Application of deterministic and stochastic geo-statistical tools for analysing spatial patterns of fish density in a tropical monsoonal estuary

Abstract

In this paper, we compared the efficiency of advanced deterministic and stochastic geostatistical techniques to predict spatial patterns of fish density in the tropical monsoonal estuary, Zuari, using the following environmental descriptors: temperature, salinity, dissolved oxygen, transparency, and geographical coordinates. The methods applied in this study were multiple linear regression (MLR), Cubist, support vector regression (SVR), random forest regression (RF), universal kriging (UK), and regression kriging (RK). Fish abundance and environmental data were collected from September 2013 to August 2016 in 48 sampling stations distributed along the estuarine gradient. The ranking procedure of various regression methods showed that the Cubist model was the best performing model based on prediction accuracy in the development phase and prediction consistency in the validation phase. Latitude, temperature, salinity, and DO had a positive influence on fish abundance, while longitude and transparency showed negative impacts. This study offers scope for refining the employed currently models to predict spatial densities of fishes using a wide range of available biotic and abiotic variables, which will enable to develop an efficient management framework for tropical monsoonal estuaries.