model (DEM) from SRTM (~30 m resolution); (b) LULC classification from ESA WorldCover 2021 (10 m) by [45]; (c) average annual precipitation (2001–2020) interpolated from meteorological station data using Inverse Distance Weighting (IDW); (d) average annual actual evapotranspiration (ETa) for 2013–2023 derived from VIIRS-SSEBop ET data. Base map: Esri.

Figure 1. Study area showing the Ili River Basin between China and Kazakhstan. (a) Digital Elevation Model (DEM) from SRTM (~30 m resolution); (b) LULC classification from ESA WorldCover 2021 (10 m) by [45]; (c) average annual precipitation (2001–2020) interpolated from meteorological station data using Inverse Distance Weighting (IDW); (d) average annual actual evapotranspiration (ETa) for 2013–2023 derived from VIIRS-SSEBop ET data. Base map: Esri.

Figure 2. Annual, monthly, and seasonal (December–January–February, DJF; March–April–May, MAM; Jun–July–August, JJA; September–October–November, SON) precipitation from rain gauges across three elevation zones: low (0–800 m), mid (800–1600 m), and high (>1600 m) elevations. The number of stations in 0–800 m, 800–1600 m, and >1600 m elevation ranges is 15, 8, and 7, respectively.

Figure 2. Annual, monthly, and seasonal (December–January–February, DJF; March–April–May, MAM; Jun–July–August, JJA; September–October–November, SON) precipitation from rain gauges across three elevation zones: low (0–800 m), mid (800–1600 m), and high (>1600 m) elevations. The number of stations in 0–800 m, 800–1600 m, and >1600 m elevation ranges is 15, 8, and 7, respectively.

3. Data and Methods

3.1. Data

We utilized monthly and annual precipitation data from approximately 30 meteorological stations spanning the period 2001–2023, comprising records from around 20 stations in Kazakhstan and 10 in China (Figure 1 and Figure 2). The data for Kazakhstan were obtained from the publicly accessible Kazhydromet platform (https://www.kazhydromet.kz/en/interactive_cards, accessed on 20 January 2025), while the Chinese station data were sourced from the China Meteorological Administration (CMA) website (http://data.cma.cn, accessed on 22 January 2025). These ground-based observations served as the reference dataset for evaluating the performance of various precipitation datasets.

In this study, we conducted a comprehensive evaluation of six precipitation datasets, including satellite-, reanalysis-, and gauge-based datasets: CHIRPS, ERA5_Land, GPCC, IMERG, PERSIANN, and TerraClimate. Detailed characteristics of these datasets are provided in

Table 1. Data types and key characteristics.

Dataset	Description	Spatial Resolution	Temporal Resolution	Coverage	Data Availability	Citation
CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data)	Merged satellite and station-based precipitation dataset	0.05° (~5 km)	Daily, Monthly	Global (50° S–50° N)	1981–present	[3]
ERA5 (ECMWF Reanalysis 5) _Land	Global reanalysis dataset from ECMWF, includes multiple atmospheric and land variables	0.1° (~9 km)	Hourly, Monthly	Global	1950–present	[7]
GPCC (Global Precipitation Climatology Centre)	Gauge-based global precipitation dataset	0.25° to 2.5°	Monthly	Global	1891–2020	[10]
IMERG (Integrated Multi-satellite Retrievals for GPM)	Satellite-based precipitation estimates using GPM and TRMM data	0.1° (~10 km)	30-min, Daily, Monthly	Global (60° S–60° N)	2000–present	[13]
PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks)	Satellite-based precipitation estimates using infrared and microwave sensors	0.25° (~25 km)	Daily, Monthly	Global (60° S–60° N)	1983–present	[17]
TerraClimate	Climate and water balance dataset derived from multiple sources	0.04° (~4 km)	Monthly	Global	1958–present	[21]

. With the exception of GPCC, these datasets, available as point or gridded raster data, can be accessed through their respective web portals or platforms such as Google Earth Engine [46] and Climate Engine [47]. GPCC point data are available for download at https://opendata.dwd.de/climate_environment/GPCC/html/download_gate.html accessed on 5 December 2024). All analyses and visualizations were conducted using R [48] programming environments, complemented by QGIS [49] for geospatial mapping and analysis.

3.2. Methods

We employed statistical metrics such as Pearson’s correlation coefficient (r), percent bias (PBIAS), and normalized root mean square error (NRMSE) to assess the relationship between observed and estimated precipitation values.

Pearson’s Correlation Coefficient (r):

Pearson’s r measures the linear correlation between two variables, indicating the strength and direction of their relationship. It ranges from −1 (perfect negative correlation) to +1 (perfect positive correlation), with 0 indicating no linear correlation, which is calculated by an equation (Equation (1)).

$$r={\displaystyle \frac{{\sum }_{i=1}^n\left(O_i-\overline{O}\right)\left(S_i-\overline{S}\right)}{\sqrt{\sum _{i=1}^n{\left(O_i-\overline{O}\right)}^2\times \sum _{i=1}^n{{(S}_i-\overline{S})}^2}}}$$

(1)

where $O_i$ is the observed precipitation value; $S_i$ is the estimated precipitation value; $\overline{O}$ is the mean of the observed values; $\overline{S}$ is the mean of the estimated values, and n is the number of observed or simulated values. Interpretation guidelines vary by field.

Percent Bias (PBIAS):

PBIAS evaluates the average tendency of model estimates to overestimate or underestimate observed values. It is calculated as follows (Equation (2)):

$$\mathrm{PBIAS}={\displaystyle \frac{{\sum }_{i=1}^n\left(S_i-O_i\right)}{\sum _{i=1}^n{O}_i}}$$

(2)

where $O_i$ is the observed precipitation value, $S_i$ is the estimated precipitation value, and n is the number of observed or simulated values.

