Variations in Tree Cover in North America since the Last Glacial Maximum

J. W. Williams

2003 Global and Planetary Change 35: 1-23

Abstract —  Accurate reconstructions of late-Quaternary land-cover change are needed to better understand past interactions of the terrestrial biosphere with other components of the earth system.  This paper presents a sequence of reconstructed needleleaved and broadleaved tree-cover densities in North America since the last glacial maximum, generated from fossil-pollen data and present-day tree-cover estimates derived from the Advanced Very High Resolution Radiometer (AVHRR).  For this study, a refined form of the modern analog technique was developed, called the hierarchical analog technique, which can constrain paleoenvironmental properties even for fossil-pollen assemblages without close analogs in the modern-pollen record.  Pollen taxa from samples that are compositionally unlike any modern-pollen samples are regrouped into plant functional categories based upon phenology, life form, leaf shape, and climatic tolerances, and the analog analysis rerun.  Reclassifying individual pollen taxa into broader functional categories enables analogs to be found when no compositional analogs exist, but at a cost of increased uncertainties in the analog estimates.  Tests of the hierarchical analog technique shows that it accurately reconstructs present-day tree-cover densities.  Median standard deviation for each individual estimate is <10%.  Tree-cover densities during the last glacial maximum were low relative to present, and have increased since then.  Lower-than-present tree-cover densities at the last glacial maximum were likely due to a combination of low temperatures, low precipitation, and low atmospheric CO2 concentrations.  By 14 ka, broadleaved tree-cover densities had begun to rise in the southeastern US and needleleaved forests grew in the western US, southeastern US, and as a belt along the southern margin of the Laurentide Ice Sheet.  By the mid-Holocene, the northern and western needleleaved forests had joined.  Needleleaved and broadleaved tree densities continued to increase until European settlement. Mapping percent tree-cover represents a useful alternative to biome-based classification schemes, enabling a fuller representation of vegetational gradients in space and time, and can be directly compared to the tree distributions simulated by dynamic global vegetation models.  In effect, by calibrating the modern-pollen data against the AVHRR-derived estimates of tree-cover, the fossil-pollen data are applied to extrapolate satellite-based observations into the Quaternary, enabling study of vegetation dynamics and land-cover change at timescales beyond the period of direct observation by engineered remote sensors.  Analog-based approaches, however, require extensive networks of surface and fossil-pollen samples, making further data collection a priority in sparsely sampled regions of the world. 

Figures

Figure 1:  Flowchart outlining the key steps of the hierarchical analog technique.  In the first level of analog-matching, the squared chord distances (SCD’s) between a fossil-pollen sample i and modern-pollen samples are measured, using the percent abundance of individual pollen taxa as the unit of comparison (Overpeck et al., 1985).  The tree-cover estimates associated with the ‘best’ modern analogs (criteria described in Methods) are averaged and assigned to the fossil-pollen sample.  If no good modern analogs exist, the fossil-pollen sample is subjected to a second iteration of the modern analog technique.  In the second level of analysis, the pollen taxa in the modern- and fossil-pollen samples are reaggregated to plant functional types, and dissimilarities recalculated.  In an experimental third level of analysis, the fossil- and modern-pollen samples were reclassified into simple life forms, but the high uncertainties associated with this level analysis meant that only the assignments made by the first two analyses were included in the paleovegetation reconstructions.  PDF (749 KB) 
Figure 2:  Broadleaved and needleleaved percent tree-cover in North America, as inferred from AVHRR observations collected between 1992 and 1993.  Redrawn from DeFries et al. (1999).  PDF (312 KB) 
Figure 3:  Isopoll maps showing percent pollen abundances for various plant taxa, interpolated from surface sediment samples collected across North America.  Darker grays correspond to higher percentages, and areas with no available pollen samples are white.  The contour intervals used for each map are indicated in its lower left corner.  The lower-left map shows the location of the surface samples used to construct the isopoll maps. PDF (1345 KB)
Figure 4:  Scatter plots comparing the needleleaved and broadleaved percent tree-cover observed at the surface pollen sites to the values estimated by the analog technique.  The results for each level of the hierarchical analog technique are presented separately.  PDF (1852 KB)
Figure 5:  Box plots showing the distribution of the standard deviations associated with the individual estimates of percent needleleaved and broadleaved tree-cover, for each level of the hierarchical analog technique.  The boxes mark the 25th, median, and 75th percentiles; the length of each whisker is 1.5 times the interquartile range.  The median standard deviation increases with successive levels of analysis.  PDF (9 KB)
Figure 6:  Maps of present-day tree-cover, spatially interpolated from the values assigned to individual pollen sites.  A)  Percent tree-cover for needleleaved and broadleaved trees, interpolated from the AVHRR-derived values assigned to each surface sample location.  Differences between A) and Figure 2 are due to the variable density of the pollen sample network and the smoothing produced by the spatial algorithm.  B) Percent tree-cover, as estimated using the hierarchical analog technique, levels 1 & 2.  The two sets of maps closely agree.  PDF (230 KB)
Figure 7:  Land-cover reconstructions for North America for 21, 16, 14, 11, 9, 6, and 0.5 ka. As in figure 3, increasing saturations correspond to higher tree densities, and areas left blank had insufficient data for mapping.  First row:  total tree-cover (sum of needleleaved and broadleaved tree cover densities), expressed in percentage form.  Second row:  percent needleleaved tree-cover.  Third row:  percent broadleaved tree-cover. Fourth row:  locations of the individual pollen assemblages used to create each set of maps.  Green dots indicate fossil-pollen samples that have a good compositional analog in the modern-pollen dataset.  Yellow dots indicate fossil-pollen assemblages with no compositional analog in the modern dataset, but with a good functional analog.  Red dots were not assigned values in the first two levels of analog analysis and were not used in the land-cover reconstruction.  PDF (2560 KB)