MODELING COFFEE YIELD IN OROMIA REGION: APPLICATION OF LINEAR MIXED EFFECT MODEL

Abstract

Coffee is an important agricultural commodity that is grown all around the world. In Ethiopia coffee is the main source of income. The objective of this study was to model coffee yield (Qu/Hect) in Oromia using exploratory data analysis (EDA) and a linear mixed effect model (LMM). The study was conducted in 50 district in Oromia, which were categorized into 10 zones, repetition. The data for this study was obtained from each zonal agricultural office. The AIC and BIC were used to determine which model better suited the data. Descriptive statistics shows that the lowest average coffee yield is 6.769785 (Qu/hect) were recorded at baseline (2005), while the highest average yields (7.164 Qu/hect) occurred at the conclusion of the study period (2021). The findings also revealed that Bale zone had the lowest average yields (5.993 Qu/hect), while Jimma zone had the highest average yields (8.001 Qu/hect). The variance and correlation pattern between measures of coffee yield within districts and districts nested inside locations were found to be best fit by AR (1) (zone). The estimated variance of random effect of (district’s) and location (zone) were 𝑉𝑎𝑟 (𝑏𝑜𝑖) = 0.021177 and 𝑉𝑎𝑟 (𝑏1𝑖) = 0.0030122, respectively. The outcomes of the linear mixed model demonstrated that, at a 5% level of significance, variables like soil witness, the quantity of phosphorus in the soil, humidity, soil moisture, annual temperature, and annual rainfall greatly affect the district's coffee yields. According to this study, the random effect and the grouping effects of districts nesting inside zones accounted for 84.14% of the overall variability in a district's coffee yield where as 11.96% of the total variances in the district's coffee yield were caused by the grouping effects of zone. Summary of goodness-of-fit statistics indicates that linear mixed model had better results than GLM model. The 𝑅2was 0.34 and RMSE was 3.87 for the mixed-effects model compared to 0.41 and 3.27 for 𝑅2 and RMSE respectively, for GLM model. The results imply that 𝑅2 increased by 20.5 % [(0.41– 0.34)/0.34) × 100] while RMSE reduced by 15.5% [((3.27− 3.87)/3.87) × 100] when mixed effects modelling approach was used to fit the same data fitted using GLM. Based on these conclusions, a comprehensive policy is required to maximize the potential of rain-fed agriculture in the face of the threats posed by climate change. Coffee modelers and managers in Oromia should consider statistical model that absorb variability in different location and correlation in order to estimate yields with higher precision and accuracy.