Analysis of spatial variation of soil fertility gradients in Vihiga and Siaya districts of Western Kenya using geostatistical techniques
In western Kenya, several soil fertility management technologies have been developed in specific benchmark areas and then recommended to the rest of the farmers. Adoption of such technologies has been minimal at best, and, one of the reasons given for this low rate of adoption is that they did not take into consideration the existing spatial variations in biophysical and socio-economic conditions within which the local smallholder farmers operate. Against this background, a study was carried out to quantify the variability of soil fertility at different spatial scales and formulate domains for better targeting of soil fertility management recommendations. Farms were selected using a hierarchical Y-frame sampling design and in each farm information on the main biophysical factors collected. Field measurements, observations and sampling were used to collect data on the biophysical conditions, while participatory rural appraisal (PRA) was used to collect socio-economic data. All fields in each farm were characterised and top soil samples collected at a depth of 020 cm. All the sample collection points were georeferenced using a GPS system. Exploratory data analysis techniques were used to assess the effects of biophysical and socio-economic parameters on soil fertility. Geostatistical techniques of semivariography and kriging were used to explore the spatial structure of soil fertility gradients. Mixed effects modelling was used to confirm relationships, while accounting for spatial correlation structures, and understanding the variance of predicted soil organic C at different spatial scales. Predicted soil organic C was found to be spatially correlated and the spatial structure was modelled using experimental semivariograms fitted with spherical, exponential and ratio quadratic models. At the Y-level, using the exponential semivariogram model, spatial structures ranged from weak in Y3 (nugget/sill ratio > 0.75), moderate in Y2, Y5, Y7, Y8 and Y9 (nugget/sill ratio 0.25 < r < 0.75) to strong in Y 1 and Y4 (nugget/sill ratio < 0.25). On average all the three variogram models gave a nugget/sill ratio of between 0.5-0.6 indicating moderate spatial correlation. The maximum range at which this spatial structure can be reliably predicted is up to 60 m beyond which correlation errors increase significantly. All the three model variogram estimates had high nugget variances which imply that the micro-scale variation (i.e. variation below the minimum sampling interval) was large. Analysis of the estimated variance components showed that the field (residual) effect accounted for the greatest percentage (62.5%) of the variation associated with random effects. After accounting for spatial variability all the other measured parameters (fixed effects) failed to explain the large local variability, thus, posing a challenge to making soil fertility management recommendations. Future soil fertility management strategies in western Kenya should target at explaining the large spatial variability of soil fertility within the smallholder farms.