Defining and validating regenerative farm systems using a composite of ranked agricultural practices

The study systems

Corn was examined on 20 farms (each with four fields) in North Dakota, South Dakota, Minnesota, and Nebraska in 2015 and 2016. Fields were a minimum of 4 ha in size, and the sampling area in each field was 61 × 61 m. All samples were taken at least 12 m into the field to minimize border effects. Only three of these farms were certified organic. Genetically modified (Bt hybrid) corn was universally treated with neonicotinoid seed treatments and was regarded as insecticide-treated. In corn, we examined soil organic matter (SOM), fine particulate organic matter (fPOM), soil bulk density, water infiltration, invertebrate communities in the soil, on the soil surface and in the plant canopy, pest abundance, yields, and profits.

Sixteen California almond orchards were studied in 2018 and 2019. Replicate plots (n = 4; 40 × 40 m) were established in each orchard. Plots were established 20 m into the field to avoid field margin effects. The orchards in the study ranged from the Northern half of the San Joaquin Valley through the Capay Valley to Chico. Orchards were 3–38 y old. Almond orchards are generally planted with at least two different varieties, with over-lapping bloom periods to promote pollination. Therefore, all of the orchards in the study contained at least two varieties, and almond varieties varied among the orchards. In each orchard, we examined a full range of soil and water characteristics, soil microbial, plant, and invertebrate community characteristics, pest damage, yields, and economics.

Dung invertebrate communities were examined on 16 ranches in a 7,935 km² region in eastern South Dakota during 2015 and 2016. All sites were grazed by cattle for at least 5 y, but annual grazing intensity and grazing period varied. Herds ranged from 20 to 120 individuals, and the cattle differed in size, breed, and administered ivermectin products. Focal pastures were at least 4 ha in size. The systems were ranked from regenerative to conventional based on several practices (Table 2). In our study, the designation of management systems was based upon the grouping of several management practices that were occasionally all present on the same cattle operation.

A second study of cattle ranches examined plant communities and helminth fecal parasites. Selected sites (n = 20 per grazing treatment) focused on rangelands in the Dakotas during 2019 and 2020. Grazing treatments were assigned based on a ranch management character matrix, which consisted of rancher-defined best management practices for regenerative and conventional ranching systems (Table 2). This character matrix was adapted from a similar study conducted in the region by Pecenka & Lundgren (2019). Grazing treatments were paired within each region and year. Each grazing system had been practiced on the site for at least 4 years.

Soil quality

Soil physical and chemical properties were assessed in corn fields and in almond orchards. SOM and bulk density (BD) were assessed in both of these systems. In cornfields and almonds, four soil cores (8.5 cm deep, 5 cm in diam) were randomly collected from a single field on each of 20 farms; cores were collected at locations that were at least 10 m apart. On each sample, vegetative material was removed from 60 g of soil by hand, and the resulting samples were stored in aluminum containers at 105°C overnight. SOM in the soil sample was measured using the weight loss on ignition (LOI) technique (Davies, 1974). To measure bulk density, four soil cores of known volume per field were weighed before and after drying at 100°C for 55 h.

In cornfields, we also measured fine particulate organic matter (fPOM). Approximately 30 g of soil was weighed in sterile aluminum pans. The soil was soaked in 90 mL of hexametaphosphate for 24 h, mixed for 5 min with a stainless-steel mixer, and were then sieved through screens with 500 μm and then 53 μm holes; the finer screen isolated the fPOM fraction. The isolated sample was then weighed, dried for 24 h at 105°C. Samples were then cooled in a desiccator cabinet, weighed, baked in a furnace for 4 h at 450°C. The samples were then cooled in a desiccator cabinet and weighed a final time.

In almond orchards, soil samples were collected at random locations in the inter-row area of each plot to determine total soil carbon (TSC) and total soil nitrogen (TSN). The probe (2.54 cm × 91.44 cm) was inserted 60 cm deep. Each core was placed into a plastic bag that was stored on ice until it could be transferred to a paper bag in the laboratory. Samples were weighed, dried, and weighed again. All visible pieces of rock and organic matter were removed from the samples. The samples were ground and were then passed through a sieve with 0.180 mm openings. Elemental analysis was conducted on three subsamples (12–15 mg) that were housed on tin capsules (5 × 9 mm) (ECS 8020, NC Technologies, Milan, Italy). To control for the relative compaction and other circumstances, the mass (Mg) of TSC per ha was assessed using the Equivalent Soil Mass (ESM) method, in which a cubic spline of Mg of TSC per depth layer was calculated (Wendt & Hauser, 2013). This resulted in the assessment of carbon as Mg of TSC/ha at 6,000 Mg (59.2 cm deep).

