Weeds are one of the major reasons for crop yield loss but current weeding practices fail to manage weeds in an efficient and targeted manner. Effective weed management is especially important for crops with high worldwide production such as maize, to maximize crop yield for meeting increasing global demands. Advances in near-sensing and computer vision enable the development of new tools for weed management. Specifically, state-of-the-art segmentation models, coupled with novel sensing technologies, can facilitate timely and accurate weeding and monitoring systems. However, learning-based approaches require annotated data and show a lack of generalization to aerial imaging for different crops. We present a novel dataset for semantic and instance segmentation of crops and weeds in agricultural maize fields. The multispectral UAV-based dataset contains images with RGB, red-edge, and nearinfrared bands, a large number of plant instances, dense annotations for maize and four weed classes, and is multitemporal. We provide extensive baseline results for both tasks, including probabilistic methods to quantify prediction uncertainty, improve model calibration, and demonstrate the approach's applicability to out-of-distribution data. The results show the effectiveness of the two additional bands compared to RGB only, and better performance in our target domain than models trained on existing datasets. We hope our dataset advances research on methods and operational systems for fine-grained weed identification, enhancing the robustness and applicability of UAVbased weed management. The dataset and code are available at https://github.com/GFZ/weedsgalore.

The synthetic aperture radar (SAR) has long been used from satellites for forest monitoring at global level. The boreal forests in Sweden are well described wall-to-wall from airborne laser scanning and airborne photography. Hence, in Sweden, SAR can often only provide a limited added value, due to the low resolution compared to other sensors, despite its all-weather acquisition capabilities. By accounting for interfering effects that currently degrade the useful information in SAR images, it can be extremely valuable for both vegetation mapping and belowground mapping (e.g., soil conditions and tree roots). In the current work, we present the configuration of the first drone-based SAR experiment that allows us to image the forest in 3D with very high spatial resolution. We have started the analyses by using tomography to derive reflectivity for the roots of single trees, and comparing these with reference root biomass. The linear relationship indicates a potential for using SAR to derive forest variables that were yet neglected or little researched. Moreover, extensive additional remote sensing data have been collected from both airborne and spaceborne platforms, and reference data for both the vegetation and soil have been inventoried in-situ using complementary measurements and sensors. Hence, this unique experimental setup enables many unprecedented analyses about SAR applied to boreal forests.

ESA Forest Carbon Monitoring project (FCM) is developing Earth Observation based, user-centric approaches for forest carbon monitoring. Forest carbon accounting based on forest inventory requires precise and timely estimation of forest variables at various spatial levels accompanied by verifiable uncertainty information. In this paper, we present the algorithm trade-off and selection approach and preliminary results of the algorithm intercomparison exercise in the FCM project. The studies were performed over 7 European test sites located in Finland, Ireland, Romania, Spain and Switzerland, and one tropical forest site in Peru. EO datasets were represented by Sentinel-1, Sentinel-2, TanDEM-X and ALOS-2 PALSAR-2 imagery. Examined approaches include popular parametric and SAR/InSAR scattering physics based approaches, and nonparametric and machine learning approaches such as k-NN, random forests, support vector regression.

Maps of above-ground biomass (AGB) maps derived from Earth Observation (EO) have been increasing as a reflection of their importance in global and national applications such as global climate modelling and carbon accounting, respectively. Here we present Plot-To-Map, a tool to validate the accuracy of AGB maps using plot data from freely available global plot networks or own user data. The tool consists of five main steps starting from quality checking and harmonizing plot data to address the spatial (e.g., forest area) and temporal gaps between plots and maps. Uncertainties of plots are assessed before map comparison so plots with high uncertainty will have less impact on the comparisons. The comparison is initiated at a coarse spatial scale e.g., 10 km to eliminate random errors and enable systematic errors (bias) to be revealed and eventually be modelled. Bias predictions from decision tree-based regression model and other ecological EO products are key for producing bias-reduced maps and estimates of country AGB with associated uncertainties. The tool can be used locally and interactively via https://github.com/arnanaraza/PlotToMap. Its access and use are also possible through the Multi-Mission Algorithm Platform (MAAP), a joint project of ESA and NASA to revolutionize AGB research.

