Within the framework of the DRAGON project, the Institute of Electronics, Chinese Academy of Sciences (IECAS) continuously had a tight collaboration with the European and the Chinese partners. Joint research is around the 3 scientific topics: hybrid-polarity (HP) architecture, vegetation classification, and multi-aspect polarimeric scattering mechanism.
As to the study on hybrid-polarity (HP) architecture, an improved version of the HP architecture is proposed in order to overcome the two implementation difficulties concerning calibration issue and transmit configuration. Polarimetric SAR (PolSAR) systems based on this improved HP architecture do not only have mitigated transmit distortions through a pre-distortion technique, but also have the ability to transmit arbitrary polarizations, thus supporting all dual- and quad-polarimetric modes which uses orthogonal linear polarization basis in reception. Here, the supported dual-polarimetric modes include the circular transmit linear receive (CTLR) mode and π/4 mode of compact polarimetry. A prototype system has been built for further experimental verifications and experiments.
In the aspect of polarimetric classification, a supervised PolSAR image classification based on nearest-regularized subspace (NRS) and the vegetation classification based on the decision tree classifier are respectively studied. NRS is a representation-based classification approach. The polarimetric SAR feature space is constructed by the components of different target decomposition algorithms for each pixel, including those scattering components of Freeman, Huynen, Krogager, Yamaguchi, Arii decomposition, as well as the eigenvalues, eigenvectors and their consequential parameters such as entropy, anisotropy and mean scattering angle. Furthermore, because that all these representation-based methods were originally designed to be pixel-wise classifiers which only consider the polarimetric features while ignoring the spatial-contextual information, Markov Random Field (MRF) is also introduced into our scheme. Preliminary experiments display that the proposed NRS-MRF method outperforms other compared approaches with few training percent, including SVM, NRS and SVM-MRF. For vegetation classification applications, in order to make full use of PolSAR features’ scattering mechanism description, a decision tree classifier comes into use due to its simple and hierarchical structure. Since a decision tree classifier is flexible in terms of the discriminant rules, a hierarchical classification process based on Fisher Linear Discriminant (FLD) is built. Furthermore, the feedback adjustments of the classification results, such as feature combinations, partition algorithms and split sequences, make it possible to improve the accuracy of some specific classes. The experiment regarding the AIRSAR-Flevoland data illustrates that this method can obtain a good classification accuracy, and that it can extract useful information from the entire classification process in order to form new expert knowledge, such as the PolSAR features, classifying algorithms, and relationships between the classes.
Referring to the multi-aspect polarimeric scattering mechanism, the anisotropic scattering property is analyzed and two parameters both derived from traditional polarimetric entropy is proposed. One is called multi-aperture polarimetric entropy(MAPE), describes both depolarisation and directivity. The other is called sub-aperture entropy(SAE), describes change of depolarisation with look angle. These two parameters are used to separate targets which have both strong directivity and specific scattering mechanism, have specific scattering mechanism but not directivity, have neither directivity nor specific scattering mechanism. The effectiveness of the new parameters is validated with real polarimetric Circular SAR data, acquired by the Institute of Electronics airborne CSAR system at P-band.

In this report, we will introduce the main research progress of forest above ground biomass (AGB) estimation study based on integration of multi-dimensional SAR data. It mainly contains the following three aspects. (1) We proposed a three-steps semi-empirical radiometric terrain correction approach for PolSAR data. The three steps of terrain effects correction are polarization orientation angle, effective scattering area, and angular variation effect corrections. Based on the LiDAR-derived forest AGB data, detailed analysis and evaluation were carried on for the three correction steps. (2) Base on the simplified InSAR decorrelation model (SINC model) and Algebraic Difference theory, we developed a terrain correction approach for coherence image of InSAR data. And the method was evaluated by space-borne and air-borne InSAR data. (3) Based on the X-band single-pass InSAR data and P-band PolSAR data acquired by multi-dimensional SAR system (CASMSAR) of China, we developed one combined estimation approach of forest AGB based on multi-dimensional SAR data by integrating the terrain correction methods proposed above.

Virtual Dispalacement Method for Tree Volume

Hysteresis refers to the asymmetric path between to alternative states. Specifically, we apply the concept to analyse differences between degradation and regeneration processes, a phenomena pointed out in the dynamics of several ecosystems [1]. The topic is quite relevant for the ongoing debate about the irreversibility of desertification, challenged by different studies that support re-greening in degraded areas [2].

It is possible to deduce the existence of hysteresis through the speed of the aforementioned opposed processes. For that, a statistically sound analysis based on 2dRUE results [3] has been implemented. Previously, 2dRUE was applied to assess land condition in China Drylands using time series NPP data computed from Envisat Meris images [4]. The method incorporates a stepwise regression to signify the effects of time and aridity on vegetation. This allows distinguishing if biomass changes are explained by the impact of wet and dry years, or by land degradation itself. In this way, 2dRUE supplies standard partial regression coefficients, an explicit and untainted measure of changes in land condition.

We have selected the Mann-Whitney U test to compare two variables: Positive and negative time regression coefficients, i.e. regeneration and degradation do not related with aridity fluctuations. The null hypothesis is that both processes happen at the same rate. Data is divided into 3 categories, each one with a different number of classes, Land-use (20); 2dRUE Assessment (8); FAO Aridity (3). Exploratory results show that in half the cases there are significant differences between the speed of degradation and regeneration. In most of them, the pace of regeneration is higher than degradation. Forthcoming works require a detailed insight within those significant categories in aim to interpret the scope of our preliminary results.