Soil macro- and micronutrients were also quantified in the almond orchards. The samples were ground to pass a 2 mm sieve and divided into three subsamples (two were 4 g each and one weighed 40 g). The 40 g soil sample is analyzed with a 24 h incubation test at 24^oC. This sample is wetted through capillary action by adding 20 mL of deionized water to a 237 mL glass jar and then capped. After 24 h, the gas inside the jar was analyzed using an infrared gas analyzer (IRGA) (Li-Cor 840A, LI-COR Biosciences, Lincoln NE) for CO₂-C. The two 4 g samples were extracted with 40 mL of deionized water and 40 mL of H³A to extract the NO₃-N, NH₄-N, and PO₄-P from the samples. The samples were shaken for 10 min, centrifuged for 5 min, and filtered through filter paper (Whatman 2V, Cytiva, Marlborough, MA). The water and H³A extracts were analyzed on a flow injection analyzer (Lachat 8000, Hach Company, Loveland CO). The water extract was also analyzed on a Teledyne-Tekmar Torch C:N analyzer for water-extractable organic C and total N. The H³A extract was also analyzed on a Thermo Scientific ICP-OES instrument for P, K, Mg, Ca, Na, Zn, Fe, Mn, Cu, S and Al. The Haney soil health score combines five independent measurements of a soil’s biological properties to provide a general estimate of the overall health of a soil system (Haney et al., 2018). It is calculated as 1-d-CO₂-Carbon/10 plus water extractable organic carbon (WEOC)/50 plus water extractable organic nitrogen (WEON)/10.

Water infiltration

We used the rainfall infiltration rate kits of the Natural Resource Conservation Service (NRCS) to sample water infiltration rates in northern cornfields and California almonds. A metal ring (15 cm diam, 13.5 cm tall) was hammered 6.5 cm into the soil. Water (444 mL) was poured into the ring, and the time it required to saturate into the soil was recorded. A second container of 444 mL of water was poured into the ring, and the time to saturation was recorded again. In cornfields, this process was recorded during anthesis.

Microbial diversity

Microbial communities were only sampled in almond orchards using the Phospholipid fatty acid (PLFA) test. Soil cores (15 cm depth, 1.9 cm diam; N = 16), were taken from four replicates per site per farm during the fruiting period. The samples were taken at random locations within each replicate, at least 5 m apart, using a transect that diagonally bisected the plot. Soil cores for each orchard were combined in a sealed plastic bag and placed in a cooler with dry ice (Drenovsky et al., 2010). Soil samples were stored at -80˚C until they could be freeze-dried and ground to 2 mm particle sizes. Phospholipid fatty acid (PLFA) testing provided an index of a soil’s microbial biomass and composition (Frostegård et al., 2011). The microbial biomass and community composition were recorded as total microbial biomass, undifferentiated microbial biomass, total bacteria, Gram-positive bacteria, Actinomycetes, Gram-negative bacteria, Rhizobia bacteria, total fungi, arbuscular mycorrhizal fungi, saprophytic fungi, and Protozoa.

Plant community structure

Plant community structure was assessed solely as ground cover in the almond study. The ground cover height and composition in each of the replicates/plots was recorded during each of the three sampling periods. The percent ground cover was categorized as 0–25%, 25–50%, 50–75%, and 75–100%. Percent ground cover in the overall orchard was assessed using visual assessments of the percent ground cover in each invertebrate quadrat.

