Conference Agenda

The AMEOS (Assimilating Multi-source Earth Observation Satellite data for crop pests and diseases monitoring and forecasting) project aims at providing a tool for pest and disease monitoring and forecast information, integrating multi-source information (Earth Observation, meteorological, entomological and plant pathological, etc.) to support decision making in the sustainable management of insect pests and diseases in agriculture.

The previous two years of the project were devoted to delineate the procedure enabling the satellite imagery based monitoring of crop pests and diseases. In particular, with reference to the Xylella fastidiosa threats of olive groves the approach consists in:

• Improving the classification of the agricultural areas devoted to olive trees, starting from what has been made available from the Corine Land Cover initiative;

• Developing an approach suitable to be automated for estimating trees by using Sentinel 2 images. This tool will be used to monitor the spread of pest.

• Developing an approach suitable to be automated for counting trees by using very high spatial resolution images in areas at high risk of infection.

In the previous year, working on Sentinel-2 images and exploiting the characteristics olive tree phenology and carotenoid indices allowed to improve, respect to the Corine Land Cover, the classification of the olive groves in the area of interest. Further, the use of a morphological approach on NDVI computed by using Sentinel-2 images of 2017 allowed to assess the olive groves density and trees number for each crop field. The quality of the results were validated by using a VHR image.

The present paper concerns the analysis of the possibility to monitor the decrease in the number of olive trees as a consequence of the spread of the xylella fastidiosa disease. The analysis has been carried out on Sentinel-2 images acquired in the second half of 2018.

The final objective of the project is the development of a satellite based system capable to assess the evolution of diseases on both permanent crops (olive groves, vineyards) and row crops (wheat, maize), in Italy and China, aiming at developing an early working tool.

Pathologically, the foliar biophysical variations are critical indicators for tracking the progressive host-pathogen interactions at different development stage. The interaction of electromagnetic radiation with plant leaves is governed by their biophysical constituents, and response to infestations. Hyperspectral- and multispectral- continuum observations could permit the acquiring of the host-pathogen processes within entire epidemic stages.

With this in mind, the launch of the hyper-spectral satellite PRISMA (PRecursore IperSpettrale della Missione Applicativa, HyperSpectral Precursor of the Application Mission, to be launched on the 15^th of March 2019), could give a significant boost to the achievement of the project objectives.

The leaf area index (LAI) is a key parameter for describing the crop canopy structure and is of great importance for crop disease monitoring and yield estimation. Light detection and ranging (LiDAR) is an active remote sensing technology that can detect the vertical distribution of a crop canopy. To quantitatively analyze the influence of the occlusion effect, three flights of multi-route high-density LiDAR dataset were acquired using a UAV-mounted RIEGL VUX-1 laser scanner at the altitude of 15m, to evaluate the validity of LAI estimation in different layers under different planting densities. The result revealed that normalize root-mean-square error (NRMSE) for the upper, middle, and lower layers were 4.0%, 9.3%, 15.0% of 88050 plants/ha, respectively. In order to investigate the effect of different incidence angle, the accuracy of the point cloud data inversion for different air routes (different angles of incidence) was compared and found that the optimal incidence angle was −12° to −5°, and the NRMSE for the upper, middle, and lower layers were 9.3%, 8.0%, 11.5%. The voxel-based method was used to invert the LAI, and we concluded that the optimal voxel size was 0.05 m, which is approximately 2.09 times of the average point distance. The detection of different layers of different planting densities, incidence angle, and optimal voxel size can provide a guideline for UAV-LiDAR application in the crop canopy structure analysis

