Effects of Habitat Heterogeneity on Crop Yield and Biodiversity

First, landscape heterogeneity analysis indicated that Farms 1 and 3 had higher land cover type richness and diversity than Farm 2 at all spatial scales ranging from one to 100-kilometer radii (Figure 1). A shortened subset of values ranging from one to 20-kilometer buffer zones are pictured in Figure 1.

For more detail see the image Landscape Heterogeneity Surrounding Each Farm attached.

Farms 1 and 3 were surrounded by higher landscape heterogeneity than Farm 2, as indicated by their higher levels of land cover type diversity, while Farm 2 existed in a more homogenous landscape matrix, as indicated by its lower level of land cover type diversity (Figure 2). Land cover type heterogeneity at the landscape scale likely plays a crucial role in supporting biodiversity within ecological refugia through source-sink dynamics. The dynamics between landscape-scale heterogeneity and patch-level biodiversity will be discussed in more depth in future work.

For more details see the image Gamma Diversity of Land Cover Types attached.

Second, observations from vegetative surveys indicated that crop fields with an ecological refuge typically hosted higher plant diversity than fields without a refuge, though diversity trends varied by growing season conditions, farm management type and refuge size. In 2020, observations from Farm 1 and 2 indicated significantly higher plant diversity in the field with a refuge than in the control field (p-value < 0.0001) (Figure 3). In 2021, Farm 1 indicated no significant difference in plant diversity between field types (p-value = 0.11) while Farm 2 indicated significantly higher plant diversity in the field with a refuge (p-value < 0.0001) (Figure 3). In 2022, Farm 1 and 2 indicated significantly higher plant diversity in the heterogeneous field than the homogeneous field (p-value < 0.0001) (Figure 3).  The lack of difference between the control and refuge field in 2021 on Farm 1 may have been the severe drought that that summer which resulted in early plant maturation, early senescence, low survival and difficulty identifying plant species in July. As for Farm 3, observations from 2020 indicated significantly lower plant diversity in the field with a refuge than in the control field (p-value = 0.01) (Figure 3). However, in 2021 and 2022, Farm 3 indicated no significant difference in plant diversity between control and refuge field types (p-value = 0.26 and p-value = 0.74) (Figure 3). The lack of diversity in the field with a refuge on Farm 3 may be explained by the small size and recent creation of the refuge. Furthermore, the lack of difference in plant diversity between control and refuge field types on Farm 1 and 3 may have been exacerbated by organic management practices, which tend to facilitate higher plant diversity within crop fields due to the prevalence of volunteers and weedy species, thus reducing any stark differences between field types. Farm management practices and their effect on plant diversity will be discussed in greater detail in the following sections.

For more detail see the image Plant Diversity Boxplots attached.

Alpha diversity is a common measure of local diversity within a sampling site. Plant alpha diversity was significantly lower on Farm 3 than the other farms over a three-year sampling period (Figure 4). Plant alpha diversity was significantly higher on Farm 2 than the other farms in 2020 and 2022. The fact that Farm 2 had the highest plant diversity, and the largest ecological refuge suggests evidence of the species-area relationship, where larger sampling areas result in higher amounts of species sampled. The consistent trend of lower plant diversity on Farm 3 may have occurred because the ecological refuge on Farm 3 was much smaller and more recently established than the other two refugia, resulting in lower species diversity due to the smaller sampling area and lower level of ecosystem complexity.

For more detail see the image Plant Alpha Diversity attached.

As general proof of concept that small patch refugia can host biodiversity, a Bayesian Kriging interpolation of sampled plant species indicated that plant species richness was likely to be higher in the refuge than in the crop field for Farm 1 (Figure 5A). With higher plant species richness in and around the ecological refuge, we would expect accompanying ecosystem services such as pollination or pest predation to also occur around the refuge. However, not all biodiversity is beneficial. To decipher between desirable and undesirable on-farm biodiversity, plant diversity was separated into native and nonnative diversity for analysis. A Bayesian Kriging interpolation indicated that native species are likely to have higher species richness in the refuge while nonnative species are likely to have higher species richness in the crop field (Figure 5B and 5C). This suggests that ecological refugia are beneficial sources of native plant diversity and not detrimental sources of weeds in crop fields.

For more detail see the image Plant Species Richness Kriging Interpolation attached.

