Farmland loss and land development: More on the misuse of statistics from the Census of Agriculture

During last week’s hearing for the Oregon Senate Committee on Natural Resources and Wildfire, Kelly Howsley-Glover (President, The Association of County Planning Directors) cited some of my blog posts concerning how to interpret the decrease in Oregon’s farmland reported in the 2022 USDA Census of Agriculture. The 2022 Census numbers show a decrease in farmland of 667 thousand acres between 2017 and 2022. On the surface, this seems like an alarming change, as it represents a loss of 4.18% of the 2017 farmland area. Although this most states lost farmland since the prior Census, the rate of farmland loss for Oregon is about double the national rate. Upon its release, the national Census statistics were cited as a cause for concern by some of the most influential people in US farm policy, including former Ag Secretary Tom Vilsack and American Farm Bureau Foundation President Zippy Duvall.

To be sure, a 4% loss in farmland over a five-year period is an alarming change. Farmland loss tends to be taken as a synonym for farmland conversion, where agricultural land is irreversibly developed for things like housing. It’s easy to read too much into these trends, and the widespread media coverage they receive doesn’t help, but the fact is that the loss of farmland accounted for in the Census is not the same as that land actually being lost forever to development. Instead, it’s a statistical loss pertaining to the land that’s tracked by the Census, which has been subject to declining response rates over the past decade. More importantly, it doesn’t match the pattern revealed in any other data source I’m aware of. On a more practical level, Oregon has a severe shortage of housing, which seems hard to square with the idea that we’ve been losing farmland at a rate of over 100 thousand acres per year.

Figure 1 shows the total amount of new land developed in Oregon in five-year increments from several different data sources. Although each measure comes from an entirely different source, they collectively paint a picture that should put to rest any notion that the Census farmland trend is indicative of land development. Below the figure I explain how each source is measured, along with its high-level pros and cons for this type of analysis.

First, note that I put Oregon’s Census-reported farmland loss (in thousands of acres) in parentheses under the years on the x-axis. Large farmland losses have been reported for Oregon in each of the past five Census years. The loss of 667,000 acres in 2022 is actually the second-highest loss reported over the past 25 years, with 681,000 acres in 2007 being the largest. In all but a few cases, the total amount of land developed in Oregon is less than 20% of the of the reported farmland loss. When we account for the fact that most land development in Oregon occurs on forestland, the fraction of farmland lost to development would be even smaller.

• NRI (National Resources Inventory): The NRI is conducted by the USDA’s Natural Resources Conservation Service in cooperation with Iowa State University. It is based on a stratified sample of over 300,000 land segments (representing over 800,000 individual sample points) throughout the United States. Land use information comes from a combination of detailed satellite and aerial imagery and local knowledge from county NRCS and Farm Service Agency offices. Importantly, the NRI is designed to capture land use (as opposed to land cover) and has tracked the same plots of land since 1982.  I’ve shown this elsewhere, but the NRI indicates a steady decline in land conversion throughout Oregon since 1997. The 2022 version is supposed to be released in Fall of 2025.

• NLCD (National Land Cover Database): The NLCD is a produced by the U.S. Geological Survey’s Earth Resources Observation and Science Center. USGS personnel use NASA’s Landsat (Land Satellite) images to develop annual maps of the US divided into 30m-resolution pixels (about 0.22 acres in size). As the name indicates, the NLCD tracks land cover, which does not always correspond with how land is used. For example, there may be houses that are obscured by surrounding trees and thus won’t be picked up in the satellite image. Similarly, forests that are harvested will appear from space as grassland or shrubland, which obviously misses how the land is actually used. The general trend shown in in the figure is similar in direction to the NRI, but less extreme and with a higher total area of development.

• HISDAC-US (Historical Settlement Data Compilation): HISDAC-US is a relatively new database developed a few years ago by geographers at the University of Colorado. It divides the US into a 250m grid (about 15 acres per grid cell) and then consults local county assessment records (and other supplemental records) for the year when individual buildings were constructed across the US. The HISDAC-US data can also be thought of as representing land use, as they represent actual buildings that have been constructed. A downside is that not all buildings have recorded construction years, especially older ones. This limitation should be minimized by focusing on a relatively recent period, but it is worth keeping in mind. In 2020, the most recent year available, about 57% of all buildings in Oregon have construction year information.

The plot is showing the HISDAC built-up area (BUA), which is the area of Oregon covered by 15-acre grid cells where at least one building is present, so if there’s one building in the cell (even if it’s a barn), the entire 15 acres is considered developed, which might exaggerate the total amount of development in more rural, low-density areas. That likely contributes to the large amounts of development shown earlier in the figure, which still represent at most 25% of Census-reported farmland loss, but the rate falls off dramatically in more recent periods.

Despite the large differences in how total land development is measured across these three sources, which again includes all land developed (not just farmland), they collectively suggest that farmland loss reported in the Ag Census should not be confused with land development.

So, what did happen to farmland between 2017 and 2022? Ideally, we could look to something like the NRI, but that isn’t currently available for 2022. Between 2012 and 2017 (Figure 2), when the Census showed Oregon losing 339 thousand acres of farmland, the NRI shows that most of the actual land-use change for agricultural land was due to moving between cropland, pasture, and range, with about 0.4% (63 thousand acres) being converted out of these three categories. Most of the agricultural land that changed use went into forest or “other” rural land, like farmsteads and farm buildings, and about 11 thousand acres were developed.

If we look at the 2017-22 change in the satellite-based NLCD land cover data (Figure 3), 95% of the 2017 cropland remained that way in 2022, as did 91% of pasture/hay. Most of the land-cover changes for two categories were movements to and from each other or movements into grassland or shrubland. Agricultural land, especially in eastern Oregon, is sometimes misclassified in the NLCD as grassland or shrubland. These categories also retained 88 and 92% of their 2017 area, with most of the remainder representing changes across the two categories or, in the case of shrubland, movements into forest cover.

As I’ve said before, the Census is a valuable resource when it is used correctly. The statistics that come out of the Census are cited widely and can inform policymaking at all levels. It provides important information on all sorts of things about farms and ranches throughout the US, but it is not a definitive source of land use information. The main issue is that it does not provide repeated information on the same farms over time, which contributes to misleadingly large statistics concerning farmland loss. To that end, it would be helpful if the USDA were to provide supplemental information on the same farms that responded to the previous Census. This would give both policymakers and the general public some context on how to square the data with reality.