Crop classification provides relevant information for crop management, food security assurance and agricultural policy design. The availability of Sentinel-1 image time series, with a very short revisit time and high spatial resolution, has great potential for crop classification in regions with pervasive cloud cover. Dense image time series enable the implementation of supervised crop classification schemes based on the comparison of the time series of the element to classify with the temporal signatures of the considered crops. The main objective of this study is to investigate the performance of a supervised crop classification approach based on crop temporal signatures obtained from Sentinel-1 time series in a challenging case study with a large number of crops and a high heterogeneity in terms of agro-climatic conditions and field sizes. The case study considered a large dataset on the Spanish province of Navarre in the framework of the verification of Common Agricultural Policy (CAP) subsidies. Navarre presents a large agro-climatic diversity with persistent cloud cover areas, and therefore, the technique was implemented both at the provincial and regional scale. In total, 14 crop classes were considered, including different winter crops, summer crops, permanent crops and fallow. Classification results varied depending on the set of input features considered, obtaining Overall Accuracies higher than 70% when the three (VH, VV and VH/VV) channels were used as the input. Crops exhibiting singularities in their temporal signatures were more easily identified, with barley, rice, corn and wheat achieving F1-scores above 75%. The size of fields severely affected classification performance, with ~14% better classification performance for larger fields (>1 ha) in comparison to smaller fields (<0.5 ha). Results improved when agro-climatic diversity was taken into account through regional stratification. It was observed that regions with a higher diversity of crop types, management techniques and a larger proportion of fallow fields obtained lower accuracies. The approach is simple and can be easily implemented operationally to aid CAP inspection procedures or for other purposes.

The high revisit time of Sentinel-1 (S1) observations enables the design of crop type mapping approaches exploiting the backscatter time series observed for the different crops. The objective of this study is to propose a supervised crop classification methodology based on the temporal signature of crops. With this aim 29 dual-pol S1 observations acquired over an agricultural area of Spain, where ground truth was available, were processed. The classification approach was based on the temporal signatures obtained for each polarization channel (VH, VV and the cross-pol ratio) for the different crops. Highest accuracies were obtained when fields were assigned to the class that minimized the RMSE, with an overall accuracy of 79% and best results for rapeseed, sunflower, alfalfa and barley.

Wheat is one of the most important crops worldwide, and thus, the use of remote sensing data for wheat monitoring has attracted much interest. Synthetic aperture radar (SAR) observations show that at C-band and VV polarization, wheat canopy attenuates the surface scattering component from the underlying soil during a significant part of its growth cycle. This behavior needs to be accounted for or corrected before soil moisture (SM) retrieval is attempted. The objective of this article is to develop a new method for wheat attenuation correction (WATCOR) applicable to Sentinel-1 VV time series and based solely on the information contained in the time series itself. The hypothesis of WATCOR is that without attenuation, VV backscatter would follow a stable long-term trend during the agricultural season, with short-term variations caused by SM dynamics. The method relies on time-series smoothing and changing point detection, and its implementation follows a series of simple steps. The performance of the method was compared by evaluating the correlation between backscatter and SM content in six wheat fields with available SM data. The water cloud model (WCM) was also applied as a benchmark. The results showed that WATCOR successfully removed the attenuation in the time series and achieved the highest correlation with SM, thus improving markedly the correlation of the original backscatter. WATCOR can be easily implemented, as it does not require parameterization or any external data, only an approximate indication of the period where attenuation is likely to occur.

Crop monitoring is essential for different applications such as food security assurance, crop management, and the design and implementation of agricultural policies. Remote sensing provides information about biophysical properties of plants and soils and their spatial variability on large areas of the territory on a periodic basis. The launch of Sentinel-1 synthetic aperture radar (SAR) satellites in 2014 and 2016 made possible the acquisition of dense time series of images with good spatial resolution and temporal resolution even in cloud-covered areas. The main objective of this thesis is the evaluation of different methodologies for agricultural applications at the field scale using Sentinel-1 time series. First, a supervised crop classification methodology based on time signatures from Sentinel-1 images was proposed and implemented in a case study with 14 crop classes and a large dataset of agricultural fields. Secondly, the influence of acquisition geometry of Sentinel-1 images over wheat fields was assessed. The influence of the incidence angle on backscatter coefficients and terrain-flattened coefficients was evaluated, and an incidence angle normalization followed by an azimuthal anisotropy correction were applied to VV polarized time series. Thirdly, the backscatter attenuation produced by wheat canopy in VV polarization was evaluated, and a new wheat attenuation correction methodology named WATCOR was proposed. Finally, four different soil moisture (SM) estimation techniques based on Sentinel-1 time series were evaluated over wheat fields, proposing different methodological alternatives for their application at the field scale. Despite the complexity of estimating SM at the field scale solely with SAR data, acceptable estimations were obtained. The results of this thesis showed that the analysis and extraction of the information contained in SAR time series is useful for various agricultural applications, foreshadowing exciting future developments in this field.