Land degradation is a process by which the land productive capacity declines or even is completely lost under the influence of natural forces and human activities. The scope of land degradation has become global in the last decades, which compromises sustainable land management and threatens the safety of food production, especially in the poverty-stricken areas of developing countries. Desertification is one kind of land degradation and mainly occupied in arid, semi-arid and semi-humid areas. China is one of the most seriously affected countries by desertification. By the end of 2014, the desertified land area of China was 2.61×10⁶km². Post-hoc mitigation approaches are expensive and often ineffective. Therefore early warning systems based on Earth Observation make the most accepted scientific basis for controlling land degradation.

With the development of remote sensing technology, long time series remote sensing data have been available for land degradation assessment and monitoring, and the vegetation indicators, such as the NDVI, NPP, Vegetation coverage and biomass were commonly used. However, time series vegetation index will fluctuate severely due to the impact of climate change, especially the fluctuation of annual precipitation, thereby the land production capacity could not be determined accurately.

Therefore, to solve the problem, Xilin Gol League, In+-ner Mongolia Autonomous Region, China, where the land degradation is prevailing in the first decade of the 21^st century was selected as the study area. Based on the annual NPP dataset estimated by 10-Day composite NDVI from Envisat-Meris data at 1.2km resolution during 2003 to 2013 and the same period meteorological raster dataset, a new Moisture-responded Net Primary Productivity (MNPP) method, for identifying areas of land degradation based on the change of annual NPP and MNPP over time and Moisture Index (MI) was developed. It was expected that provide technical support and scientific reference data for land degradation assessment and monitoring in study area, even in the whole drylands in China.

Desert vegetation plays significant roles in securing the ecological integrity of oasis ecosystems in western China. Timely monitoring of photosynthetic/non-photosynthetic desert vegetation cover is necessary to guide management practices on land desertification and research into the mechanisms driving vegetation recession. In this study, nonlinear spectral mixture effects for photosynthetic/non-photosynthetic vegetation cover estimates are investigated through comparing the performance of linear and nonlinear spectral mixture models with different endmembers applied to field spectral measurements of two types of typical desert vegetation, namely, Nitraria shrubs and Haloxylon. The main results were as follows. (1) The correct selection of endmembers is important for improving the accuracy of vegetation cover estimates, and in particular, shadow endmembers cannot be neglected. (2) For both the Nitraria shrubs and Haloxylon, the Kernel-based Nonlinear Spectral Mixture Model (KNSMM) was the best unmixing model. In consideration of the computational complexity and accuracy requirements, the Linear Spectral Mixture Model (LSMM) could be adopted for Nitraria shrubs plots, but this will result in significant errors for the Haloxylon plots because of the stronger nonlinear spectral mixture effects. (3) The vegetation canopy structure (planophile or erectophile) determines the strength of the nonlinear spectral mixture effects. Therefore, nonlinear spectral mixing effects for Nitraria shrubs and Haloxylon were validated to be different, additional research is necessary to validate their performance from the canopy to the landscape scale.

Among those diseases threatening human health and well-being, many epidemic and infectious diseases are closely related to natural environment due to the presence, breeding and evolution of their pathogens or reservoir hosts, especially vector-borne diseases (e.g. schistosomiasis, malaria and dengue, etc.) which rely heavily on their vectors. Therefore, monitoring the diseases' vector is an important way to prevent and control the vector-borne diseases.Because of complex spatial distribution and dispersion of typical diseases and their vectors, it is difficult to acquire relevant environmental factor data by traditional in-situ measurements. Remote sensing technology provides the capability of obtaining temporal-spatial variations of ground environmental factors. However, remote sensing experts may not exactly know what environmental factors are required to identify the incubators of vector-borne diseases. On the other hand, effective RS data processing and parameters retrieval techniques are also challenges for hygiene experts who are lack of experience of remote sensing applications. Taking into account of different type of massive data are involved, computing scientists with substantial intelligent data analysis expertise is crucial to successfully incorporate advance intelligent data analysis, such as data mining, pattern analysis. Consequently, any single of these disciplines is insufficient, it is essential to bring together scientists from computing science and remote sensing along with domain experts to foster a substantial collaboration.This proposed project aims to apply advanced remote sensing and computing technologies into monitoring and early warning of vector-borne diseases, e.g. schistosomiasis, malaria and dengue. First is to reveal environmental factors which have significant influences on the breeding of epidemic disease and its vectors. Then the project will make full use of the advantage of European and Chinese earth observation resources and the partners capability to develop parameter inversion, feature extraction and pattern analysis methods that will be used to characterise environmental features and habitants that are mostly suitable for the growth and dispersion of vector-borne disease and dynamic monitoring. Furthermore, temporal-spatial models of the distribution of vector-borne diseases will be developed by data mining techniques. Finally, the driving mechanism and data assimilation methods of land surface process model will be explored in order to implement identification and early warning of vector-borne disease transmission areas. All the institution of the project can provide sufficient funding to run the whole project successfully. The outcomes of the project will help to decrease the scope and extent of vector-borne diseases, and improve prevention & control capabilities to vector-borne diseases. Additionally, the research results can be used to assess environmental characteristics around the sits of major infrastructure and facilities, and provide the suggestion on site selection and implementation of infrastructures. The synthetic feature extraction techniques developed for multi-source multi-level remote sensing data can also be applied to other service fields, sustainably making contribution to knowledge within the communities.

