Landscape context affects patch habitat contributions to biodiversity in agroecosystems
Data files
Apr 18, 2024 version files 989.68 MB
-
AllFarmsScaled.csv
170.28 KB
-
Cropscape_2021_big_UTM.tif
583.76 MB
-
Cropscape_2021_NAD.tif
405.74 MB
-
README.md
10.11 KB
Abstract
Effective conservation schemes are needed to advance dual objectives of biodiversity conservation and agronomic production in agricultural landscapes. Understanding how plant and arthropod taxa respond to both local habitat patch characteristics and landscape complexity is crucial for planning effective agri-environment schemes. This study investigated the relative effects of local (≤ 100 m from patch habitat center) and landscape (≤ 5 km from patch habitat center) variables on diversity of plants and arthropods within non-crop habitat patches (i) at different spatial extents ranging from 0.1 km to 5 km, while (ii) quantifying differential effects of local and landscape variables on unique components of diversity (i.e. species richness and abundance), and accounting for (iii) unique components of landscape extent (0.1, 0.5, 1, 2 and 5 km radii) and complexity (i.e. landscape composition and configuration). Landscape variables were significantly correlated with local plant and arthropod species richness and abundance at all spatial extents. Biodiversity responses to landscape variables were largely scale-dependent, as pairwise comparisons were significantly different between all spatial extents except between 1-km and 2-km extents, and correlations were lowest at the 5-km extent. Partial R-squared values for predicting local biodiversity were highest when both local and landscape variables were included as predictors of species richness and abundance, underscoring the importance of considering both local and landscape effects on local diversity. Landscape configuration variables accounted for more variation in plant and arthropod species richness than composition variables. However, models performed best when composition and configuration were considered together rather than alone, suggesting that both components of landscape complexity should be considered for identifying and managing conservation areas in crop fields. Conservation schemes that incentivize farmers to create or conserve small patch habitat within crop fields may be more effective when combined with landscape-scale designs that enhance landscape complexity across the Northern Great Plains. Local conservation efforts should be coordinated with landscape-level efforts to ultimately enhance biodiversity and desired ecosystem service outcomes across agricultural landscapes.
README: Landscape context affects patch habitat contributions to biodiversity in agroecosystems
https://doi.org/10.5061/dryad.pk0p2ngww
Datasets include local biodiversity observations obtained from field surveys and landscape data obtained from the Cropland Data Layer (USDA, 2023). Local variables included percent cover of ecological refuge (ER) and crop field, plant species richness, plant abundance (i.e. percent cover), arthropod species richness, and arthropod abundance within 100 meters from the ER center. All data from each transect were summed by taxonomic group (plants or arthropods) across the two years of data collection, and placed in the response variable matrix for analysis. Local biodiversity datasets were analyzed using the vegan package in R. Landscape variable data were obtained from the Cropland Data Layer and extracted for the five landscape spatial buffers surrounding each ER using the LandscapeMetric package in R. Landscape variables grouped under landscape composition included land cover type diversity, number of land cover types, and the percent cover of agricultural cultivated land, undeveloped land, or developed land within each spatial extent. Landscape variables grouped under landscape configuration included patch cohesion as a metric of connectedness within each spatial extent, patch division as a metric of patchiness within each spatial extent, total edge amount as a metric of fragmentation within each spatial extent, total number of patches as a metric of patchiness within each spatial extent, and Large Patch Index as a measure of the percentage of landscape covered by the dominant patch type within each spatial extent.
Description of the data and file structure
Prior to PLS analysis, local and landscape variable data were obtained and analyzed as following:
1. Obtained local biodiversity data from field surveys in 2020 and 2021
a. Combined local biodiversity data for 2020 and 2021 for all sites
b. Subsetted local biodiversity data in nested buffers around each ER (at 20, 40, 60, 80 and 100 m buffers)
c. Calculated diversity and richness within each extent using vegan package (species richness and Shannon's Diversity)
2. Obtained landscape scale data from Cropland Data Layer
a. Downloaded data from Cropland Data Layer at a 200 km radius each site
b. Created nested buffers around each ER at landscape scale (0.1, 0.5, 1, 2, and 5 km buffers)
c. Used LandscapeMetric package to calculate landscape composition and configuration metrics within each spatial extent
3. Standardized the combined local and landscape dataset to mean of 0 and variance of 1.
Local Data
Local data included plant and arthropod abundance and species richness sampled in 2020 and 2021 at 20, 40, 60, 80, 100 m buffers from ER center.
