Spectral assessment of nutrient limitation in the savanna landscape: selection of spectral indices towards Sentinel-2 upscaling

Abstract

Nutrient limitations can significantly impact the ecosystem services provided by the savanna biome, potentially leading to degradation and reduced grazing capacity if not detected in time. A key indicator of growth-limiting nutrients is the Nitrogen to Phosphorus (N:P) ratio. However, grass foliar phosphorus content has rarely been studied in African savannas, especially using remote sensing approaches. As a result, there is limited information on the spatial distribution of nutrient limitations in these ecosystems. This study aimed to develop a Sentinel-2-based machine learning regression model to predict and map the distribution of the N:P ratio in the northern region of Kruger National Park (KNP), South Africa, which is dominated by the savanna rangeland biome. Fieldwork was conducted between 15 March and 30 April 2008 to collect grass samples and spectral data using an Analytical Spectral Device (ASD). The hyperspectral field data were then resampled to match the multispectral configuration of Sentinel-2 imagery. A Random Forest Regression (RFR) technique was applied to the simulated Sentinel-2 datasets to develop predictive models of the N:P ratio. Model accuracy was evaluated using the Root Mean Square Error (RMSE) Relative Root Mean Square Error (RRMSE), Percent Bias (PBIAS), and the coefficient of determination (R2). The results showed that vegetation indices (VIs), particularly the Normalized Difference Red Edge (NDRE) derived from Sentinel-2 bands B8 and B5, was optimal for estimating N:P ratio. This index explained over 80% of the N:P variability, with the lowest PBIAS of 0.02%. The best-performing model was used to map nutrient limitations across the study area using Sentinel-2 imagery. The spatial analysis indicated consistent nitrogen limitation and co-limitation across the investigated regions, with no evidence of phosphorus limitation. The high-accuracy models demonstrate the effectiveness of Sentinel-2 imagery for estimating nutrient limitations in heterogeneous savanna landscapes. This study offers a cost-effective, scalable tool for decision-makers involved in the management, sustainability, and restoration of the savanna biome. Future research should consider incorporating textural and environmental variables to enhance model performance and understanding of nutrient dynamics.