Fig 1.
Conceptual model of Tree Crown Health (TCH).
(A) Visible damage, as illustrated in the aerial photo on the left, is typically captured by red and gray crowns, whereas healthy crowns appear green, and crown shadows appear dark/black. (B) Photo interpreters perceive these crown color classes and collect training data in red, green, blue (RGB) color space. (C) TCH models are trained after converting RGB to hue, saturation, value (HSV) color space. (D) The classification model is projected back onto the image, where orange = red model class, light gray = gray model class, dark green = green model class, black = shadow or dark object model class, and white = background or areas unclassified by the model. Credits: Photo (A) courtesy of Alaska Division of Forestry; HSV diagram (C) from SharkD, CC BY-SA 3.0 <https://creativecommons.org/licenses/by-sa/3.0>, via Wikimedia Commons.
Fig 2.
Tree Crown Health (TCH) model classifications in hue, saturation, value (HSV) color space.
Plots in top row show color in relation to hue (x-axis), saturation (y-axis), and three levels of value, corresponding to the three columns of plots (V = 0.12 [left column], 0.35 [middle column], 0.75 [right column]). Plots in rows two through five (bottom row) show how HSV color space is classified using the TCH model equations and different sets of model constants, shown to the left of each row of plots (orange = red model class [red tree crowns], light gray = gray model class [gray tree crowns], dark green = green model class [green tree crowns], black = shadow or dark object model class [crown shadows], and white = background or areas unclassified by the model [i.e., X ≤ t, in Eq (8), where t = 0.1]). The left and middle columns of plots have fewer points because, at lower values, hue and saturation are compressed into smaller portions of HSV space. The bottom row of plots corresponds to the median optimal model constants across all NAIP state x year combinations, and thus is most representative of the best fit models.
Fig 3.
Mortality in a mixed stand of northern pin oak (Quercus ellipsoidalis E. J. Hill) and northern red oak (Q. rubra L.) caused by oak wilt (Bretziella fagacearum).
(A) NAIP imagery from 2019. (B) TCH model results (orange = red crowns, light gray = gray crowns, dark green = green crowns, black = shadowed crowns). Location: North of Minneapolis, Minnesota. Insect and Disease Survey database DAMAGE_AREA_ID: {e3260d8f-6e10-4eb1-8e51-23e4a3105d7b}. NAIP imagery (public domain) provided by USDA-FSA-APFO.
Fig 4.
Mortality in a mixed stand of white ash (Fraxinus americana L.), green ash (F. pennsylvanica Marsh), and black ash (F. nigra Marsh) caused by emerald ash borer (Agrilus planipennis).
Left panels are NAIP imagery from 2016 (A) and 2012 (C). Right panels (B, D) are the corresponding TCH model results from each year (light gray = gray crowns, dark green = green crowns, black = shadowed crowns). Note: None of the red crown color class existed in this area. Location: Eastern Michigan, near Lake Huron. Insect and Disease Survey database DAMAGE_AREA_ID: {7a2df48e-6e22-45bd-b9cf-b0e8d5a28536}. NAIP imagery (public domain) provided by USDA-FSA-APFO.
Fig 5.
Mortality of tanoak (Notholithocarpus densiflorus) caused by sudden oak death (Phytophthora ramorum).
Left panels are NAIP imagery from 2018 (A) and 2012 (C). Right panels (B, D) are the corresponding TCH model results from each year (orange = red crowns, light gray = gray crowns, dark green = green crowns, black = shadowed crowns). Location: Northern California. Insect and Disease Survey database DAMAGE_AREA_ID: {015c6d5a-ee89-4f6c-a3ea-7799b6401bfc}. NAIP imagery (public domain) provided by USDA-FSA-APFO.
Fig 6.
Mortality of tamarack (Larix laricina) caused by Eastern larch beetle (Dendroctonus simplex).
Left panels are NAIP imagery from 2019 (A), 2017 (C), and 2013 (E). Right panels (B, D, F) are the corresponding TCH model results from each year (orange = red crowns, light gray = gray crowns, dark green = green crowns, black = shadowed crowns). Note: The red crown color class is difficult to see at this map scale, but it mostly exists in panel D. Location: Northern Minnesota. Insect and Disease Survey database DAMAGE_AREA_ID: {83accf7a-04b6-4f34-8e8e-4ab2770b42df}. NAIP imagery (public domain) provided by USDA-FSA-APFO.
Fig 7.
Mortality of ponderosa pine (Pinus ponderosa) caused by western pine beetle (Dendroctonus brevicomis).
Left panel is NAIP imagery from 2016 (A). Right panel (B) shows the damaged tree crowns as points on top of the 2016 NAIP imagery (orange = red crowns, light gray = gray crowns). Location: Sequoia National Forest, California. Insect and Disease Survey database DAMAGE_AREA_ID: 20165031067. NAIP imagery (public domain) provided by USDA-FSA-APFO.
Fig 8.
Mortality in a mixed stand of mostly loblolly pine (Pinus taeda L.) caused by southern pine beetle (Dendroctonus frontalis).
Left panels are NAIP imagery from 2019 (A), 2017 (C), and 2015 (E). Right panels (B, D, F) are the corresponding TCH model results from each year (orange = red crowns, light gray = gray crowns, dark green = green crowns, black = shadowed crowns). Location: William B. Bankhead National Forest in northern Alabama. Insect and Disease Survey database DAMAGE_AREA_ID: {d9b4b4b9-855f-4a9f-9d87-87b7c95c211e}. NAIP imagery (public domain) provided by USDA-FSA-APFO.
Fig 9.
A) shadows (South Dakota 2018), B) absence (non-treed areas, Minnesota 2019), C) green healthy crowns (Oklahoma 2019), D) red damaged crowns (Alabama 2017), and E) gray damaged crowns (Virginia 2012). The chips are 59 x 59 pixels in size, and the green points in the center of each chip are the NAIP digitized points collected using the NPD tool in Google Earth Engine, for each respective color class. Red outlines show the results of the algorithm used to “grow” points into patches, with patch labels determined from the color classes of the digitized points. NAIP imagery (public domain) provided by USDA-FSA-APFO.