The potential for using remote sensing to quantify stress in and predict yield of sugarcane (Saccharun spp. hybrid)

Abstract

South Africa is the leading producer of sugarcane in Africa and one of the largest sugarcane producers in the world. Sugarcane is grown under a wide range of climatic, agronomic, and socio-economic conditions in the country. Stress factors such as water and nutrient deficiencies, and insect pests and diseases are among the most important factors affecting sugarcane production in the country. Monitoring of stress in sugarcane is therefore essential for assessing the consequences on yield and for taking action of their mitigation. The prediction of sugarcane yield, on the other hand is also a significant practice for making informed decisions for effective and sound crop planning and management efforts regarding e.g., milling schedules, marketing, pricing, and cash flows. In South Africa, the detection of stress factors such as nitrogen (N) deficiency and sugarcane thrips (Fulmekiola serrata Kobus) damage and infestation are made using traditional direct methods whereby leaf samples are collected from sugarcane fields and the appropriate laboratory analysis is then performed. These methods are regarded as being time-consuming, labour-intensive, costly, and can be biased as often they are not uniformly applied across sugarcane growing areas in the country. In this regard, the development of systematically organised geo-and time-referenced accurate methods that can detect sugarcane stress factors and predict yields are required. Remote sensing offers near-real-time, potentially inexpensive, quick and repetitive data that could be used for sugarcane monitoring. Processing techniques of such data have recently witnessed more development leading to more effective extraction of information. In this study the aim was to explore the potential use of remote sensing to quantify stress in and predict yield of sugarcane in South Africa. In the first part of this study, the potential use of hyperspectral remote sensing (i.e. with information on many, very fine, contiguous spectral bands) in estimating sugarcane leaf N concentration was examined. The results showed that sugarcane leaf N can be predicted at high accuracy using spectral data collected using a handheld spectroradiometer (ASD) under controlled laboratory and natural field conditions. These positive results prompted the need to test the use of canopy level hyperspectral data in predicting sugarcane leaf N concentration. Using narrow NDVI-based vegetation indices calculated from Hyperion data, sugarcane leaf N concentration could reliably be estimated. In the second part of this study, the focus was on whether leaf level hyperspectral data could detect sugarcane thrips damage and predict the incidence of the insect. The results indicated that specific wavelengths located in the visible region of the electromagnetic spectrum have the highest possibility of detecting sugarcane thrips damage. Thrips counts could also adequately be predicted for younger sugarcane crops (4–5 months). In the final part of this study, the ability of vegetation indices derived from multispectral data (Landsat TM and ETM+) in predicting sugarcane yield was investigated. The results demonstrated that sugarcane yield can be modelled with relatively small error, using a non-linear random forest regression algorithm. Overall, the study has demonstrated the potential of remote sensing techniques to quantify stress in and predict yield of sugarcane. However, it was found that models for detecting a stress factor or predicting yield in sugarcane vary depending on age group, variety, season of sampling, conditions at which spectral data are collected (controlled laboratory or natural field conditions), level at which remotely-sensed data are captured (leaf or canopy levels), and irrigation conditions. The study was conducted in only one study area (the Umfolozi mill supply area) and very few varieties (N12, N19, and NCo 376) were tested. For practical and operational use of remote sensing in sugarcane monitoring, the development of an optimum universal model for detecting factors of stress and predicting yield of sugarcane, therefore, still remains a challenging task. It is recommended that models developed in this study should be tested – or further elaborated – in other South African sugarcane producing areas with growing conditions similar to those under which the predictive models have been developed. Monitoring of sugarcane thrips should also be evaluated using remotely-sensed data at canopy level; and the ability of multispectral sensors other than Landsat TM and ETM+ should be tested for sugarcane yield prediction.