A percent bias (PBIAS) value close to 0% indicates minimal bias, meaning the model’s estimates align closely with observed values. Positive PBIAS values suggest a tendency for the model to underestimate observed data, while negative values indicate a propensity to overestimate observed data. Understanding PBIAS is crucial for assessing and improving model accuracy in representing observed phenomena.

Normalized Root Mean Square Error (NRMSE)

The normalized root mean square error (NRMSE) is calculated by dividing the root mean square error (RMSE) by the range of the observed values. A smaller NRMSE value indicates a better model fit.

Equation:

NRMSE = RMSE/(Max(observed) − Min(observed))

(3)

where NRMSE is the normalized root mean square error; RMSE is the root mean square error, calculated as the square root of the mean of the squared differences between predicted and actual values; Max(observed) is the maximum value in the observed data; Min(observed) is the minimum value in the observed data.

$$\mathrm{RMSE}=\sqrt{{\displaystyle \frac{1}{n}}\sum _{i=1}^n{(Si-Oi)}^2}$$

(4)

To analyze the spatial variability of precipitation across the study area, we applied Inverse Distance Weighting (IDW) interpolation using observed precipitation data from ground-based meteorological stations. IDW is a commonly used spatial interpolation method that estimates values at unsampled locations by weighing surrounding data points inversely to their distance. The interpolation was performed using QGIS, with the power parameter set to 2 and a fixed search radius to balance local detail and smoothness.

To evaluate the spatial performance of gridded precipitation datasets, we selected one dry year (2008) and one wet year (2016) for detailed anomaly analysis. Precipitation anomalies were computed for each gridded dataset by subtracting the long-term mean precipitation (2001–2023) from the annual total for each selected year. All anomaly calculations and map visualizations were performed using QGIS, enabling spatial overlay, raster calculations, and comparison of anomaly distributions.

4. Results and Discussion

4.1. Spatiotemporal Characteristics of Observed Precipitation

Precipitation is distributed unevenly across the basin, with annual totals exceeding 1000 mm in the mountainous regions and falling below 200 mm in the arid lowlands (Figure 1 and Figure 2). Based on aggregated data (Figure 2) from 30 rain gauges, the lowland zone (0–800 m) receives an average annual precipitation of 260 mm, with the highest seasonal precipitation occurring in spring (March–April–May, MAM, 80 mm) and the lowest in winter (December–January–February, DJF, 53 mm); summer (Jun–July–August, JJA) contributes 60 mm to the annual total. In the mid-elevation zone (800–1600 m), the average annual precipitation increases to 576 mm, peaking in spring (MAM, 220 mm) and reaching a minimum in winter (DJF, 83 mm); summer precipitation totals 143 mm. In contrast, the high-altitude zone (>1600 m) receives an average of 661 mm annually, with the highest precipitation in summer (JJA, 251 mm) and the lowest in winter (DJF, 60 mm).

4.2. Point-Based Evaluation

4.2.1. Long-Term Monthly Mean and Anomaly

The long-term analysis of monthly mean precipitation (Figure 3) reveals distinct seasonal patterns across both ground-based observations and satellite-derived datasets in the basin. Observational data indicate that precipitation peaks during the warm season, particularly from April to June, followed by a gradual decline from July to September, and winter months (December to February). All satellite and reanalysis datasets successfully capture this general seasonal cycle; however, notable discrepancies remain in terms of magnitude and timing. For instance, GPCC and CHIRPS exhibit strong agreement with in situ observations during the peak precipitation months, accurately representing both seasonal peaks and transitions. In contrast, ERA5_Land and PERSIANN systematically overestimate monthly precipitation during spring and early summer, but winter estimates align more closely with observed values.

These discrepancies can largely be attributed to the inherent differences in data acquisition methods, spatial resolutions, and underlying retrieval algorithms among the datasets. Reanalysis datasets such as ERA5_Land often rely on numerical weather prediction models, which may not fully resolve localized orographic influences and convective precipitation, particularly in data-scarce and topographically complex regions [2,11]. PERSIANN, being based on infrared satellite imagery, may misclassify cold cloud tops associated with non-precipitating systems (e.g., cirrus or shallow convection) as rainfall, especially during spring when cloudiness is high but actual precipitation may be low [18]. The biases are particularly pronounced at high elevations and during transitional seasons, underscoring the limitations of model-driven datasets in mountainous basins like the Ili River Basin.

The comparative analysis of monthly climatology not only highlights the relative strengths and weaknesses of each dataset but also demonstrates the importance of bias correction and multi-source data fusion to improve precipitation estimates [50,51]. This is especially critical in hydrologically sensitive regions where sparse rain-gauge coverage limits the reliability of single-source data.

The evaluation of long-term annual precipitation anomalies, referenced against the 2001–2023 climatological baseline, highlights pronounced interannual variability in observed precipitation across the basin (Figure 4). Most of the examined datasets successfully capture the temporal dynamics and relative magnitudes of these anomalies. This includes the accurate identification of 2008 as the driest year and 2016 as the wettest year across the basin. While PERSIANN markedly overestimates precipitation during the wet year (2016), it closely captures the magnitude of the dry year (2008). In contrast, ERA5_Land exhibits limited skill in representing the magnitude of both dry and wet years, indicating potential challenges in reproducing extreme precipitation variability in the region.

The ability of most datasets to replicate such hydrologically critical years demonstrates their applicability for a wide range of environmental and water management applications. Accurate anomaly detection is particularly vital for drought monitoring, agricultural planning, and the design of early warning systems in data-scarce, transboundary basins such as the Ili River Basin [11,52]. Among the datasets, GPCC and TerraClimate most accurately estimated the mean and range monthly precipitation patterns based on percent anomaly data (Figure 4). CHIRPS and IMERG accurately estimated the mean, but were less dynamic in the ranges of precipitation that they simulated. In contrast, ERA5_Land and PERSIANN tended to either overestimate or smooth the extremes, which may limit their effectiveness in capturing localized climatic shocks, particularly in mountainous or snow-influenced areas. Nevertheless, the observed differences in anomaly magnitude and spatial detail among datasets suggest that careful dataset selection and, where possible, bias correction remain essential for operational applications.

