Landscape reconstruction and analysis of historical structure and disturbances in dry western forests using General Land Office survey data

Management of forest ecosystems, such as dry forests of the western U.S., can be enhanced by knowledge of the historical spatial structure across large landscapes. Historical information about forest structure before widespread Euro-American land uses is often limited to a few small remnant patches, but one source, the General Land Office (GLO) survey, has potential for systematic information across large land areas. The surveys, largely completed in the mid- to late-1800s in the West, recorded information on trees at surveyed corners, including tree species, size and distance from the corner. However, there are limitations to using GLO survey data that need to be addressed before widespread reconstructions can be performed. Limitations include potential selection bias, unknown accuracy of surveyor measurements, potential recording errors, and unknown accuracy of survey reconstruction methods.Bias and accuracy of survey methods were assessed in three landscapes: the Mogollon Plateau, Arizona, the Front Range, Colorado, and the Blue Mountains, Oregon. A fourth landscape, Black Mesa, Arizona, was added for reconstructions. Accuracy of survey reconstruction methods was assessed by comparing modern survey data with plots (considered truth) at the same location in a sampling grid replicating the sampling intensity of the surveys. Surveyor selection bias, measurement accuracy and recording errors were assessed by directly observing the location of historical trees at survey corners, remeasuring surveyor measurements and reconstructing tree diameters by extracting tree cores. I found that surveyors selected the closest tree to the corner 95 to 98% of the time thus, selection bias was low. Most trees selected against were <30 cm diameter at breast height. When trees were selected with bias, the mean tree distance increased by 4.2 m but diameter was statistically unchanged and species changed only 23% of the time. Bearing tree measurements appear to have been made with survey instruments. Bearing was measured to within 5 degrees of the truth and distance was measured to within 2 to 4 links (0.4 to 0.8 m, 1 link is the resolution of measurement) of the truth. Diameter was visually estimated by surveyors and within 7 to 14 cm of the truth most estimated diameter at stump height. Recording errors were more common but most do not affect reconstruction calculations. Omission errors, where less than the required number of trees are recorded, were the most serious error identified (0 to 15%), but can be corrected.New estimators were developed for reconstructions that utilized the mean Voronoi area of a tree. Data at any one corner were insufficient to reconstruct forest parameters pooling was necessary to increase accuracy. New Voronoi-based estimators were superior to most existing estimators, and yielded accurate estimates at smaller pooling levels. The best Voronoi estimators were within 14 to 23% of the truth at the 6-corner pooling level there was little gain in accuracy at higher pooling levels. Basal area required pooling to the 9-corner level and was accurate to within 23% of the truth. Composition also required a 9-corner pool to obtain accuracy there was a mean 91% similarity in the estimated and true distributions. For diameter-class distribution there was an 88% similarity in estimated and true distributions at the 12-corner pool. Missing trees had modest impacts on density and basal area calculations but little impact on composition and diameter-class distributions. A correction for missing trees is recommended for estimating density and basal area. Historical survey data and tree-ring reconstructions were also compared. Density and basal area accuracy were better than with modern data estimates but composition and diameter-class distribution accuracy were poorer.Landscape reconstructions of four dry western forests using the new methods with survey data showed structural variability not previously revealed by tree-ring reconstructions. There were large patches of both low and high tree density across landscapes. Tree density varied from 20 to over 1000 trees/ha mean tree diameter averaged 22.4 cm to 46.5 cm. A developed understory of small trees and shrubs was common in some areas and rare in others. The dominant paradigm suggested that most forests were park-like and low density, composed of large, old trees a structure maintained by low-severity fire. Using structural clues, I estimated that 62% of the Mogollon Plateau, 40% of the Blue Mountains, 12% of the Black Mesa and 3% of the Front Range fit the low-severity fire model. Thus, only one landscape was dominated by low-severity fire, the others showed more evidence of mixed- or high-severity fires. These new estimates of fire severities in historical landscapes suggest that some modern large fires within dry western forests are not anomalous in severity. GLO-based landscape reconstructions can provide land stewards with spatially comprehensive and accurate information upon which to make management decisions concerning forest restoration prescriptions, watershed protection, and fire/fuels management.