Application of optimal data-based binning method to spatial analysis of ecological datasets

TL;DR

This paper evaluates a data-driven binning method for spatial analysis of ecological datasets, demonstrating its effectiveness in identifying spatial features and outperforming kernel methods in ecological pattern detection.

Contribution

It introduces and tests Knuth's optimal bin size rule for spatial data analysis, showing its advantages over traditional kernel methods in ecological applications.

Findings

Better detection of spatial features like anisotropy and clustering.

Outperforms kernel methods in estimating process intensity.

Effectively reveals ecological spatial structures.

Abstract

Investigation of highly structured data sets to unveil statistical regularities is of major importance in complex system research. The first step is to choose the scale at which to observe the process, the most informative scale being the one that includes the important features while disregarding noisy details in the data. In the investigation of spatial patterns, the optimal scale defines the optimal bin size of the histogram in which to visualize the empirical density of the pattern. In this paper we investigate a method proposed recently by K.~H.~Knuth to find the optimal bin size of an histogram as a tool for statistical analysis of spatial point processes. We test it through numerical simulations on various spatial processes which are of interest in ecology. We show that Knuth optimal bin size rule reducing noisy fluctuations performs better than standard kernel methods to infer the intensity of the underlying process. Moreover it can be used to highlight relevant spatial characteristics of the underlying distribution such as space anisotropy and clusterization. We apply these findings to analyse cluster-like structures in plants' arrangement of Barro Colorado Island rainforest.