Spatiotemporal dynamics of stress factors in wheat analysed by multisensoral remote sensing and geostatistics

Abstract

Plant stresses, in particular fungal diseases, basically show a high variability in space and time with respect to their impact on the host. Recent ‘Precision Agriculture’ techniques allow for a spatially and temporally adjusted pest control that might reduce the amount of cost-intensive and ecologically harmful agrochemicals. Conventional stress detection techniques such as random monitoring do not meet demands of such optimally placed management actions. The prerequisite is a profound knowledge about the controlled phenomena as well as their accurate sensor-based detection. Therefore, the present study focused on spatiotemporal dynamics of stress factors in wheat, Europe’s main crop. Primarily, the spatiotemporal characteristics of the fungal diseases, powdery mildew (Blumeria graminis) and leaf rust (Puccinia recondita), were analysed by remote sensing techniques and geo-statistics on leaf and field scale.<br /> Basically, there are two different approaches to sensor-based detection of crop stresses: near-range sensors and airborne-/satellite-borne sensors. In order to assess the potential of both approaches, various experiments in field and laboratory were carried out with the use of multiple sensors operated at different scales. Besides the spatial dimension of crop stresses, all studies focussed on the temporal dimension of these phenomena, since this is the key question for an operational use of these techniques. In addition, a comparison between multispectral and hyperspectral data gave an indication of their suitability for this purpose. <br /> The results exhibit very high spatiotemporal dynamics for both fungal diseases. However, powdery mildew and leaf rust showed different characteristics, with leaf rust showing a more systematic temporal progress. The physiological behaviours of the phenomena, which are strongly influenced by various environmental factors, define the optimal disease detection date as well as the temporal resolution required for sensor-based disease detection. Due to the high spatiotemporal dynamics of the investigated diseases, a general recommendation of optimal detection periods can not be given, but critical periods are highlighted for each pathogen. <br /> The results indicate that multispectral remote sensing data with high spatial resolution shows a high potential for quantifying crop vigour by using spectral mixture analyses. Simulated endmembers for the identification of stressed wheat areas were utilized, whereby promising results could be achieved. However, due to the low spectral resolution of these data, a discrimination of stress factors or early disease detection is not possible. Hyperspectral data was therefore used to point out the potential of early detection of crop diseases, which is a crucial and restrictive factor for Precision Agriculture applications. In a laboratory experiment, leaf rust infections could be detected by hyperspectral data five days after inoculation. In a field experiment with respect to early stress detection, it could be demonstrated that hyperspectral data outperformed multispectral data. High accuracy for the detection of powdery mildew infections in the field was thereby achieved.<br /> Due to the fact that typical spatiotemporal characteristics for each pathogen were found, there is a high potential for decision support systems, considering all variables that affect the disease progress. Besides the further analysis of hyperspectral data for disease detection, the development of a decision support system is the subject of the upcoming last period of the Research Training Group 722.