Deep learning (DL) has proven to be a suitable approach for despeckling synthetic aperture radar (SAR) images. So far, most DL models are trained to reduce speckle that follows a particular distribution, either using simulated noise or a specific set of real SAR images, limiting the applicability of these methods for real SAR images with unknown noise statistics. In this article, we present a DL method, deSpeckNet,1that estimates the speckle noise distribution and the despeckled image simultaneously. Since it does not depend on a specific noise model, deSpeckNet generalizes well across SAR acquisitions in a variety of landcover conditions. We evaluated the performance of deSpeckNet on single polarized Sentinel-1 images acquired in Indonesia, The Democratic Republic of Congo, and The Netherlands, a single polarized ALOS-2/PALSAR-2 image acquired in Japan and an Iceye X2 image acquired in Germany. In all cases, deSpeckNet was able to effectively reduce speckle and restore the images in high quality with respect to the state of the art.

Estimates of the area of land cover classes or land change are frequently calculated from land cover classification maps by counting the pixels labeled as each class in the map. This procedure is known to produce biased estimates of area for many widely used classification algorithms, including random forests. Poststratification estimation using the mapped classes as strata has been proposed to obtain unbiased estimates of the class areas. Still, the method requires additional sampling units, which may not be available or be the most efficient method depending on the application. Alternatively, consistent estimates of class areas can be obtained using class membership probabilities estimates from a random forest classification. This article demonstrates that, for a large sample and proper set of explanatory variables, the error of the predicted class membership probabilities obtained from a random forest classification converges to zero. Therefore, the expected class areas calculated from these probabilities converge to the population class areas. On average, the relative error of the expected class proportions computed by class membership probabilities from a random forests model was 40% points lower than the proportions estimated by pixel counting. Our proposed approach is also comparable to the area-adjusted method, which is currently considered the best practice by the remote sensing community. We recommend that class probability estimates area always retained and used for calculating expected class areas or area proportions based on our results. Our method reduces bias compared to statistics calculated by pixel counting and circumvents the need for poststratification area estimates under certain conditions.

Maps of above-ground biomass (AGB) using remote sensing (RS) are valuable to countries like the Philippines for multi-purposes including national greenhouse gas reporting, carbon accounting and even reforestation monitoring. As RS data increase, both optical and radar satellite data have been combined to produce higher quality AGB maps, rather than using individual satellites alone. AGB is then produced after establishing a statistical relationship between combined RS data and plot-based AGB often using machine learning (ML) regression tasks. Here we model a country-wide AGB map of the Philippines while assessing the effects of combining freely and easily accessible satellite data as AGB predictors: Landsat-8 (optical RS), and ALOS-2 PALSAR-2 and Sentinel-1 (radar RS). Using annual composites of each satellite data along with other ecological variables and the National Forest Inventory (NFI) 2014–2015, we trained, cross-validated and compared ML models: Random Forest (RF), Support Vector Machine (SVM) and Neural Network (NN). Model evaluation results were found similar between RF and SVM models i.e., 63 and 65 Mg ha−1RMSE respectively; while the NN model showed the lowest accuracy i.e.,$\text{RMSE} =84$Mg ha−1, likely a consequence of limited training data. The effect of combining Landsat-8 and ALOS-2 PALSAR-2 was indicated by better AGB predictions in the upland forests, while the inclusion of Sentinel-1 improved the AGB estimates in the agricultural lowland. Using the three satellite data and an RF model not only provides the map with the least bias, but also complements the ridge-to-reef topography of the country where woody vegetations exist from forested mountains down to agricultural lands and mangroves. This study is helpful also to other tropical countries in addressing the problem on expensive forest inventories as well as inaccessible forest lands and conflict areas, while providing spatially explicit and reliable AGB estimates.

As Forest Monitoring (FM) requirements for international policies and activities such as REDD+ increase, Earth Observation data becomes more and more available and initiatives to support FM such as Copernicus are developing. In this context, user needs for data and methods are varied, with several research gaps remaining. Based on policy requirements, user needs, and available research, priorities for research were identified: extraction of map statistics, uncertainty/accuracy assessment methods, analysis ready high-resolution sampling data, and information on Forest biomass stocks and change, and Forest related GHG emissions and removals information.

Despite the relevance of historical land cover maps for scientists and policy makers, an accurate high resolution record is currently lacking over the Sudano-Sahel. In this study, 30m resolution historically consistent land cover and cover fraction maps are provided over the Sudano-Sahel for the period 1986–2015. These land cover/cover fraction maps are achieved based on the Landsat archive preprocessed on Google Earth Engine and a random forest classification/regression model, while historical consistency is achieved using the hidden Markov model. Using these historical maps, a multitude of variability in the dynamic Sudano-Sahel region over the past 30 years is revealed. These include cropland expansion and the re-greening of the Sahel, forest degradation & the detection of fine-scale changes, such as smallholder or subsistence farming. The historical land cover / cover fraction maps are made available via an open-access platform.