Approximately half of the world’s population is at the risk of at least one vector-borne parasitic disease. The survival of intermediate hosts of vector-borne parasitic diseases is governed by various environmental factors, and remote sensing can be used to characterize and monitor environmental factors related to intermediate host breeding and reproduction, and become a powerful means to monitor the vector-borne parasitic diseases. In this research, satellite remotely sensed data has been used to obtain the environmental factors (vegetation, soil,temperature, terrain et al.), which are related to the living, multiplying and transmission of intermediate host. Then based on ground truth data, the remote sensing monitoring model of the intermediate host has been developed, which can enhance the remote sensing monitoring capabilities of the vector-borne parasitic disease and provide the theoretical foundation and technical support for diseases prevention and control.

Accurate and recursive maps of crop and weeds at the field-scale could be of great interest with major economic and environmental impacts, including competition with native plant species, choking irrigation infrastructure, reducing agricultural yields and affecting the health of livestock.

The development of new generation multispectral satellite sensors and hyperspectral UAVs sensors has resulted in significant interest in their use for crop and weeds classification applications in view of their high spatial/spectral resolution. Small UAVs are more suited to site-specific weed management applications, as they can collect data at high spatial resolutions, which is essential for the classification of small or localized weed outbreaks. The extraction of features (spatial and/or spectral) that discriminate between the weeds of interest and background objects (i.e. different soils and crops) is crucial to map and monitor weeds in agricultural fields. Weed classification is an essential requirement for site-specific weed management in the context of precision agriculture, allowing a considerable reduction of herbicide spraying, with favorable environmental consequences. However, the spectral discrimination between crop and weeds from multi and hyperspectral remote sensing has yet to solve several issues, which is mainly due to the small spectral differences between crops of different species with also the requirement of a very high spatial resolution. Soil background effects are another problem that complicates weed detection in post-emergence row crops. Currently, no established methodology has been widely accepted, but different authors have employed successfully multivariate and machine learning algorithms such as PLSR-DA (Partial Least Squares Discriminant Analysis) and Support Vector Machine with a Radial Basis Function Kernel (Gaussian SVM) to discriminate and classify weeds from crops (Hermann et al., 2013; Hadoux et al., 2014) in using imaging spectroscopy field-based studies.

In the WP1 of the Topic 1, the main research activities are focused on: (a) setting up a spectral library of the most common crops and weeds in the investigated agricultural fields; (b) testing the capability of images acquired from both Sentinel 2 and UAVs imagery using different classification tools for crop mapping and weed detection; (c) mapping crops and weed patches in operational situations of site-specific crop management.

Improved classification methods are applied to Sentinel-2 and UAVs imagery for crop and weed detection: Minimum Distance (MD), Support Vector Machine (SVM), Neural Network (KNN), Random Forest (RF) and Spectral Feature Fitting (SFF).

Performance of the applied crop and weeds classification methods are compared on Sentinel-2 and UAVs’ data sets related to agricultural fields in Central Italy. Classification performances that can be obtained depends not only on the scenario and, thus, data set, but also on the specific sensor and platform as well as on the classification methods employed.

Moreover, with global revisit times of five days from next months onwards, Sentinel-2 based classifications can probably be further improved by using (a) temporal information in addition to the spectral signatures and (b) textural as well as canopy height information from Sentinel-1 radar images.

References

Herrmann, I., Shapira, U., Kinast, S., Karnieli, A., Bonfil, D.J. (2013). Ground-level hyperspectral imagery for detecting weeds in wheat fields. Precision Agriculture, 14, 637-659.

Hadoux X., Gorretta N., Roger J.-M., Bendoula R., Rabatel G. (2014). Comparison of the efficacy of spectral pre-treatments for wheat and weed discrimination in outdoor conditions. Computers and Electronics in Agriculture, 108 , 242-249.

As field measurement of massive germplasm for complex traits in breeding is challenging, there's a strong demand for real-time, fast and nondestructive phenotyping to accelerate the breeding efficiency. Unmanned aerial platforms‑based remote sensing can be used to rapidly and cost‑effectively phenotype large numbers of plots and field trials. Strategies for high-throughput field-based phenotyping in breeding was investigated by National Engineering Research Center for Information Technology in Agriculture (coordinated by NERCITA) in recent years, where proximal remote sensing is the deployment of sensors using aerial platforms. Strategies include the following: 1) Selection and specification of UAV platforms for application of phenotyping in breeding; 2) Rapid processing of multi-source remote sensing data for high-throughput phenotyping; 3) Analysis of phenotypic information for soybean, maize and wheat in breeding (over 10,000 plots) by proximal remote sensing under different growth periods; 4) Determining the optimal growth stage, indexes and algorithm model for crop yield prediction; 5) Validation of the phenotypic information resolution and yield prediction using agricultural UAV for breeding plots to ascertain its stability and accuracy; 6) Genome-wide association study of morphological indicators and genotype of maize and the identification of candidate gene.

