Biome and campaign dependence limits transfer of airborne-lidar structure-productivity relationships across Brazilian woody vegetation

Authors:

Abstract
Lidar-derived canopy structure is increasingly used to infer ecosystem function, but whether structure-productivity relationships transfer across vegetation domains and independent survey campaigns remains uncertain. We combined wall-to-wall airborne lidar from 2,336 natural woody-vegetation tiles across Amazônia, Cerrado, and Mata Atlântica with year-matched MODIS gross and net primary productivity (GPP and NPP), precipitation, vapor pressure deficit, leaf area index, and land cover. Lidar canopy-top height was independently evaluated against field measurements from 80 matched Amazônia tiles (r = 0.726; site-cluster bootstrap 95% confidence interval, 0.53-0.84). Primary inference used standardized climate-adjusted models with campaign fixed effects and 50-km Conley spatial covariance, supported by residual spatial tests, 10-100 km aggregation, land-cover and point-count sensitivities, and campaign- and spatial-blocked prediction. In Amazônia, foliage height diversity, canopy cover, and height percentiles were positively associated with GPP (standardized slopes, 0.156-0.213). In Cerrado, only median canopy height had a significant negative slope (-0.093; 95% confidence interval, -0.163 to -0.024); the remaining structural slopes were indistinguishable from zero. Cerrado-Amazônia GPP slope contrasts were negative for all five metrics and persisted across spatial aggregation and sensitivity analyses. Median-height slopes were negative in both Cerrado campaigns, although only the larger campaign was individually significant. NPP relationships were weaker and inconsistent. Adding structure improved campaign-blocked GPP prediction by Delta R2 = 0.038 and spatially blocked prediction by 0.067-0.094, but the full model did not transfer reliably between the two Cerrado campaigns. These results show that lidar structure provides useful GPP information while its functional relationship is biome and campaign dependent. A true sign reversal is supported most clearly for median canopy height; other metrics primarily lose predictive sensitivity in Cerrado. Structural remote-sensing models should therefore be stratified by vegetation domain and validated using independent campaigns before regional transfer.

This publication has 17 references indexed in Scilit: