I performed this analysis in Ghana during my capacity as a Database Administrator and GIS Specialist at the International Fertilzier Development Center (IFDC), to assist Samuel Bua, an intern who was doing his MSc project work titled Yield Response of Maize to Fertilizers in Ghana with objectives including analyzing the yield response of different fertilizer treatments on maize. The exploratory data analysis was done in ArcGIS whilst the spatial analysis was done in R programming environment.
The study area is Ghana as shown in the map. The points are field observations of grain yield secondary data obtained from research papers, and are grouped into 5 classes for visualization.
The spatial distribution was analyzed using the Average Nearest Neighbor Analysis in ArcMap which showed that the data points were clustered which indicated that interpolation of the grain yield values over Ghana could produce reasonable results.
The data points were further analyzed to determine whether there was the tendency that points that were close to each other had similar values as opposed to points that were far apart. This was analyzed using the Spatial Autocorrelation Analysis tool in ArcMap and the results showed that near points had similar values as indicated as clustered in the Spatial Autocorrelation Report chart.
The Spatial Structure was further analyzed to understand the extent of autocorrelation effect in the grain yield values using the Semivariogram tool in ArcMap which showed a rather short range of autocorrelation effect as shown in the semivariogram chart. This informed the decision to use IDW as the best interpolation technique since statistical prediction techniques such as Kriging requires a strong spatial structure.
Using the Inverse Distance Weight (IDW) function of the gstat package, a IDW interpolation was performed for all fertilizer treatment types with the results as shown in the map.
Using the Inverse Distance Weighted (IDW) function of the gstat package, a IDW interpolation was performed for zero fertilizer treatment types with the results as shown in the map.
A regression-kriging technique was explored using parameters including Climate (Annual Precipitation, Annual Transpiration), Remote Sensing (EVI, NDVI), Soil (Ca, K, Mg, P, Zn, N, OCS, Clay, Sand, Silt, pH, AWHC, RZDM, DEM) as covariates to regress with the maize yield as the dependent variable. Below is the grid created for covariate data extraction and kriging. The yellow dots are the locations of observed maize data. The procedure is to use Multiple Linear Regression (MLR) to predict over the grid, and then the residuals also predicted over the grid using Ordinary Kriging (OK). The final prediction at each grid location then becomes the addition of the MLR prediction + OK prediction
After fitting the Multiple Linear Regression model to the data, the residuals variogram was created and used for the residuals ordinary kriging.
The final prediction at each grid location then becomes the addition of the MLR prediction + OK prediction