Many studies have demonstrated the advantage of the exploitation of multi-temporal and multi-sensor remote sensing data for the improvement of crop classification methods. However the methods based on optical and SAR data for monitoring arable crops, such as cereals and forage crops (e.g. wheat, barley, oats, triticale, ryegrass...), both at the local and regional level, are still far from being effective and operational. This is because usually small differences in the canopy reflectance and SAR backscatter occurs among these crops, hampering a clear and robust discrimination. There is however, an increasing requirement, e.g. in the context of the European agricultural policy of "coupled subsidy" (payment related to real crop cultivation on parcel), to identify individual crops. For example in the "crop diversification" measure, imposed by the “Greening Policy”, it is necessary to separate barley and wheat on the same farm, which is rather difficult when using only a type of data. The objective of WP1 of the topic 1 of the Dragon4 project ID. 32275 "Combined Exploitation Of Sino EU Earth Observation Data for Supporting The Monitoring and Management of Agricultural Resources", is the development of classification algorithms based on multitemporal optical and radar data allowing the incorporation of cropping systems dynamics information. In this context, a study was carried out in the Maccarese farmland test site (Central Italy), in which a quite extensive dataset, including both optical (RapidEye, Landsat8 and Zy-Yuan 3) and radar (Cosmo SkyMed and Sentinel-1) data, was collected during the 2015 crop growth season. These data were used to test a multi-temporal phenology-based classification algorithm, based on pixel-level decision tree (DT) incorporating information on temporal crop dynamics. The results were validated using the ground data available within a farm management geographic information system (GIS).

The same dataset was also employed for field-based wheat yield estimation, which is the objective of WP5 of topic 1 of the mentioned Dragon4 project. For this purpose, the remote sensing data were converted into biophysical variables (LAI and above ground biomass), using empirical relationships with ground data obtained during the 2015 crop season, in field campaigns carried close to the satellite acquisitions (on wheat, barley, alfalfa, broad bean and maize). For optical data, an algorithm based on the training of artificial neural networks with PROSPECT+SAIL model simulations was employed to retrieve biophysical canopy variables. These variables were then assimilated into the SAFY model (Duchemin et al., 2008, Environ Model Softw. 23, 876-892), using the Ensemble Kalman Filter (EnKF) method, to estimate grain yield. The models were re-calibrated on the basis of a preliminary sensitivity analysis study. Two versions of the SAFY model were compared, i.e. the original SAFY and a modified version (SAFYE) which includes a description of the soil water balance and water stress factors (Veloso, 2014, PhD Thesis, University of Toulouse, France). The results were validated using yield map data collected using a yield monitor system available on the farm's combine harvester machine. A very small difference emerged between the results of the two model versions, suggesting that the original SAFY version, which includes less parameters and input variables than SAFYE, is a suitable and practical choice for farm-based grain yield estimation, although water stress is inferred indirectly, from the processes regulating leaf growth.

Yellow rust is one of the most important fungal diseases of wheat and have caused serious yield loss of winter wheat in China. Studies focused on detecting and monitoring of rust infestation highly rely on field investigation and visual judgement with an advanced pathogen attack. Automatic methods based on remotely sensed techniques and spectral absorption features for an early detection of rust disease are rarely discussed. On the other side, in Italy, the Province of Lecce, located in the region (Apulia) where the 35% of the Italian olive oil is produced has been significantly affected by the Xylella fastidiosa disease which caused a rapid decline in olive plantations, the so-called olive quick decline syndrome (OQDS, in Italian: complesso del disseccamento rapido dell'olivo). By the beginning of 2015 it had infected up to a million trees in the Lecce Province. Then this paper has two main objectives:

- determine the most sensitive spectral vegetation indices (SVIs) for characterizing specific pathological lesions of winter wheat leaves infected with rust disease, and then to propose a hyperspectral analysis procedure for the early detection and identification of rust diseases before specific symptoms became visible. Based on in situ hyperspectral data measured in experiments of 2002 and 2003 with a total of 122 samples, 14 pro-existing SVIs that are related to foliar physiological and photochemistry variations were assessed at different pathogen stages. And then, in order to extract the subtle spectral features between the healthy and diseased winter wheat leaves, an enhanced feature space (EFS) were developed by the non-linear combination and transformation of the identified SVIs. Based on this feature space, early differentiation of healthy leaves and leaves infected with yellow rust could be achieved by a Support Vector Machine (SVM) classifier. Finally, for validation and extension, this approach was also successfully implemented on MODIS data for mapping the rust occurrence conditions in the regional level in the major winter wheat planting area.

- presenting the preliminary results of the analysis of a time series of Landsat images of the Province of Lecce (Italy) spanning a period of seven years on the way to analyze the possibility of using available satellite images (e.g. L8 and Sentinel-2) datasets to assess the evolution of diseases on permanent crops (olive groves, vineyards), for example, the spread of phytosanitary threats as the Xylella fastidiosa (olive groves) or fungal trunk diseases (vineyards) in Italy.

Keyword: Yellow rust; Spectral vegetation indices; Early detection; Support vector machine

Nitrogen (N) is an important indicator of the plant nutritional status and then affects the end of wheat production. Rapid real-time monitoring of wheat N status is crucial for precision N management during wheat growth. Traditional inversion methods by remote sensing are mostly based on its relationship with vegetation index or spectral band. Some models are not suitable as the change of time and location. N-PROSAIL model (Yang et al., 2016) was developed by replacing the absorption coefficient of chlorophyll in the original PROSAIL model (with an equivalent N absorption coefficient, and it gave us a physical method to estimate canopy nitrogen density (CND, g m^-2). Therefore, the objective of this study was to estimate CND at different growth stages by inverting the N-PROSAIL model.