Yield estimation and forecasting, at the field scale, would allow farm managers to plan their agronomic operations, e.g., sowing, tillage or fertilization, on the basis of expected yield potential. At the district scale, it would be useful for public and private organization, for commercial or planning purposes. In order obtain in-season crop yield predictions, biophysical variables retrieved from remotely sensed data can be assimilated into crop growth models. Data assimilation is a group of methods allowing to combine in an optimal way different information types, dynamically integrating mathematical descriptions of processes embedded into deterministic models and physical information obtained from observations. It allows to update model simulations with observed data, e.g. from remote sensing, within a systematic estimation structure. The most advanced assimilation methods include a framework for the incorporation of model and observation errors and the quantification of prediction uncertainty.
The Ensemble Kalman Filter (EnKF) method has been shown to be particularly suitable for the assimilation of remotely sensed data into crop models and has been extensively assessed for this purpose. The efficiency of EnKF has proved to be affected by a range of factors, such as the number of observations ingested into the process. Many studies have shown that using a higher number of observations brings about an improved accuracy of estimation. Other factors such as errors and uncertainties in the remote sensing observations, the variables that are retrieved as well as crop models formulation errors and parameters uncertainties play also an important role.
The project presented in this study, carried out in the framework of the Dragon 4 Programme supported by the European Space Agency (ESA) and the Ministry of Science and Technology (MOST) of China, aims at targeting these methodological issues, affecting the operational application of data assimilation schemes to the precision farming context.
A study was carried out in Maccarese (Central Italy) for the purpose of assessing the effect of temporal and spatial resolution of multispectral satellite imagery on the accuracy of durum wheat yield data estimation at the field scale from EnKF data assimilation. Field campaigns to measure biophysical variables were carried out from January 2018 to April 2018, for different growing conditions of the crop, at dates close to Sentinel-2 and Venus satellite acquisitions. The latter is a French-Israeli mission combining a high spatial (5 m) and temporal (2 days) resolution, with spectral bands quite similar to Sentinel-2. Initially, an assessment of different hybrid methods, i.e. based on the training of non-parametric regression algorithms with PROSAIL model simulations, was made, in order to identify the most accurate technique for the retrieval of biophysical variables from satellite imagery. Subsequently to a validation study with ground data, the retrieval technique was applied to all the cloud-free images available for the wheat season, notably 25 for Sentinel-2 and 26 for Venus. Among all the biophysical variables, fractional vegetation cover (FCOVER) was estimated with a smaller error (RRMSE of 9.5%) as compared to LAI (RRMSE of 19%). The former variable was thus employed for the data assimilation using the Acquacrop crop model developed by the Food and Agriculture Organization of the United Nations (FAO), since this model uses FCOVER as the main dynamic crop state variable. Recently an open source code version, Aquacrop-OS, has been released, thus allowing the application of sequential data assimilations methods, such as the Ensemble Kalman Filter, which was previously impossible.
EnKF was thus applied to Aquacrop-OS using a) only observations from Sentinel-2, b) only observations from Venus, or c) observations from both satellites, for durum wheat fields of the Maccarese farm. The results, in terms of grain yield prediction, were compared to fields where yield map data had been collected from the combine harvester at the end of the season. In general, benefits from an increased spatial and temporal resolution for field scale yield prediction mapping could be observed, but these came at the cost of an increased computational demand, therefore suggesting that a balance would be required for an operational application in the context of precision agriculture.

The development of high-quality specialty grain and protein-based classification of different types of grain is an important measure, which accelerates the shift from agricultural production to quality improvement. Remote sensing technology had achieved the prediction of grain protein content (GPC), but there were still large deviations in the interannual expansion and regional transfer. The study was conducted in wheat growing areas of Beijing, Renqiu and Jinzhou in Hebei Province. Firstly, Spectral consistency of Landsat-8 and RapidEye respectively compared with Sentinel-2 satellites at the same ground point in the same period. Then, based on the hierarchical linear model (HLM) method, the GPC prediction model was constructed by coupling the vegetation index with the meteorological data obtained by the European Center for Mediumrange Weather Forecasts (ECMWF). The prediction of regional GPC and its spatial expansion were validated. Results were as follows: (1) spectral information calculated from Sentinel-2 imagery were highly consistent with that from Landsat-8 (R² = 1.00) and RapidEye (R² = 0.99) imagery could be jointly used for GPC modeling. (2) Predicted GPC by using HLM method (R² = 0.55) demonstrated higher accuracy than empirical linear model (R² = 0.23) and showed higher improvements across inter-annual and regional scales. (3) The GPC prediction results of the verification samples in Xiaotangshan and Renqiu City were ideal with RMSEv of 0.91% and 0.49%, respectively. This study had great application potential for regional quality prediction and inter-annual quality prediction.

Wheat powdery mildew is caused by the fungus Blumeria graminis and is one of the most common diseases that result in significant loss of crop yield and quality in China. Accurate monitoring of wheat powdery mildew at the regional level is important for food security and environmental protection. Traditionally, wheat powdery mildew is monitored by visual inspection of individual plants, which is time-consuming and inefficient. In recent years, a new satellite-based remote sensing technology has become a more viable option for managing and controlling agricultural practices.

