<p>This research introduces a novel spatial downscaling and correction framework designed to produce historical seasonal and annual precipitation estimates at high spatial resolution under data-scarce conditions. This is crucial for climatologists and water resource managers seeking to understand the annual precipitation cycle at the basin scale. Leveraging a weighted deep neural network (DNN) model, denoted as <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({DNN}_{w}\)</EquationSource> </InlineEquation>, in conjunction with the Kolmogorov–Smirnov normality test, the framework aims to capture intricate precipitation patterns at a finer scale within the humid southern region of mainland China. By integrating geospatial and environmental predictors into <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\({DNN}_{w}\)</EquationSource> </InlineEquation>, the high-resolution precipitation data (0.01° × 0.01°) is derived from the Global Precipitation Mission’s (GPM’s) coarser Integrated Multi-satellitE Retrievals for GPM (IMERG) dataset (0.1° × 0.1°). The correction process involves a two-step strategy: (1) validating the downscaled <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\({DNN}_{w}\)</EquationSource> </InlineEquation> (<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\({DNN}_{dw}\)</EquationSource> </InlineEquation>) against a limited number of rain gauge (<i>n</i> = 5) data, and (2) establishing a relationship between the resampled <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\({DNN}_{dw}\)</EquationSource> </InlineEquation> at varying resolutions and the validated data using statistical tests. Notably, the 1.25° × 1.25° resolution exhibits the strongest agreement with the rain gauge-verified data. Regression equations derived from this relationship are then applied to enhance and reconstruct the downscaled data, resulting in the corrected <InlineEquation ID="IEq6"> <EquationSource Format="TEX">\({DNN}_{dw}\)</EquationSource> </InlineEquation> (<InlineEquation ID="IEq7"> <EquationSource Format="TEX">\({DNN}_{cdw}\)</EquationSource> </InlineEquation>) outperforming comparative models. This adaptable framework shows great potential for generating high-resolution precipitation data in data-scarce environments, providing a robust solution for a variety of climatic conditions and water resource management applications, such as irrigation planning, reservoir management, and long-term water availability studies.</p> Graphical Abstract <p>This research develops an advanced framework that transforms coarse-resolution GPM IMERG satellite precipitation data (0.1° × 0.1°) into high-resolution estimates (0.01° × 0.01°) for data-scarce regions in humid southern China. The innovative approach combines a weighted deep neural network (<i>DNN</i>_w) with key high resolution predictors (i.e., geospatial and environmental) to achieve high accuracy (r² &gt;0.9999, RMSE &lt; 2 mm). The framework's two-stepped process first downscales the data using<i> DNN</i>_w, then applies a rigorous statistical correction validated against limited ground observations (n = 5 rain gauges). The Kolmogorov-Smirnov test identifies 1.25° × 1.25° resolution as optimal for calibration (p-value 0.44–0.77), ensuring reliable correction across all temporal scales. The final corrected product (<i>DNN</i>_cdw) significantly outperforms conventional models while maintaining computational efficiency. The study's key innovation lies in its ability to generate regional-scale precipitation estimates with remarkable precision despite minimal ground data, addressing a critical challenge in water resource management. The framework's adaptability makes it particularly valuable for practical applications including irrigation scheduling, reservoir operations, and long-term water availability assessments. Latitude emerges as the most influential predictor, demonstrating the method's sensitivity to geographic factors. Validation results confirm the approach's robustness, with consistently high correlation to in-situ measurements across monthly, seasonal and annual timescales. This advancement represents a significant leap in precipitation mapping technology, offering water managers and climatologists an unprecedented tool for decision-making in data-limited environments. The framework's modular design allows for future expansion to other climatic regions, promising global applicability.</p> <p></p>