Plant community diversity and green canopy cover were examined during the 2019 and 2020 grazing seasons of the rangeland study, while vegetation biomass was assessed only during the 2020 season. All three plant community metrics were measured during three periods of the grazing season; early June, mid-July, and late-August/early-September. At each pasture site, two 50 m transect lines were established 15 m apart and perpendicular to the slope of the land. Plant community diversity was assessed using the belt transect method for monitoring native prairie vegetation in the Dakotas (Grant et al., 2004). Briefly, plant community structure was recorded for a randomly selected 5 × 0.5 m² belt along each transect line. Vegetation is classified in each 5 m segment according to a hierarchical breakdown of plant groups common to the region (Grant et al., 2004). Green canopy cover was documented using the smartphone app Canopeo, which quantifies fractal green vegetation canopy cover area per unit area (Patrignani & Oschner, 2015). Quadrats (50 × 25 cm²) were established at 0, 25, and 50 m along transect lines to quantify green canopy cover with Canopeo. Lastly, a disc pasture meter was used to estimate above-ground vegetation biomass along each transect line and within their immediate surrounding area (<15 m). Compression height of vegetation was recorded by dropping a 0.13 m² plate weighing 1.34 kg from a height of 183 cm.

Insect diversity

Insect surveys in cornfields and almonds involved sampling the epigeal community from 0.25 m² sheet-metal quadrats (15 cm tall) inserted into the soil (Lundgren et al., 2006). Quadrats were placed in randomly selected locations between corn rows (n = 5 per plot) during anthesis. Invertebrates were exhaustively collected from the surface of the soil and beneath plant debris using handheld mouth aspirators. In almonds, sampling of the invertebrate communities occurred during the bloom, fruit development, and harvest periods. The invertebrate communities that could be collected from the soil surface and top 2 cm of the soil with mouth-operated aspirators in 15 min were preserved in 70% ethanol.

In cornfields, we also sampled the foliar invertebrate community. The foliar invertebrate community was sampled using a destructive whole plant assessment. Corn plants (n = 25 per plot) were randomly selected, thoroughly examined, and invertebrates that were not sight-identified were collected using mouth aspirators. Plants were then severed at the ground level and were transported to the field margin where their leaves, stems, ears, whorls, and tassels were inspected and dissected on white sheets.

Soil invertebrate and dung invertebrate communities were assessed using soil cores in corn fields and rangelands. In corn fields, five soil cores (10 cm diameter, 10 cm height) were collected per field in randomly selected locations within and between corn rows during anthesis. In rangelands, the same core approach was used to collect invertebrates from 2–5 d old dung pats and the soil directly beneath the pat (n = 10 per ranch). Samples were collected monthly from May to September. All cores were refrigerated for < 36 h, and then were placed into a Berlese collection system over 7 d, when they reached a constant weight.

All extracted invertebrates were identified microscopically and cataloged; each unique specimen was identified to the lowest taxonomic level possible representing a functional morphospecies. Each morphospecies was placed into a trophic guild (coprophage, predator, parasitoid, herbivore, or pest) based upon previous descriptions of the biology of arthropod community in these systems. Voucher specimens are housed in the Mark F. Longfellow Biological Collection at Blue Dasher Farm, Estelline, SD, USA.

Pests

In cornfields, pests included aphids, lepidopteran larvae, and corn rootworm adults that were identified from the foliar bioinventories. In almonds, pest incidence was assessed on 500 almonds per farm in 2018 and 600 almonds per farm in 2019 (< 20 from any one tree) (Bentley et al., 2001; Doll, 2009; Legner & Gordh, 1992). The almonds were each categorized as having no pest damage, navel orange worm damage (Amyelois transitella [Walker]; Lepidoptera: Pyralidae), ant damage (Formicidae), oriental fruit moth damage (Grapholita molesta [Busck]; Lepidoptera: Tortricidae), peach twig borer damage (Anarsia lineatella Zeller; Lepidoptera: Gelechiidae), leaf footed plant bug or stinkbug damage (Hemiptera: Coreidae, Pentatomidae), and unknown pest damage (Bentley et al., 2001; Doll, 2009; Symmes, 2018).

Fecal oocyst counts of coccidia and egg counts of gastrointestinal nematodes belonging to the superfamilies Trichostrongyloidea and Strongyloidea were conducted on herds from the 2019 – 2020 rangeland study. Three grams of dung was collected from 2 – 4 d old cattle dung pats present in pastures (n = 5 pats sampled/site). Fecal samples were processed according to the Modified Wisconsin Sugar Float Technique (Cox & Todd, 1962). Coccidia oocysts and eggs of Trichostrongyles and Strongyles were quantified for 1 g of fecal sample. Sampling for internal parasites in the herds was conducted twice in 2019 (mid-July and late-August/early September) and three times in 2020 (early-June, mid-July, and late-August/early September).