4.2.2. Basin Scale Performance

At the basin scale, most precipitation datasets demonstrate strong temporal agreement with station observations. Annual correlations exceed r > 0.74, and monthly correlations are similarly high, with r > 0.68 across all datasets (

Table 2. Median correlation coefficients (r) of different precipitation datasets in relation to station observations across temporal and spatial scales. The number of stations in the entire basin, 0–800 m, 800–1600 m, and >1600 m elevation ranges is 30, 15, 8, and 7, respectively.

Spatial	Temporal	Number of Observations	CHIRPS	ERA5_Land	GPCC	IMERG	PERSIANN	TerraClimate
Basin	Annual	690	0.76	0.74	0.87	0.85	0.8	0.8
Basin	Monthly	8280	0.68	0.76	0.86	0.83	0.71	0.76
Basin	DJF	2070	0.74	0.82	0.89	0.82	0.45	0.72
Basin	MAM	2070	0.74	0.8	0.83	0.85	0.78	0.8
Basin	JJA	2070	0.69	0.63	0.77	0.79	0.56	0.77
Basin	SON	2070	0.72	0.78	0.87	0.82	0.79	0.72
0–800 m	Annual	345	0.76	0.79	0.89	0.87	0.81	0.8
0–800 m	Monthly	4140	0.68	0.73	0.86	0.82	0.64	0.75
0–800 m	DJF	1035	0.75	0.85	0.93	0.86	0.54	0.72
0–800 m	MAM	1035	0.69	0.8	0.88	0.86	0.81	0.82
0–800 m	JJA	1035	0.69	0.71	0.78	0.8	0.59	0.77
0–800 m	SON	1035	0.73	0.8	0.93	0.91	0.82	0.76
800–1600 m	Annual	184	0.79	0.77	0.87	0.87	0.82	0.8
800–1600 m	Monthly	2208	0.77	0.81	0.84	0.86	0.73	0.78
800–1600 m	DJF	552	0.79	0.8	0.86	0.83	0.37	0.71
800–1600 m	MAM	552	0.84	0.86	0.86	0.87	0.78	0.82
800–1600 m	JJA	552	0.73	0.6	0.8	0.81	0.57	0.78
800–1600 m	SON	552	0.8	0.75	0.86	0.81	0.85	0.77
>1600 m	Annual	161	0.77	0.57	0.75	0.78	0.65	0.72
>1600 m	Monthly	1932	0.53	0.75	0.85	0.85	0.71	0.76
>1600 m	DJF	483	0.72	0.68	0.63	0.7	0.34	0.68
>1600 m	MAM	483	0.74	0.72	0.8	0.83	0.75	0.75
>1600 m	JJA	483	0.63	0.37	0.7	0.73	0.38	0.6
>1600 m	SON	483	0.63	0.66	0.72	0.78	0.72	0.63

and Figure 5). Notably, GPCC and IMERG show particularly high annual correlations (r > 0.80), underscoring their strong performance in capturing temporal precipitation variability. With the exception of ERA5_Land, all datasets show relatively low annual and monthly precipitation bias (PBIAS < ±5%) (

Table 3. Median PBIAS (%) of different precipitation datasets in relation to station observations across temporal and spatial scales. The number of stations in the entire basin, 0–800 m, 800–1600 m, and >1600 m elevation ranges is 30, 15, 8, and 7, respectively.

Spatial	Temporal	Number of Observations	CHIRPS	ERA5_Land	GPCC	IMERG	PERSIANN	TerraClimate
Basin	Annual	690	0.9	33.6	4.6	−3.1	4.4	−3.6
Basin	Monthly	8280	0.9	33.6	4.6	−3.1	4.4	−3.6
Basin	DJF	2070	9.1	22.6	2.4	8.8	8.3	0.2
Basin	MAM	2070	9.7	40.8	6.5	−5.3	−4.5	−2.4
Basin	JJA	2070	−10.1	31.8	6.9	26.2	24.6	−6.3
Basin	SON	2070	10.5	30.6	4.3	−3.4	−1.4	3.5
0–800 m	Annual	345	17.5	18.4	5.8	43.3	51.2	10.5
0–800 m	Monthly	4140	17.5	18.4	5.8	43.3	51.2	10.5
0–800 m	DJF	1035	19.3	20.4	4.4	25.3	29.6	13.3
0–800 m	MAM	1035	18	51.5	8.1	48.2	54.4	17.8
0–800 m	JJA	1035	18.7	8.3	5.8	49.4	92.7	3.7
0–800 m	SON	1035	9.7	17	5.4	33.3	29.6	10.8
800–1600 m	Annual	184	−14.8	33.6	2.5	−14.6	−21.5	−20.7
800–1600 m	Monthly	2208	−14.8	33.6	2.5	−14.6	−21.5	−20.7
800–1600 m	DJF	552	−14.7	11.3	−9	−15	−23	−16.1
800–1600 m	MAM	552	−15.4	27.5	−3.2	−24.7	−36.9	−33.3
800–1600 m	JJA	552	−18.3	52.4	16.2	−1.3	1.4	−20.2
800–1600 m	SON	552	−13.8	31	−1.4	−17	−26.2	−13.5
>1600 m	Annual	161	−6	76.4	−0.3	−25.9	−21.9	−33.3
>1600 m	Monthly	1932	−6	76.4	−0.3	−25.9	−21.9	−33.3
>1600 m	DJF	483	23.4	63.4	18.1	10.1	38.2	20.5
>1600 m	MAM	483	17.9	61.3	7.5	−28.6	−29.8	−34.8
>1600 m	JJA	483	−29.9	53.7	−1.9	−36	−28.3	−36.1
>1600 m	SON	483	26.6	96.2	3.9	−17.9	−18.5	−18.1

