A generic composite measure of similarity between geospatial variables

Authors: LUI, Y - Seoul National University; KIM, K - Seoul National University; BERESFORD, R - New Zealand Institute For Crop & Food Research; Fleisher, David

Submitted to: Ecological Informatics
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 7/28/2020
Publication Date: 10/2/2020
Citation: Liu, Y., Kim, K.S., Beresford, R.M., Fleisher, D.H. 2020. A generic composite measure of similarity between geospatial variables. Ecological Informatics. https://doi.org/10.1016/j.ecoinf.2020.101169.
DOI: https://doi.org/10.1016/j.ecoinf.2020.101169

Interpretive Summary: Geographic maps are often used to help visualize important topics such as spatial variation in yield, rainfall, and soil quality. The maps help scientists understand where, when, and why, limitations to crop productivity occur. Sometimes, the connection between two or more variables, such as yield and soil quality, is needed to further to understand these limits. Spatial scales for these data can range across county, state or international boundaries. These data also vary temporally, with day, month, and year, adding to the complexity and difficulty of this process. Statistics, or metrics, are frequently used to simplify the way in which scientists can more easily identify these multi-variable relationships. Existing methods, however, are very limited in scope. A new geospatial metric was developed that improves on these other approaches. It was tested using datasets for gross primary production and rainfall over several countries. This new metric provide a more accurate way to quantify and understand relationships between crop yield and limiting production factors across many different countries, as well as linkages with monthly and annual variation. It will be of use for researchers who focus on agricultural models and ecological studies that address problems in food security.

Technical Abstract: The ability to compare spatiotemporal patterns of variables is often necessary for model evaluation and change detection in ecological studies. The statistics developed for image quality assessment, e.g. the structural similarity (SSIM) and composite similarity measure based on means, standard deviations and correlation coefficient (CMSC), have been recommended for comparing ecological variables. However, these statistics can only be used where there is a positive association between variables and the data are at the same scale. We propose a new approach, the generic composite similarity measure (GCSM), to overcome the limitations of existing statistics. A set of numerical experiments was performed to illustrate the properties of GCSM in comparison with SSIM and CMSC. Two case studies were conducted to examine the consistency between two sets of gross primary production (GPP) estimates and the association between GPP and precipitation. The GCSM had advantages over the SSIM and the CMSC, including a higher sensitivity and the ability to quantify negative association between two variables, for which the latter two cannot be used. In addition, with the application of fuzzy sets, it is able to capture the relative association between variables with different scales. Our results indicated that the GCSM provides a comprehensive measure of association between spatial, temporal or spatiotemporal patterns. It could be applied to research topics including crop and ecological modeling and limitations to yield potentials.