Machine Learning for Biodiversity Scientists
An opinionated primer
Preface
Data science is now an established methodology to study biodiversity, and this is a problem.
This may be an opportunity when it comes to advancing our knowledge of biodiversity, and in particular when it comes to translating this knowledge into action (Tuia et al. 2022); but make no mistake, this is a problem for us, biodiversity scientists, as we suddenly need to develop competences in an entirely new field. And as luck would have it, there are easier fields to master than data science. The point of this book, therefore, is to provide an introduction to fundamental concepts in data science, from the perspective of a biodiversity scientist, by using examples corresponding to real-world use-cases of these techniques.
But what do we mean by data science? Most science, after all, relies on data in some capacity. What falls under the umbrella of data science is, in short, embracing in equal measure quantitative skills (mathematics, machine learning, statistics), programming, and domain expertise, in order to solve well-defined problems. A core tenet of data science is that, when using it, we seek to “deliver actionable insights”, which is MBA-speak for “figuring out what to do next”. One of the ways in which this occurs is by letting the data speak, after they have been, of course, properly cleaned and transformed and engineered beyond recognition. This entire process is driven by (or subject to, even) domain knowledge. There is no such thing as data science, at least not in a vacuum: there is data science as a methodology applied to a specific domain.
Before we embark into a journey of discovery on the applications of data science to biodiversity, allow me to let you in on a little secret: data science is a little bit of a misnomer.
To understand why, it helps to think of science (the application of the scientific method, that is) as cooking. There are general techniques one must master, and specific steps and cultural specifics, and there is a final product. When writing this preface, I turned to my shelf of cookbooks, and picked my two favorites: Robuchon’s The Complete Robuchon (a no-nonsense list of hundreds of recipes with no place for improvisation), and Bianco’s Pizza, Pasta, and Other Food I Like (a short volume with very few pizza and pasta, and wonderful discussions about the importance of humility, creativity, and generosity). Data science, if it were cooking, would feel a lot like the second. Deviation from the rules (they are mostly recommendations, in fact) is often justifiable if you feel like it. But this improvisation requires good skills, a clear mental map of the problem, and a library of patterns that you can draw from.
This book will not get you here. But it will speed up the process, by framing the practice of data science as a natural way to conduct research on biodiversity.