and Figure 6), indicating reasonable agreement with ground-based observations. CHIRPS exhibits the lowest annual and monthly biases (~1%), making it the most consistent with observations. In contrast, ERA5_Land shows a positive bias (~30%), a trend aligned with previous studies highlighting its tendency to overestimate orographic precipitation in mountainous regions [6,50,53]. IMERG and TerraClimate display moderate negative biases overall; CHIRPS, GPCC, and PERSIANN tend to overestimate precipitation slightly, although their seasonal dynamics differ. A similar pattern is noted in South Asia where IMERG and TerraClimate exhibit slight negative biases, whereas other datasets exhibit slight positive biases [54]. NRMSE results align with correlation and bias analyses that GPCC and CHIRPS report the lower annual error magnitudes at 17.6% and 25.1% (

Table 4. Median NRMSE (%) of different precipitation datasets in relation to station observations across temporal and spatial scales. The number of stations in the entire basin, 0–800 m, 800–1600 m, and >1600 m elevation ranges is 30, 15, 8, and 7, respectively.

Spatial	Temporal	Number of Observations	CHIRPS	ERA5_Land	GPCC	IMERG	PERSIANN	TerraClimate
Basin	Annual	690	25.1	44.1	17.6	36.8	45.1	35.4
Basin	Monthly	8280	14.1	15.1	10	13.2	15.8	13.7
Basin	DJF	2070	25.9	23.2	16.4	23	46.5	30.6
Basin	MAM	2070	24.7	35.6	16.9	31.2	40.2	31.5
Basin	JJA	2070	24.2	28.5	16.7	24.4	27	27.6
Basin	SON	2070	24.4	33.6	21	25.9	40.5	21.5
0–800 m	Annual	345	22.2	24.1	15.7	41	51.6	28.3
0–800 m	Monthly	4140	12.8	13.9	9	15	18	12.9
0–800 m	DJF	1035	26.8	19.2	10.8	22.5	46.9	28.1
0–800 m	MAM	1035	25.7	35.6	15.5	36.4	42	23
0–800 m	JJA	1035	22.7	21.8	16.8	28.9	53.7	17.7
0–800 m	SON	1035	21.7	18.5	15	24.1	25	25.9
800–1600 m	Annual	184	25.3	45.8	23.8	24.8	35.2	33.9
800–1600 m	Monthly	2208	13.9	16	13.2	12.6	15.3	14.3
800–1600 m	DJF	552	26.4	24.2	25.1	21	41.9	26
800–1600 m	MAM	552	22	27.2	19.1	28.1	38.1	33.9
800–1600 m	JJA	552	21.6	40.4	28.6	24.3	24.2	23.2
800–1600 m	SON	552	25.3	29.4	16.3	23	29.3	27.4
>1600 m	Annual	161	36.5	91.2	18.4	50.6	48.1	59.2
>1600 m	Monthly	1932	16.7	20.2	9.7	12.8	15.6	14.9
>1600 m	DJF	483	24	59.1	18.3	28.2	30.3	47
>1600 m	MAM	483	23.4	53	25.2	23.9	48.4	44.1
>1600 m	JJA	483	37.9	44.2	22.1	35.1	38.8	31.3
>1600 m	SON	483	29.1	78.6	20.1	24.6	26.7	30.1

and Figure 7), respectively, while the remaining datasets range from 35% to 45%, indicating higher levels of dispersion relative to observed values. In general, all evaluated precipitation datasets demonstrate sufficient reliability for direct application in water resource management, hydrological modeling, and climate impact assessments at both monthly and annual timescales. However, ERA5_Land, despite exhibiting relatively high correlation with observed data, presents notably high PBIAS and NRMSE values, indicating a consistent overestimation of precipitation. This suggests that ERA5_Land may require bias correction prior to its use in applications where accurate precipitation magnitude is critical, such as drought monitoring, flood risk assessment, or hydrological modeling for planning and infrastructure development.

Seasonal accuracy was more variable across datasets than basin scale performance While most datasets maintain relatively high seasonal correlations (r > 0.69) with rain-gauge observations (Table 2 and Figure 5), discrepancies become evident under specific seasonal conditions. For instance, PERSIANN shows relatively weak correlations during both winter (DJF) and summer (JJA), with correlations r < 0.60. Similarly, ERA5_Land exhibits comparatively low correlation during JJA (r = 0.63), reflecting its difficulty in resolving convective precipitation patterns typical of the warm season. These findings are consistent with previous evaluations highlighting the seasonal weaknesses of satellite and reanalysis datasets, particularly in mountainous or semi-arid regions [6,50].