Next, linear regression of plant diversity by distance from refuge demonstrated that plant diversity was highest within refugia and carried into the surrounding crop field. Plant diversity declined significantly with distance from refugia on Farm 1 (p-value = 0.13, p-value = 0.07, p-value <0.0001) and Farm 2 (p-value = 0.01, p-value < 0.0001, p-value < 0.0001 ) in all years, and did not drop off abruptly at the crop field boundary (Figure 6). However, plant diversity on Farm 3 increased significantly with distance from the refuge in 2020 (p-value = 0.01), decreased insignificantly in 2021 (p-value = 0.87) and decreased significantly in 2022 (p-value = 0.04) (Figure 6). The nonconforming diversity trends for Farm 3 emphasize the important role of ecosystem structure in supporting ecosystem function for refugia, as this refuge is by far the youngest, least established, and least diverse of the three refugia. The three-year trend of increasing plant diversity on Farm 3 indicate that the refuge may increase in plant diversity as it continues to establish. The refuge on Farm 3 was a bare patch of unproductive land that is still in the process of being restored. Thus, the trend in plant diversity should continue declining with distance from refuge as diversity within the refuge is established and enhanced with native plantings. In contrast, Farm 2 consistently demonstrated dramatic differences in plant diversity between fields with and without an ecological refuge. The contrast may be due to the fact that Farm 2 is conventionally managed, meaning that weedy species are managed, and overall plant diversity is reduced within the crop field. The field with a refuge on Farm 1 is organically managed and has the highest overall plant diversity. However, while both Farms 1 and 3 are managed organically, they display different plant diversity trends due to differences in their ecological refugia. The size and age of refugia tend to overpower the farm management style, as the larger and more established refuge on Farm 1 hosted significantly more biodiversity than the organic crop field, while the smaller, younger refuge on Farm 3 only hosted higher plant diversity than the crop field after three years of restoration. This indicates that regardless of organic or conventional farm management practices, ecological refugia may become increasingly beneficial to farmers over time as they undergo restoration and establishment.

For more detail see the image Plant Diversity Trends attached.

Third, insect sweep net samples indicated that winged insect diversity declined significantly with distance from refuge for all refugia in all years sampled (p-value < 0.0005) (Figure 7). These results suggest a decline in ecosystem complexity and insect habitat with distance from the refuge and into the more homogeneous crop field.

For more detail see the image Insect Sweep Net Diversity Trends attached.

In addition, pitfall samples indicated that ground-dwelling insect diversity declined significantly with distance from refuge for all farms (p-value < 0.001) (Figure 8).  Declines in ground-dwelling insect diversity may be accompanied by a reduction of beetle-mediated services such as weed seed predation, soil bioturbation and nutrient cycling.

For more detail see the image Pitfall Trap Diversity attached.

However, as stated previously, not all biodiversity is beneficial to crop production. Further insect analysis was conducted to distinguish between beneficial insect and pest insect services associated with ecological refugia. A pairwise comparison indicated that the field with a refuge on Farm 1 had a significantly higher abundance of insects from the orders Hymenoptera, Araneae and Odonata which were associated with the beneficial ecosystem service of pollination and pest predation (p-value < 0.05) (Figure 9A). On Farm 2, the field with a refuge had a higher abundance of insects from the orders Diptera and Hymenoptera, which were associated with beneficial ecosystem services such as pollination and pest predation (p-value < 0.001). In addition, the field without a refuge on Farm 2 had a significantly higher abundance of pest insects from the orders Orthoptera and Hemiptera, which were associated with the disservice of yield reduction (p-value < 0.05) (Figure 9B). The field with a refuge on Farm 3 had a significantly higher abundance of insects from the order Coleoptera, which were associated with beneficial ecosystem services such as weed seed predation and nutrient cycling. However, the same field had a significantly higher abundance of insects from the orders Orthoptera and Hemiptera, which were associated with yield-reducing pest insects like grasshoppers and aphids (Figure 9C). Once again, these results highlight the mixed services provided by the recently created refuge on Farm 3 in contrast to the benefits from well-established, naturally occurring refugia on Farm 1 and 2. These differing results suggest that structural and temporal factors determine the ability of a refuge to host biodiversity and provide ecosystem services.

For more detail see the image Insect Abundance by Order and Field Type attached.

