2.3 Remotely-sensed data as model predictors
As a cloud-based platform, Google Earth Engine (GEE) provides easy
access to an extensive catalog of satellite imagery and other geospatial
data for scientific, business and government users (Gorelick et al.,
2017). We obtained a combination of topographic, climatic, and
vegetation derived variables with pixel sizes of 1000 m (Supplementary
Table S1) for the period of 2009 to 2014 from GEE to assemble a
nation-wide geospatial dataset to use as predictors in tree height and
tree density models.
Datasets included WorldClim V1; a set of bioclimatic variables derived
from the monthly temperature and rainfall (Hijmans, 2005); time-series
analysis of Landsat images from the Hansen Global Forest Change v1.8
(2000-2020) dataset (Hansen et al., 2013); 4-day composite dataset from
Moderate Resolution Imaging Spectro-radiometer (MODIS) sensors with
fraction of photosynthetic active radiation and leaf area index at 500-m
resolution (Myneni, Ranga et al., 2015) and the Advanced Spaceborne
Thermal Emission and Reflection Radiometer Global Emissivity Database
(2000-2008) (Hulley et al., 2009, 2012, 2015; Hulley & Hook, 2008,
2009, 2011; NASA JPL, 2014) (Supporting Information, Table S1). All
covariates were resampled to 1000 m. The resampling was done with
conventional bilinear interpolation as implemented in GEE. Data
available from Zenodo under the name “Nationwide geospatial dataset of
environmental covariates at 1km resolution in Mexico”
(https://doi.org/10.5281/zenodo.7130164) (Barreras & Guevara, 2022).
We reduced the number of potential predictor layers to 6, through a
culling process guided by an analysis of the correlation between each
potential predictor and the target variables (tree height and density).
We ranked the variables based on the magnitude of their correlation
coefficients. The intent of this data reduction step was to improve the
efficiency of our modeling framework. These univariate correlation
results are only included to give a sense of the directionality of the
relationships with target variables, but do not suggest causality.