Yield and profit

In cornfields and almonds, responses from producer surveys were used to determine management practices, costs, and revenues that went into the direct net profitability of each operation. Gross profit analysis included yield, return on grain, and additional revenue that included livestock grazing and production. In cornfields, yield information provided by the farmers were confirmed by yields that were hand-gathered from three 3.5-m row sections from each replicate-field. Costs associated with corn production included corn seed, cover crop seed, drying/cleaning grain, crop insurance, tillage, planting corn, planting cover crop, fertilizers, herbicides, and irrigation (additional information on profit/loss analysis can be found in LaCanne & Lundgren 2018). Operating costs in almonds included winter sanitation, sampling for tree nutrient status and soil salinity, pH, and nutrient levels, irrigation and frost protection, fertilizers, insecticides, herbicides, fungicides, disease treatment sprays, trapping vertebrate pests, cover crop seed, tillage, mowing, flamers, grazers, and harvest.). Harvest in the almonds included the hourly labor to conduct the harvest and the price paid to external contractors (additional information on the almond profit/loss analysis can be found in Fenster et al. in prep). In both systems, only direct costs and revenues were used to calculate profitability.

Data analysis

Unless otherwise described, we used linear regressions to search for correlations between the regenerative matrix scores and response variables. In the almond study, clay percentages of the soils were considered as co-factors in all models that examined TSC. Plot values were composited across plots into single values in the almond study and linear regression analyses were used to compare regenerative scores and response variables. In instances where resampling a single farm was performed (either spatially or temporally) Linear Mixed Models were used to remove pseudoreplication and account for this dependence; field and farm were included as random factors in the corn models, and month was included as a fixed factor and farm as a random factor in the rangeland studies. All statistics were conducted using Systat 13.1 (Systat Software Inc., San Jose, CA)

Soil physical and chemical properties

In cornfields, fPOM of the soil was strongly and positively associated with regenerative matrix scores (F_{1, 17} = 4.61; P = 0.047) (Figure 1A), but SOM (F_{1, 17} = 0.11; P = 0.75), bulk density (F_{1, 17} = 2.32; P = 0.15), and water infiltration rates (F_{1, 11} = 0.47; P = 0.51) of the soil were not correlated with matrix scores. The slowest water infiltration rates in the higher regenerative scoring farms were those that practiced tillage.

Figure 1.

The relationships of regenerative farming intensity on fine particulate organic matter in cornfields of the Upper Midwest (A) and percent soil organic matter in California almond orchards (B).

In almond orchards, higher matrix scores correlated to significantly higher levels of TSC and TSN (500–6000 ESM layers) and SOM (15 cm depth). At the 0–6000 Mg ESM layer, there was a significant correlation between the orchards’ matrix scores and TSC (model: F_{2, 13} = 13.77, P = 0.001; score: t = 2.26, P = 0.04; clay: t = 4.73, P < 0.001) (Figure 2). At the 6000 Mg ESM layer there was a significant correlation between the orchards’ matrix scores and TSN (model: F_{2, 13} = 17.73, P < 0.001; score: t = 1.94, P = 0.08; clay: t = 5.63, P < 0.001). In contrast to what we found in cornfields, in the top 15 cm of soil there was a significant correlation between the orchards’ SOM%, matrix scores, and clay percentage (model: F_{2, 13} = 47.56, P < 0.001; score: t = 5.50, P < 0.001; clay: t = 8.05, P < 0.001) (Figure 1B).

Figure 2. Total soil carbon in California almond orchards as the orchards increase the number of regenerative practices.

In almonds, higher matrix scores and soil clay percentages correlated to higher levels of WEON (F_{2, 13} = 9.16, P = 0.003) and WEOC (F_{2, 13} = 15.61, P = 0.001) in the organic matter of the soils. Total phosphorus (model: F_{2, 13} = 4.32, P = 0.04; score: t = 2.65, P = 0. 02, clay: t = -1.28, P = 0.23) and inorganic phosphorus (model: F_{2, 13} = 4.56, P = 0.03, score: t = 2.74, P = 0.02, clay: t = -1.28, P = 0.22) were significantly and positively correlated with an orchard’s matrix score. Higher matrix scores correlated to higher levels of Calcium (F_{2, 13} = 5.97, P = 0.01) and Sulfur (F_{1, 14} = 10.80, P = 0.01), but lower levels of Aluminum (F_{1, 14} = 6.13, P = 0.03). There were no correlations between other micronutrients and the regenerative matrix score (P > 0.05). Haney soil health test scores were positively correlated with regenerative matrix scores (model: F_{2, 13} = 7.82, P = 0.01, Score t = 2.99, P = 0.01, clay: t = 2.59, P = 0.02) (Figure 4). Soil respiration was unaffected by regenerative matrix scores (model: F_{2, 13} = 2.91, P = 0.09, Score t = 1.62, P = 0.13, clay: t = 1.77, P = 0.10).

