By Jason Curtis Droboth
March 5, 2020
A short paper for GLGY 699: Philosophy of Geoscience
Over the last few years I’ve become an avid gardener. I mostly grow plants that I can eat - like kale, beans, and lettuce - but I also play around with exotic plants - coffee, bananas, kiwis, pomegranates, lemons, and oranges - that may not yield a harvest for years, if ever. Over the years, thanks to lack of experience and knowledge, I’ve killed many plants. Yet, I’m always learning more and improving my skills. I’m not out to kill the plants! I’ve noticed over the years that there are generally two different approaches I take to gardening: the experimental and the predictive. Though not mutually exclusive, they do represent important methodological differences. Let me explain.
One day I was spritzing a lemon and out popped a seed. I happened to be near a pot containing soil so I stuck it in to see what would happen. To my excitement it germinated and began growing. It grew for a couple years but then died and I’ve since started over with new seeds, determined not to kill anymore. This is the experimental approach where, through trial and error, I learn how to grow plants and how to kill them too.
In early spring I begin to plan my outdoor garden and while trial and error will still play a role, I want to minimize the risk of the whole garden dying. So, I use the predictive approach. Primarily, I look up the date of the last and first frosts along with the average light the garden will receive including the variable sun path. Based mostly off of those conditions I draw a map of the garden outlining which plants to plant, when and where to plant them, how much water they’ll need, and so on. Through this process I’ve created a model, the map, which should allow me to make better predictions about the success of my garden and thus save time, money, and heartache. This is, in essence, what Gregory V. Jones, Nicholas Snead, and Peder Nelson attempt to do in their paper Geology and Wine 8. Modeling Viticultural Landscapes: A GIS Analysis of the Terroir Potential in the Umpqua Valley of Oregon (2004).
In their study, Jones et al. (2004) try to assist existing or aspiring farmers to classify and choose the best locations in the Umpqua Valley, Oregon where they can grow the best grapes and run a successful winery. Essentially, they hope to identify ideal terroirs - a term referring to “a holistic concept that relates to both environmental and cultural factors that together influence the grape growing to wine production continuum” - by modeling climate and landscape data (Jones et al., 2004). They assume that the key to good wine are good grapes and the key to good grapes are suitable topography, soils, land zoning, and climate. Thus, using data and simple models they develop Geographic Information System (GIS) maps that outline the optimal growing locations.
In the most basic of terms, the theoretical premise of the article is essentially that topography, soils, land zoning, and climate - the central components of terroir - are the essential influences on grape growth, which, is the key determinant of good wine. They make this theory accessible to the average farmer or investor by constructing a final model in the form of a digital map that essentially says: if you plant your grapes here, it’s highly probable that your wine will be excellent! The distinction between these two, the theory and the model, are not easy to explain. A proponent of the syntactic view of theories might say that models are nothing but allegories, mental crutches that help a material human understand an abstract idea. But, a proponent of the semantic view of theories would say quite the opposite. In fact, theories are not sets of true or false propositions but sets of models and these models are, according to Nancy Cartwright, autonomous agents that mediate between theory and world (Psillos, 2007). Or as Gimbel says, “[they] are a legitimate part of the full meaning of the theory” (Gimbel, 2011). The word ‘model’ or ‘modeling’ occurs more than a dozen times in the Jones et al. study. Clearly, models play a central role in this study, so let’s dawn our semantic goggles and see how models function in their scientific method.
The authors appear to apply models in many different ways, they use data that was produced from existing models, they use existing models, they create new models or tweak models, and they combine models to create new overarching models. But are they referring to the same type of thing every time they use the term model? Model is not a new, confusing, or complicated term in popular knowledge. Ever heard of America’s Next Top Model? In that case, it’s referring to fashion models. Does that even hold the same sentiment in fashion as it does in the philosophy of science? Let’s consider what the term means in the fashion industry. A common critique of fashion models is that they cause body image issues because they don’t represent what real people look like. They accentuate certain desirable features that represent an ideal of reality, not reality itself. Positive or not, it is indeed a useful abstraction (the fashion industry is big business).
This use is somewhat similar in the philosophy of science in that a model is a representation of reality, not reality itself, and it exists to do something. It’s important to note, however, that there is little agreement within the philosophy of science and even the semantic view of theories as to what models really are. Max Black for example described at least three different types of models: scale, analog, and mathematical (Gimbel, 2011). Patrick Suppes saw models as “a structure that makes a theory true” (Psillos, 2007). Cartwright, mentioned earlier, saw models as “devices employed whenever a mathematical theory is applied to reality” (Psillos, 2007). Though Black distinguished between mathematical, analog, and scale models, it seems that even the analog and scale models only maintain representational meaning to the world in terms of mathematics. A map is only useful, and a model fire truck is only accurate in terms of its mathematical scale! Moving forward then, I’ll use Cartwright’s conception of models in that they are essentially structures that mediate between theory and the world and do so largely through mathematical language.
The primary output of the Jones et al. study were maps. Maps are, in my opinion, the ideal type of model to help illustrate what a model actually is. The paper contains 6 maps of the same geographical area but representing different types of data: topographic suitability, soil suitability, land-use suitability, composite suitability, climate maturity, and composite landscape suitability. The composite landscape suitability map essentially represents the averaging of all the geographic factors that affect grape growth. This is the map that says if you set up your vineyard here, you’ll succeed! But, as the authors state, this is not all inclusive. One should check, for instance, each site’s access to a well or municipal water supplies. To gain a better understanding of how the semantic view of theories might apply to this study, let’s focus on how they determined the soil and climate suitability.
