The Forest Inventory and Analysis Program (FIA) provides information about the Nation’s forests such as forest land area, tree growth, and tree mortality. FIA’s estimates are based on a sample design that is unbiased and representative of all lands, including privately owned forest lands. When a large proportion of plots in a given area cannot be measured, for example when private landowners do not grant permission to access plots on their property, the area of forest land may be underestimated--a problem is referred to as nonresponse bias.
Nonresponse bias in forest monitoring programs can result in under-estimation of forest attributes. This study presented a modified stratification scheme for New Mexico that compensated for nonresponse bias and produced more accurate forest inventory.
FIA’s annual inventory in New Mexico served as an ideal test case for compensating for nonresponse bias. From 2008-2012, nearly 12,000 forest inventory plots were sampled, and more than 1,000 of these were nonresponse plots. In New Mexico, the traditional FIA stratification underestimated forest land area by about one million acres compared to the modified stratification that compensated for nonresponse.
One unexpected consequence of accounting for high nonresponse is that strata with very low percentages of nonresponse, such as those with a majority of plots on National Forest lands, were assigned disproportionately high weights under the traditional stratification. Thus failure to address high nonresponse may lead to underestimation at the state level, as well as overestimation for specific areas with a low percentage of nonresponse.
Investigation of the impact of stratifying to compensate for nonresponse on tree-level attributes, such as growth and mortality, is ongoing. An accurate baseline of New Mexico's forests and future estimates now exists and can be compared against this baseline to detect changes over time. In other states where a large percentage of plots cannot be measured, the stratification key helps improve the accuracy of FIA's estimates by compensating for nonresponse bias. As a result, forest managers now have higher quality data.