Bias patterns are similarly more pronounced at the seasonal level. All datasets output tend to overestimate winter (DJF) precipitation (Table 3 and Figure 6), which may stem from challenges in accurately detecting snow events using satellite-based sensors, especially where ground station density is low at high altitudes. This behavior is echoed in regional studies across Central Asia and the Himalayas, where solid precipitation remains a significant source of uncertainty in remote sensing datasets [4,55]. In contrast, during summer (JJA), CHIRPS and TerraClimate consistently underestimate precipitation (−10.1% and −6.3%, respectively), while IMERG and PERSIANN substantially overestimate it (+26.2% and +24.6%, respectively). These deviations likely indicate difficulties in capturing intense and short-lived convective rainfall, which is a dominant process during warm seasons in semi-arid and mountainous terrains. Similar seasonal patterns were reported by [11] in arid basins and [56] in Central Asia, emphasizing the need for regionally adapted validation strategies. Among all evaluated datasets, TerraClimate demonstrates the most consistent seasonal accuracy, maintaining biases within ±6.3%, thereby supporting its robustness across different climatic regimes. This aligns with global studies [53,57] which have recognized TerraClimate as a reliable dataset in both tropical and mountainous settings due to its hybrid data assimilation framework. Normalized root mean square error (NRMSE) results also underscore the seasonal limitations of some datasets. While most datasets show lower seasonal NRMSE values compared to annual values, PERSIANN reports high error rates during DJF and SON exceeding 40% (Table 4 and Figure 7), reinforcing its limited accuracy during cold and transitional seasons. This is consistent with findings by [6,56,58] who found high dispersion in PERSIANN outputs during low-precipitation or snowfall-dominated periods. Overall, except for ERA5_Land and the JJA seasons of IMERG and PERSIANN, which exhibit large bias and may require correction, most precipitation datasets display satisfactory seasonal skill and can be applied directly for hydrological modeling, water resource management, and climate assessments at sub-seasonal to annual timescales.

These findings align with similar validation studies in data-scarce and topographically complex regions. For example, Refs. [6,23] found limited seasonal performance of PERSIANN and ERA5_Land in CA, particularly in high-elevation and arid zones. Ref. [59] emphasized IMERG’s strong performance across Central Asian basins but noted seasonal inconsistencies. Globally, Refs. [53,60] identified CHIRPS and GPCC as highly consistent across diverse hydroclimatic regions, while Ref. [50] documented ERA5_Land’s bias in the Andes. These cross-regional comparisons highlight the importance of region-specific validation when selecting precipitation datasets for hydrological and climate studies.

4.2.3. Elevation-Dependent Performance

Low elevations (0–800 m)

Most precipitation datasets are strongly correlated with rain-gauge observations across annual and monthly timescales, with correlation coefficients exceeding r > 0.76 annually and r > 0.64 monthly (Table 2 and Figure 5), indicating a robust ability to capture temporal precipitation variability in lowland regions. On seasonal timescales, all datasets maintain correlations above r > 0.69, except PERSIANN, which exhibits weaker correlations during winter (DJF) and summer (JJA) (r < 0.60). This seasonal reduction in correlation is likely due to PERSIANN’s known limitations in detecting solid precipitation during winter and its tendency to overestimate convective rainfall during the summer months [61,62].

Bias analysis reveals pronounced overestimations of precipitation among most datasets, particularly IMERG and PERSIANN, which demonstrate the largest positive seasonal biases, especially in JJA, exceeding +50% (Table 3 and Figure 6). These overestimations are likely due to challenges in accurately capturing intense convective events common in summer, a well-documented issue in satellite-derived datasets lacking sufficient ground-based calibration [14]. Similarly, ERA5_Land shows a marked spring overestimation (MAM), exceeding +50%, which may stem from its coarse-resolution land-atmosphere coupling and limitations in representing transitional season dynamics in semi-arid settings [6,50]. These patterns align with findings from other semi-arid lowland regions, such as the Himalaya Mountains Basin, where satellite datasets like IMERG and PERSIANN were found to significantly overestimate summer rainfall due to convective misrepresentation [63]. In contrast, GPCC and TerraClimate exhibit relatively low seasonal biases, remaining below 8% and between 3.7 and 18% (Table 3 and Figure 6), respectively, suggesting rain-gauge observations in lowland zones were consistently and accurately estimated. CHIRPS also accurately estimated precipitation consistently across seasons with biases typically ranging from 10% to 20%, attributed to its gauge-blending algorithm that integrates station data with satellite observations to enhance regional accuracy [3,53]. NRMSE results further support these findings. While most datasets show acceptable annual NRMSE values (<30%); however, the NRMSEs for PERSIANN and IMERG exceeded 40% (Table 4 and Figure 7) with largest exceedances occurring during DJF and SON, demonstrating substantial dispersion in seasonally dynamic climates.

Overall, these findings underscore that GPCC, CHIRPS, and TerraClimate are suitable for hydrological modeling, agricultural planning, and water management in lowland areas of the basin, where accurate precipitation estimates are critical for irrigation scheduling, drought risk monitoring, and long-term climate adaptation strategies.

Mid-elevations (800–1600 m)

Most precipitation datasets output were strongly correlated with rain-gauge observations, with correlation coefficients (r) exceeding 0.77 annually and 0.73 monthly (Table 2 and Figure 5). On seasonal scales, most datasets maintain r > 0.73, again except PERSIANN, which drops below r = 0.60 during DJF and JJA. This seasonal reduction in correlation for PERSIANN likely stems from persistent difficulties in capturing snowfall during winter and convective rainfall during summer, limitations frequently observed in PERSIANN and similar infrared-based satellite datasets [6,56,62].

Bias patterns become more variable at mid-elevation. GPCC records a moderate positive summer bias (+16.2%) and slight negative in winter bias (−9.0%) (Table 3 and Figure 6), suggesting it adapts well to modest terrain complexity. CHIRPS consistently maintains biases <15%, similar to its demonstrated accuracy in similar topographies such as the Ethiopian Highlands and Central Asian piedmont [6,64]. IMERG reasonably estimates precipitation but tends to overestimate summer rainfall, a recurring issue also observed in the mountainous southwestern China [65]. TerraClimate displays seasonal biases often exceeding 20% that demonstrate limitations in its empirical downscaling techniques in terrain with steep gradients [21,22]. NRMSE values remain <30% for most datasets (Table 4 and Figure 7), except for PERSIANN and ERA5_Land, which exceed 40% during DJF and JJA, respectively. This reduced the accuracy of these datasets during key hydrometeorological transition periods.

