GPC/m - Global Precipitation Climatology by Machine Learning - Quasi-global, Daily, and One Degree Spatial Resolution

Aug 3, 2026

This paper introduces a daily precipitation dataset with quasi-global coverage and one-degree spatial resolution, spanning over 42 years, built using machine learning. The dataset aims to deliver a homogeneous daily record for multi-decade climate analysis with minimal gaps. A first 42-year daily dataset is generated from supervised learning using reference precipitation estimates from 2001–2020. Random forest, gradient-boosted decision trees, and convolutional neural networks are trained with satellite observations and reanalysis atmospheric circulations, then used to predict back to 1979 where daily global precipitation was scarce.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages22

Reading Time55 min

AccessFree to download

Introduction

Motivation and existing datasets

Challenges in long-term homogeneous quality

Approaches using machine learning

Dataset goals and evaluation status

arXiv:2409.09639v1 [[physics. ao-ph](physics. ao-ph)] 15 Sep 2024 GPC/m: Global Precipitation Climatology by Machine Learning; Quasi-global, Daily, and One Degree Spatial Resolution Hiroshi G. Takahashi Tokyo Metropolitan University [hiroshi3@tmu.ac.jp](hiroshi3@tmu.ac.jp) September 17, 2024 Abstract This paper presents a new precipitation dataset that is daily, has a spatial resolution of one degree on a quasi-global scale, and spans more than 42 years, using machine learning techniques. The ultimate goal of this dataset is to provide a homogeneous daily precipitation dataset for several decades without gaps, which is suitable for climate analysis. As a ﬁrst step, 42 years of daily precipitation data was generated using machine learning techniques. The machine learning methods are supervised learning, and the reference data are estimated precipitation datasets from 2001 to 2020 . Three machine learning methods are used: random forest, gradient-boosted decision trees, and convolutional neural networks. The input data are satellite observations and atmospheric circulations from reanalysis, which are somewhat modiﬁed based on knowledge of the climatological background. Using the trained statistical models, we predict back to 1979, when daily precipitation data was almost unavailable globally. The detailed procedures are described in this paper. The produced data have been partially evaluated. However, additional evaluations from diﬀerent perspectives are needed. The advantages and disadvantages of this precipitation dataset are also discussed. Currently, this GPC/m precipitation dataset version is GPC/m-v1-2024 . 1 Introduction The importance of precipitation datasets has increased in recent years as they are extensively used for climate and meteorological analyses, such as precipitation variability, water budget, evaluation of climate model, and related research (e.g., Trenberth, 2011) . However, the available precipitation data have some limitations, especially diﬃculties in using precipitation data globally, including over the ocean overlong periods (Becker et al., 2013) . This study produces a prototype of the precipitation dataset using machine learning methods. This machine learning-based precipitation can be used in climate and related studies. Additionally, this study can contribute to discussing not only the advantages but also the problems of global precipitation climatological datasets using machine learning methods. In general, these advantages are easy to understand, but this dataset also focuses more on limitations and issues with the use of machine-learning-based precipitation data as climate data (e.g., Kumar et al., 2015; Takahashi et al., 2023) . Global Precipitation Climatology Centre (GPCC) and Global Precipitati,on Climatology Project (GPCP) are often used as de facto standard precipitation datasets with a focus on long-term climate variability. These datasets provide monthly moderate-resolution outputs that are suﬃcient for analyses on global and continental scales (Schneider et al., 2014) . The typical scales are monthly and with a 1◦ spatial resolution. Land datasets are available starting from the early 20th century although the spatial coverage and density of the observations are limited. High-resolution but periodlimited datasets derived from rain-gauge observations became available starting in approximately 1950. Global datasets, including sea and no-ground observation areas, generally began in 1979, stemming from advances in satellite observations (e.g., Adler et al., 2003) . Currently, the two dataset have been updated positively. Recent datasets such as IMERG and GSMAP focus on severe meteorological disasters and have very high temporal and spatial resolutions. Typical scales are shorter than one hour and have 0.1◦ order resolution (e.g., Huﬀman et al., 2015b) . Although precipitation data are widely available, these datasets are mainly from 2000 onward, with limited data

FAQs

What kind of precipitation dataset is proposed in this paper?

The paper presents a new precipitation dataset that is daily, quasi-global, and has one-degree spatial resolution, covering more than 42 years.

Which machine learning models are used to generate and extend the dataset?

Three supervised methods are used: random forest, gradient-boosted decision trees, and convolutional neural networks.

How does the method extend daily precipitation data to earlier years?

Trained statistical models use satellite observations and modified reanalysis atmospheric circulations to predict daily precipitation back to 1979.

Aug 3, 2026

Upload by Theodora