Modelling spatial non-stationarity of Oreochromis karongae in South-East Arm of Lake Malawi

Fisheries researchers are faced with the challenging task of studying complex patterns and processes in aquatic resources. Analysis of such patterns is mostly performedundertheassumptionthatecologicalrelationshipsdonotvarywithin management areas (i.e. assuming spatially stationary processes). This assumptionwasquestionedbystudyingthedistributionofatargetfishpopulation,(Oreochromis karongae) commonly known as Chambo from South East Arm (SEA) of Lake Malawi where it is in abundance. Presence/absence of Chambo is not equally distributed as factors like depth or distance from shore line to where fish is caught are not the same in the whole of SEA.Survey data from June 1999 and October 2007 were used firstly in the comparison of fish abundance and only data from 2007 was used in the modeling of spatial distribution. Global logistic regression (GLR), generalized additive logistic model (GAM) (both global) and geographicallyweightedlogisticregression(GWR),alocalmodelingtechnique, were run to explore the best model that can explain spatial non-stationarity and how they affect fish distribution. Akaike Information Criterion (AIC) was used for best model selection depending on the lowest possible deviance value by comparison with other models. The best model was used in further analysis in mapping the model coefficients. Results from the global model on abundance indicate that there was less likelihood for finding Chambo in 1999 of 12.6% as compared to 2007. Results from the GWR (AIC = 18.62) model explained significantly morevariabilitythantheglobalmodelsGLR(AIC=40.84)andGAM (AIC = 40.22). Adjusted R2 explained 62.8% in GWR against 41.4% for GAM model. The significant local parameter estimates and t-values for depth were mapped and they provided a visual of their non-stationarity and reduction in the spatial autocorrelation of its model residuals.