In summary, CHIRPS and GPCC continue to demonstrate the highest accuracies at mid-elevations, offering a balanced performance in terms of correlation, bias, and error dispersion. Their reliability makes them suitable for hydrological modeling, flood forecasting, and watershed planning in montane environments of the basin.

High elevations (>1600 m)

The performance of most precipitation datasets declines substantially at high elevations, although annual and monthly correlations remain moderate to high for some datasets (r > 0.65 annually and r > 0.71 monthly) (Table 2 and Figure 5). Notably, ERA5_Land demonstrates reduced correlation (r = 0.57) compared to correlations at low and mid-elevations, and CHIRPS demonstrates reduced monthly correlation (r = 0.53) compared to correlations at low and mid-elevations (Table 2). Seasonal correlation is even more challenging: ERA5_Land drops to r ≈ 0.40 during JJA, while PERSIANN correlation coefficients are low in both DJF and JJA (r < 0.40).

Biases at high elevations are large and complex. ERA5_Land and TerraClimate exhibit substantial annual deviations (+76% and −33%, respectively) (Table 3 and Figure 6), driven by misrepresentation of orographic uplift, snow/rain partitioning issues, and the paucity of high-altitude stations. Similar issues have been identified in the Tianshan Mountains [23], Andes [50], and Qilian Mountains [66]. These challenges are compounded by data scarcity and extreme precipitation heterogeneity in alpine microclimates highlighting the importance of bias correction [67]. CHIRPS shows relatively low annual and monthly bias (<−10%), and GPCC outperforms all datasets at this elevation with a near-zero annual bias (<1%). This supports prior studies that highlight the strength of gauge-based and hybrid datasets for high-elevation hydrology [3,53,68].

Seasonal bias patterns are notably divergent. Most datasets display positive biases during winter (DJF) (Table 3 and Figure 6), likely stemming from the overestimation of snowfall, misclassification of the precipitation phase (rain vs. snow), or limitations in accurately representing snow accumulation and sublimation processes under frozen conditions. For instance, DJF biases range from a modest +10% for GPCC to a pronounced +63% for ERA5_Land, suggesting that reanalysis and satellite-driven datasets may not accurately simulate cold-season hydrometeorology. These winter biases align with documented issues in cryospheric regions, where passive microwave retrievals struggle with snow-covered terrain and reanalysis datasets tend to overestimate precipitation due to coarse resolution and simplified land–atmosphere coupling [53,66,69]. In contrast, summer (JJA) biases are predominantly negative, demonstrating widespread underestimation of convective precipitation. GPCC remains relatively stable (−2%); however, most datasets indicate strong negative biases from −28.3% for PERSIANN to −36% for both TerraClimate and IMERG. This pattern suggests that many datasets do not capture the short-lived, high-intensity summer storms typical of continental interior and mountainous regions [6,13,18]. These storms are often spatially localized and driven by thermal convection and orographic uplift, phenomena poorly resolved by coarse-scale satellite sensors or gridded interpolation datasets [6,23,50]. Additionally, ERA5_Land exhibits the highest seasonal bias during autumn (SON), reaching +96%, a substantial overestimation likely caused by its known issues with transitional seasons, where model physics overpredict post-summer convective activity and soil moisture feedback [70,71]. Such seasonal discrepancies are consistent with previous evaluations in other alpine and cryospheric regions, including the Hindu Kush–Karakoram Shan–Himalaya Mountains, European Alps, and Andes, where satellite and reanalysis datasets consistently underestimate summer precipitation due to their limitations in capturing high-frequency convective events and fine-scale orographic effects [53,68].

NRMSE further demonstrates degradations in accuracy at high elevations, exceeding 30% for all datasets except GPCC, which remains below this threshold (Table 4 and Figure 7). The elevated NRMSE values highlight compounded uncertainties from both bias and random errors, especially in regions with high spatial and temporal variability in precipitation regimes [72].

Ultimately, all datasets face substantial limitations in high-altitude settings. However, GPCC and CHIRPS demonstrate high accuracy annually, monthly, and across seasons, making them suitable for cryosphere-related hydrological modeling, glacier melt estimations, and climate vulnerability assessments in mountain headwater of the basin.

4.2.4. Cumulative Distribution Analysis of Annual Precipitation

The empirical cumulative distribution function (ECDF) analysis provides a robust, non-parametric approach for assessing how well satellite- and reanalysis-based precipitation datasets replicate the long-term distribution of observed rainfall. In this study, ECDFs were used to evaluate the cumulative probabilities of annual precipitation from six datasets, CHIRPS, ERA5-Land, GPCC, IMERG, PERSIANN, and TerraClimate, against observed station data across the basin. As shown in Figure 8, the observed ECDF curve provides a benchmark to compare the fidelity of each dataset in capturing both central tendencies and variability.

At the basin-wide scale, most datasets demonstrate reasonable agreement with observations at the median (50th percentile) value of annual precipitation. Notably, while ERA5-Land exhibits noticeable deviation, all other datasets closely follow the observed distribution near the central quantiles. This suggests that, with the exception of ERA5-Land, the datasets generally perform well in reproducing the mean hydroclimatic conditions. Such alignment in the mid-range of the distribution indicates an accurate capture of the long-term climatological average [73].

However, when stratified by elevation bands, more pronounced discrepancies emerge between precipitation datasets and rain-gauge observations. In the low-elevation zone (<800 m), all datasets consistently overestimate precipitation compared to observations. In the mid-elevation range (800–1600 m), all datasets except ERA5-Land underestimate precipitation, potentially demonstrating the under-capture of orographic effects in satellite retrievals [74]. Conversely, in the high-elevation zone (>1600 m), CHIRPS and ERA5-Land overestimate precipitation, while the other datasets underestimate it, indicating varying sensitivity to complex topography and snow-dominated processes.

