Abstract:
Nutrient limitations may impact the ecosystem services the savanna biome provides. It may lead to degradation and, consequently, reduce the grazing capacity of the savannas if the necessary control measures are not implemented in time. The key indicator of the growth-limiting nutrients is the Nitrogen to Phosphorus (N:P) ratio. Grass foliar phosphorus content had rarely been investigated in African savannas, especially with remote sensing. Hence, information on the distribution of nutrient limitation is very limited. This study aimed to develop a Sentinel-2-based N:P predicting model and map the spatiotemporal variations of the N:P ratio in the Kruger National Park (KNP) area in the Northern part of the South African savanna biome. This was achieved by simulating the Analytical Spectral Device (ASD) reflectance data from 49 sampling points to Sentinel-2 MultiSpectral Instrument (MSI) configuration dataset. Laboratory-based chemical analysis was conducted to extract the concentrations of N and P from the grass samples. Partial least squares regression (PLSR) and random forest regression (RFR) techniques were used to develop the N:P prediction models from the simulated Sentinel-2 datasets. Results show that the best predicting RFR model explained over 80% of N:P variability with the lowest relative root mean square error (RRMSE) of 14%, with a p-value of less than 0.05. The optimal-predicting model was used to map the distribution of nutrient limitation using Sentinel-2 images across KNP and surroundings. Different parts of the KNP area are either N-limited or co-limited. The observed variations may result from varying environmental factors and anthropogenic activities. The Sentinel-2 N:P ratio estimation accuracies were then compared to the ratio of N:P of data from commercial multispectral (RapidEye and WorldView-2) and hyperspectral (Hyperion and EnMap) sensors. There is no vast difference between the estimation accuracy of these commercial sensors and that of the freely available Sentinel-2 when using RFR. However, when using PLSR, Sentinel-2 produced improved N:P ratio estimation accuracy than the commercial sensors with the highest R2 value of 0.66 and an RRMSE of 20.696%. This makes Sentinel-2 a cost-effective means for estimating nutrient limitation in a heterogeneous savanna landscape. This study provides decision-makers with a cost-effective tool for managing, sustaining, and restoring the savanna biome. The inclusion of textural information is recommended for future research.