To conclude the results from objective one, it is important to reiterate that results from Farm 3 were inconsistent with results from the other two farms. The nonconforming diversity trends on Farm 3 can be explained by two main factors. The first factor is a difference in farm management. Organic management of Farm 1 and 3 explained higher crop field plant diversity due to weeds, cover crops and volunteers not found in high abundance on farms managed with herbicide like Farm 2. Furthermore, organic management is commonly associated with higher insect diversity due to enhanced habitat from higher plant diversity, and lack of insecticide application. The second factor explaining nonconforming diversity trends on Farm 3 was a difference in timescale. The refuge at Farm 3 was recently converted from poor, alkaline soil to a patch habitat, while the other two refugia were pre-existing, relatively undisturbed natural habitat. While plant diversity trends demonstrated a direct relationship between diversity and distance from refuge on Farms 1 and 2, the results were inconsistent on Farm 3. Furthermore, Farms 1 and 2 hosted a higher abundance of beneficial insects in the field with a refuge, while Farm 3 hosted a higher abundance of pest insects in the field with a refuge. These divergent plant and insect trends on Farm 3 emphasize the important role of ecosystem structure in supporting ecosystem function for on-farm refugia, as this refuge is by far the youngest, smallest, least established, and least diverse of the three refugia. This relatively young experiment in small patch habitat restoration should continue to reveal interesting results regarding how plant and insect-mediated ecosystem services change over time as the refuge establishes. Before recommending that farmers remove land from production to convert into on-farm habitat, further research is needed to understand the amount of time, amount of habitat area and amount of biodiversity needed to provide beneficial ecosystem services in on-farm patch habitat. 

Transitioning to objective two, seed trap data indicated that seed predators provided a tradeoff in ecosystem services to farmers. In 2021, seed predators equally selected between the two crop seeds, common wheat and Arvika green pea. In terms of weed seeds, seed predators largely preferred field pennycress over wild oat. In 2022, seed predators preferred common wheat over Arvika green pea, and selected kochia over wild oat. Weed seed predation of field pennycress, kochia and wild oat was classified as a beneficial service to famers, while crop seed predation of wheat and pea was considered a disservice to farmers. Therefore, seed predators created a tradeoff for farmers by providing a beneficial service of weed suppression and a disservice of reducing crop yield. Seed predators associated with an ecological refuge provided two beneficial services to Farm 1 in 2021, as weed seed predation was higher and crop seed predation was lower in the field with a refuge than in the control field (Figure 10). However, in 2022, Farm 1 received two disservices of higher crop seed predation and lower weed seed predation in the field with a refuge. In 2021, Farm 2 received two disservices of higher crop seed predation and lower weed seed predation in the field with a refuge (Figure 8). However, in 2022, Farm 2 received mixed services of higher weed seed predation but also higher crop seed predation in the field with a refuge. On Farm 3, the field with a refuge received mixed services as it had higher weed seed predation and higher crop seed predation for both years (Figure 10). Overall, crop seed predation was higher than weed seed predation on all three farms. However, weed seed predation was higher on organic farms than on the conventional farm. Mixed results for seed predation services and disservices suggest that seed predator activity may be explained by differences in farm management practices. Farm 1, which received two beneficial seed predation services, had the advantage of starting with a large, naturally occurring refuge. In addition, organic management practices, including polycropping and lack of chemical inputs, may be another factor contributing to higher biodiversity and seed predation services on this organic farm. In contrast, Farm 2, received only disservices from seed predators, despite having a large, naturally occurring refuge. These disservices may be explained by differences in conventional farm management, as monocropping and application of herbicides and insecticides likely reduced beneficial seed predator activity. Farm 3, which received both seed predation services and disservices, was also under organic management. It experienced higher weed seed predation but also higher crop seed predation. The higher amount of crop seed predation on Farm 3 may be attributed to the lack of seed and other food sources provided by the small and recently established ecological refuge.

For more detail see the image Seed Predation By Seed Type attached.

Additionally, seed predation trends changed with distance from refuge on all three farms. On Farm 1, crop seed and weed seed predation trends were weak and increased insignificantly with distance from refuge (p-value = 0.31 and p-value = 0.18) while overall seed predation increased significantly with distance from refuge (p-value = 0.09) (Figure 11).  These results could be interpreted as an indication of lower seed predator activity near and in the ecological refuge. However, due to the fact that Farm 1 is managed organically, has a large ecological refuge and hosts high levels of biodiversity, these results may suggest evidence for the alternative hypothesis, where lower seed predation occurred near the refuge due to its abundance of food sources, and seed trap predation was higher in the crop field due to a lack of food sources typically found there. On Farm 2, crop seed predation, weed seed predation and overall seed predation decreased significantly with distance from refuge (p-value = 0.03, p-value = 0.01 and p-value < 0.01) (Figure 11). Due to the fact that Farm 2 is managed conventionally, the ecological refuge is large and the crop field is a monocrop with little food source diversity, these results indicated a higher abundance of seed predators living near and in the ecological refuge. Seed predation trends on Farm 2 indicated tradeoff for farmers in terms of beneficial weed seed predation but detrimental crop seed predation associated with the refuge. On Farm 3, seed predation trends were weak, as crop seed and overall seed predation decreased insignificantly (p-value = 0.29 and p-value = 0.57), while weed seed predation increased insignificantly (p-value = 0.98) (Figure 11). The weak trends between seed predation and distance from refuge on Farm 1 and 3 may be explained by the interaction between organic management, which would be expected to boost overall seed predation activity, and the young, small refuge, which would be expected to support less seed predator activity than the more established refugia.