These systematic differences demonstrate elevation-dependent biases and limitations in each dataset’s retrieval algorithms, spatial resolution, and assimilation methods. The ECDF approach highlights these biases effectively by providing insights into the entire distribution, not just mean or total values. Therefore, the ECDF approach can be used for evaluating the performance of precipitation datasets in representing interannual variability and extreme events [23].

Overall, ECDF-based analysis complements traditional performance metrics (e.g., correlation, PBIAS, RMSE) by focusing on distributional similarity. This is especially important in hydroclimatic applications such as drought monitoring, water resource modeling, and climate change impact assessments, where accurate representation of long-term precipitation variability and extremes is critical. For instance, deviations in the tails of the ECDF curves could indicate underrepresentation of drought years or extreme wet events, both of which are essential for risk assessment and adaptation planning in climate-sensitive transboundary basins like the basin.

4.3. Basin-Wide Gridded Evaluation

The gridded evaluation of long-term mean precipitation reveals substantial spatial variability in the performance of satellite and reanalysis datasets across the basin. All datasets broadly replicate the spatial distribution of precipitation, with higher values in mountainous regions and lower values in the plains. However, the accuracy and bias of these estimates vary significantly among datasets.

Among the datasets, GPCC substantially underestimates precipitation and exhibits a strong negative bias in the high-altitude eastern regions (Figure 9). This underestimation is likely due to the scarcity of ground stations in alpine zones and the interpolation techniques used by GPCC, which rely heavily on in situ data [53]. This pattern is consistent with other evaluations in mountainous regions, where GPCC was found to underrepresent orographic precipitation due to sparse station density and limited representativeness at high elevations [56,69].

CHIRPS demonstrates the closest agreement with the spatial pattern and magnitude of observed precipitation, particularly in its ability to replicate orographic gradients and localized high-elevation maxima. This can be attributed to CHIRPS’ semi-empirical design, which blends high-resolution satellite imagery with ground-based observations and incorporates terrain-sensitive calibration algorithms [3,64]. A similarly high performance of CHIRPS has been reported in other topographically complex regions, including the Ethiopian Highlands [64], Central Asia [6], and the mountainous basin of Turkey [75], where CHIRPS consistently demonstrated higher spatial fidelity compared to other gridded datasets.

TerraClimate demonstrates good spatial agreement with moderate biases but tends to smooth extreme values and underrepresent peak precipitation zones in mountainous areas. This smoothing effect is a known limitation of its downscaling approach, which relies on climatological norms and interpolation that may dilute localized extremes [21,70]. While suitable for broad climatological assessments, such smoothing limits its applicability for high-resolution hydrological modeling in rugged terrain.

ERA5_Land exhibits a systematic positive bias across most of the basin, particularly in the high-altitude southern ranges. This overestimation is consistent with other evaluations of ERA5_Land and reanalysis datasets in complex terrain, where model dynamics often overestimate precipitation due to unresolved topographic detail and convective parameterization biases [23,66,71].

PERSIANN tends to overestimate precipitation in lowland and mid-elevation zones and underestimate precipitation in upper elevations. These patterns are typical of infrared-based satellite retrievals, which often fail to detect cold cloud-top precipitation associated with snow or stratiform rainfall in mountainous areas [18,63].

IMERG accurately estimates broad spatial patterns. However, IMERG underestimates precipitation in high-altitude zones as observed in other high-mountain evaluations, where IMERG struggled to resolve localized convective and orographic precipitation due to sensor limitations and insufficient gauge calibration [24,50,71].

In summary, the CHIRPS dataset estimates long-term mean precipitation with the highest spatial accuracy across the basin, followed by IMERG, and TerraClimate. These findings emphasize the importance of spatial resolution, topographic sensitivity, gauge integration, and calibration algorithms in determining the reliability of gridded precipitation datasets. The consistency and robustness of CHIRPS demonstrates its suitability for basin-scale applications, including hydrological modeling, water resource management, and transboundary climate risk assessments in mountainous, data-scarce environments of the basin.

The anomaly-based evaluation of annual precipitation over the basin confirms 2008 as a distinctly dry year and 2016 as an anomalously wet year (Figure 10), with consistent signals across all datasets. Anomalies were computed relative to the 2001–2023 climatological mean, and gridded anomaly maps were compared with station-based interpolations and a topographically distributed reference map to assess spatial consistency and elevation-dependent patterns.

In 2008, station observations uniformly indicated significant drought conditions, with negative precipitation anomalies ranging from 19% to 60%, affirming it as one of the driest years in the study period. Gridded datasets generally reproduce this dry signal, although they differ in the magnitude and spatial extent of anomalies. CHIRPS indicates relatively moderate drought conditions, with most basin areas showing precipitation deficits below 20%, while PERSIANN indicates more severe dryness, with widespread anomalies exceeding 20%. IMERG indicates slightly wetter conditions in small portions of the high-altitude terrain, and GPCC shows limited wetness in the eastern highlands but still indicates overall dryness below the 40% threshold. ERA5_Land, in contrast, highlights stronger drought conditions in the lowland regions, with deficits over 20%, and minor wet anomalies in select high-altitude zones. TerraClimate exhibits a distinct spatial pattern, emphasizing substantial dryness in wetland areas (anomalies > 20%) and more subdued deficits in parts of the eastern highlands. Additionally, TerraClimate captures severe dryness (>40%) in certain northern high-elevation areas. Most datasets consistently depict greater drought severity in the lowland and mid-altitude areas compared to high-altitude regions, likely due to elevation-dependent error propagation and the influence of orographic precipitation processes. These findings collectively confirm that 2008 was a year of basin-wide drought, though spatial distributions and magnitudes vary significantly among datasets.

