Computing Trees

QC 20260923

Abstract

This dissertation examines how forests have been made governable through measurement, following three monitoring infrastructures across a century: the Swedish National Forest Inventory from 1923, the satellite-based kNN forest maps developed in Sweden during the 1990s, and the Global Forest Change product built at the University of Maryland and used in REDD+ carbon accounting in the Democratic Republic of Congo. Forest monitoring has run since its beginnings on what the study calls the information promise: that one cannot govern what one cannot see, and that better governance therefore waits on better information. The dissertation asks what such infrastructures actually produce. Drawing on field manuals, method reports, technical publications, government inquiries and press debates, and on a conceptual apparatus developed from Bruno Latour and Charles Sanders Peirce, it reconstructs the specific operations through which measurements become claims capable of governing. The finding is that these infrastructures did not deliver progressively sharper pictures of an existing forest. Each produced a new kind of object for governance to work on. The Swedish inventory aggregated hundreds of thousands of field measurements into an estimate of how much the nation's forests grow in an average year. In 1979, that estimate was used to set a national harvest target, and the target was written into law. The kNN maps combined satellite images with the inventory's field plots to predict what kind of forest stood at every location in the country, so that conservation administrators could search for old and valuable stands on a screen rather than discover them in the field. The Global Forest Change product used machine-learning classification of a vast archive of satellite images to produce a global, yearly record of forest loss. In the Democratic Republic of Congo, the loss recorded over one period was projected forward as a prediction of the next, and payments for avoided deforestation were calculated against that prediction. Read in sequence, the cases show a single development. In each, the work of deciding that scattered measurements could count as measurements of the same thing was handed over to a mathematical or computational procedure: a growth function, a matching algorithm, and a classification model. With each case, that procedure operated at a greater remove from the forests it made claims about. The Swedish inventory generalized from plots its own fieldworkers had stood on. The kNN maps still depended on those plots, but used them to make claims about every location in the country, the overwhelming majority of which no fieldworker V had visited. The final case generalized from a dataset comprising a decade's worth of pixels, and the researchers who produced it had no presence in the Congo Basin, yet their figures anchored a decade of contracts and financial commitments there. The promise of digital monitoring was better information about the forest. What it delivered was something else: authoritative figures about forests that the people producing them had never set foot in.

Link to DiVA