TEC Map Completion Using DCGAN and Poisson Blending

Abstract

Because of the limited coverage of global navigation satellite system (GNSS) receivers, total electron content (TEC) maps are not complete. The processing to obtain complete TEC maps is time consuming and needs the collaboration of five international GNSS service (IGS) centers to consolidate final completed IGS TEC maps. The advance of deep learning offers powerful tools to perform certain tasks in data science, such as image completion (or inpainting). Among them, deep convolutional generative adversarial network (DCGAN) is capable of learning the properties of the objects and recovering missing data effectively. With years of IGS TEC maps for training, the combination of DCGAN and Poisson blending (DCGAN-PB) is able to effectively learn the completion process of IGS TEC maps. Both random and more realistic masks are used to test the performance of DCGAN-PB. The results with random masks (15-40% missing data) show that DCGAN-PB can achieve better TEC map completion than DCGAN alone, and more training data can significantly improve its generalization. For the cross-validation experiment using the realistic mask from Massachusetts Institute of Technology (MIT)-TEC data (~52% missing data), DCGAN-PB achieves the average root mean squared error about three absolute TEC units (TECu) for high solar activity years and less than two TECu for low solar activity years, which is about 50% reduction of means and more than 50% reduction on standard deviations compared to two conventional single-image inpainting methods. The DCGAN-PB model can lead to an efficient automatic completion tool for TEC maps by minimizing the manual work. Plain Language Summary The limited number of global positioning system (GPS) receivers onEarth's ground leads to the incomplete original total electron content (TEC) maps, which are the GPS measurements of ionosphere. International global navigation satellite system (GNSS) service (IGS) TEC maps are completed with a joint effort of different observation stations involving a lot of manual work. In this work, we propose a deep learning method to learn the completion process of IGS TEC data to facilitate the TEC map completion. The proposed method can automatically recover the incomplete TEC maps with much reduced errors compared to two conventional automatic methods. Key Points: • We proposed a deep convolutional generative adversarial networks model to achieve the automatic completion of total electron content maps • Large amount of training data and Poisson blending are two important elements to achieve the superior performance of the proposed deep learning model • With 18 years of training data, the proposed deep learning model is able to recover the missing data more accurately than two traditional image inpainting methods

TEC Map Completion Through a Deep Learning Model: SNP‐GAN

“…Similar to our previous work (Pan et al., 2020), we use a 10‐fold cross‐validation with pairs of low and high solar activity years (low solar activity: ; high solar activity: ) to systematically evaluate the inpainting performance for different models. Specifically, the IGS‐TEC data from 18 out of 20 years (i.e., 90% of all data) were randomly selected for training, and the rest two (i.e., 10% of all data), one low solar activity year and one high solar activity year, were used for test.…”

“…With an additional reference discriminator using IGS‐TEC maps, R‐DCGAN shows good completion performance for MIT‐TEC maps with data gaps (Chen et al., 2019). In another work, Poisson blending following DCGAN (DCGAN‐PB) has achieved excellent inpainting results of IGS‐TEC maps for both small and dispersive gaps and large and continuous gaps (Pan et al., 2020). The pix2pix image translation model (Isola et al., 2017) was adopted to develop a DeepIRI model (Ji et al., 2020) to obtain improved International Reference Ionosphere (IRI) TEC maps.…”

A Deep Learning Model for the Thermospheric Nitric Oxide Emission

“…The global three‐dimensional (3‐D) image of NO emission, constructed by SABER NO emission, contains a large number of missing data that are unobserved since there is only one orbital observation every ∼1.6 h. Such an incomplete 3‐D image brings significant difficulties for modeling. The image completion has been successfully applied in space physics based on deep learning (Chen et al., 2019; Pan et al., 2020). However, there is no attempt to build a 3‐D prediction model using incomplete 3‐D images.…”

Modified iterative method with red-black ordering for image composition using poisson equation

“…By combining the efficient edge-preserving filters and optimization-based smoothing, the approach obtained comparable runtime to the fast edge-preserving filters and also overcame many limitations of the existing local filtering approaches. Yang Pan, Mingwu Jin, Shunrong Zhang, et al, [28] proposed an approach that combines deep convolutional generative adversarial network (DCGAN) and Poisson blending (PB) to perform image completion tasks. Most recently, Nordin Saad, Andang Sunarto, and Azali Saudi [29] employed a modified method by combining red-black strategy with accelerated over-relaxation, known as MAOR method, demonstrated encouraging results.…”