Fig 1.
Summary of data processing, cleaning and analysis methods for the 2005 and 2015 datasets.
The number of trees used at each step is shown for before and after alignment.
Table 1.
Parameters used in this paper to better understand the controls on tree growth.
The numbers next to each name indicate the source of the data.
Fig 2.
Histograms of tree growth from the different datasets used in this study.
If a tree had three or more points the Kendall tau is included to show the goodness of fit for all trees. A +1 indicates a monotonic increase in DBH when growth rates are calculated and a -1 indicates a monotonic decrease. Bins of 0.1 in/yr were chosen as this is the smallest possible increment for the NYC Tree census data. (A) Tree growth in trees from the NYC Tree censuses. (B) NYC ground-truthed validation, growth rates of 86 locations with tree records in both the 2005 and 2015 datasets that were revisited and DBH measured. (C) NYC temporal validation, growth rates of 44 trees in a four-block zone in the 10027 ZIP Code measured approximately annually between 2015 and 2022.
Table 2.
Summary of the newly created NYC tree growth dataset.
The growth dataset grouped by species is compared to the top trees in the 2015–2016 TreesCount! Street Tree Census (termed 2015 data) as taken from NYC data [1]. The growth database is missing two trees that are in the top 15 of the 2015 data: Cherry (rank = 7) and Sophora (rank = 10).
Fig 3.
Diameter at breast height (DBH) versus time for select trees from the different studies to highlight the variety of responses observed.
Orange points are from the NYC Tree censuses and blue points were measured in ground truthing and temporal validation. The color of the best fit lines represents the points they were fit to. A-C) represent trees that were remeasured in ground truthing. D-F) represent trees that were measured multiple times in temporal validation of growth rates. The points and darker line represent a typical example tree and the lighter lines show the remainder of the data for each type. (A) Trees with positive growth between the 2005 and 2015 NYC datasets; representative of the majority of cases. (B) Trees with a DBH decrease of less than -5 in between the 2005 and 2015 NYC datasets. (C) Trees with a DBH decrease of between 0 and -5 in for 2005 and 2015 NYC datasets. (D) A typical tree that was monitored annually and has linear growth along with all other samples. (E) The only tree for which we have two NYC Tree census points and annual data points. It is not clear why only one tree had 2005 data. (F) A tree that was measured annually but did not grow highlighting that zero growth is possible.
Fig 4.
Maps of NYC tree data at the ZIP Code level.
The large map shows mean growth rates whereas the two insets show the number of trees per km2 and the mean diameter at breast height (DBH). The data is only for the trees in the NYC Tree Growth dataset. The five boroughs of New York City are labeled and outlined in thicker black for reference for subsequent maps. The shapefiles are available from the NYC Open Data Portal (https://opendata.cityofnewyork.us/).
Fig 5.
Forest plot to summarize impact of categorical data (Table 1) on growth rates.
The vertical gray line is the overall mean of 0.275 in/yr. Each horizontal line goes from the 25th to 75th percentile with the point being the mean. The letters on the right are the statistically significant groups as determined with an ANOVA and a Tukey post-hoc test within that subgroup if there are three or more parameters or a t-test when two parameters. Significant differences are determined based on p<0.05.
Table 3.
Pearson correlation coefficients.
The data is between growth rates from the NYC growth database and continuous data (Table 1) that may impact growth rates.
Fig 6.
Percentage of trees measured by volunteers, TreesCount paid staff and NYC parks staff for each borough.
Fig 7.
Ordinary least squares regression of factors controlling growth rate data.
The parameters are shown by Effect Estimate on Tree Growth (in/yr). If the Effect Estimate intersects zero, it is not significant and grayed; if the Effect Estimate is significant at the 0.05 level, it is shown in Black. Analysis includes the 15 most common species in the growth database (shown here and Table 2). All other species were grouped into ‘other’ and used as the reference. The adjusted R2 was 0.252 with a root mean square error (RMSE) of 0.202 in/yr.
Fig 8.
Residuals between observed and predicted growth rates averaged by ZIP code.
The Predicted growth rates were based on ordinary least squares (OLS) regression excluding borough. Positive residual values indicate faster than expected growth rates and negative values indicate slower than expected growth rates. The inset shows the CDC/ATSDR Social Vulnerability Index (SVI) with higher numbers indicating more vulnerable neighborhoods. The 19 ZIP codes with less than 100 trees and ZIP codes 11224 and 11691 which were heavily flooded during Hurricane Sandy were removed leaving a total of 155 ZIP Codes. The spatial data are publicly available from the NYC Open Data Portal (https://opendata.cityofnewyork.us/ and Center for Disease Control Agency for Toxic Substances and Disease Registry Social Vulnerability Index (http://svi.cdc.gov).
Fig 9.
The CDC/ATSDR social vulnerability index (SVI) versus growth rate residuals.
The symbols are colored by growth rate residuals to match the y-axis and the colors on the main map from Fig 8.