Wheat powdery mildew is monitored by remote sensing technologies based on changes in transpiration rate, chlorosis, leaf color, and morphology in infected plants, which in turn affects the spectral reflectance properties of wheat. The existing methods includes statistical analysis and machine learning methods. In general, wheat powdery mildew can be detected visually for only a short period of time and the field sampling is often expensive. Thus, these methods do not always meet the needs of agricultural management due to the difficulty of collecting field samples.

Limited training data is a common problem in remote sensing applications and many studies have applied transfer learning in remote sensing to increase the quality and quantity of samples. However, thus far, the use of transfer learning techniques for crop disease monitoring has received limited attention. In this study, an instance-based transfer learning method, i.e., TrAdaBoost, was applied to improve the monitoring accuracy with limited field samples by using auxiliary samples from another region.

This study was carried out in Gaocheng city, Hebei province, and the auxiliary field survey samples were acquired from western Guanzhong Plain, Shaanxi province. The samples were categorized into three groups, i.e., normal, slightly diseased and seriously diseased, according to the disease index (DI). The normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), wetness and greenness were extracted using Landsat-8 OLI images to indicate the growth status of wheat and the field habitat characteristics. With these VIs, TrAdaBoost, which using the support vector machine with radial basis function kernel (RBFSVM) as weak learner, was used to develop the wheat powdery mildew monitoring model. At the initialization of the algorithm, an initial weight vector was given, and the maximum number of iterations was also determined. Per iteration, some samples were taken out from the auxiliary samples and study area samples, respectively. The removed samples were used to train a RBFSVM, and the weight of a sample indicated the probability of this sample to be selected to train the RBFSVM. At the end of each iteration, the error of RBFSVM on each training sample was calculated. If an auxiliary training sample was wrongly predicted, the sample likely conflicted with the study area sample. Then, we decreased its weight to reduce its effect, where in the next round, the misclassified auxiliary training sample would affect the learning process less than the previous round. In contrast, if a study area training sample is wrongly predicted, we increased its weight to improve its effect in the next round. Through this mechanism, the auxiliary samples that were useful for improving the monitoring accuracy of wheat powdery mildew in the study area were selected.

The model was tested using a dataset with 53 study area samples and 39 auxiliary samples. The overall monitoring accuracy was 83%, and the kappa coefficient was 0.69. Moreover, TrAdaBoost was also compared with four algorithms that are commonly used to monitor wheat powdery mildew at the regional level, and TrAdaBoost performed better than other algorithms. Experimental results demonstrated that TrAdaBoost was effective in improving the accuracy of monitoring wheat powdery mildew using limited field samples.

Since March the 22th 2019 the PRecursore IperSpettrale della Missione Applicativa – PRISMA is in orbit on a sun-synchronous orbit at 615km. The PRISMA mission, fully developed by the Italian Space Agency (ASI), combines two payloads: a hyperspectral and a panchromatic camera. PRISMA, together with the Chinese Gaofen-5 (GF-5), are the only hyperspectral sensor in orbit with similar characteristics in term od GSD and spectral resolution.

The PRISMA hyperspectral payload is a pushbroom scanner covering the full range (VNIR-SWIR) from 400 to 2500nm with 239 spectral bands at a spatial resolution of 30 m with a swath of 30 km, while the panchromatic camera provides 5m pixel images co-registered with hyperspectral imagery. PRISMA is a prism spectrometer and it is characterized by a variable bandwidth across the nominal spectral range, nevertheless the band width is less than 12nm (i.e. between 7.3 and 11.04nm). The PRISMA coverage is global with a 29 days (orbit repeat period), while the revisit time on a specific area of interest is of 7 days thanks to the off-nadir angle of up to ± 20.7°.