For more detail see the image Seed Predation by Distance from Refuge attached.

Next, linear and nonlinear regression were used to assess how proximity to the ecological refuge and nitrogen fertilizer rates impacted crop yield. A nonlinear regression indicated that every pound of nitrogen fertilizer initially increased yield by 0.78 bushels per acre, but that the effects of nitrogen fertilizer plateaued when 38.17 pounds of nitrogen were applied. Without considering the effect of nitrogen application, linear regression indicated that crop yield significantly declined approximately 1.2 bushels per acre with every 20 meters from the refuge (p-value < 0.0001) (Figure 12).

For more detail see the image Yield Decreased with Distance From Refuge on Farm 2 attached.

The nonlinear regression also indicated that distance from refuge negatively impacted yield even after accounting for the effects of nitrogen fertilizer. Yield declined approximately 0.115 bushels per acre every meter from the refuge, or approximately 2.4 bushels every 20 m from the refuge while the maximum fertilizer effect plateaued at 53.06 bushels per acre (Figure 13). As a result, both the linear and non-linear characterizations of yield indicated significant declines in yield with distance from the ecological refuge on Farm 2.

For more detail see the image Wheat Yield with Distance from Refuge and Nitrogen Fertilizer Rate.

However, the opposite trend occurred on Farm 3, as yield significantly increased about 0.6 bushels per acre with every 20 meters from the edge of the refuge (p-value < 0.0001) (Figure 14). No effects of nitrogen fertilizer were accounted for as the farm was managed organically. As noted previously, the length of time required to establish native plant species and associated ecosystem services may explain the difference in yield trends surrounding the newly created refuge on Farm 3 versus those surrounding the well-established, naturally occurring refugia on Farm 2.

For more detail see the image Yield Increased with Distance From Refuge on Farm 3 attached.

Next, a random forest model was built using distance from refuge, soil characteristics, topography and weather data as explanatory variables for crop yield response. The x-axis of the variable importance plot indicated the decrease in accuracy the model would experience if a particular variable was removed from the model. The y-axis ranked the most important variables from top to bottom. The variable importance plot indicated that distance from refuge was the most important explanatory variable in the model, followed by nitrogen application and then NDVI (Figure 15). In addition, R-squared values were calculated for the model with and without distance from refuge as a variable to quantify each model's ability to explain the variability in yield. A random forest model without distance from refuge as an explanatory variable explained only 57% of the overall variability (R-squared = 0.57) while a random forest model that included distance from refuge as a variable explained 70% of the variability (R-squared = 0.70) model. The vast improvement of the R-squared value and the high ranking of distance from refuge in the variable importance plot indicated that ecological refugia have a significant impact on crop yield.

For more detail see the image Variable Importance Explaining Yield attached.

To clarify what sort of effects ecological refugia have on crop yield, multiple linear regression was used to characterize the relationship between each modeled variable and the yield response. Regression coefficients indicated that yield was negatively related to elevation, distance from refuge, sin of aspect, and minimum daily temperature (p-value < 0.0001) (Table 2). Yield was positively related to NDVI, nitrogen application and soil organic carbon content (p-value < 0.0001) (Table 2).

Table 2. Multiple linear regression coefficients indicated that crop yield was negatively related to elevation, distance from refuge, sin of aspect, and minimum daily temperature.

