Hideaki Takenaka is an Associate Professor at the University of Yamanashi and holds a Ph.D. in Science. He specializes in solar radiation analysis using geostationary meteorological satellites, satellite remote sensing, atmospheric radiation, machine learning, and renewable energy applications. By integrating first-principles physical models with machine-learning-based computational intelligence, he has developed analysis algorithms that rapidly and accurately estimate physical quantities from satellite observations.
He is the developer of AMATERASS, a quasi-real-time solar radiation analysis system for geostationary satellites, and has continuously developed, implemented, and operated the system since its analysis began on July 7, 2007.
Takenaka conducted research on atmospheric radiation and satellite remote sensing at the Center for Environmental Remote Sensing (CEReS), Chiba University, where he received his Ph.D. in Science in 2009. He subsequently pursued the physical analysis of satellite observations and research on the global environment at institutions including the Climate System Research Division of the Atmosphere and Ocean Research Institute, The University of Tokyo (AORI/CCSR), and the Earth Observation Research Center of the Japan Aerospace Exploration Agency (JAXA/EORC). In 2026, he joined the Faculty of Engineering, Graduate Faculty of Interdisciplinary Research, University of Yamanashi, as an Associate Professor.
In the field of ground-based observations, which complement satellite measurements, he participated in SKYNET, a radiation and aerosol observation network spanning East Asia. He helped maintain observation sites, including the station at Cape Hedo, and took part in a range of field campaigns. By combining ground-based and satellite observations, he contributed to research on the characteristics of clouds, aerosols, and solar radiation over East Asia. From 2003 to 2013, he was responsible for administering SKYNET's main server, which collected, transferred, and archived observational data from participating countries.
This experience of bringing together ground-based observations, radiative transfer calculations, and satellite sensor calibration within a unified analytical framework later formed the foundation for the development of AMATERASS.
Physical analysis of satellite observations has traditionally relied on the Look-Up Table (LUT) method, in which radiative transfer calculations are performed in advance and stored in reference tables. With the LUT method, however, the required database expands dramatically as the number of atmospheric and surface parameters increases, making it difficult to address high-dimensional problems.
To overcome this limitation, Takenaka developed a method in which the results of radiative transfer calculations are learned by a neural network, which then functions as a high-speed radiative transfer solver. His original Distortion Back-Propagation method (Distortion-BP) introduces multiple learning parameters to deform the error-function landscape, reducing the tendency of training to become trapped in low-accuracy local solutions. The method also incorporates the "sudden death" and "slow death" of neurons, allowing the network structure to be adjusted dynamically during training.
This neural-network solver accelerates radiative transfer calculations by approximately 1,000 times while retaining numerous physical parameters — including aerosols, clouds, absorbing gases, and surface properties — without simplification. It enables higher-level products, such as surface global, direct, and diffuse solar irradiance and top-of-atmosphere radiative fluxes, to be generated from satellite-observed radiances within approximately ten minutes of observation.
This approach integrates physical models and machine learning in a way that anticipates what is now known as physics-informed machine learning. Rather than performing a purely statistical regression on observed values, the neural network learns the physical processes represented by a radiative transfer model, enabling physical quantities to be estimated over wide areas at high temporal frequency.
On July 7, 2007, Takenaka began a quasi-real-time analysis to estimate solar radiation at the Earth's surface and the top of the atmosphere from geostationary meteorological satellite observations. This analysis system is AMATERASS.
AMATERASS was developed and implemented as an integrated analysis platform that continuously performs satellite data acquisition, preprocessing, radiative-transfer-based physical analysis, geolocation correction, conversion to latitude–longitude grids, product generation, and data publication. The system has been repeatedly updated through successive generations of the Himawari satellites — from Himawari-6 and Himawari-7 to Himawari-8 and Himawari-9 — and its quasi-real-time analysis continues today.
Applying the algorithms developed for AMATERASS to observations from Japan's Himawari geostationary satellites has produced new scientific findings, including an inverse relationship between the seasonal variations of direct and diffuse solar irradiance over East Asia. AMATERASS products are now used across a wide range of fields, including weather and climate, renewable energy, agriculture and vegetation, water resources, and the social sciences. By March 2026, the data had been downloaded more than 200 million times.
Accurate sensor calibration and stable geolocation are essential when geostationary satellite observations are used for quantitative physical analysis.
In collaboration with the Japan Meteorological Agency's Meteorological Satellite Center, Takenaka contributed to the development of a vicarious calibration method for the visible and near-infrared sensors aboard the Himawari satellites. The method uses a radiative transfer model to reproduce the conditions of the surface, atmosphere, clouds, and aerosols, and compares the simulated radiances with satellite observations to continuously evaluate the calibration coefficients of sensors in orbit.
The Japan Meteorological Agency presented this work to the World Meteorological Organization's Global Space-based Inter-Calibration System (GSICS) as a Japanese satellite calibration technique. It is used in the vicarious calibration processing of visible and near-infrared bands, including those aboard Himawari-8 and Himawari-9.
Geostationary satellites can observe the same region at high temporal frequency. However, variations in satellite attitude and the movement of scanning mirrors can introduce geolocation errors that vary from one observation time to another. These errors are a major source of uncertainty when satellite data are used to derive physical quantities quantitatively.
To address this problem, Takenaka developed a Geolocation Correction method based on Phase-Only Correlation. The method rapidly and accurately determines the correspondence between satellite imagery and geographic information, corrects geolocation errors in geostationary satellite images, and generates stable products on latitude–longitude grids.
The geolocation correction and gridding algorithms were provided to NASA NEX through an international collaboration with NASA Ames Research Center. Following their integration by the GeoNEX research team, the corrected data were released as NASA GeoNEX L1G products.
The Great East Japan Earthquake of 2011 and the electricity supply challenges that followed led Takenaka to extend AMATERASS-based solar radiation analysis into research on photovoltaic power estimation, electric power systems, and distributed cooperative energy management.
Under the JST CREST research area for distributed cooperative energy management systems, he developed and operated an analysis platform that generated solar irradiance and photovoltaic power estimates from satellite data in quasi-real time. For approximately seven and a half years, the platform provided foundational data for interdisciplinary research spanning control engineering, information science, electric power systems, and the social sciences. It also served to connect the work of the participating research teams in areas such as control theory, electricity markets, demand analysis, and social science.
Through JST SATREPS, Takenaka expanded the geographical scope of the analysis to Central Asia using observations from the European Meteosat geostationary meteorological satellites. AMATERASS analysis technologies were thereby extended into international collaborative research on water resources, agriculture, and climate resilience in the Aral Sea region.
Takenaka's current research builds on geostationary satellite data analysis, radiative transfer models, and machine learning. His work now encompasses the spatiotemporal variability and potential of photovoltaic power generation, the optimal placement of distributed energy resources, energy resilience during disasters, freshwater production powered by photovoltaics, and water resource assessment.
Beginning with fundamental research on ground-based observations and atmospheric radiation, he has developed analysis algorithms that integrate physical models and machine learning and implemented them as an information infrastructure capable of sustained long-term operation. Through this work, he aims to connect Earth science with a broad range of interdisciplinary research.
