Vincent de Leijster

41 Agroecological management improves ecosystem services in almond orchards within one year 3 stored per cluster in 70% ethanol. After field sampling, individuals were identified based on morphotype to the level of sub-family, family or more detailed taxonomic level if possible. Then we classified the following three taxonomic groups as pollinators based on existing studies on pollinators of almonds: the superfamily Apoidea and the order Diptera (Klein et al., 2012; Ortiz-Sánchez and Tinaut, 1993; Saunders et al., 2013) (Appendix Table A3-3). 3.2.3.5 Statistical analyses Ecosystem service index To compare the different ecosystem services, we rescaled each ecosystem service indicator to range between 0.1–1 using the transformation proposed by Kearney et al. (2017): Where i is the ecosystem service indicator index, Y is the response index value of i, and max and min represent the maximum and minimum of i. For pest control we applied the reverse transformation of pest abundance to ‘pest absence’ to maintain a more-is-better scale. Therefore the ecosystem service indicator index is subtracted from 1.1, as proposed by Kearney et al. (2017). The values of all ecosystem service indicator indices were then averaged per ecosystem service and per treatment to obtain the so-called ‘single ecosystem service index’. Then, we averaged the data of all single ecosystem service indices to get an ‘overall ecosystem service index’ per treatment. The rescaled data was used in a principal component analysis and used to test the effect of treatment on ecosystem service index with Generalized Linear Mixed Models (GLMM). Generalized Linear Mixed Models (GLMM) To assess the effect of treatment on ecosystem service indicators (Table 3-2) we fitted GLMMs using maximum likelihood and a frequentist approach for hypothesis testing (Faraway, 2016). The indicators for ecosystem services were considered the response variables and treatment the fixed factor, while farm location was included as a random factor. We did not include an interaction between treatment and farm because of the small sample size. Additionally, we applied a GLMM on the ecosystem service index to test whether treatments differed significantly from each other. Therefore, we made a model with treatment as the fixed factor and both farm and the ecosystem service indicator as categorical random factors. We first tested if the assumptions of a Gaussian distribution in the response variables were met, by analyzing homoscedasticity and normal distribution of the residuals. If the model