STA-GAN: A Spatio-Temporal Attention Generative Adversarial Network for Missing Value Imputation in Satellite Data

Abstract

Satellite data is of high importance for ocean environment monitoring and protection. However, due to the missing values in satellite data, caused by various force majeure factors such as cloud cover, bad weather and sensor failure, the quality of satellite data is reduced greatly, which hinders the applications of satellite data in practice. Therefore, a variety of methods have been proposed to conduct missing data imputation for satellite data to improve its quality. However, these methods cannot well learn the short-term temporal dependence and dynamic spatial dependence in satellite data, resulting in bad imputation performance when the data missing rate is large. To address this issue, we propose the Spatio-Temporal Attention Generative Adversarial Network (STA-GAN) for missing value imputation in satellite data. First, we develop the Spatio-Temporal Attention (STA) mechanism based on Graph Attention Network (GAT) to learn features for capturing both short-term temporal dependence and dynamic spatial dependence in satellite data. Then, the learned features from STA are fused to enrich the spatio-temporal information for training the generator and discriminator of STA-GAN. Finally, we use the generated imputation data by the trained generator of STA-GAN to fill the missing values in satellite data. Experimental results on real datasets show that STA-GAN largely outperforms the baseline data imputation methods, especially for filling satellite data with large missing rates.

A Comprehensive Bibliometric Analysis of Missing Value Imputation

“…It is labeled as missing value by both log-likelihood ratios (LLR) and Latent Semantic Indexing (LSI), and as analysis ( Some study opportunities that received less attention comprised missing data and the class center method, as initially explored by Nugroho et al [169]- [172], building upon the work initiated by Tsai et al [173]. More recently, the GAN method for data imputation was developed and studied [133], [137], [147], [148], [174]- [200]. To identify emerging areas of research and trends, the CiteSpace algorithm-dependent analytical tool generated keywords burst map from the strongest citation bursts in scientific literature.…”

“…Centrality served as a metric for gauging a chosen topic's significance, whereas density measured its level of development. These fundamental themes and evolving topics describe crucial facets of the discipline that had not yet reached their full potential, presenting opportunities for further investigation, such as missing data imputation using GAN [133], [137], [147], [148], [174]- [200], and deep learning [10], [205]- [216].…”

Missing Data Imputation Method Combining Random Forest and Generative Adversarial Imputation Network

“…The work of Bernardini [ 15 ] and others is based on generative adversarial networks under clinical conditions, using nonlinear and multivariate information between data to fill missing values, which is more robust than other algorithms. Wang [ 16 ] and others proposed a spatiotemporal attention generative adversarial network algorithm (STA-GAN) to learn short-term correlation and dynamic spatial correlation in satellite data in order to improve the filling effect when the missing data rate is large. Zhang et al [ 17 ] proposed a self-attention generative adversarial network filling model (SA-GAIN) to fill in missing data of traffic flow.…”

Improvement in Spatiotemporal Chl-a Data in the South China Sea Using the Random-Forest-Based Geo-Imputation Method and Ocean Dynamics Data

“…Despite the important advantages of using remote sensing data for monitoring marine Chl-a concentrations, several factors can affect the accuracy and integrity of such data. These include sensor resolution, atmospheric disturbances, tides and waves, reflectance absorption, and so on [14,15]. Researchers have noted that because of the complexity of the marine environment and the inherent limitations of remote sensing techniques, there may be incomplete and sporadic data on Chl-a concentrations, both in terms of spatial and temporal coverage [16,17].…”

“…Recent research demonstrates the efficacy of Machine Learning (ML) methodologies as viable alternatives to traditional statistical approaches in the realm of spatiotemporal imputation [14][15][16]20,21]. He Qian compared the interpolation of Chinese temperature data based on three machine learning methods (random forest, support vector machine, and Gaussian process regression) and three traditional interpolation methods (inverse distance weighting, ordinary kriging, and ANUSPLIN), and found that the machine learning algorithms performed better at interpolating temperature prediction compared to the traditional algorithms [22].…”