Fusing multidimensional hierarchical information into finer spatial landscape metrics.

Gang Fu,Juyi Xia,Junsheng Li,Wei Wang,Caiyun Zhao,Ming Cao,Nengwen Xiao,Yue Qi

doi:10.1002/ece3.8206

Abstract

One of the core issues of ecology is to understand the effects of landscape patterns on ecological processes. For this, we need to accurately capture changes in the fine landscape structures to avoid losing information about spatial heterogeneity. The landscape pattern indicators (LPIs) can characterize the spatial structures and give some information about landscape patterns. However, researches on LPIs had mainly focused on the horizontal structure of landscape patterns, while few studies addressed vertical relationships between the levels of hierarchical landscape structures. Thus, the ignorance of the vertical hierarchical relationships may cause serious biases and reduce LPIs' representational ability and accuracy. The hierarchy theory about the landscape pattern structures could notably reduce the loss of hierarchical information, and the information entropy could quantitatively describe the vertical status of landscape units. Therefore, we established a new multidimensional fusion method of LPIs based on hierarchy theory and information entropy. Here, we created a general fusion formula for commonly used simple LPIs based on two‐grade land use data (whose land use classification system contains two grades/levels) and derived 3 fusion landscape pattern indicators (FLIs) with a case study. The results show that the information about fine spatial structure is captured by the fusion method. The regions with the most differences between the FLIs and the traditional LPIs are those with the largest vertical structure such as the ecological ecotones, where vertical structure was ignored before. The FLIs have a finer spatial representational ability and accuracy, not only retaining the main trend information of first‐grade land use data, but also containing the internal detail information of second‐grade land use data. Capturing finer spatial information of landscape patterns should encourage the application of fusion method, which should be suitable for more LPIs or more dimensional data. And the increased accuracy of FLIs will improve ecological models that rely on finer spatial information.

Highlights

The research of ecology increasingly requires multidimensional spatial data across broad extents, it is necessary to capture their fine spatial structures to avoid losing of information on spatial heterogeneity (Graham et al, 2019)
We found that regions with the most differences between the fusion landscape pattern indicators (FLIs) and the traditional landscape pattern indicators (LPIs) were those with the largest vertical structure such as ecological ecotones, where their vertical structures were ignored before
We found that the FLIs' information mainly came from the G1LIs', because the regression coefficients in the equations illuminated that the information volumes of the G1LIs played a more significant role than that of G2LIs in explaining the information volumes of the FLIs (Table S5)

Summary

| INTRODUCTION

The research of ecology increasingly requires multidimensional spatial data across broad extents, it is necessary to capture their fine spatial structures to avoid losing of information on spatial heterogeneity (Graham et al, 2019). 2. Describing the configuration association of lower landscape patch classes within upper ones by information entropy theory, with the first-grade LULC as the target layer. In the study of landscape patterns, the information entropy can measure the degree of disorder of the system (Brunsell, 2010), which can be used to quantitatively describe the states of landscape patch classes between different levels with subordinate relationships (Wang & Zhao, 2019). The continuous gradient grid set of LPIs can be obtained by using the moving window approach, which is a common method in the GM framework to visualize the demonstration of FLIs. First, according to the calculation process of the moving window approach, we fused the G2 data in the window into G1. The complete code has been provided as a Python supplemental archive

| MATERIALS AND METHODS

| DISCUSSION