Caracterización de las royas del trigo en Andalucía y uso de sensores remotos para su detección temprana

Abstract

Las royas son una enfermedad importante en el cultivo del trigo, generando pérdidas de producción y, por tanto, económicas, en el sector cerealista. Podemos distinguir principalmente tres especies que causan tres enfermedades diferentes: roya de la hoja (o parda), amarilla (o lineal) y del tallo (o negra). Hasta la temporada 2019-2020, la roya de la hoja estaba controlada, ya que los cultivares más sembrados en el sur de España tenían genes de resistencia contra ella, y los cultivares susceptibles estaban protegidos con fungicidas. Pero un problema surgió en la primavera de 2020, cuando todos los cultivares de trigo duro comenzaron a infectarse con una roya parecida a la de la hoja. La roya del tallo apenas está presente durante la temporada de cultivo regular, pero puede ser una enfermedad importante en el futuro. Se identificaron las principales razas como Clade IV-B y Clade IV-F, al igual que en la mayor parte de Europa y partes del norte de África. Para esta roya, las resistencias siempre son incompletas.

Rusts are a significant disease in wheat cultivation, causing yield loss and economic losses in the cereal sector. We can distinguish three rust species causing three different diseases: leaf (or brown), yellow (or stripe) and stem (or black) rust. Until the 2019-2020 season, leaf rust was controlled since the most planted cultivars in southern Spain had resistance genes against it, and susceptible cultivars were protected with fungicides. A problem arose in the spring of 2020 when all durum wheat cultivars began to become infected with a kind of leaf rust. The leaves had large pustules, different from those of typical rust caused by P. triticina, and the teliospores sprouted rapidly after a few days of infections. The symptoms fitted P. tritici-duri, another wheat leaf rust species already reported in the western Mediterranean basin. This study, evaluated the severity of leaf rust in durum wheat field trials during the 2019-20, 2020-21 and 2021-22 seasons in Huelva, Seville, Cádiz and Córdoba provinces. Additionally, during the spring of 2020 and 2021, isolates from individual leaf rust pustules were collected from different durum wheat cultivars. Inoculation of the isolate into a differential set of lines showed that five different races were present, of which two were of the P. tritici-duri subspecies. This subspecies is not new in southern Spain. Still, in the last 25 years, it has not been observed with such severity in almost all durum wheat cultivars (susceptible and resistant to P. triticina). Cultivar Calero constantly resisted all races of P. tritici-duri used in this study. Furthermore, effective mitigation of the current threat of yellow rust and the potential threat of stem rust to wheat production in southern Spain requires the characterization of the breeds currently present in the region. The results of this study indicated that the primary races of yellow rust now present in southern Spain are PstS10, PstS13 and PstS14, to which several widely planted commercial cultivars (possessing, among others, Yr27 gene) are resistant. Stem rust does not pose a severe threat yet during the regular wheat growing season but can be a potential biotic stress soon. The primary races were Clade IV-B and Clade IV-F, as in most Europe and North Africa. For this rust species, the available resistances are always incomplete. The evaluation of differential series and unique breeding lines with known genes in local conditions (such as Sr27 and Sr35) has indicated the availability of several genetic options that could be used in breeding/selection programs to provide adequate levels of resistance to stem rust. However, for any of the three rusts affecting wheat, in undertaking these efforts, it is essential to consider not only the races currently present in the region but also to consider effective resistance options against races that are being developed elsewhere and that they could very probably reach the south of Spain soon. The other way to control rust in wheat is through chemical control; for this, early detection of the disease is essential. Therefore, this work studied the detection of yellow and leaf rust using hyperspectral data. An experiment was carried out in a greenhouse, with inoculated and non-inoculated wheat plants, with this disease. Each wheat plant was planted in separate pots, and their spectral reflectance was measured. Data pre-processing included standardization and the synthetic minority oversampling technique (SMOTE) for data balance. Four machine learning models (ANN, SVM, RF, GNB) were tested with different SMOTE applications, focusing on accuracy and F1 score. The SVM and RF models showed the highest accuracies, especially with SMOTE (enhanced datasets). However, concerns were raised about real-world applications due to the use of synthetic data. The study highlighted the mixed impact of SMOTE on the data and identified RF and SVM as the best models for yellow and leaf rust detection, respectively. The findings emphasize the potential of spectral data and machine learning in crop disease detection and the need for data processing in future research. Another part of the study focuses on an alternative method to detect rust in wheat plants early, taking advantage of 3D LiDAR technology to discern the impact of this disease on wheat cultivars from the moment of its infection. Despite rust inoculation, variations in plant height and biomass were negligible. However, notable drops in grain yield were recorded, especially in cultivars susceptible to rust. LiDAR-derived data demonstrated a substantial correlation with disease severity, underscoring its promise for reliable biomass assessment and early detection of rust. The study affirms the vital role of LiDAR in precision agriculture, offering a proactive approach to disease management and safeguarding the stability of wheat production yields. Using new technologies for the early detection of rust in wheat plants, the precision, and results of the F1 score have been analyzed in different models and data sets. The data sets used can be considered comparatively small compared to other studies; this challenge is related to the current insufficiency of the size and diversity of the data sets in the application of plant disease classification models In addition, it is essential to consider all possible capture conditions, symptom variations and sensors. For these reasons, it is inevitable to have incomplete data sets and therefore the models in their application are limited in scope. There are several alternatives to address the effects of using incomplete data sets. One of the most common is data augmentation through various techniques that can be combined. However, each data augmentation technique can affect model accuracy differntly. Finding a design pattern to select the most optimal method presents a challenge. In this study, the SMOTE algorithm is applied to varying degrees to a small data set for a single disease with three categories. This demonstrates how, in general, precision increases with the application of data augmentation techniques. Regarding the data set, where the SMOTE algorithm was applied only to the training set, the accuracy for yellow rust detection decreased, and the accuracy for leaf rust detection increased. The classification models that achieved the highest accuracy values in this study were SVM and RF. In the study with LiDAR sensors, the estimation of disease severity using LiDAR reflectance intensities presents a novel method that aligns with the growing interest in precision agriculture, and the need for timely disease management. The findings of this study reveal that disease severity can be quantitatively assessed through LiDAR. In conclusion, while applying LiDAR technology for severity estimation is promising, it is imperative to consider the variability between different wheat cultivars and stages of disease progression. Future studies should focus on refining estimation models and exploring integrating LiDAR data with other remote sensing modalities to improve the accuracy and reliability of disease severity assessments.