Use of geographically weighted regression to enhance the spatial features of forest attribute maps

Abstract

Geographically weighted regression (GWR) procedures can be adapted to enhance the spatial features of low spatial resolution maps based on higher resolution remotely sensed imagery. This operation relies on the assumption that the GWR models developed at low resolution can be proficiently applied to higher resolution data. An example of such an application is presented for downscaling a forest growing stock map which has been recently produced over the Italian national territory. GWR was applied to a Landsat Thematic Mapper image of Tuscany (Central Italy) for downscaling the growing stock predictions from a 1-km to a 100-m resolution. The accuracy of the experiment was assessed versus the measurements of a regional forest inventory. The results obtained indicate that GWR can enhance the spatial features of the original map depending on the spatially variable correlation existing between the forest attribute and the ancillary data used. A final ecosystem modeling exercise demonstrates the utility of the spatially enhanced growing stock predictions to drive the simulation of the main forest processes.

Spatial variation of earthworm communities in the motorway proximity

“…Environmental parameters derived from remote sensing data, as a comprehensive representation of climate and environmental factors, are used in large-scale biodiversity studies (Liu et al, 2019). Satellite remote sensing methods are a valuable alternative source of information on the characteristics of forest ecosystems at different spatial and time scales (Maselli et al, 2014;Furukawa et al, 2020). Equally effective is the use of remote sensing data for the analysis of flora species diversity (Coops et al, 2019;Damtew et al, 2021).…”

A novel approach to produce NDVI image series with enhanced spatial properties

“…The current iterative algorithm has provided satisfactory results, improving the spatial properties of the LR images in practically all cases. Other algorithms could be envisaged and tested, for example applying to the HR imagery classification or regression techniques trained on the LR abundance images [Maselli et al, 2014]. The final results are affected by the temporal proximity of the base and synthetic HR images but are mainly dependent on the information content of the HR data, which varies during the growing season.…”