.png){kind=icon}
.png){kind=icon}
Return
PERSIANN-U-Net: A Global Deep Learning Framework for Near-Real-Time Precipitation Estimation Using Infrared Data
.png){kind=icon}
PRE
AI
P
T
K
S
A
G
F
DOI：10.1175/JHM-D-25-0162.1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Access to high-quality, high-resolution, near-real-time precipitation data is essential for hydrological and meteorological research and disaster response. Traditional tools such as rain gauges and radar networks, though effective, are limited by sparse coverage in remote areas and radar constraints such as beam blockage and increasing beam height with range, which reduce near-surface accuracy. Satellite observations address these challenges by providing global coverage with fine spatial and temporal resolution. Many precipitation products combine geosynchronous thermal infrared (IR) and passive microwave (PMW) data. PMW sensors offer detailed atmospheric profiles but are restricted to infrequent overpasses and increasing reliance on smaller satellites with higher-frequency channels, which are less sensitive to liquid precipitation. In contrast, IR sensors provide consistent, high-frequency global observations, making them valuable for near-real-time estimation. Recent advances in deep learning, particularly convolutional neural networks (CNNs), have further improved satellite precipitation retrievals. This study introduces Precipitation Estimation from Remotely Sensed Information Using Artificial Neural Networks (PERSIANN)-U-Net (PU-Net or PERSIANN V3), a quasi-global algorithm covering 60 degrees N-60 degrees S that combines IR data, monthly climatology, and the U-Net architecture to produce half-hourly precipitation estimates at 0.04 degrees resolution. The product is evaluated against Hydro Estimator (HE), Integrated Multi-satellitE Retrievals for Global Precipitation Measurement (IMERG), and PERSIANN Dynamic Infrared-Rain Rate (PDIR-Now) for 2022-23. Results show that PU-Net closely matches its training target, IMERG V07 Final, at the global scale, and its performance is further evaluated against Stage IV as a reference over contiguous United States (CONUS). Training PU-Net on IMERG (2016-21) leverages a high-quality, integrated PMW-IR-gauge precipitation product while developing an IR-based framework not reliant on PMW availability. By operating on a single global image, PU-Net avoids tile partitioning and blending steps, reducing edge discontinuities, and produces more spatially consistent precipitation fields across hemispheres.
Keywords:
Precipitation
Remote sensing
Satellite observations
Deep learning
Neural networks
Journal
IF:
2.9
Papers:
2.9K
Citations:
1.1W
Organization
Cited Papers