Field experiments were carried out in 2012-2013, 2013-2014 and 2015-2016 at the National Precision Agriculture Experimental Base of Xiaotangshan town, Changping district, Beijing, PR China. Data acquisition included canopy hyperspectral reflectance, leaf area index (LAI, m² m^-2), and leaf nitrogen density (LND, µg cm^-2), which were measured at four wheat growth stages, jointing (Z.S. 31), heading (Z.S. 47), anthesis (Z.S. 65), and filling (Z.S. 75). First, parameters of N, Cm, Cw, LAD, hspot, psoil, and solar zenith angle in N-PROSAIL model (Jacquemoud et al., 2009) were calibrated at different growth stages. Second, LAI and LND were inversed by Shuffled Complex Evolution (SCE-UA) algorithm and the estimation accuracy was tried to improve by using the calibrated parameters set at various growth stages. Finally, CND was calculated by LAI multiplied LND.

The inversion results with and without considering calibrating parameters at different growth stages were compared (Fig. 1). Without considering calibrating parameters at different growth stages (unified parameters at all growth stages, Li et al., 2015), The R² and RMSE values for LAI and LND were 0.67 and 0.74, and 0.30 and 58.49 μg cm^-2, respectively. Relationship between the simulated and measured CND showed well (R² = 0.66, RMSE = 2.45 g m^-2). After considering calibrating parameter at different growth stages, the estimation accuracy of LAI was improved (R² =0.75, RMSE = 0.73), and LND estimation showed a significant improvement (R² = 0.59, RMSE = 17.43 μg cm^-2). In the end, the CND estimation performed better than CND estimation with unified parameter set at all growth stages, with R² and RMSE values of 0.75 and 1.32 g m^-2, respectively. These results confirm the potential of using N-PROSAIL model for CND retrieval in winter wheat at different growth stages and under variables climatic conditions.

Wheat powdery mildew is one of the serious crop diseases which affects the food safety of China. Integrating multi-source information (Earth Observation-EO, meteorological, etc.) to support decision making in the sustainable management of wheat powdery mildew in agriculture is demanded. In this study, the Landsat8 remote sensing image of May 22nd, 2014 was used to extract the land surface temperature (LST) and many vegetation indices including normalized difference vegetation index (NDVI), enhanced vegetation index(EVI), triangular vegetation index (TVI), wetness index and renormalized difference vegetation index (RDVI). Site daily meteorological data including temperature, humidity, rainfall, sunshine hour from April to May was used to get the parameters delineating the environment condition such as average temperature from April 22nd to May 22nd, average humidity from April 22nd to May 22^nd, total sunshine hour from April 22nd to May 22nd and number of rainy days with more than 0.1 mm rainfall from April 22nd to May 22nd. Corresponding space meteorological features were got by combining remote sensing image and site daily meteorological data. In the field work, 148 field sites were collected in study area and the degree of wheat powdery mildew in these sites was recorded. An independent t-test analysis was used to test the difference between disease and healthy sites based on calibration data. Those vegetation indices and environmental factors that failed to show a statistical significant(p<0.001) were eliminated. Five variables including RDVI, temperature, humidity, average temperature from April 22nd to May 22^nd and average humidity from April 22nd to May 22^nd were identified as optimal explanatory variables for developing the powdery mildew forecasting model. The artificial neural network was trained using feedforward neural network learning algorithm in combination with simulated annealing technique to learn the relationship between selected factors and powdery mildew. The powdery mildew forecasting model based on artificial neural network (ANN) was established to predict powdery mildew occurrence of wheat in Gaocheng, Jinzhou and Zhaoxian County, Shijiazhuang City, Hebei Province. The results obtained from the ANN model were compared with prediction model developed using support vector machine(SVM) technique. The accuracy of models respectively based on validation samples were obtained to evaluate the difference of performance of the models. The result showed that the overall accuracy of the ANN model was 86.49%, which is higher than SVM model(78.38%). The result reveals that ANN model could be used to forecast the occurrence of wheat powdery mildew and compared with the traditional forecasting model based on support vector machine, ANN model has a better performance.

Crop yield is one of the most concerned complex traits in crop research, which is linked to CO₂ fixation through the photosynthetic process and partitioning of photoassimilates to the harvested part of the plant. The traditional methods for measuring crop yield are the use of manual sampling or establishing the relationship between agronomic factors or climatic factors and crop yield using statistical analysis methods. Many observations and samplings in field experiments are required to determine the parameters of the yield prediction model, which is time-consuming, low efficiency, and incomplete spatial coverage. Unmanned aerial vehicle remote sensing platforms (UAV-RSPs) equipped with different sensors have recently become important means for fast and non-destructive access to complex traits and have the advantages of flexible and convenient operation, on-demand access to data and high spatial resolution, which provide a new way for studying phenomics and genomics. As improving the accuracy and adaptability of the yield estimation model is a prerequisite for the application of UAV remote sensingthe objective of the research is to build crop yield prediction models that combine crop physiology and remote sensing parameters to improve the accuracy of yield prediction by UAV-RSP.

