Leveraging Google Earth Engine to estimate foliar C: N ratio in an African savannah rangeland using Sentinel 2 data

Abstract

Rangelands are important fodder for livestock and wildlife, and provide a range of ecosystem services to the environment. Foliar nutrients such as nitrogen, carbon, and plant pigments such as chlorophyll can be used as indicators of rangeland stress, and play a vital role in determining their health and productivity. The C:N ratio is a key factor in regulating nutrient utilization efficiency and productivity in plants. Understanding the C:N ratio in rangelands could therefore help herders understand the nutrient limitations, and herbivores distribution to facilitate strategic grazing plans and management. Therefore, there is a need for spatially accurate and up-to-date information on C:N ratio to understand and monitor rangeland health for proactive rangeland management. Remote sensing approaches are spatially explicit, cost-effective, and efficient in monitoring foliar nutrient ratio in rangelands. Whereas, the new generation and advanced Sentinel 2 multispectral sensor has the potential to monitor vegetation health, the strength of its spectral settings in relation to predicting the C:N ratio in rangelands remains largely unexplored. Advanced and freely available Sentinel 2 multispectral sensor (MSI) with specialized red edge bands offer unprecedented opportunities in mapping and monitoring rangeland nutrients. Hence, this study examined the prospect of combined Sentinel-2 (MSI) spectral bands and vegetation indices, and the random forest algorithm to map the C: N ratio within a rangeland. To determine the C:N ratio distribution, the Random Forest and the Boruta variable selection were employed to assess the performance of the combined Sentinel 2 spectral bands and vegetation indices models. Results show an estimated accuracy R2 of 81 and 74, with RMSE of 2.38 and 2.68 for calibration and validation datasets of the C:N ratio model established by combining the spectral bands and vegetation indices. The random forest variable selection model indicates that the red edge bands, and near-infrared were the most valuable in predicting the C:N ratio. The red edge and near-infrared (Inverted Red-edge Chlorophyll Index) and near-infrared and red band (Enhanced Vegetation Index) vegetation indices were important predictor variables for estimating the C:N ratio. This study demonstrates the prospects and value of mapping the geographic distribution of the C:N ratio in rangelands using high spatial resolution Sentinel 2 MSI. This information will not only help determine nutrient deficiencies in rangelands but will also provide informed recommendation in mitigating landscape degeneration to allow for rangeland regeneration.