Leveraging of hyperspectral remote sensing on estimating biomass yield of Moringa oleifera Lam medicinal plant. South African Journal of Botany 140, 37–49. https://doi.org/10.1016/j.sajb.2021.1003.1035.

Abstract

Moringa oleifera Lam. is a functional plant considered to be high in nutrients as well as medicinal properties, largely utilised in most of the developing countries. Therefore, early prediction of M. oleifera biomass yield is a valuable pre- and post-harvest planning strategy for ensuring a reliable supply of the plant products and for marketing purposes. Subsequently, the objective of the current study was to explore the potential use of hyperspectral data in predicting biomass yield of different cultivars of M. oleifera. Canopy hyperspectral data were collected on five M. oleifera cultivars when they were one month and two months old using a hand-held spectroradiometer. The M. oleifera plants were harvested after two months and their leaves as well as stems were separated and immediately (i.e. within a minute) weighed (g plant −1). First-order derivative was used to transform the reflectance spectra and analysis of variance (ANOVA) as well as random forest (RF) regression and classification algorithm were used to analyse the data. The results showed that the first-order spectra of the five cultivars were significantly different (p ≤ 0.05) from each other in most portions of the electromagnetic spectrum. Furthermore, the results indicated that the studied M. oleifera cultivars can be discriminated from each other using their first-order derivative of reflectance and RF classifier. RF regression models developed to predict the biomass yield of a two-month old M. oleifera crop were more accurate compared to the models developed when the plant was month old. The relative root mean square error of validation (RRMSEV) for biomass models of two-months old plants ranged between 23 and 37%, while the RRMSEV for biomass models of one-month-old plants ranged between 30 and 59%. Cultivar Tanzania obtained the most accurate whole plant biomass yield prediction model with a RRMSEV of 23.20%. When the data aggregated across the five cultivars to develop a universal model, the results showed R2 = 0.59: RMSEV = 6.47 g plant−1; RRMSEV =29.42% for the whole plant biomass yield, R2 = 0.52, RMSEV = 4.78 g plant−1RRMSEV =29.51% for the leaves biomass yield and R2 = 0.42; RMSEV = 2.17 g plant−1 RRMSEV = 37.35% for the stem biomass yield. Based on current results, the study infers that combined spectral data across different cultivars may not be useful in estimating M. oleifera biomass yield. The results underscore the use of hyperspectral data as a quick non-destructive means to measure the yield of M. oleifera medicinal plant. Furthermore, the present study gives an insight into the potential of hyperspectral data to be extrapolated to larger scales for prediction of biomass yield of medicinal plants using spaceborne and airborne acquired data.