Field experiment tested 98 breeding materials of soybean was conducted by NERCITA in 2015. Unmanned aerial platform equipped with digital camera, multi spectral camera and hyper-spectrometer was used for field-based high-throughput phenotyping of soybean in breeding plots. Ten vegetation indices (VIs) including NDVI, RVI, GNDVI, PVI, OSAVI, EVI, DVI and NDVI₇₀₅ and plant height, combining algorithms including partial least-squares regression (PLS), multiple linear regression (MLR) and multiple stepwise regression (MSR) have been adopted for predicting yield of soybean in breeding plots. The results showed that MLR performed best, with R², NRMSE and d values of 0.83, 6.61 and 0.95, respectively, while MSR had the lowest accuracy, with R², NRMSE and d values of 0.72, 8.59 and 0.91, respectively.

The UAV-based field phenotyping platform can be as a preliminary method for screening cultivar in soybean breeding. The UAV platform equipped with multi-sensors was able to identify the differences of yields among the cultivars of soybean. Combing proximal sensing data and crop physiological traits can improve the accuracy of yield prediction in soybean breeding. The UAV-based proximal sensing platform provided novel insights in accelerating the breeding efficiency.

Leaf area index(LAI) and above ground biomass (AGB) are considered as two most important physiological parameters for crops, its correct estimation has been the focus in the crops growth monitoring and yield prediction. With the development of remote sensing technology and its application in agriculture, LAI and ABG estimation have been one of the main research areas of agricultural remote sensing. In recent years, unmanned aerial vehicle (UAV) technology has emerged as a new platform for remote sensing sensors, can provide remote sensing image in higher temporal resolution and spatial resolution. This study provides insight into the LAI and ABG estimation using an ASD Field Spec 3 spectrometer on the ground and applied to mapping using an UHD 185 hyperspectral sensor on board an UAV. The UAV was an 8-propellered UAV platform with 6 kg take-off weight, 50 meters flying height and 8m/s speed. The canopy spectral of winter wheat was measured by two methods, a DJI-S1000 UAV equipped with UHD 185 hyperspectral sensor fly 50m high, and a ASD Field Spec 3 spectrometer on the ground. By using 550nm, 680nm and 800nm spectral from ASD Field Spec 3 spectrometer, a linear model for LAI and ABG estimation was established. Results indicate that the R² was 0.78 and 0.60, RMSE was 0.60m²/m² and 1.46 t/ha, MAE was 0.48 m²/m² and 1.19 t/ha, respectively. After the hyperspectral image fusion and mosaic processing, hyperspectral image of whole study area was obtained. We applied the models for LAI and ABG estimation on the hyperspectral image of whole study area, and completed the winter wheat LAI and ABG monitoring, with results of LAI and ABG as follows: R² was 0.76 and 0.44, RMSE was 0.64 m²/m² and 1.48 t/ha MAE was 0.54 m²/m² and 1.15 t/ha. The results suggest the UAV UHD 185 hyperspectral system and the models have high application potential.

The satellite data application in agricultural has a long history. The new
developments of satellite series in China and Europe are bringing new opportunities
to advance the agricultural monitoring with abundant satellite data. The Sentinel
and GF are both quite similar high resolution satellite series onboard European
and Chinese satellites, respectively. The Proba-V and FY3-MERSI both have quite
similar channels and their own advantages in the medium to low resolution
satellite. This project is going to focus on the crop mapping, crop condition
monitoring and crop drought monitoring with both satellite data. The Ningxia
Hui autonomous region, one of small size provinces in China, was selected as
the study area for the orchard and crop mapping study with GF and Sentinel
optical satellite data. The field survey was conducted in June, 2016 and will
be organized again in June this year. The GF satellite data were collected as
much as possible for the coverage of Ningxia in the growing season. The processing
method of GF data is now developing in order to automatically ingest large
volume data in the next step. The preliminary results will be hopefully presented
at the symposium. Another two major agricultural production areas in China, North
China Plain and Northeast China Plain were also selected for the crop
monitoring and crop mapping with both medium to low resolution satellite data. The
field surveys were conducted in summer 2016 and spring in 2017. The further
field surveys will be organized in summer this year. The relevant Proba-V satellite
data have been downloaded and a processing code was developed to extract the Proba-V
data for the area of interesting. The FY-MERSI process chain has been developed
in the past. The classification method is still in developing. Hopefully the
methods and the preliminary results may be reported at the symposium.

Abstract: The orchard is developing very fast in Ningxia Hui autonomous region in Northwest China with the development of wine brewing Companies in recent year. Agrometeorological monitoring for the growth of grapes and other fruit trees is one of key process to identify the high quality fruit in the region but there is lack of the baseline map of orchard in Ningxia. This study aims at developing an orchard map in Ningxia in support of the agrometeorological service. The study collected high resolution GF satellite as much as possible which resolutions are around 8 t0 16 meters for multispectral bands. The decision tree was used to classify the images. All classified images were merged and finally an orchard map was formed. The validation was conducted with the field data.