Next, the impact of ecological refugia on crop quality were assessed using winter wheat grain analysis results from 2021 for Farm 2 and 3.  Barley samples were collected from Farm 1 but could not be analyzed due to drought-induced lack of grain formation. When comparing field types on Farm 2 and 3, winter wheat protein content was significantly higher and moisture content was insignificantly lower in the field with a refuge on Farm 2 (p-value < 0.001 and p-value = 0.3) (Figure 16). These trends were reversed on Farm 3, where winter wheat grain moisture content was significantly higher while grain protein was significantly lower in the field with a refuge (p-value < 0.0001 and p-value = 0.014) (Figure 16). Typically, grain protein content is strongly influenced by environmental factors such as nitrogen application, temperature and water availability. Higher precipitation at the beginning of the grain-fill phase can positively affect protein formation, as it tends to improve the uptake of nitrogen. However, water stress during early grain development can reduce overall grain weight, and increase relative grain protein content. Thus, water stress effects might explain the inverse relationship between grain moisture and grain protein content reported on both farms.

For more detail see the image Grain Moisture and Protein Content attached.

On both Farms 2 and 3, grain polyphenol content was significantly lower in the field with a refuge and grain iron content was significantly higher in the field with a refuge on Farm 2 (p-value = 0.003 and p-value = 0.039) (Figure 17). Grain iron content was significantly higher in the field with a refuge on Farm 2 (p-value = 0.011) while grain iron content did not significantly differ between field types on Farm 3 (Figure 17).

For more detail see image Grain Iron and Polyphenol Content attached.                    

On Farm 2, grain total polyphenol content significantly increased with distance from the refuge (p-value = 0.04) while grain moisture, protein and iron content did not change significantly with distance from the refuge (Figure 18). On Farm 3, grain quality metrics did not differ with distance from the refuge, likely due to the low sample size and lack of grain development. As for differences in grain polyphenols, the synthesis and bioaccumulation of phenolic compounds often depend on environmental factors such as temperature and precipitation during grain kernel development. The phenolic pathway in plants can also be stimulated by stressful conditions, such as herbivory or drought. Consequently, crops grown in organic or low-input systems often contain more phenolic compounds than those grown in conventional systems with high inputs. Thus, higher concentrations of plant polyphenols tend to appear in stressed plants. One explanation of the increase in grain polyphenols with distance from refuge is that plants near the refuge underwent less stress than plants far from the refuge. Less stress near the refuge may be due to beneficial services associated with the refuge such as beneficial insects reducing pests that could stress and damage crops, seed predators reducing crop competition with weeds or even ground-dwelling insects providing bioturbation that reduced crop stress by improving soil water holding capacity.

For more detail see the image Grain Total Polyphenol Content with Distance from Refuge on Farm 2 attached.

Using the additional 60 plant tissue samples, a linear regression on Farm 2 indicated that grain protein and iron content significantly decreased with distance from the refuge (p-value < 0.0001) while grain polyphenol content increased with distance from the refuge (p-value < 0.0001) (Figure 19). Grain protein decreased by 0.16% every 20 meters from the edge of the refuge, while grain iron content decreased by 0.011 mg/L every 20 meters from the edge of the refuge (p-value <0.0001) (Figure 19). Grain polyphenols increased by 0.224 mg/L every 20 meters from the edge of the refuge. One explanation for the higher grain protein content near the refuge may be due to an interaction between environmental factors and nitrogen application. The lower topography and potentially increased soil water holding capacity near the refuge could lead to enhanced nitrogen uptake near the refuge, subsequently enhancing grain protein formation in crops closer to the refuge. However, the relationship between precipitation, temperature, nitrogen uptake and grain protein formation are complex and difficult to parse out without an empirical study of how they specifically interacted on Farm 2. The increase in grain polyphenol concentration with distance from the refuge offers further evidence that plants near the refuge underwent less stress than plants far from the refuge. This could suggest that beneficial ecosystem services proximal to the refuge created conditions that are more conducive to crop growth. 

For more detail see the image Grain Protein, Polyphenol, and Iron Content attached.

To explain the relationship underlying growth conditions and grain quality metrics, a variable importance plot for the random forest model indicated that distance from refuge, followed by nitrogen application, were the most import variables in explaining grain protein, iron and polyphenol concentration (Figure 20). R-squared values were calculated to quantify each model's ability to explain variation in grain quality. A random forest model explained 54% of the overall variability in grain protein content (R-squared = 0.54), 72% of the variability in grain iron content (R-squared = 0.72) and 58% of the variability in grain polyphenol content (R-squared = 0.58) (Figure 20). While the models captured some of the variability underlying grain quality, these R-squared values indicate that interactions between environmental conditions and other variables such as management practices that affect grain quality should be studied in more depth in future research.

For more detail see the image Variable Importance Explaining Grain Quality attached.