Figure 3. Water infiltration rates in the soils of California almond orchards in relation to the number of regenerative practices on each orchard.

Figure 4. Haney test scores on soils from California almond orchards with increasing numbers of regenerative practices.

Higher regenerative-conventional matrix scores in almonds correlated to lower bulk densities (model F_{2, 13} = 8.95, P = 0.004; score: t = -3.91, P = 0.002, clay: t = -1.61, P = 0.13). Water infiltration rates increased alongside matrix scores for these orchards (model: F_{2, 5} = 11.43, P = 0.01; score: t = -4.50; P = 0.009, clay: t = 2.56, P = 0.05) (Figure 3).

Soil microbiology

In almond orchards, higher matrix scores correlated to higher amounts of bacterial biomass (F_{1, 14} = 6.25, P = 0.03) and higher Gram (+) biomass (F_{1, 14} = 9.12, P = 0.01). The percent composition in the microbial community was not affected by regenerative score (F_{1, 14} = 2.47, P = 0.14), indicating greater overall microbial biomass rather than bacterial dominance. The remainder of these microbial community metrics were not correlated to the regenerative conventional matrix score (P > 0.05).

Plant community

More regenerative cropland and rangeland supported more diverse and robust plant communities. In almond orchards, higher matrix scores correlated to more ground cover (F_{1, 14} = 127, P < 0.001), more plant species (F_{1, 14} = 27.61, P < 0.001) (Figure 5A), and greater biomass height (F_{1, 14} = 9.90, P = 0. 01). In rangelands of the Northern Plains, there was a significant relationship between matrix score and plant biomass (Score: F_{1, 53} = 8.47, P = 0.01; month: F_{2, 53} = 0.45, P = 0.64) (Figure 5B), plant diversity (Score: F_{1, 106} = 14.22, P < 0.001; month: F_{3, 106} = 1.02, P = 0.39) (Figure 5C), and ground cover (Score: F_{1, 106} = 10.06, P = 0.002; month: F_{3, 106} = 14.76, P < 0.001).

Figure 5.

Plant species richness on the orchard floor in California almonds (A), and plant diversity (Shannon H; B) and biomass index (C) in rangelands of the Northern Plains as they relate to how regenerative a farm is.

Invertebrate community

Across cropland and rangeland, more regenerative farms had more robust invertebrate communities. Several aspects of invertebrate community structure in cornfields were strongly and positively correlated with how regenerative a field was scored. The abundance (F_{1, 55} = 6.53, P = 0.01), species richness (F_{1, 55} = 38.31, P < 0.001), and diversity (F_{1, 55} = 7.34, P = 0.01) (Figure 6A) of the epigeal invertebrate community was positively affected by matrix score. Similarly, matrix scores were positively associated with the species richness of the invertebrate community found in the soil column (F_{1, 56} = 12.96, P = 0.001), but the abundance was only marginally associated with matrix score (F_{1, 56} = 3.41, P = 0.07), and the diversity of this community was unaffected by matrix score (F_{1, 56} = 2.45, P = 0.12). Foliar invertebrate abundance (F_{1, 55} = 0.99, P = 0.32), species richness (F_{1, 55} = 1.35, P = 0.25), and species diversity (F_{1, 55} = 0.20, P = 0.66) were not related to the matrix scores.

Figure 6.

The invertebrate species diversity (Shannon H) on the soil surface in cornfields of the Upper Midwest (A), and on the orchard floors of California almonds (B), as they relate to the number of regenerative practices implemented on farms.