The discomfort index (DI) is an important indicator that measures human heat sensation for different climatic conditions. Currently, the DI of a city is usually calculated using a few weather stations and hence does not accurately represent various thermal discomfort states of the city as a whole. This is a considerable drawback taken the importance of the index for assessing the quality of life within urban agglomerations and thus facilitating measures for smart urbanization. In this study a technique to produce fine-scale DI maps is proposed and applied accordingly. The technique is based on the combination of Sentinel-2 and Landsat 8 images with in-situ measured meteorological data. The DI map clearly reveals the spatial details of the DI in different locations of the city and thus supports focused interventions with the potential to support the operation of a city as smart.

Urban land surface temperature (LST) is an important parameter. However, due to the contradiction between the temporal and spatial resolution of thermal infrared remote sensing data, it’s difficult to obtain high temporal- spatial resolution LST data simultaneously. In order to solve this problem, this author considered multiple city characteristical parameters and used a simple model that can adjust the spatial distribution of scaling factors, combining with the SVM model to establish a new surface temperature spatial downscaling method which was suitable for city surface, named A-SVM.

In this paper, Beijing was chose as research area. The author divided the study into four steps. Firstly, correlation analysis and PCA (Principal Component Analysis) were used to select the parameters strongly related-to LST but independent respectively as the regression kernels (scaling factors) to do downscaling regression. The result shows that there are 6 factors were selected to do SVM regression, namely NDVI, UI, MNDWI, VAP, BAP and WAP. Screening of scale factors is the basis of SVM and A-SVM methods.

Secondly, three methods were used to verify the accuracy of the A-SVM model, which are direct resampling, TsHARP method and SVM model. The direct resampling method is a strategy that does not use any auxiliary data to directly resample a low-resolution image to a high spatial resolution. The TsHARP method is an improvement to the DisTrad quadratic model, using the negative correlation between NDVI and LST, and its basic assumption is that the linear relationship between LST and NDVI is scale invariant. SVM is a newer statistical learning theory proposed by Vapnik. Using the SVM model for low resolution to high resolution, replacing the low-resolution scale factors to high-resolution scale factors and further adding the model estimation error，high-resolution LST can be obtained.

There is a simple model to dynamically adjust the scale factor spatial distribution according to the Standard Deviation of scale factors from high resolution and low resolution LST. A-SVM model was built based on this method and SVM model. By considering the relationship between high-resolution and low-resolution images, the purpose of the model-building is to make the remote-sensing indices of the high spatial resolution image have the same spatial distribution as those of the low spatial resolution image, which will reduce the downscaling estimation error of different resolutions.

Finally, the author got the results of retrieved LST from four methods and the scatter diagram of estimated 120m LST from four methods against the LST received from TM. The 120m LST which was directly retrieved from TM thermal infrared band was selected as the real LST and verification data. Results shows that the improved method could obtain ideal downscaling results, whose RMSE was 1.82 K, R² was 0.85, and ERGAS index achieved 0.07, and it was superior to the direct resampling method, the classical method TsHARP and the SVM method whose scaling factors were unadjusted.

Through this research, this author got two important conclusions:

(1)To reflect the complex characteristics of urban surface, multiple related parameters with LST including NDVI, UI, MNDWI,VAP,BAP,WAP were chosen as the scaling factors based on the correlation and PCA analysis, which can partly show the change of urban LST and avoid redundancy phenomenon between each other.

(2) Based on the 6 kinds of scaling factors, a new adjusted-SVM spatial downscaling method for urban land surface (A-SVM) was developed combining the SVM model with a space distribution adjustment model, which could avoid the ideal assumptions that the relation between scaling factors and LST was invariant. The validation result shows A-SVM method can get the better prediction precision.

Urbanization affects considerably the thermal environment of cities and influence the spatial and temporal consumption of energy for heating and cooling. The increase of impervious surfaces alongside with the reduction of vegetated areas lead to increased air and surface temperatures. Remote sensing data is suitable for up-to-date urban land use mapping and for the assessment of the thermal environment of urban areas. In this study a statistical approach is developed on the basis of satellite data, in order to identify the areas where maximum and minimum temperature values are observed; the approach is tested for the city of Athens, Greece. The recognition of such areas is important as they reflect areas where immediate interventions are necessary to ameliorate the thermal environment, whereas the knowledge of their spatial distribution and temporal variations is needed for smart urbanization.

This abstract presents a detailed application of feature extraction methods on PolSAR records for built-up area identification. This work aims at providing a thorough description of the multivariate patterns insisting on roll-variant and roll-invariant quad-POLSAR features.

Indeed, the detection and classification of urban areas in the instantaneous field of view can take a great advantage from processing the whole amount of records and parameters obtained from POLSAR images. In fact, multivariate processing is used to deliver a thorough understanding of the relationships and regularities that can be retrieved over the selected areas for each target class. Hence, the rationale for employing both roll-variant and roll-invariant features is the intimate involvement of all the attributes acquired by POLSAR remote sensing with the anthropogenic extents distribution, which is used to provide a complete characterization of the built-up areas and of the hidden regularities lying within the records as well.

In order to provide such a multivariate analysis for feature extraction, two methods which aim at retrieving a solid recognition of significant patterns within the dataset have been taken into account. First, we employed a method named Hierarchical Binary Decision Tree (HBDT) [1], which combines different processing chains (made by a feature selection and a classification step) and automatically adapts to the spatial (and spectral) properties of the classes available in the scene. Typically, in homogeneous decision trees, at each node the same algorithm is used for separation between groups of classes. In our case, the most useful processing chain, composed by the most suitable feature set and the most efficient classifier, is selected per each node. This selection is performed by computing an intermediate accuracy assessment for each processing chain. Only the feature set (or its most noiseless subset) and classifier pair that produces the best result is selected and assigned to the node. The procedure is then repeated removing the already identified class/classes until all the nodes are identified.

