Automatic cloud classification of whole sky images

Abstract

Abstract.The recently increasing development of whole sky imagers enables temporal and spatial high-resolution sky observations. One application already performed in most cases is the estimation of fractional sky cover. A distinction between different cloud types, however, is still in progress. Here, an automatic cloud classification algorithm is presented, based on a set of mainly statistical features describing the color as well as the texture of an image. The k-nearestneighbour classifier is used due to its high performance in solving complex issues, simplicity of implementation and low computational complexity. Seven different sky conditions are distinguished: high thin clouds (cirrus and cirrostratus), high patched cumuliform clouds (cirrocumulus and altocumulus), stratocumulus clouds, low cumuliform clouds, thick clouds (cumulonimbus and nimbostratus), stratiform clouds and clear sky. Based on the Leave-One-Out CrossValidation the algorithm achieves an accuracy of about 97%. In addition, a test run of random images is presented, still outperforming previous algorithms by yielding a success rate of about 75%, or up to 88% if only "serious" errors with respect to radiation impact are considered. Reasons for the decrement in accuracy are discussed, and ideas to further improve the classification results, especially in problematic cases, are investigated.

Coupling sky images with radiative transfer models: a new method to estimate cloud optical depth

“…3a) for any given day, allowing the calculation of the ratio of measured RBR to clear-sky RBR. The red-blue difference (RBD; Heinle et al, 2010) uses the same principles as the RBR for cloud detection but attempts to mitigate the strong directional variability in the RBR due to variability in the radiance, I λ , of the blue chan- Figure 2. Diagram of the UCSD sky imager (USI) and related solar and sky geometries.…”

“…Other ground-based sky imaging designs have also been developed (Seiz et al, 2007;Souza-Echer et al, 2006;Calbo and Sabburg, 2008;Cazorla et al, 2008;Heinle et al, 2010;Román et al, 2012;Gauchet et al, 2012) with the most dissimilar design consisting of a downward-pointing camera capturing the sky from a reflection off a spherical mirror (Pfister et al, 2003;Kassianov et al, 2005;Long et al, 2006;Mantelli et al, 2010;Martínez-Chico et al, 2011). Most ground imaging devices follow a relationship between the camera's signal and radiance similar to Eq.…”

Compact Sparse Coding for Ground-Based Cloud Classification

“…These methods include local binary pattern [5], spectral-texture feature [7], census transform-structure feature [1], traditional sparse coding (T SC) [8], the nonnegative sparse coding (NN SC) [12], laplacian sparse coding (LSC) [13], and K-SVD sparse coding [14]. For all the coding models, we use SVM as the classifier.…”

“…While the algorithms based on texture features include co-occurrence and auto-correlation matrices [4], and local binary patterns and it extension [5], [6]. In addition, several algorithms [1], [7] have been proposed that fuse these two characteristics as the final representation. Although these works are suggestive, they fail to extract really useful information from cloud images.…”

Learning Discriminative Features for Ground-Based Cloud Classification via Mutual Information Maximization

“…Calbó et al [2] employed Fourier transformation to represent the cloud images. Heinle et al [3] utilized spectral and textural features, such as energy and entropy for cloud classification. Liu et al [4] proposed a tensor ensemble of ground-based cloud sequences (eTGCS) model to represent the cloud sequences in a tensor manner.…”