These findings qualitatively support the findings of Tu et al. (2020) who found that urban expansion primarily consumed agricultural land. However, unlike in China where urbanization was also the primary driver of cropland loss, in Chicago urbanization is only one of many drivers of cropland loss. The other drivers of cropland loss include conversion to wetland, forest, and grassland. Currently, I cannot conclude that urbanization is the primary driver of cropland loss in Chicago, but is a significant contributor to cropland loss. More research may show that the underlying drivers of cropland becoming forest, wetland, or grassland in Chicago are also related to urbanization.

Limitations

This project uses land cover data to approximate land use, but land cover is not the same as land use and the two do not always line up. The amount of error going from land cover to land use depends on the specific characteristics of the study site and is not known for Chicago. The effects of this error can be seen in the apparent transition from developed to forested land being much more common than I would have expected (Table 4).

Another limitation is the resolution of the data. The 2016 data set is provided at a much higher resolution than the 1992 data set (30m vs 1km). In the analysis I had to use the coarsest resolution (1km) which is rather coarse for the kind of specific, detailed land use change I wanted to map. Similarly, there is error mapping finer resolution data onto coarser resolution data because each raster cell is given one value regardless of the underlying variation in the landscape. This is most evident when comparing the shoreline of Lake Michigan between the 1992 and 2016 data.

There is also error going from the satellite imagery to land cover, regardless of the resolution of the data set. For example, the NLCD data set is known to poorly capture tree canopy cover out side of urban areas (Kim, 2015). Satellite imagery is more practical to use, but limited in how accurately it can describe the landscape.

Finally, my decision to simplify the land classes into broad categories is useful for analytical purposes, but makes the results far less descriptive of reality. For example, the distinctions between developed land classes are important. A suburb with tall trees, houses with urban gardens, and a city downtown were all grouped together even though they function very differently ecologically, socially, and economically.