Similar to demeaning of the dependent variable, this method uses only the variation in the data that is not explained by differences in pesti- evaluate if crop diversity, field size, or cropland extent drives in- cide use between years, regions, or crop types . In contrast, the pooled OLS pendent on the classification and taxonomic level at which diversity model clumps all observations together and does not distinguish is measured. More taxonomically similar crops may be expected to unobservable differences in crops, years, or regions. However, share more pests, yet taxonomically similar crops can be used for crops differ dramatically in average insecticide use , and very different products that are as- failing to control for these differences may bias the statistical es- sociated with different levels of pesticide use and different timing of timation. In other words, the effects observed may be driven by planting and harvesting. We therefore calculate diversity at different differences in crop types in highly diverse versus less diverse areas, taxonomic levels to understand better which, if any, is most relevant not by diversity perse. To control for the possibility that farmers for pest control decisions. Finally, we probe crop-specific relation- plant low-insecticide crops in highly diverse crop landscapes, we ships,40 litre pot focusing on almonds, grapes, oranges, pistachios, carrots, and included crop type dummy variables to com- wine grapes, which account for over 80% of insecticide use in Kern pare how insecticide use varies with cropland diversity, extent, and County. For each of these crops, we again evaluate the influence of field size for a given crop.
We also tested models with diversity, field size, and cropland extent on the magnitude of regional fixed effects to account for time-invariant pesticide applications. characteristics, such as soil quality or cultural norms, that are shared by all fields in a region. Here, region was defined by the Results 93-km2 Public Land Survey Township. Annual variability in sec.In general, we find that increasing crop diversity reduces inticides , diversity, or weather could also be important determinants of the relationship insecticide use. These relationships, as well as the relationship between landscape characteristics and insecticides. As such, we between cropland extent and insecticide use, are strongly included year dummy variables in both the crop and region fixed enced by crop type. effects models.Across the different all-crop models, the estimated effect of diversity decreased with taxonomic level. The effect of diversity calculated at the species or genus level had an estimated relationship roughly twice the magnitude as diversity calculated at the family level . For diversity calculated at the species level, an increase in diversity of 1 standard deviation resulted in reduced insecticide use of about −5.0 kg/ha, −3.0 kg/ha, and −2.1 kg/ha for the pooled OLS, region, and year fixed effects model and for the crop and year fixed effects model, respectively. Average field level insecticide use across all crops and years is 16.1 kg/ha; thus, a 2.1-kg/ha reduction is equivalent to a 13% decrease in average use. For all models, the coefficients for diversity calculated at genus and commodity followed a similar pattern to the coefficients for diversity calculated at the species level. In contrast, the larger aggregations of family and agricultural class had diversity coefficients of smaller magnitude relative to species . For cropland extent, defined as the proportion of area in agriculture within a 2500-m radius of the focal field, increasing extent was non-significant in all models, with coefficients smaller than 1 kg/ha in magnitude .
In contrast, field size led to a consistent increase in insecticide use regardless of statistical approach, with an increase in field size of 1 SD resulting in an ∼1- to 2-kg/ha increase in insecticides . To explore the heterogeneity over space, we evaluated diversity and cropland extent in five concentric circles of 500-m distances from the focal field, creating five annuli at distances of 0–500 m, 500–1,000 m, 1,000–1,500 m, 1,500–2,000 m, and 2,000–2,500 m. We again included crop and year fixed effects. The average field size was 33 ha; thus, the smallest annuli was about 2.5-fold greater in area than the average field and contained an average of about three surrounding fields, whereas the largest annuli was nearly 60-fold greater in area and contained an average of 21 surrounding fields. We find an important distance component to our results. Cropland extent significantly decreases insecticide use by 1 kg/ha in the 0- to 500-m annuli, yet trends toward a positive and marginally not significant relationship of similar magnitude at 2,000–2,500 m. Crop diversity hints at a nonlinear pattern over space, with diversity decreasing insecticide use at 500 m and 1,500 m, but not at 1,000 m. After 1,500 m, the relationship is smaller, it is marginally not significant at 2,000 m, and it is non-significant at 2,500 m . The relationship between landscape simplification and insecticides could alternatively be affected by economic changes that ultimately drive the simplification. For example, larger farms could lead to both landscape simplification and increased use of insecticides because a single owner decides the area to plant, what to plant, and treatment levels. To test the impact of farm structure on insecticide use, we tested models that included a covariate for the proportion of the surrounding area owned by the owner of the focal field. However, including this variable did not change the patterns we observed for crop diversity, field size, or cropland extent, indicating that ownership was not driving our results. Other economic factors, such as crop and pesticide prices, certainly also affect pesticide use; however, because the relevant prices become observable only after the land use decision, they add to the noise but do not bias the estimates. Nevertheless, we also evaluated crop-by-year fixed effects models to account for year shocks that may be unique to individual crops .
Doing so did not change the patterns observed in the crop and year fixed effects model . To account for the possibility that different crops had different relationships between diversity and insecticides, we reran the models for the top six crops representing about 83% of insecticides used. We again included year fixed effects, in this case,to account for year-specific shocks shared by all fields of the focal crop . As anticipated, there was substantial heterogeneity across the different crops. The largest negative and significant relationship between diversity and insecticides was observed for grapes, where an increase in species diversity of 1 SD reduced insecticide use by nearly 8 kg/ha . Diversity surrounding almonds and pistachios also significantly reduced insecticide use on those crops by ∼2 kg/ha and ∼4 kg/ha, respectively. For oranges and wine grapes, the diversity relationship was negative, but not statistically significant, and less than 2 kg/ha in magnitude, and for carrots, the relationship was near zero. Surprisingly, the crops for which diversity had little effect on insecticides were also the crops for which surrounding cropland extent led to an increase in insecticides. Here, wine grapes and oranges were the only crops with nearly significant increases in insecticides as a result of increasing cropland extent . Thus, for a given level of crop diversity, an increase of 1 SD in cropland in the 2500-m radius area surrounding wine grapes or oranges resulted in an additional 6–8 kg/ha of insecticides. However, for crops such as grapes and pistachios,collection drainage the coefficients were near zero and non-significant. Carrot was the only crop of the top six to have a negative coefficient , although this negative coefficient was not significant. Interestingly, despite wine grapes having the largest increase in insecticide as a result of surrounding cropland, they had a sizeable significant decrease in insecticides as field size increased. For all other crops, field size led to an increase in insecticide use, which was significant for almonds and pistachios . Finally, to confirm that neither pesticide outliers nor decisions on model functional form were driving our results, we repeated the main analysis removing the top 1% of pesticide use and evaluated the effect of a logged dependent variable specification. Our results were robust to both modifications .Landscape complexity has long been shown to increase natural enemy abundance, and, as such, it has been considered a mechanism of ecological pest control.
However, how different aspects of landscape complexity function with respect to chemical pest control has remained poorly understood. In part, the lack of conclusive evidence is due to the complexity of ecological and economic factors, as well as crop-specific heterogeneity, annual conditions, and/or regional characteristics . Using a combination of crop-specific, region-specific, and year-specific controls, we parsed apart the individual effects of crop diversity, cropland extent, and field size on insecticide use. In general, we find diversity reduces insecticide use. However, interestingly, including crop type fixed effects modifies the diversity–insecticides relationship substantially, indicating that crop types with lower insecticide requirements are planted in areas of higher crop diversity. This correlation may explain the ambiguous literature regarding diversity. For example, our results suggest that in a random sample of fields within an agricultural landscape, crop diversity would be correlated with lower insecticide use because high-diversity areas are composed of low-insecticide use crops.Here, we observed a larger effect of diversity at the species or commodity scale than at larger aggregations. At first, this result seems surprising. The suggested mechanism for the impact of landscape simplification on insecticide use is its impact on habitat suitability for pests and their natural enemies. This suitability is largely determined by the suitability of the crop as a habitat, but also by the timing of the planting, pest control, and harvesting decisions that may interrupt species interactions and population demographics. Even within one species, agricultural practices can be very different depending on the agricultural product, such as the difference between wine and table grapes. We suggest here that diversity at lower taxonomic levels better captures the biological and management differences relevant to pest control. With respect to the spatial scale, we observed that diversity within 2,500 m significantly decreases insecticide use, yet this effect is heterogeneous over varying distances. A large effect of surrounding diversity was observed in the nearest annuli, 0–500 m from the focal crop. Nearby fields would be expected to have a stronger influence on insect and enemy spillover than farther away fields, holding the diversity in other annuli constant. However, the relationship between diversity and insecticides does not appear to change linearly with distance. Rather, our results tentatively suggest the existence of an inverted U-shaped relationship, where diversity at 1,000–500 m is also statistically important with similar magnitude, whereas further annuli tend to be less so. More robust spatial analysis is necessary to confirm this result. A nonlinear response of pests to distance has been noted elsewhere and may reflect dispersal characteristics or patch use of different pests, although the existence and exact underlying causes of this pattern deserve further attention. Evaluating individual crops hints at underlying mechanisms for the relationship between landscape characteristics and insecticides. We find large differences in the slope of the relationship between diversity and insecticides among the top six insecticide-use crops, with almonds, grapes, and pistachios having large decreases in insecticides in highly diverse landscapes, whereas oranges, carrots, and wine grapes show a smaller and non-significant response. The underlying mechanisms could be related to the specialist or generalist nature of the specific crop pests. For example, a crop for which the most damaging pest was a specialist would benefit from surrounding diversity, because diversity would dilute available host crops. Thus, we would expect diversity to reduce insecticides, whereas cropland extent would have little effect. In contrast, if the pest were a generalist, diversity would be expected to have little effect, but extent would be important because other crops would function as suitable hosts. Indeed, we find some evidence of this trade off, where crops with strong responses of insecticides to diversity have weak responses to extent, and vice versa. Beyond diversity, we find surrounding cropland extent decreases insecticide use on the focal field by ∼1 kg/ha when situated nearby yet increases insecticides by ∼1 kg/ha when located at larger distances . This spatial trend is particularly interesting, given the ambiguous literature between cropland extent and insecticides .