Fires are disruptive events that should be carefully studied. To this date, the monitoring at large scale of fire events is mainly performed using low spatial resolution sensors such as MODIS and VIIRS and ancillary data like active fire maps. However, the data produced by the new generation of spaceborne multispectral missions can be used for burned area detection at a finer spatial and temporal scale with respect to existing methods. In this paper we present a method that analyzes time-series of optical images acquired by Landsat-8 and Sentinel-2 to detect burned areas in a fully automatic and unsupervised way. The method analyzes each image in the time-series to extract Candidate Burned Areas (CBAs). CBAs are analyzed accounting for the temporal correlation to reduce the number of false alarms. In particular, only pixels that have a temporal profile consistent with the physical event of a fire are selected. Moreover, the time variable is used to compute the confidence of the detection. The method has been tested on a time-series of the Attica region, Greece, which was affected by large fires in 2018. Preliminary experimental results showed that the method correctly identified the burned areas of the region.

The scope of this dissertation is to present and discuss novel paradigms and techniques for the extraction of information from long time series of remotely sensed images. Many images are acquired everyday at high spatial and temporal resolution. The unprecedented availability of images is increasing due to the number of acquiring sensors. Nowadays, many satellites have been launched in orbit around our planet and more launches are planned in the future. Notable examples of currently operating remote sensing missions are the Landsat and Sentinel programs run by space agencies. This trend is speeding up every year with the launch of many other commercial satellites. Initiatives like cubesats propose a new paradigm to continuously monitor Earth’s surface. The larger availability of remotely sensed data does not only involve space-borne platforms. In the recent years, new platforms, such as airborne unmanned vehicles, gained popularity also thanks to the reduction of costs of these instruments. Overall, all these phenomena are fueling the so-called Big Data revolution in remote sensing. The unprecedented number of images enables a large number of applications related to the monitoring of the environment on a global and regional scale. A non-exhaustive list of applications contains climate change assessment, disaster monitoring and urban planning. In this thesis, novel paradigms and techniques are proposed for the automatic exploitation of the information acquired by the growing number of remote sensing data sources, either multispectral or Synthetic Aperture Radar (SAR) sensors. There is a need of new processing strategies being able to reliably and automatically extract information from the ever growing amount of images. In this context, this thesis focuses on Change Detection (CD) techniques capable of identifying areas within remote sensing images where the land-cover/land-use changed. Indeed, CD is one of the first steps needed to understand Earth’s surface dynamics and its evolution. Images from such long and dense time series have redundant information. So, the information extracted from one image or a single image pair in the time series is correlated to other images or image pairs. This thesis explores mechanisms to exploit the temporal correlation within long image time series for an improved information extraction. This concept is general and can be applied to any information extraction process. The thesis provides three main novel contributions to the state of the art. The first contribution consists in a novel framework for CD in image time series. The binary change variable is modeled as a conservative field. Then, it is used to improve the bi-temporal CD map computed between a target pair of images extracted from a time series. This framework takes advantage of the correlation of changes detected between pairs of images extracted from long time series. The second contribution presents an iterative approach that aims at improving the global CD performance for any possible pair of images defined within a time series. The results obtained by any bi-temporal technique, either binary or multiclass, are automatically validated against each other. By means of an iterative mechanism, the consistency of changes is tested and enforced for any pair of images. The third contribution consists in the detection of clouds in long time series of multispectral images and in the restoration of pixels covered by clouds. The presence of clouds may strongly affect the automatic analysis of images and the performance of change detection techniques (or other processes for the extraction of information). In this contribution, the temporal information of long optical image time series is exploited to improve the identification of pixels covered by clouds and their restoration with respect to standard monotemporal approaches. The effectiveness of the proposed approaches is proved on experiments on synthetic and real multispectral and SAR images. Experimental results are accompanied by comprehensive qualitative and quantitative analysis.

Missing information in optical remote sensing data due to sensor malfunction or cloud cover reduces their usability. Although several methods in the literature allow an accurate image restoration, they are usually time-consuming and thus impractical for operational applications. In this paper, we propose a Fast Cloud Removal Approach (FCRA) that exploits the multitemporal information in an efficient way to sharply reduce the computational burden. In greater details, the proposed unsupervised method: (i) generates a short time series (TS) of cloud free images temporally close to the cloudy one (i.e., the target image), (ii) detects ground-clear pixels in the target image similar to the missing pixels using the multitemporal information (i.e., similar temporal patterns), and (iii) restores the occluded pixels using the most similar ground-clear spectral values from the same image. The method reduces the computational effort by focusing on short image TS and by detecting similar temporal patterns in a fast and effective way by using the KD-trees. Moreover, since the method restores the missing pixels with ground clear pixels belonging to the same image, no computational demanding approaches are needed to adjust the replacement information (i.e., regression). Experimental results obtained on Sentinel-2 images on simulated clouds acquired over two test areas located in Italy demonstrate the effectiveness of the proposed method.

This paper presents a multitemporal technique for multi-class Change Detection (CD) between pairs of images of a satellite image time series. Changes between different pair of images within a time series must be consistent with each other since images acquired over the same scene are causally related with one another. The temporal consistency of the pixel status can be used to formulate a principle that constrains the CD results within the series to be mutually consistent. This principle coincides with the conservative property of the change variable and it allows the unsupervised validation of changes detected between arbitrary image pairs. Thus, all images in the series, rather than a single couple, are used in the pair-wise CD. The proposed technique was applied to a dataset of dual-polarized terrain-corrected SAR images acquired by Sentinel-1. Experimental results show the validity of the proposed multitemporal approach in improving the CD results.