Local Biodiversity Data Files:
Corresponding shapefiles biodiversity and shapefiles for each ER (Site.Local.Long.csv and Site20m.shp, Site40m.shp) were not included as they could not be properly anonymized to respect the participating producers' privacy and comply with human subjects data regulations. Certain data may be made available upon request.
Code Example for Subsetting Local Biodiversity Data by Local Spatial Buffers:
df <- read_excel("LocalSiteData.csv")
#Subset by spatial extent
Site20m <- subset(df,df$EXTENT (m)==20)
Site20m <- Site20m[,2:8]
Site40m <- subset(df,df$EXTENT (m)==40)
Site40m <- Site40m[,2:8]
Site60m <- subset(df$EXTENT (m)==60)
Site60m <- Site60m[,2:8]
etc. for 80m and 100m extents.
Landscape Data
Cropland Data Layer Files:
Cropscape_2021_NAD.tif is a shapefile of 2021 Cropland Data Layer for all crops clipped in a 200 km radius around site 1.
Cropscape_2021_big_UTM.tif is a shapefile of 2021 Cropland Data Layer for all crops clipped in a 200 km radius around sites 2 and 3.
Method for Obtaining Cropland Data Layers:
#Defined area of interest using "square" tool (defined 200 km radius around the ER)
#Defined year of interest under "layers" tab (selected 2021)
#Selected all crops (value)
#Exported the selected crops from mapping button (downloaded as .tif)
Spatial Buffers
Method for Creating Local Spatial Buffers:
Local spatial buffers were manually created around each ER at 20, 40, 60, 80 and 100 m radii and combined into one dataframe in R.
Code Example:\
#20m Buffer Ring
#SiteLocalBuffers <- readOGR('Site20m.shp')
#Site20m <- crop(Cropscape_2021_NAD.tif, SiteLocalBuffers)
#Site20m <- mask(Site20m, SiteLocalBuffers)
#40m Buffer Ring
SiteLocalBuffers <- readOGR('Site40m.shp')
Site40m <- crop(Cropscape_2021_NAD.tif, SiteLocalBuffers)
Site40m <- mask(Site40m, SiteLocalBuffers)
Method for Creating Landscape Spatial Buffers:
Spatial buffers were manually created around each ER at 0.1, 0.5, 1, 2, and 5 km radii and combined into one dataframe in R. Shapefiles for each ER (SiteBuffer100m.shp) were not included to respect the participating producers' privacy.
Code Example:
#0.1 km buffer ring
Sitebuffers <- readOGR('SiteBuffer100m.shp')
Site0.1km <- crop(crops2021, Sitebuffers)
Site0.1km <- mask(Site0.1km, Sitebuffers)
#0.5 km buffer ring
Sitebuffers <- readOGR('SiteBuffer0.5km.shp')
Site0.5km <- crop(crops2021, Sitebuffers)
Site0.5km <- mask(Site0.5km, Sitebuffers)
Land Cover Classification:
Land cover types obtained were manually classified from the Cropland Data Layer in R as agricultural land, wild land, or developed land.
Code Example:
percentage_class_Site0.1km <- as.data.frame(percentage_class_Site0.1km)
AGLAND.0.1km <- c(1,4,5,6,12,21,22,23,24,25,27,28,29,30,31,32,33,35,36,37,38,39,41,42,43,44,47,51,52,53,57,58,59,61,66,67,71,205,224,228)
WILDLAND.0.1km <- c(111, 112, 131, 141, 142, 143, 152, 176, 190, 195)
DEVLAND.0.1km <- c(121, 122, 123, 124)
The vegan package was used to calculate land cover type species richness and diversity at each extent.
