Methods
We analyzed data collected by NWERN between 2015 and 2023 (Webb et
al., 2016), including data from thirteen plots, each with at least
five data collection events for a total of 151 sampling events
(Figure 1). The NWERN plots represent a diversity of rangeland
(CPER, Holloman, Jornada, Lordsburg, Moab, Red Hills, San Luis
Valley, Twin Valley), pastureland (El Reno), and cropland (Akron,
Mandan, Morton, Pullman) ecosystems (Table S2). Akron, CPER, El
Reno, Jornada, Morton, and Mandan are also part of the USDA’s
Long-Term Agroecosystem Research network (Kleinman et al., 2018).
Each NWERN plot represents one hectare with a meteorological tower
at the center (Webb et al., 2016). Three, 100-m transects
intersect at the meteorological tower at 60-degree intervals
(Figure 1B). Transect measurements follow Herrick et al. 2018: 1)
canopy gap intercept between all plants (annuals and perennials)
with a 5-cm minimum gap, 2) LPI measured every 0.25 m and 3)
vegetation height measured every two meters. All data collectors
were trained and calibrated by NWERN staff. Data were accessed via
the Landscape Data Commons (McCord et al., 2023).
2_2 Subsampling
All analyses were conducted in R 4.3.3 (R Core Team 2024). To
test the impacts of different sampling approaches, we assumed
that the NWERN plot sample design represented the upper limit of
feasible data collection for most management and research
projects. We subsampled the NWERN sites to create nine different
sampling scenarios of combinations of one to three transects of
25, 50, and 100 m (Figure 1B). First, we reduced transect
length, where transect lengths less than 100 m were set up to
run from the plot center towards the edge of the plot (Figure
1B). Second, we randomly reduced the number of transects from
three to two or one. Third, we reduced measurement intensity for
LPI and vegetation height by varying pin drop measurement
intensities at 0.25-, 0.5-, 1-, and 2-m intervals for LPI and
2-, 4-, 6-, 8-, and 10-m intervals for vegetation height. These
subsampling scenarios were selected as they include close
approximations of common radial monitoring plot sample designs,
such as BLM AIM (three 25-m transects sampled every 0.5 m for
LPI and every 2.5 m for height) and NRCS NRI (two 45.72-m
intersecting transects sampled every 91.44 cm for LPI and every
304.8 cm for height).
We calculated indicators for each subsampling scenario and
method using the terradactyl R package (McCord et al., 2022).
For canopy gap intercept, we calculated the percent of the plot
covered in all-plant canopy gap size classes (5-24 cm, 25-50 cm,
>50-100 cm, >100-200 cm, >200 cm). We used the LPI
observations to calculate percent total foliar cover and to
count the number of species detected. For vegetation height, we
calculated the mean plant height.
2_3 Limits of agreement analysis
We conducted a limits of agreement analysis (LoA) to determine
the effects of reducing sampling effort compared to NWERN
sampling. Limits of agreement evaluates the differences between
two measurement approaches with respect to their means (Martin
Bland & Altman, 1986). In doing so, we determined the level
of agreement between two approaches and detected bias exhibited
by any one approach. For each indicator, we used the SimplyAgree
R package (Caldwell, 2022) to conduct a nested LoA analysis that
accounted for the repeat measures within NWERN sites, as well as
the response across all sites using the MOVER method for
calculating prediction intervals (Donner & Zou, 2012; Zou,
2013) at 80% and 95% confidence levels. All sites and indicators
met the assumption of normality for the distribution of the
differences. We compared the bias and agreement intervals across
sampling approaches and to acceptable difference criteria. These
criteria match the acceptable difference limits for field crew
calibration (Herrick et al., 2018), including 1) bias was not
significantly different to zero, and 2) agreement intervals and
their prediction intervals were less than or equal to 5% for
total foliar cover and the canopy gap indicators, five cm for
vegetation height, and two species for species count. We then
identified the lowest effort sampling scenario required to meet
the acceptable criteria at both 80% and 95% confidence levels
for each of the core methods.