Deep Learning Model for Global Spatio-Temporal Image Prediction

Abstract

Mathematical methods are the basis of most models that describe the natural phenomena around us. However, the well-known conventional mathematical models for atmospheric modeling have some limitations. Machine learning with Big Data is also based on mathematics but offers a new approach for modeling. There are two methodologies to develop deep learning models for spatio-temporal image prediction. On these bases, two models were built—ConvLSTM and CNN-LSTM—with two types of predictions, i.e., sequence-to-sequence and sequence-to-one, in order to forecast Aerosol Optical Thickness sequences. The input dataset for training was NASA satellite imagery MODAL2_E_AER_OD from Terra/MODIS satellites, which presents global Aerosol Optical Thickness with an 8 day temporal resolution from 2000 to the present. The obtained results show that the ConvLSTM sequence-to-one model had the lowest RMSE error and the highest Cosine Similarity value. The advantages of the developed DL models are that they can be executed in milliseconds on a PC, can be used for global-scale Earth observations, and can serve as tracers to study how the Earth’s atmosphere moves. The developed models can be used as transfer learning for similar image time-series forecasting models.

A Comparative Evaluation of Self-Attention Mechanism with ConvLSTM Model for Global Aerosol Time Series Forecasting

“…In the previous study, the authors developed models to forecast AOT sequences [3]. The sequence-to-one ConvLSTM model had the smallest errors and represents a basic model in this study.…”

“…The model input was sequence of 10 images temporally distributed by 8 days, and output is one image that represent 11th image as prediction for next 8 day. Difference and improvement to previous research [3] was 10 times increasing of the database by overlapping technique which shift input sequence of 10 images for one image (8 days) instead for 10 images as in [3]. Larger database provides better and more accurate machine learning results.…”

“…In the MODAL2_E_AER_OD dataset black pixels of the images represent spots where the sensors could not make measurements, and the difference in data preparation compared to the previous study [3] is that the unread i.e., black pixels were converted to white color which indicate a zero AOT concertation. Therefore, the problem with loading and predicting unknown values has been reduced on possible difference in useful range and the obtained results are more comparative.…”

“…The evaluation of the ConvLSTM-SA model was done by several evaluation metrics such as Cosine Similarity CS, Root Mean Squared Error RMSE and Euclidean Distance EUCD. In previous research we used RMSE and CS [3], and a new metric is EUCD. The purpose of a similarity measurement is to compare two vectors of pixels in order to compute a single number which evaluates their similarity.…”

Literature Review on Integrating Generalized Space-Time Autoregressive Integrated Moving Average (GSTARIMA) and Deep Neural Networks in Machine Learning for Climate Forecasting

“…The model used diverse climate data sources, prioritizing the integration of DL models without considering location parameters [20,23,[58][59][60]79]. The performance analysis of DL integration yielded quite good results based on empirical simulation.…”