On the other hand, humans can distinguish fairly easily between these types of phenomena, in large part by recognizing different types of texture and its temporal evolution. Rather than letting the machine learn texture, here we take a step back and explore using deterministic mathematical methods which can distinguish texture to generate features which can then be used as input to NNs, in addition to the raw imagery. Our motivation is that mathematical methods can be developed with few labeled samples – since they generally only require tuning of a few parameters -- and that this combination of texture detection followed by NNs results in simpler, more interpretable algorithms that can be trained on a small number of samples but nevertheless can identify complex patterns.
A set of methods with particular promise to detect texture is based on the Gray Level Co-Occurrence Matrix (GLCM) [2] used routinely in remote sensing, especially for detecting land cover changes [3]. Previous studies [4,5], have explored these methods for cloud classification in satellite imagery over two decades ago, but they never entered widespread usage for this application. We seek to re-apply these methods to satellite weather data by using them in conjunction with modern machine learning algorithms with the goal of creating more transparent techniques. This presentation will apply the GLCM approach to a dataset of GOES-16 visible images of convection and show how different textural features derived from the GLCM relate to satellite signatures of convection. The strengths and weaknesses of this approach will be discussed, and the most promising avenues for future research will be identified.
References:
[1] Lee, Y., Kummerow, C.D. and Ebert-Uphoff, I., 2021. Applying machine learning methods to detect convection using Geostationary Operational Environmental Satellite-16 (GOES-16) advanced baseline imager (ABI) data. Atmospheric Measurement Techniques, 14(4), pp.2699-2716.
[2] Haralick, R.M., Shanmugam K., Dinstein I., 1973. Textural Features for Image Classification. IEEE Transactions on Systems, Man, and Cybernetics, vol. SMC-3, no. 6, pp. 610-621.
[3] Kupidura, P., 2019. The comparison of different methods of texture analysis for their efficacy for land use classification in satellite imagery. Remote Sensing, 11(10), p.1233.
[4] Tian, B., Shaikh, M.A., Azimi-Sadjadi, M.R., Haar, T.H.V. and Reinke, D.L., 1999. A study of cloud classification with neural networks using spectral and textural features. IEEE Transactions on Neural Networks, 10(1), pp.138-151.
[5] Christodoulou, C.I., Michaelides, S.C. and Pattichis, C.S., 2003. Multifeature texture analysis for the classification of clouds in satellite imagery. IEEE Transactions on Geoscience and Remote Sensing, 41(11), pp.2662-2668.

