Core Concepts
The author proposes a spatially varying autoencoder to merge climate data products, ensuring interpretability and capturing spatial patterns effectively.
Abstract
The research introduces a novel approach using an identifiable and interpretable autoencoder to combine climate data products. The study focuses on the High Mountain Asia region, where in situ measurements of precipitation are limited, relying heavily on digital data products. Various digital data products often disagree due to unique construction methods for different scientific purposes. The proposed autoencoder aims to create a consensus product by merging multiple data sources through spatial variation and Bayesian statistical framework. By imposing constraints on the autoencoder, the resulting consensus includes essential features from each input while maintaining interpretability. The study demonstrates the utility of the autoencoder by combining information from various precipitation products in High Mountain Asia.
Stats
Nearly 650 glaciers feed water into major rivers.
In situ measurements of precipitation are sparse in HMA.
Digital data products commonly disagree in measurements.
APHRODITE relies on statistical interpolation of rain gauge data.
MERRA-2 utilizes data assimilation techniques.
ERA5 is global and relies on weather forecasting models.
TRMM focuses on tropical area precipitation.
Quotes
"The proposed autoencoder utilizes a Bayesian statistical framework."
"Constraints are placed on the autoencoder as it learns patterns in the data."
"The primary contribution is proposing an autoencoder to fuse precipitation data."