The authors deemed soil characterization as essential to their work and obtained soil data from the State Soil Geographic (STATSGO) database, focusing on four main soil properties: drainage, depth to bedrock, available water-holding capacity, and pH. Here they’re using open data which has likely already been influenced by other models. They determined that “depth to bedrock gives an indication of how well vines can cope with dry periods, with a minimum of 30-40 inches generally needed”. Using the STATSGO data, they obtained bedrock depth weighted averages for each map unit. The scale or resolution of these map units is unclear; however, the authors somehow convert this into a 9-acre scale. They then created 3 different classes in which they would group the values (lower threshold is 25-inch depth, upper threshold is 65-inch depth) and gave this soil property a weight of 20% in relation to the other soil properties. They deemed drainage to be twice as important, at 40% of the total soil suitability. In this way they created a new model. An abstraction, not reality. Is it impossible for grapes to grow in soil with a depth less than 25 inches? Not necessarily. Is drainage always exactly twice as important as soil depth? Why not 1.999999 times more important or 1.5438764958? These aren’t really natural cut offs or laws, but limitations that the authors built into the model for general ease of use and piece of mind.
With a look into the climate suitability factor we begin noticing some issues or limitations with the semantic view of theories. The following excerpt highlights a few:
“this analysis uses the PRISM (Parameter-elevation Regressions on Independent Slopes Model) model, which is derived from a combination of point data, a digital elevation model, and other spatial data sets to create estimates of monthly and annual climate variables that are gridded at roughly 1.45 miles resolution” (Jones et al., 2004).
First, the climate suitability model that the authors create is built on the PRISM model which is itself built on data and other models. To me, this actually supports the semantic view that models are at the center of science. However, it certainly doesn’t make it any easier to understand how the model works or to assess its accuracy since one must understand the models on which the model is based and the model on which that is based and so on. There may be a fix to this though. One way to assess the accuracy of a model is through the following hypotheses: if the predictions of the model come true, then maybe the model is true or at least accurate and useful. Formally, the hypothesis states that “the physical system X is, or is very close to, M – where M is the abstract entity described by the model” (Psillos, 2007). So, if the model accurately predicts the actual behaviour of the physical system, then we might say that the model is accurate. If growers plant grapes where the map tells them to, or I plant kale where my map recommends, and the yields seem to be better than expected, the model is accurate! Second, the authors talk about their roughly 1.45-mile resolution. From my experience growing I know a little bit about microclimates. Essentially, the area beside the wall of my house is a little bit warmer than the average of my neighbourhood. For my purposes, 50 centimeters is a useful resolution, not 1.45 miles. Context matters! In this case, the authors state that the resolution was limited by the availability of the data. However, to a new farmer on untested land this resolution may prove to be useless. In fact, the authors actually provide this disclaimer. A model then, is a representation of reality that is constructed by someone with fundamental limitations and biases. So, while hypotheses allow us to test the accuracy of the model in relation to reality, is it only subjective and non-structured factors that regulate the original construction of the models?
What elements in these researcher’s methodology does a semantic view of theories not account for? To me, it’s motive, values, norms, and goals. The semantic view is capable of explaining how Jones et al. were able to describe and predict the world in terms of the centrality of the creation of various models. However, these models were designed to perform a function in support of achieving a goal. I think this is key! A map is only useful if it is, well, useful. By that I mean that the creator of a map already has a goal in mind which fundamentally influences the design of the map so that it helps achieve the goal. In fact, building a map is impossible without having a goal in mind, or at least, without determining the important characteristics to include and discriminating others based on pre-existing motives, values, norms, and limitations. Why did the authors not consider the likely geographic spread and severity of pests or weeds as key controls on grape growth? Maybe they did but deemed it to be out of scope, or maybe it never crossed their minds. Why did they focus on wine not beer? Maybe they’re avid wine drinkers but don’t like beer, or maybe they could not get access to the data that describe wheat and barley growth. Why are they assuming that producing more wine is a positive venture? Maybe they, like many, are less likely to associate wine with destructive binge drinking and its societal harm.
The semantic view of theories does not help to explain which factors, observations, limits, or variables can and cannot enter into a model. Does the theory define how the models are built? According to the semantic view, theories are based models. So how can the theory structure the models if the models first need to provide structure to the theories? To me, I’m looking for a starting point. What comes first the theory or the model? Does the physical world construct the model? Or is it cultural pressure, scientific institutions, personal bias, or restrictions of time and resources? By deciding which factors to include, how to rank them, and which factors to exclude, Jones et al. are making value judgements which fundamentally alter the models and thus the results. This may be more of a sociological critique, but it’s an important one (Allchin, 2014).
When I plant my garden this year, I hope that my map will help. Does it matter how strait the lines are or whether it’s hand drawn or digital? Did I choose the right environmental factors, sunlight and sun path? Is it really accurate? If you find me eating heaps of fresh kale each week, then you’ll know!
Allchin, D. (2014). Teaching the Nature of Science: Perspectives & Resource. Science Education, 98(6), 1111–1113. https://doi.org/10.1002/sce.21131
Gimbel, S. (2011). Exploring the scientific method : cases and questions. The University of Chicago Press.
Jones, G. V., Snead, N., & Nelson, P. (2004). Landscapes : A GIS Analysis of the Terroir. Geology and Wine 8. Modeling Viticultural Landscapes: A GIS Analysis of the Terroir Potential in the Umpqua Valley of Oregon, 31(4) https://journals.lib.unb.ca/index.php/GC/article/view/2779.
Psillos, S. (2007). Philosophy of science A-Z. Edinburgh University Press.