To clarify what sort of effects ecological refugia have on each metric of grain quality, multiple linear regression was used to characterize the relationship between each modeled variable and grain protein, iron and polyphenol content. A multiple linear regression indicated that grain protein content was negatively related to soil clay content, slope, and soil organic carbon content (p-value < 0.0001 and p-value = 0.0025) (Table 3). Protein content was positively related to elevation, distance from refuge, maximum daily temperature, nitrogen fertilizer application, NDVI, cos of aspect and topographic positioning index p-value < 0.0001, p-value = 0.0086 and p-value = 0.0001) (Table 3).

Table 3. Multiple linear regression coefficients indicated that grain protein content was negatively related to soil clay content, slope and soil organic carbon and positively related to elevation, distance from refuge, maximum daily temperature, nitrogen fertilizer application, NDVI, cos of aspect and topographic positioning index.

Multiple linear regression indicated that grain iron content was negatively related to elevation, nitrogen fertilizer application and slope (p-value < 0.0001) (Table 4). Grain iron content was positively related to soil clay content, distance from refuge, cos of aspect, soil organic carbon content and topographic positioning index (p-value < 0.0001) (Table 4).

Table 4. Multiple linear regression coefficients indicated that grain iron content was negatively related to elevation, nitrogen fertilizer application and slope and positively related to soil clay content, distance from refuge, cos of aspect, soil organic carbon content and topographic positioning index.

A multiple linear regression indicated that grain polyphenol content was negatively related to elevation, nitrogen fertilizer application and slope (p-value < 0.0001 and p-value = 0.0301) (Table 5). Grain polyphenol content was positively related to distance from refuge, sin of aspect, NDVI, cos of aspect, soil organic carbon content and topographic positioning index (p-value < 0.0001, p-value = 0.038, p-value = 0.0147, p-value = 0.0222, p-value = 0.0001 and p-value < 0.0001) (Table 5). 

Table 5. Multiple linear regression coefficients indicated that grain polyphenol content was negatively related to elevation, nitrogen fertilizer application and slope and positively related to distance from refuge, sin of aspect, NDVI, cos of aspect, soil organic carbon content and topographic positioning index.

In conclusion, ecological refugia were characterized as a proxy for habitat heterogeneity and assessed for their capacity to support biodiversity, enhance beneficial ecosystem services, and benefit food production on three farms in Montana. A combination of field surveys, precision agriculture technology and remote sensing data was used to quantify the tradeoff between the conservation benefits and food production impacts that refugia provided. Contrary to common assumptions, native plant species richness was highest in the crop field and lowest in the refugia, indicating that ecological refugia were not sources of weedy species but hosts of native plant diversity. Plant diversity declined significantly with distance from naturally occurring refugia into the crop field and from a newly created refuge in the third year of data collection.  Insect diversity declined significantly with distance from refugia for all refugia. In addition, fields with naturally occurring refugia had a higher abundance of beneficial insects while the created refuge and control field without naturally occurring refugia had a significantly higher abundance of pest insects. Seed trap data indicated that beneficial weed seed predation services were higher on organic farms while crop seed predation disservices were higher on a conventionally managed farm. Crop yield declined significantly with distance from a naturally occurring refugia and increased significantly with distance from a created refuge. A variable importance plot indicated that distance from refuge was the most importance predictor of yield in a random forest model and that incorporating distance from refuge in the model raised the R-squared value from 0.57 to 0.70. Lastly, grain protein concentration had an inverse relationship with grain moisture content, though the relationship between proximity to an ecological refuge, moisture retention, nitrogen uptake and grain protein formation remained uncertain. Grain phenolic compounds tended to decrease with distance from the refuge, signaling that crops proximal to ecological refugia may undergo less stress than plants growing far from them. In addition, plants grown near ecological refugia contained a higher concentration of phenolic compounds, which contain antioxidants and tend to have positive human health outcomes.

Future implementation of ecological refugia as an on-farm conservation practice will depend on site-specific evidence that refugia can maintain or increase crop yields, decrease inputs, increase beneficial biodiversity, enhance ecosystem services, and increase farmers' net return. To do so, future research must clarify the relationship between on-farm biodiversity and habitat fragments at multiple scales, test the association between on-farm biodiversity and specific ecosystem services, and explore ways to increase affordability and application of on-farm technology like yield monitors that can provide farm-specific data to aid decision making. The adoption of ecological refugia as an on-farm conservation strategy provides a novel opportunity to optimize conservation and food production objectives at the agroecosystem scale.