The invertebrate community on the soil surface was positively affected by the number of regenerative practices on almond orchards. Invertebrate abundance (F_{1, 14} = 10.28, P = 0.01), biomass (F_{1, 14} = 70.10, P < 0.001), species richness (F_{1, 14} = 14.44, P = 0.002) and species diversity indices (H: F_{1, 14} = 12.37, P = 0.003; DS: F_{1, 14} = 11.11, P = 0.01) (Figure 6B) were positively correlated with matrix score. There was a positive correlation between earthworm (square rooted to normalize residuals) abundance and matrix score on these orchards (F_{1, 14} = 9.87, P = 0.007).

In rangelands of the Northern Plains, the invertebrate community associated with cattle dung was strongly and positively associated with the matrix score attained. The species richness (Score F_{1, 74} = 15.38, P < 0.001; month F_{4, 74} = 6.54, P < 0.001), and species diversity (Score F_{1, 74} = 16.10, P < 0.001; Date: F_{4, 74} = 1.50, P = 0.21) (Figure 7A) of the dung invertebrate community were positively associated with matrix score. Abundance of dung invertebrates was uncorrelated with matrix score (Score F_{1, 74} = 0.01, P = 0.99; month: F_{4, 74} = 4.83, P = 0.001). The abundances of two critical functional groups, dung beetles (Score F_{1, 60} = 10.05, P = 0. 006; month: F_{4, 60} = 5.70, P = 0.001) (Figure 7B) and invertebrate predators, (Score F_{1, 74} = 7.41, P = 0.01; month: F_{4, 74} = 4.09; P = 0.01) increased as a rangeland became more regenerative.

Figure 7.

Invertebrate species diversity (Shannon H) in dung pats (A) and the abundance of dung beetles (B) as they relate to regenerative intensity on ranches in the Northern Plains.

Pest populations

In all of these systems, pest populations were uniformly below any recognized economic thresholds. Pests were not correlated with the number of regenerative practices used in cornfields (pest abundance: F_{1, 55} = 2.18, P = 0.15) or in almond orchards (% damaged nuts: F_{1, 14} = 0.69; P = 0.42). In rangelands, there was no relationship between the matrix score and Trichostrongyles (Score: F_{1, 88} = 0.09, P =0.77; month: F_{3, 88} = 0.39, P = 0.76) or Coccidia (Score: F_{1, 88} = 1.78, P = 0.19; month: F_{3, 88} = 0.70, P = 0.56) fecal parasite loads.

Yields

Corn yields were negatively correlated with the regenerative matrix score (F_{1, 56} = 3.88, P = 0.05). Despite this, three of the top ten yielding cornfields were regenerative, indicating that regenerative practices are not necessarily tied with yield reductions. Standard deviations became greater proportions of the mean as a farm became more regenerative (variability in yields were higher among regenerative farms) (R² = 0.79; F_{1, 4} = 5.12, P = 0.11). In almonds there was no relationship between almond yield and how regenerative an orchard was (F_{1, 12} = 0.86, P = 0.37). On a treatment level, regenerative farms had twice the variance in yield relative to the conventional orchards.

Economics

Although there was a positive relationship between matrix score and profitability in cornfields, this relationship was not significant (F_{1, 53} = 2.49, P = 0.12). In cornfields, there was a significant increase in the standard deviation of profit as a farm became more regenerative (e.g., standard deviation as a proportion of the mean; R² = 0.72; F_{1, 4} = 7.68, P = 0.07). This indicates more variability in profits among the more regenerative producers. The top ten highest netting farms were all regenerative farms (scores of four and five). In almonds, profit was strongly and positively correlated with matrix scores (F_{1, 11} = 7.41, P = 0.02) (Figure 8). Operating costs were unaffected by matrix score (F_{1, 13} = 0.70, P = 0.42), but net income increased as regenerative scores increased (F_{1, 11} = 8.50, P = 0.01).

Figure 8. Net profit of California almonds as it relates to the number of regenerative practices on orchards.

Underlying data

Open Science Framework: Matrix paper, https://doi.org/10.17605/OSF.IO/G697Y (Lundgren, 2021) (registered 15^th January 2021 https://osf.io/7v4z2).

This project contains the following underlying data:

Extended data

Open Science Framework: Matrix paper, https://doi.org/10.17605/OSF.IO/G697Y (Lundgren, 2021) (registered 15^th January 2021 https://osf.io/7v4z2).

This project contains the following extended data:

Data are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).