Métodos de estimativas para biomassa radicular em floresta tropical sazonalmente seca no semiárido brasileiro

Abstract

Seasonally Dry Tropical Forests (FTSS), such as the Caatinga, play a relevant role in the global carbon cycle, with root biomass being a fundamental component both for subsurface carbon storage and for the physiological support of plants in environments with high water seasonality. Despite their ecological importance, quantifying root biomass still presents methodological challenges. In this context, this study aimed to evaluate and compare the performance of different approaches for estimating root biomass in a fallow Caatinga area in the state of Ceará. Three strategies were analyzed: allometric equations based on root morphological attributes, a Partial Least Squares Regression (PLSR) model obtained from hyperspectral data, and simple regression models based on vegetation indices derived from orbital images (MIO). The reference biomass was obtained through the destructive ingrowth core method, with collections carried out in distinct climatic periods. The results indicated that the allometric equations, when applied to independent data, showed intermediate coefficients of determination (R²pred=0.66) and relatively high prediction errors (RMSEP=2.39 Mg ha⁻¹). The PLSR model demonstrated potential for estimating root biomass when applied to independent data (R²pred=0.37; RMSEP=3.26 Mg ha⁻¹), highlighting the need for model improvement. Although the orbital image model, based on the VARI index, presented the best statistical indicators of fit and validation, with higher R² values (0.975) and lower prediction errors (RMSE=0.371 Mg ha⁻¹), the small sample size (n=6) limits the robustness of the results, requiring caution in their interpretation due to the risk of overfitting. It is concluded that, although all approaches show potential for estimating root biomass up to a depth of 30 cm in the Caatinga, each one has specific limitations that must be considered.