Like an "espresso" but not like a "cappuccino": landscape metrics are useful for predicting coffee production at the farm level but not at the municipality level.

Coffee farms receive ecosystem services that rely on pollinators and pest predators. Landscape-scale processes regulate the flow of these biodiversity-based services. Consequently, the coffee farms' surrounding landscape impacts coffee production. This paper investigates how landscape structure can influence coffee production at different scales. We also evaluated the predictive utility of landscape metrics in a spatial (farm level) and aspatial approach (municipality level). We tested the effect of landscape structure on coffee production for 25 farms and 30 municipalities in southern Brazil. We used seven landscape metrics at landscape and class levels to measure the effect of landscape structure. At the farm level, we calculated metrics in five buffers from 1 to 5 km from the farm centroid to measure their scale of effect. We conducted a model selection using the generalized linear model (GLM) with a Gamma error distribution and inverse link function to evaluate the impact of landscape metrics on coffee production in both spatial and aspatial approaches. The landscape intensity index had a negative effect on coffee production (AICc = 375.59, p < 0.001). The native forest patch density (AICc = 390.14, p = 0.011) and landscape diversity (AICc = 391.18, p = 0.023) had a positive effect on production. All significant factors had effects at the farm level in the 2 km buffer but no effects at the municipality level. Our findings suggest that the landscape composition in the immediate surroundings of coffee farms helps predict production in a spatially explicit approach. However, these metrics cannot detect the impact of the landscape when analyzed in an aspatial approach. These findings highlight the importance of the landscape spatial structure, mainly the natural one, in the stability of coffee production. This study enhanced the knowledge of coffee production dependence on landscape-level processes. This advance can help to improve the sustainability of land use and better planning of agriculture, ensuring food and economic safety. Furthermore, our framework provides a method that can be useful to scrutinize any cropping system with census data that is either spatialized or not.