A grid of 2.5 m 2.5 m was superimposed on 80 coffee agroforestry farms and five plots were randomly selected from each resulting into 400 plots in both coffee regions. The soil
samples were collected using a soil auger with 0.083 m diameter at two depths; 0–15 cm and 15–30 cm. Three soil cores within each plot were bulked by depth to obtain a representative sample for the plot. After collection, soil samples weighing 1.5 kg were transported to the laboratory, air dried and run through a 2 mm sieve to remove stones and root fragments. The soil carbon concentrations were analyzed by the Walkley-Black method and Nitrogen by Kjedahl procedure. Available Potassium was measured using flame photometer, phosphorous using extraction (Mehlich-3 solution) and calorimetric procedures and pH by a pH meter using 1:2.5 soil water extract. Bulk density was determined by collecting eighty samples of undisturbed soil from the study sites at two different soil depths using a core sampler. The samples were weighed to obtain the wet weight and later oven dried at a temperature of 105 C for two days to constant weight. Bulk density was calculated using the volume of the inner core sampler and the oven dry weight of the soil. All soil analysis was done at the National Agricultural Research Laboratories, Kawanda, Uganda.
2.4. Statistical analysis
The data were analysed using exploratory statistics (i.e., line graphs, bar graphs, and summary statistics) to assess the differences and the trends in SOC among the two coffee species. The SOC that were collected at two depths on the same plot were expected to be correlated, hence resulting into a repeated measure. Mixed model with repeated measurements was used to account for the correlation between depths; assess whether there were differences in SOC stored under
the two coffee species and also assess whether the incorporation of trees in coffee farms stored more organic carbon than monoculture farms. The random factors included the farmers and the fixed effects included the type of coffee, type of trees present on farm and soil depths. Under the mixed model the autoregressive covariance structure was used since the data were equally spaced in terms of depth. Soil bulk density data were analysed using the descriptive statistics and mixed model analysis, specifying autoregressive covariance structure. Before specifying
the autoregressive covariance structure, and to account for the correlation due to repeated measurements in space (depth),different covariance structures [Compound symmetry (CS), 1st
order autoregressive (AR (1)), and unstructured (UN)], were fitted to SOC data. Covariance structures were assessed using Akaike’s Information Criterion (AIC) and Schwarz Bayesian Information Criterion (BIC). The covariance structures with small AIC and BIC was selected which is 1st order autoregressive (AR (1))