After the end of the commission phase (scheduled in the 2^nd semester 2019), it is foreseen a structured three years CAL/VAL activity, which will be performed on scattered instrumented sites in Italy. The test sites have been selected according to the peculiar thematic areas of interest for the mission (e.g., topsoil characteristics, vegetation biophysical parameter retrieval, snow and coastal waters), moreover international test site are still under definition. The Cal/Val activities includes: airborne surveys with VNIR-SWIR scanner eventually coupled with thermal LWIR multispectral data, field activities contemporary to the PRISMA acquisitions. CAL/VAL activities will be planned, whenever possible, in synergy with ESA (i.e. Fluorescence Explorer – FLEX and the candidate CHIME hyperspectral missions) and the actual ASI missions’ development (hyperspectral mission SHALOM).

The PRISMA acquisition plan is an opportunity to strengthen the ongoing collaborations between the Authors and the RADI Chinese colleagues in the framework of the active international collaborations programs (ESA Dragon-4 and CNR-CAS agreement). The opportunity to foster a synergy between the Italian PRISMA and the Chinese GF-5 missions, in order to increase the possibility to have more consistent hyperspectral time series suitable to monitor biophysical variables and agronomical processes.

This work aims to analyze the feasibility of application of multispectral imagery to detect anomalies caused by tree pests, towards an early warning monitoring system. The analysis take into account multispectral images from satellites as Sentinel-2 (10 mts. spatial resolution) and PlanetScope (3 mts. spatial resolution) imagery, and images obtained from sensors mounted on UAV (5 cms. spatial resolution), as MicaSense RedEdge sensor on board a UAV SR-SF6.

The scope of this study is the evaluation of this set of images to locate tree crowns, and: 1) distinguish between symptomatic and apparently healthy canopies, and 2) between symptomatic and affected asymptomatic trees.

The main pest of interest analyzed in this study is the Xylella fastidiosa (Xf), a vector-transmitted bacterial plant pathogen associated with serious diseases in a wide range of plants. Xf was detected on olive trees in Puglia, southern Italy, in October 2013, the first time the bacterium has been reported in the European Union. Since then it has also been reported as present in France, Spain and Germany. Controls are in place to prevent the bacterium from spreading.

In 2016, the European Food safety Authority (EFSA) concluded that research being carried out in Apulia showed that certain treatments reduce the symptoms of disease caused by Xf but do not eliminate the pathogen from infected plants (EFSA 2016). A significant difficulty that need to be taken into account for the containment of Xf, comes from its very wide host range (in September 2018 the list includes more than 500 plant species), and since infections that do not cause symptoms in some host–strain combinations, despite the infected hosts continuing to act as inoculum sources.

In (Zarco-Tejada et al. 2018) a multi-temporal trees inspection and airborne imaging data in several orchards has been carried out. The analysis found the physiological alterations caused by Xf infection at the pre-visual stage were detectable in functional plant traits assessed remotely by hyperspectral and thermal sensors. And it was confirmed the presence of Xf infection in the selected orchards by the quantitative Polymerase Chain Reaction (qPCR) assay.

The main area of interest for this study is region of Apulia, and as a reference, the current analysis took into account datasets and geo-locations of trees affected by this pest in olives orchards, identified during the field campaign carried out in (Zarco-Tejada et al. 2018). Furthermore, the study includes 3000 geo-locations of olive trees extracted from public records distributed by Apulia region, in cooperation with the Public Network of Research Laboratories (SELGE).

The framework to carry out the multi-temporal analysis with Sentinel-2 imagery was carried out in the Data Integration and Analysis System (DIAS). DIAS archive includes satellite data and ground observation data. These datasets are stored in an large volume disk array accessible via API requests.

The time lapse of the current analysis included Sentinel-2 images since 2016 until 2018, PlanetScope images for the period 2017-2018, and a UAV field campaign carried out in 2018.

As in (Zarco-Tejada et al. 2018), common multispectral vegetation indices, did not differ significantly between asymptomatic and symptomatic trees, so unable to detect non-visual symptoms of Xf infection. In the present study, despite it was carried out an identification and masking of background soil, some differences could be mainly associated to changes on soil background more than with tree anomalies, since spatial resolution of satellite imagery seems to be scarce to be used for a tree focused study but for an analysis at a parcel level, where at the same time the heterogeneity of trees status could differ.

Detailed analysis of multi-temporal vegetation indices profiles showed that remote sensing observations, can help to identify unexpected phenology patterns or anomalies that could be related to tree pests. Despite the capability detect individual trees vary according image spatial resolution, remote sensing techniques help to put into evidence a particular parcel where a further analysis need to focus.