Exploring the use unmanned aerial vehicle rgb data for crop monitoring and mapping within a smallholder setting: a case study in Swayimane.

Abstract

Smallholder farms within Sub-Saharan Africa (SSA) are the backbone of agricultural systems due to their significant contribution to food production, thus rendering them essential for enhancing food security across the region. Despite this, their limited financial capacity, lack of resources and access to technology and information pose significant challenges for optimal agricultural production. The detrimental consequences of climate change and population expansion further impede their capacity to keep up with food demands. Advancements in precision agriculture (PA) such as the use of unmanned aerial vehicles (UAVs) can provide, near real-time data collection and adequate spatiotemporal resolution for smallholder heterogeneous farms thereby enabling informed and customised agriculture management to optimise agricultural productivity and resource use. This study investigates the potential of UAV-RGB data as a reliable and cost-effective solution to facilitate PA in smallholder farms. Specifically, it assessed UAV RGB data for land use classification and evaluated its effectiveness in estimating Leaf Area Index (LAI). While various sensors such as multi-spectral and hyperspectral sensors, offer significant spectral depth, their high costs limit the applicability for smallholder farmers. In contrast, UAV-RGB sensors are more affordable, promoting wider adoption. These sensors coupled with machine learning algorithms within cloud computing environments are more commonly appearing as accurate alternatives, particularly for processing large complex agricultural remote sensing datasets. Subsequently, this study utilizes machine learning classification approaches, comparing the two commonly used UAV multi-spectral and RGB sensor data for cropland mapping and crop monitoring. The Random Forest (RF) classifier effectively classified agricultural land with UAV-RGB data, achieving an area under the curve receiver operating characteristic (AUC-ROC) value of 0.75, while the UAV multi-spectral data yielded a marginally higher AUC-ROC of 0.77. For crop monitoring, we assessed LAI as a key growth metric, where the RF ensemble produced UAV-RGB LAI predictions with a root mean square error (RMSE), mean absolute error (MAE) and R-squared (R2) of 0.45, 0.31 and 0.73 respectively, which was less accurate but still reliable when compared to the UAV multi-spectral predictions (RMSE = 0.37, R2 = 0.81 and MAE = 0.24) The findings underscore the effectiveness of UAV-RGB data as a low-cost alternative that enhances the accessibility for smallholder farms, promoting the widespread adoption of precision agriculture practices. By enabling the accurate classification of agricultural land and monitoring of crop growth through reliable LAI predictions, this technology facilitates tailored solutions, improved decision-making and resource management, ultimately optimising agricultural practices and productivity. These advancements hold significant implications for developing nations such as SSA where smallholder farming systems are vital for sustaining food production which strengthens food security thereby resulting in a domino effect on various socio-economic factors.