Moreover, we considered a method based on mutual information maximization to explore the dataset in order to detect relevant patterns for built-up area extraction [2]. Specifically, the method that has been developed aims at providing a identifying affinity patterns, which better describe each class within the considered dataset. Good classification performances are obtained by selecting patterns fulfilling the Pareto optimality and by properly modeling a combination of the information provided by each pattern. Since the proposed approach is completely data driven and relies on information theory-based quantities, it is very flexible and totally independent from the statistics of the classes, and allows exploring datasets consisting of heterogeneous features.

Two datasets acquired by Radarsat-2 Quad-PolSAR sensor over the San Francisco area have been considered for testing. Specifically, they consist of 28 and 113 roll-invariant and roll-variant features, respectively. The aforementioned algorithms have been used to investigate the POLSAR records in order to obtain a precise characterization of the patterns that describe anthropogenic extents in the considered area. In order to accurately assess the detection and classification performance of the aforesaid methods, we analyzed the considered datasets by means of algorithms relying on a different approach. Specifically, we used methods based on ensemble learning [3] (such as random cluster ensemble and recursive feature elimination scheme) in order to explore the relationships among the data. Results show that the methods relying on the multiple feature pattern recognition are able to provide accurate extraction of built-up areas over both the considered datasets. Indeed, they are able to outperform existing algorithms while guaranteeing acceptable computational complexity costs, so that they represent a valid option for POLSAR feature extraction applied to identification of built-up areas.

Land subsidence induced by excessive groundwater withdrawal has caused serious social, geological, and environmental problems in Beijing. Rapid increases in population and economic development have aggravated the situation. Monitoring and prediction of ground settlement is important to mitigate these hazards. In this study, we combined persistent-scatterer interferometric synthetic aperture radar (PS-InSAR) with Grey system theory to predict the evolution of land subsidence in the Beijing plain. Land subsidence during 2003–2014 in the Beijing plain was determined based on 39 ENVISAT Advanced Synthetic Aperture Radar (ASAR) images and 27 RadarSat-2 images. Results were consistent with global positioning system (GPS), leveling measurements at the point level and TerraSAR-X subsidence maps at the regional level. It was demonstrated that the land surface in the Beijing plain is settling at an accelerating rate. The average deformation rate in the line-of-sight (LOS) was from −124 mm/year to 7 mm/year during 2003–2014; accumulative displacement was up to 1426 mm by the end of 2014. To predict future subsidence, the evolution of deformation was used to build a prediction model based on an improved Grey-Markov model (IGMM), which adapted the conventional GM(1,1) model by utilizing rolling mechanism and integrating a k-means clustering method in Markov-chain state interval partitioning. Evaluation of the IGMM at three representative points showed good accuracy of simulated subsidence values (root-mean-square error <3 mm). Simulated deformation during 2013 and 2014 agreed well with the observed deformation during each year based on PS-InSAR (R² = 0.94 and 0.91). Finally, InSAR measurements from 2003–2014 were used to predict subsidence in 2015–2016. It was calculated that the maximum cumulative deformation will reach 1717 mm by the end of 2016 in Beijing plain. The promising results indicate that this method provides an alternative to the conventional numerical and empirical models in order to predict short-term deformation when there is lack of detailed geological or hydraulic information.

Change detection of urban extents is now a very important topic for urban remote sensing mapping at regional and global scales. Indeed, since more and more single-date and global data sets are becoming available, change detection of urban areas is feasible. However, so far it ahs been focused on binary changes, i.e. changes that affect (either positively or negatively) the urban area extents, by considering new urbanized areas or zones given back to the natural environment.

This work aims at providing a few preliminary results of a more accurate analysis, where the density of built-up areas, and the change from low-rise to high-rise buildings (and vice versa) is also considered. The rationale for this analysis is that these situation correspond to change that mostly go unnoticed when urban extents only are considered. Still, these changes are extremely important to understand phenomena related to change in population density, leading often to overcrowding effects, and subsequently to risk management issues.

To find a sound methodology and extract this more complex changes, we explore the use of SAR, and specifically polarimetric SAR, data, under the assumption that the electromagnetic field backscattered from built-up area is affected by the geometric properties of the illuminated structures, and therefore it is different according to the type of built-up elements located in a specific geographical area. Accordingly, a segmentation of the SAR image is first performed at the first date of thr temporal sequence to be analyzed, and then multiple polarimetric parameters (in addition to the backscattered intensity) are considered, to understand which are the parameters that are more useful to analyze to track not only 2D, but also 3D changes and density changes in urban areas.

The test area for this work is the city of Nanjing, and the data sets that were used are obtained from Radarsat-2 fully polarimetric images. A complete analysis of segmentation using different object size, applied to different polarimetric decomposition parameters, and validate using both positive and negative 2D and 3D changes in urban areas have been performed. They show that it is possible to extract a richer analysis for urban changes than the simple detection of changes in urban extents in a bi-dimensional sense. However, a clear and quantitative description of the 3D change is still to be obtained, and the methodology needs to be improved to reach this goal.