A sheer amount of data is collected everyday by a large variety of remote sensing sensors and new acquisitions are continuously added to data records of existing Earth Observation archives. The unprecedented amount of information acquired on a study area can be useful for many remote sensing applications, but requires the solution of Big Data challenges. Based on these observations, the Circular change detection has been proposed as a novel framework that redefines the bi-temporal change detection problem to take advantage of a full image time series. It evaluates the binary change variable in circular closed paths where it is a conservative quantity. In this paper, a strategy inspired by the 2D phase unwrapping is applied to this conservative variable to locate CD errors in pairs of images within the time series. The effectiveness of the proposed approach is established by experimental results obtained on synthetic and real datasets.

Change detection (CD) between a pair of images is a popular problem in remote sensing. Despite a large amount of data is acquired every day by remote sensing satellites, standard CD methods usually consider only the two target images between which we desire to detect changes. The aim of this work is to present a novel framework in which the bi-temporal CD is redefined by evaluating the consistency of the changes occurred in the target image pair with all the other changes of images within the considered time series. Our approach evaluates pixel-wise the changes in temporal closed-loops that include the two target images where the resulting binary change/no-change sequences can be processed by strategies inspired to the error-control-coding theory. Unreliable CD results for the target images can be identified and corrected. The experimental results on both a synthetic and a real dataset demonstrate the effectiveness of the proposed framework.

Forest height is one of the most important parameters for (forest) stand characterization, and it is closely related to biomass. It provides information on stand condition and site index, and it allows characterizing the successional state of the forest. At the same time, the distribution of forest heights within a stand may be used to assess the disturbance regime and to detect logging activities. In the characterization of dynamic forest processes, the knowledge of forest height changes is even more important than absolute height measurements itself. Indeed, forest height changes can be directly used to characterize forest growth, mortality and deforestation, and to conclude about the associated carbon fluxes independently from the successional status of the forest. Forest densdormetric variables are usually measured by airborne laser scanning by using LiDAR instruments that are very precise in characterizing the tree heights and the main related parameters. However, LiDAR technology is expensive and in many cases not affordable, especially in a multitemporal monitoring scenario. Low frequency synthetic aperture radar (SAR) sensors in a polarimetric-interferometric configuration represent a very promising alternative tool to LiDAR for estimating forest height and its dynamics as they can provide large continuous coverage and high spatial and temporal resolution with a limited cost. From the methodological point of view, polarimetric SAR interferometry (PolInSAR) combines the inherent sensitivity of the interferometric coherence to the vertical structure of volume scatterers combined with the potential of SAR polarimetry to interpret and characterise the individual scattering processes. Exploiting this combination, forest height estimations obtained in the last years from a pre-operational to an operational PolInSAR product that have been validated in the frame of several campaigns over a wide range of forests, terrains and environmental conditions. The overall obtained estimation error is in the order of 10% or smaller. The objective of this work is to assess qualitatively and quantitatively the capability of monitoring forest dynamics by using PolInSAR height maps. To this purpose, a multibaseline-multitemporal PolInSAR L-band airborne dataset over the temperate forest site of Traunstein (South of Germany) has been used. Data have been acquired since 2003 with the DLR’s E- and F-SAR radar platforms covering more than a decade, with temporal baselines ranging from hours to years. For each acquisition date, a multibaseline PolInSAR inversion with short temporal baseline (in the order of one hour) has been carried out. Accordingly, a time series of accurately estimated forest heights has been obtained without disturbances due to (long-term) temporal decorrelation. Depending on the time lag between two different height maps, several comparisons could be carried out in order to characterize height differences induced by scattering changes within the same stand and/or variations of the baseline distribution. Such scattering changes can in general be caused by different sources, like the water content of the vegetation (e.g. before and after a rain event), the presence of wind, the seasonal cycles, disturbances and management activities. These components could be observed in the data and their effects characterized. Extensive results of this analysis will be presented, and the capability of distinguishing between different sources of PolInSAR height changes will be discussed together with their potentials for identifying and monitoring ecosystem changes.

Microwaves can propagate through vegetation layers, allowing the radar signal to interact with the different physical forest structure elements. Exploiting this property, SAR Tomography techniques provide a 3D profile of the backscattered radar power providing a proxy to the physical forest structure. This capability has been demonstrated by means of several (long wavelength) airborne campaigns, which have triggered the development of a new generation of forest structure monitoring applications by means of SAR remote sensing. At the same time, the progress in SAR technology makes possible the implementation and operation of a new generation of spaceborne SAR configurations able to realize these applications on a global scale with a high spatial and temporal resolution. The objective of this work is to present the status and to address potentials and challenges related to applications of SAR Tomography for 3D forest structure characterization and monitoring. A number of issues will be addressed, in particular: 1.the performance of several reflectivity estimators (model-based and not) with respect to acquisition configuration (e.g. geometry and number of baselines, availability of multiple polarization channels, frequency, etc.), and different seasonal and environmental conditions (leaves on/off, dry/wet acquisition days, etc.); 2.the interpretation of the reflectivity profiles, by means of a comparison with ground measurements, lidar profiles and forest simulations. In this, a key role is played by the variability of the profiles with seasons and environmental conditions; 3.the possibility and the methodologies to distinguish between different physical forest structure types by using the reflectivity profiles. In the presentation, several experimental results will be presented. Airborne (L-band polarimetric repeat-pass interferometric) SAR data will be used. Data have been collected under different environmental and seasonal conditions systematically over a period of 12 years (2003-2015) over the Traunstein test site by means of the DLR’s airborne platforms E-SAR and F-SAR. Traunstein is a heterogeneous temperate forest located in southern Germany, with a wide range of stand compositions, stand heights (from 10 to 40m) and stand biomass (mean biomass level of 200 Mg/ha and stands up to 500 Mg/ha). Radar results have been validated by means of extensive ground measurements and lidar profiles.