Code Example:
#made dataframe with columns: Spatial Extent, Species Richness, Shannon Diversity
df.Site0.1km <- cbind.data.frame(0.1, SRSite0.1km, HSite0.1km)
names(df.Site0.1km) <- c('radius.km', 'SR', 'ShanDiv')
df.Site0.5km <- cbind.data.frame(0.5, SRSite0.5km, HSite0.5km)
names(df.Site0.5km) <- c('radius.km', 'SR', 'ShanDiv')
etc. for 1, 2 and 5km extents
The LandscapeMetric package was used to calculate configurational landscape variables at each extent.
Code Example for variable COHESION:
lsm_l_cohesion(Site0.1km, directions = 8)
lsm_l_cohesion(Site0.5km, directions = 8)
lsm_l_cohesion(Site1km, directions = 8)
lsm_l_cohesion(Site2km, directions = 8)
lsm_l_cohesion(Site5km, directions = 8)
Standardizing the Dataset
The local and landscape variables were combined in a dataset and scaled and logged to a mean of 0 and variance of 1.
Code Example:
#Standardized local and landscape dataframes
#Test <- as.data.frame(SiteLanddf[,3:18])
#SiteLanddfLogged <- log(as.data.frame(Test))
#SiteLanddfLoggedScaled <- as.data.frame(scale(SiteLanddfLogged))
#SiteLanddfScaled <- cbind(SiteLanddf[,1:2], SiteLanddfLoggedScaled)
Standardized Data File:
AllFarmsScaled.csv is the combined dataset for all scaled local and landscape variables for all sites.
Variable Description:
PLANTSR: Mean plant species richness, or number of plant species recorded within each spatial extent
PLANTABUND: Plant abundance, or percent cover of non-crop vegetation within each spatial extent including weeds in the cropped portion
FIELD: Percent cover of each spatial extent that was cultivated for crops
PATCH: Percent cover of each spatial extent that was semi-natural non-crop patch habitat
ARTHSR: Mean arthropod morphospecies richness, or mean arthropod morphospecies richness recorded within each spatial extent
ARTHABUND: Mean arthropod abundance, or number of arthropod individuals recorded within each spatial extent
LANDDIV: Land cover type diversity within each spatial extent, measured by Shannon Diversity Index
LANDRICH: Land cover type richness, or number of land cover types within each spatial extent
WILDLAND: Proportion of each spatial extent that was wildland
AGLAND: Proportion of each spatial extent that was cultivated agricultural land
COHESION: An aggregation metric that characterizes the connectedness of patches belonging to class; i.e. COHESION = 0 was isolated and COHESION = 100 was well-connected
DIVISION: An aggregation metric that characterizes the probability that two cells are not located in the same patch class; i.e. DIVISION = 0 if only 1 patch was present and DIVISION = 1 if all patches were single cells
TOTALEDGE: An area and edge metric that measures all edges in meters; i.e. TE = 0 if all cells were edge cells; TE increases as landscape becomes more fragmented
NP: An aggregation metric that measures the number of patches; i.e. NP = total number of patches in a landscape; i.e. NP increases as landscape becomes more fragmented
LPI: A dominance metric that measures the percentage of the landscape covered by the largest patch in the landscape; i.e. LPI = 100 when one patch covers the entire landscape
Farm: An ID variable that identifies each site in the dataset i.e. Farm 1, Farm 2, and Farm 3
EXTENT (km): Spatial extent of each spatial buffer; i.e. 0.1 km, 0.5 km, 1km, 2km, and 5km
Sharing/Access information
Cropland data were derived from the following sources:
Code/Software
PLS analysis were conducted in R. Code may be made available upon request.
Code Example:
#Added plsrLocal (local variables at 0.1 km) and plsr01 (landscape variables at 0.1 km)
plsrLocal <- plsr(ARTHSR ~ PLANTSR + PLANTABUND + FIELD + PATCH, ncomp = 4, data = Site05, validation = "LOO")
plsr01 <- plsr(ARTHSR ~ LANDDIV + LANDRICH + WILDLAND + AGLAND + DEVLAND + COHESION + DIVISION + TOTALEDGE + NP + LPI, ncomp = 4, data = Site05, validation = "LOO")
etc. for each variable combination and spatial extent at 0.5, 1, 2, and 5 km.
