So far I have explored the production process behind automated cloud classification and evaluated the accuracy of these results. I have shown how the words we use to describe clouds reflect how we interact with them and therefore inform automated cloud classification. Distanciation in the form of standardization tries to capture the boundless iterations of the sky with a limited vocabulary. Even with standardization, different instruments as well as humans observe clouds in different and sometimes incompatible ways. While instruments that observe clouds differently than humans are not necessarily incorrect, for cloud type observations, human observations are considered objectively true. Using that standard for classifying images from the TSI, the algorithm performs with 89% accuracy. However, this percentage is for images that have been selected as example images for these cloud types and included easily classifiable clear skies. Just as the human observations disagree for complicated cloud images, the algorithm would perform much more poorly if the images were completely random. The research question asking How accurate is our understanding of classification? is slightly misleading as we first have to identify the measure for truth. We also have to ask ourselves about the uses of the data, as some observation and instruments may be better suited to different tasks. Given the subjectivity and assumptions built into this process of cloud identification, it may be that an objective cloud reality is elusive. However, the references we create with the help of instrumentation and algorithms help us to create meaningful and useful representations of the real world.
From my own reflection on the production process, I realize how personal and nuanced writing code, making decisions about the methodology, and manually classifying clouds is. There is no cut and dry path even when I follow other scholars. Somehow I am bringing together all these different pieces from different disciplines and producing knowledge. Having been part of the process, I am both more skeptical of other scientific results and impressed by others’ processes. While I am part of this bigger picture, I am distinctly and singularly aware of my personal experience. So while much of what I am doing could fit into the context of distanciation, the actual process is still intimate and tied to the physical and conceptual place that I occupy.
Parts of the process of cloud classification exhibit distanciation, hybridity, and issues of scale, but these are not specific to meteorology or climate science. Scientific discovery arises from a social context, one which helps construct meaning from material reality. The production process takes this into account along with the location, instruments, and algorithms that knowledge arises from. Latour’s concept of references explains how science understands the physical world through constructed frameworks. Since our understanding of objects like clouds is co-produced from the material reality and human observations, it is beneficial to think of them as hybrids. This idea helps explain a paradox in which scientific representation pushes the real world farther away but also brings it closer (Latour 1999). My experience producing knowledge supports this apparent paradox since I simultaneously distanced myself to create representations while being ensconced in my personal physical and conceptual place