Wed, Aug 04, 2021:On Demand
Background/Question/Methods
Post-fire recovery of plant communities is often assessed by changes over time in understory plant cover. Recently, digital image analysis (DIA) of plant cover, in which digital images are analyzed by computer software, is increasingly used as a rapid and convenient estimation method in the field. Previous studies have concluded that complex, multi-layered vegetation may reduce the visibility of understory plant species in digital imagery used for DIA, resulting in underestimated plant cover. Cover underestimation may be especially pronounced in seral forest communities recovering from fire, where tree regeneration, particularly subcanopy trees (height >137 cm), coarse downed wood (CDW), and shadows may reduce vegetation visibility, although their association with digitally-derived cover has not been previously evaluated. We estimated recent (~30 years post-fire) plant cover on permanent plots (n = 142) established in two study areas following the 1988 Yellowstone fires. We determined the differences between conifer regeneration, CDW, and shadow cover in each study area and assessed their relationship with plant cover. Additionally, we explored the effects of these differences in our ability to estimate plant cover by comparing DIA to two visual estimation methods: plot-level (20m2) estimation (PLE) and quadrat-level (1m2) estimation (QLE).
Results/Conclusions DIA confirmed major differences between study areas in conifer regeneration canopy cover (odds ratio = 8.34). Additionally, the magnitude and direction of differences between digital and visual estimates of plant cover also varied by study area. At the study area with low conifer cover, DIA estimated, on average, 9% (95% CI = 3 – 14%) and 16% (95% CI = 10 – 21%) more plant cover than PLE or QLE, respectively. However, the variability in plant cover was similar among all three methods. At the study area with high conifer cover, DIA estimated less plant cover than either PLE or QLE by 28% (95% CI = 24 – 32%) and 22% (95% CI = 18 – 26%), respectively, and had more variability. Furthermore, subcanopy conifer density was negatively associated with plant cover estimated by DIA but showed no relationship with PLE/QLE. We conclude that conifer canopy cover severely hindered the detection of plant cover by DIA. Consequently, we recommend visual estimation of plant cover in seral communities with complex vegetation structure, but digital estimation may be advantageous early in succession and in simple vegetation communities.
Results/Conclusions DIA confirmed major differences between study areas in conifer regeneration canopy cover (odds ratio = 8.34). Additionally, the magnitude and direction of differences between digital and visual estimates of plant cover also varied by study area. At the study area with low conifer cover, DIA estimated, on average, 9% (95% CI = 3 – 14%) and 16% (95% CI = 10 – 21%) more plant cover than PLE or QLE, respectively. However, the variability in plant cover was similar among all three methods. At the study area with high conifer cover, DIA estimated less plant cover than either PLE or QLE by 28% (95% CI = 24 – 32%) and 22% (95% CI = 18 – 26%), respectively, and had more variability. Furthermore, subcanopy conifer density was negatively associated with plant cover estimated by DIA but showed no relationship with PLE/QLE. We conclude that conifer canopy cover severely hindered the detection of plant cover by DIA. Consequently, we recommend visual estimation of plant cover in seral communities with complex vegetation structure, but digital estimation may be advantageous early in succession and in simple vegetation communities.