Methods
This landscape-scale study was conducted within large-scale wheat production systems in the Northern Great Plains. To quantify the relative importance of local and landscape variables explaining local diversity, local response variables and local explanatory variables were collected at the local scale, which was defined as ≤ 100 m from the patch habitat, or ecological refuge (ER) center. Landscape explanatory variables were extracted for spatially nested buffers at the landscape scale, which was defined as ≤ 5 km from ER center. As buffers were nested circles rather than concentric rings, larger spatial extents included smaller spatial extents. At the local spatial extent, arthropod and plant species richness and abundance were observed at 20, 40, 60, 80, and 100 m from the center of the ER, which included the crop field margin. Percent cover of crop field and ER were calculated at each local spatial extent. For landscape analysis, five nested circular buffers were created around each ER at spatial extents of 0.1, 0.5, 1, 2, and 5 km radii, respectively. Landscape composition data, including land cover type diversity, richness, and percent cover of each land cover type, and landscape configuration data, including cohesion, division, large patch dominance, total edge and number of patches, were extracted for every landscape spatial extent.
Explanatory and response variables observed at the local spatial extent for this study included percent cover of ER and crop field, plant species richness, plant abundance (i.e. percent cover), arthropod species richness, and arthropod abundance within 100 meters from the ER center. Local plant percent cover and species richness variables were collected along six 100-meter transects in a radial design from the ER center using ocular percent canopy cover for each species within 0.1m2 circular frames every 20 meters. Ocular percent canopy cover was evaluated by averaging two observers’ visual estimates of plant percent cover in each sampling frame. Arthropod data were collected using sweep net sampling at 20-meter increments along the same transects at each site and mean plant species richness and diversity were aggregated at the same 20-m scale. In the lab, arthropods were divided into taxonomic groups including Orthoptera, Hemiptera, Coleoptera, Diptera, Hymenoptera, Araneae, Odonata, and Lepidoptera (Duff, Maxwell, and Debinski, 2024). Arthropod individuals were identified as morphospecies within each taxonomic group based on visual similarities in form and structure and used as a practical surrogate for species classification. Plant and arthropod species richness were calculated for each 20-meter increment vegetative observation or sweep net sample using the vegan package in R (Oksanen et al., 2017). All data from each transect were summed by taxonomic group (plants or arthropods) across the two years of data collection, and placed in the response variable matrix for analysis.
Landscape variable data were obtained from the Cropland Data Layer (USDA, 2023) and extracted for the five landscape spatial buffers surrounding each ER. The variables selected to represent landscape compositional heterogeneity in this study were land cover type diversity, number of land cover types, and the percent cover of agricultural cultivated land, undeveloped land, or developed land within each spatial extent. All landscape composition metrics were calculated using the vegan package in R (Oksanen et al., 2017). Land cover type richness was calculated as species richness, where each land cover type was considered a “species” type. Landcover type diversity was calculated using Shannon Diversity Index, where each land cover type was considered a “species” type, and pixel abundance was considered as “species” abundance (Reynolds et al., 2018). Land cover types were manually classified as agricultural land, wildland or developed land.
The landscape variables selected to represent landscape configurational heterogeneity in this study included patch cohesion as a metric of connectedness within each spatial extent, patch division as a metric of patchiness within each spatial extent, total edge amount as a metric of fragmentation within each spatial extent, total number of patches as a metric of patchiness within each spatial extent, and Large Patch Index as a measure of the percentage of landscape covered by the dominant patch type within each spatial extent (McGarigal et al., 2012). All landscape configuration metrics were calculated from the Cropland Data Layer using the LandscapeMetric package in R (McGarigal et al., 2012) at 30-meter (0.09-hectare pixel) resolution for each landscape extent buffer.