In 2016, the precipitation anomalies reveal a basin-wide wet year, confirmed by station observations showing anomalies from 30% to 82% above the long-term mean. All gridded datasets successfully captured this hydrologically wet period, although the magnitude and spatial distribution of anomalies varied substantially. ERA5_Land indicates relatively moderate increases in precipitation, with most basin areas exhibiting anomalies below 40% and some highland and wetland zones remaining below 20%. PERSIANN, in contrast, indicates a more pronounced and widespread wet signal, with anomalies exceeding 40% across nearly the entire basin, including high-altitude regions. CHIRPS indicates moderately wet conditions, with anomalies mostly below 40%, except for a small area in the southeastern highlands showing weaker signals (<20%) and distinct high anomalies (>40%) in some wetland zones. IMERG indicates a clear elevation-dependent gradient, with the strongest wet anomalies (>40%) concentrated in lowland regions and much weaker or even negative anomalies in localized high-elevation areas. GPCC indicates a heterogeneous anomaly pattern, with strong wetness in certain lowland areas, reduced anomalies in the mid-altitudes, and weak or slightly negative anomalies in parts of the northern and southeastern highlands. TerraClimate indicates a distinct anomaly distribution, with moderate wetness (<40%) in the eastern lowlands and stronger anomalies (>40%) across the central and western Basin.

Overall, all datasets classify 2016 as a wet year but differ in their portrayals of the spatial variability and intensity of precipitation surpluses. These differences are likely attributable to variations in algorithm design, calibration data, spatial resolution, and the treatment of orographic and convective precipitation processes.

4.4. The Role of Accurate Precipitation Data for Sustainable Water Management Under Climate Stress

Accurate precipitation data are critical for sustainable water management in regions facing climate variability, water scarcity, and data scarcity, such as Central Asia and other mountainous or arid areas globally. Advancing sustainable water management under climate stress in Central Asia and similar global regions requires prioritizing high-accuracy precipitation datasets, strengthening in situ monitoring, and applying regionally validated datasets tailored to seasonal and elevation-specific dynamics [23,53]. Such approaches will support effective drought preparedness, flood risk reduction, and long-term adaptation strategies.

Global assessments confirm that runoff and evapotranspiration models are highly sensitive to input precipitation, with uncertainty cascading through water balance estimates [2,76]. This sensitivity underscores the critical role of accurate precipitation data in ensuring reliable hydrological modeling, which is essential for sustainable water management under climate variability and change. However, scholars argue that no single dataset is universally reliable; instead, integrated approaches, combining satellite, reanalysis, and ground observations with regional bias corrections, are essential for developing robust, climate-resilient water governance systems [6,60]. These integrated approaches are apparent in policy frameworks which underscores the need for improved hydroclimatic data infrastructure to support sustainable land and water management in vulnerable regions [77]. Ultimately, improving precipitation data accuracy through rigorous evaluation enables more robust hydrological modeling and climate impact assessments. This supports adaptive water governance, drought risk reduction, and equitable transboundary cooperation across Central Asia and other vulnerable mountainous or arid regions worldwide [23].

Our study provides a foundational evaluation of satellite and reanalysis-based precipitation datasets, offering critical insights that enhance decision-making for sustainable water resource management and climate adaptation. The findings are relevant for the basin and offer transferable value to other mountain-fed transboundary river systems across Central Asia, where data scarcity and climatic variability pose significant challenges to water security and cooperative basin management.

5. Conclusions

At basin-wide stations, all datasets demonstrated strong annual correlations with ground observations (r > 0.75), although correlation coefficients slightly decreased for seasonal analyses. PERSIANN outputs were more weakly correlated with rain-gauge observations during winter and summer, likely due to limitations in capturing convective and snow-related precipitation events. Percent bias (PBIAS) results revealed that most datasets exhibited low annual biases (<5%), but errors increased seasonally, particularly in summer, with IMERG and PERSIANN exhibiting the highest positive biases. The accuracies of precipitation predictions generally declined in the high-elevation zone (>1600 m) for all datasets, highlighting the challenges of monitoring precipitation in complex mountainous regions.

Basin-wide gridded evaluations highlighted substantial differences in the spatial distribution of precipitation among datasets. CHIRPS most accurately estimated topographically driven precipitation patterns, particularly orographic gradients, more accurately than others. GPCC showed excellent point-based correlation with station data but was less accurate for predicting precipitation patterns in gridded spatial evaluations, especially in the eastern high-altitude regions. This is attributed to the scarcity of ground stations in those areas, leading to weak interpolation performance.

All datasets successfully identified 2008 as a drought year and 2016 as a wet year, confirming their general capability to detect interannual variability. However, differences in the magnitude and spatial detail of anomalies were evident. PERSIANN often overestimated wet or dry conditions, limiting its reliability for extreme event analysis.

In conclusion, CHIRPS, GPCC, and TerraClimate emerge as the most spatially and temporally accurate datasets across diverse spatial and temporal conditions in the basin (based on Pearson’s correlation coefficients, percent bias, and normalized root mean square error), followed by IMERG, PERSIANN, and ERA5_Land. These datasets provide inputs that can be used for hydrological modeling, climate impact assessments, and water resource management, particularly in data-scarce, topographically complex regions. However, no single dataset performs optimally under all conditions, highlighting the need for multi-source integration approaches to improve precipitation monitoring.

To enhance the performance of remote sensing and reanalysis datasets, future efforts should prioritize expanding the network of ground-based observation stations, especially in high-altitude areas. These regions are critical for water supply through snow and glacier melt that substantially contributes to early runoff and summer irrigation demands. Improved in situ data coverage would enhance the calibration and validation of satellite datasets, ultimately supporting more accurate hydrological modeling, water sustainability, and flood–drought mitigation activities for communities within the Ili River Basin.