Mapping Drivers of Greenness: Spatial Variable Selection for MODIS Vegetation Indices

TL;DR

This work introduces a Bayesian spatially varying coefficient framework that combines tensor-product B-spline representations with a group lasso prior to jointly smooth spatial surfaces and screen predictors for environmental drivers of vegetation indices. By producing spatial significance maps and a spatial coverage probability, the method identifies predictors with widespread, credibly nonzero effects and discards negligible ones, enhancing interpretability. In simulations and a MODIS EVI application, BSGL achieves accurate prediction, strong discrimination between signal and noise, and ecologically meaningful insights about drivers like NIR/Red reflectance and productivity measures. The approach offers a scalable, open-source tool for spatially explicit variable selection with broad applicability to vegetation monitoring and ecosystem modeling.

Abstract

Understanding how environmental drivers relate to vegetation condition motivates spatially varying regression models, but estimating a separate coefficient surface for every predictor can yield noisy patterns and poor interpretability when many predictors are irrelevant. Motivated by MODIS vegetation index studies, we examine predictors from spectral bands, productivity and energy fluxes, observation geometry, and land surface characteristics. Because these relationships vary with canopy structure, climate, land use, and measurement conditions, methods should both model spatially varying effects and identify where predictors matter. We propose a spatially varying coefficient model where each coefficient surface uses a tensor product B-spline basis and a Bayesian group lasso prior on the basis coefficients. This prior induces predictor level shrinkage, pushing negligible effects toward zero while preserving spatial structure. Posterior inference uses Markov chain Monte Carlo and provides uncertainty quantification for each effect surface. We summarize retained effects with spatial significance maps that mark locations where the 95 percent posterior credible interval excludes zero, and we define a spatial coverage probability as the proportion of locations where the credible interval excludes zero. Simulations recover sparsity and achieve prediction. A MODIS application yields a parsimonious subset of predictors whose effect maps clarify dominant controls across landscapes.

Figures (4)

• Figure 1: Spatial coefficient estimates for signal variables $\beta_1$-$\beta_3$ as shown in top three panels and noise variable $\beta_5$ in bottom panel at $n=1000$, $m=5$. All methods capture the true spatial patterns for signal variables, with BSGL providing sharper boundary estimates. For the zero predictor $\beta_5$, BSGL correctly shrinks coefficients to near-zero across the domain while GGP-GAM and MGWR fit spurious spatial patterns.
• Figure 2: Spatial significance maps with 95% credible intervals. Solid points indicate locations where credible intervals exclude zero. Signal variables in the top panel achieve SCP exceeding 0.90. Noise variables in the bottom panel achieve SCP below 0.03.
• Figure 3: Spatial distribution of Enhanced Vegetation Index across the test region showing observed values, model predictions, and residuals.
• Figure 4: Spatial coverage probability maps for predictor variables. Dark blue regions indicate locations where 95% credible intervals exclude zero. Spectral reflectance (red, NIR, blue) and productivity variables (GPP, LE) exhibit high spatial significance across the study region, while observation geometry and land cover variables show limited or spatially localized patterns.