ResLap: Generating High-Resolution Climate Prediction Through Image Super-Resolution

Abstract

In recent years, many models based on the convolutional neural network have achieved highquality reconstruction for single image super-resolution. Meanwhile, many researches on image superresolution have been applied to various fields. However, only a few research works have been applied to climate prediction. In this paper, we present ResLap to achieve high-resolution climate prediction. ResLap is a spatial downscaling method that converts low spatial resolution climate data into high-resolution regional climate forecasts. This method mainly introduces a novel residual dense block (RDB) into the Laplacian pyramid super-resolution network (LapSRN). Among them, we use LapSRN to achieve upsampling image reconstruction, and adopt RDB to fully extract the hierarchical features from all the convolutional layers. Extensive experimental results on benchmark climate datasets show that our new proposed model performs better than many super-resolution methods. Besides, the climate data are more complicated than the general image, because of its dynamic and chaotic nature. To facilitate model training, we integrate original climate data provided by the China Meteorological Administration, then convert it into trainable climate images. We also publish some climate image datasets online for research. Finally, we avoid the checkerboard artifacts in the generated high-resolution climate images. INDEX TERMS Super-resolution, climate image, checkerboard artifacts, convolutional neural network.

Augmented Convolutional LSTMs for Generation of High-Resolution Climate Change Projections

“…Section V describes the experimental settings. In Section VI, we compare our results with the current state-ofthe-art ResLap [27], DeepSD [22] and widely used quantile mapping technique [28]. In Section VII, we briefly discuss results, limitations, and scalability potential of our work to generate high-resolution outputs that can be leveraged by stakeholders and policymakers for design and adaptation planning.…”

“…Research efforts in the field of space-time precipitation downscaling have considered surface and atmospheric variables including sea-level pressure, wind velocities in the three orthogonal directions, relative and specific humidity, and surface temperature as potential predictors [33]. On the other hand, Previous works in the field of image super-resolution and its adaptation for statistical downscaling including DeepSD and ResLap used either only coarser resolution of the target variable [27] or surface elevation in addition to the coarser variable [34]. In this study, we explore how the performance of augmented Convolutional LSTMs will improve by inclusion of various auxiliary variables including elevation, humidity, atmospheric pressure, and wind-velocity components in the three directions.…”

“…DeepSD [34] consists of convolution neural networks based super-resolution architecture SRCNN [43] to capture the spatial dependencies. Inspired by this, other researchers have proposed ResLap [27] which uses laplacian pyramid based super-resolution network [44] for further improvement in the quality of the generated climate change projections. ResLap [27] extracts hierarchical features from the original precipitation projection and used a series of transposed convolution networks to upscale the low-resolution projection to the target scale.…”

“…Inspired by this, other researchers have proposed ResLap [27] which uses laplacian pyramid based super-resolution network [44] for further improvement in the quality of the generated climate change projections. ResLap [27] extracts hierarchical features from the original precipitation projection and used a series of transposed convolution networks to upscale the low-resolution projection to the target scale. On other hand, given the long short term memory processes that are inherent to the system, researchers have also used recurrent and vanilla Long Short Term Memory (LSTM) [45] for climate downscaling.…”

CliGAN: A Structurally Sensitive Convolutional Neural Network Model for Statistical Downscaling of Precipitation from Multi-Model Ensembles

“…Despite several limitations in their experiments, the result shows the efficiency and robustness of the approach over other methods in predicting extremes. [53] also recently introduced a novel residual dense block (RDB) into the Laplacian pyramid super-resolution network (LapSRN) to generate high-resolution precipitation forecast. [54] used super-resolution techniques to simulate high-resolution urban micrometeorology, while [55] proposed several CNN-based architectures to forecast high-resolution precipitation.…”