Many land cover changes are a direct threat to nature, especially through its negative effects on ecosystem services and thus, poses a risk to long-term human health and wellbeing. Land cover data is requested by different users, i.e. it is key for many SDG indicators and therefore accurate and continuously updated land cover information is more essential than ever. However, the “let's produce a map and someone will use it” approach is not helpful as users' have their own requirements and needs in terms of detail, quality and repeatability. Hence, developments for mapping approaches and data sharing (satellite imagery as well as reference data) should focus on distributed, flexible, user-oriented approaches. Data providers and especially data distributors should put the users in charge in terms of data and processing access, geographic scope and customization of map products (thematic classes) and next generation land cover services have to offer a much more flexible and targeted land cover characterization framework than in the past.

Obtaining accurate and detailed structural forest information has been revolutionized with the emergence of laser scanning. The sampling limitations and potential of the different laser scanning platforms (e.g. TLS, UAV -LS) have, however, not been fully explored for dense tropical forests. We fused laser scanning data from the terrestrial (TLS) and drone (UA V -LS) platform for two dense tropical forest plots and calculated their vertical point density profiles to gain insight in their sampling abilities. Our results reveal the limitations of TLS to fully sample the top of the canopy of a dense tropical rainforest. We also demonstrate how multiple returns but also cheaper single returns UAV -LS systems can be applied to sample the forest structure.

A polarimetric Synthetic Aperture Radar (PoISAR) image is able to capture target backscattering properties in different polarimetric states, making it a rich source of information for target characterization. However, as with any SAR image, PolSAR images are affected by speckle. Therefore, to extract useful information about targets, the polarimetric covariance matrix has to be first estimated by reducing speckle. In this paper, we use a deep neural network to estimate the dual PolSAR covariance matrix. This application was compared against the state of the art PolSAR despeckling methods. Even if the method is agnostic on the structure of the covariance matrix, the deep learning based PolSAR covariance matrix estimation performed better than the state of the art PolSAR despeckling methods. These results showcase the potential of supervised deep learning for the improvement of PolSAR despeckling pipelines.

Large lithium-ion battery systems rely on battery monitoring and management systems to ensure safe and efficient operation. Typically the battery current, the cell voltages, and the cell temperatures are monitored. This paper describes the use of gas sensors in large lithium-ion battery systems in addition to conventionally used means of battery monitoring. An undetected electrolyte leak in a cell can pose a serious threat to users and maintenance personnel. Experiments described in this paper show that a gas sensor can easily detect volatile organic compounds (VOC) from the leaking electrolyte, whereas standard cell monitoring methods can only detect a leak indirectly over premature cell performance degradation. Therefore, gas sensors offer a fast, simple, and cost efficient way to increase the safety of battery systems. This paper gives a description of a suitable gas sensor and its application in a battery system, followed by the identification of relevant use cases. In the experimental section the performance of the gas sensor in these use cases is investigated and evaluated. The paper ends with a summary of the results and a short outlook.

The present study assesses the remote sensing challenges experienced by the project developers for operational monitoring of REDD+ (Reducing Emission from Deforestation and Degradation) projects. The study was carried out at a sample of 20 REDD+ projects in Brazil, Peru, Cameroon, Tanzania, Indonesia and Vietnam using a questionnaire survey and regional workshops on MRV (Monitoring, Reporting and Verification). The assessment showed that eleven project developers (55%) showed high or very high remote sensing and GIS capacity, seven (35%) were ranked medium, and two (10%) were ranked low. At the regional level, capacity tended to be highest in the projects in Brazil and Peru and somewhat lower in Cameroon, Tanzania, Indonesia and Vietnam. The study calls for an increased investment and capacity building to meet the various remote sensing challenges in REDD+ projects.

The ESA Climate Change Initiative land cover project focuses on the deriving land cover information driven by requirements for observing Essential Climate Variables. Consultation mechanisms were established with the climate modelling community in order to identify its specific needs in terms of satellite-based global land cover products. Key findings were the needs for successive land cover maps stable over time. As response, an innovative global land cover mapping approach, based on multi-year MERIS and SPOT-Vegetation datasets is proposed. Pre-processing and classification chains able to handle huge amount of data have been developed and a first global land cover map associated to the 2008-2010 epoch is being produced.

Empirical relations are frequently used to derive leaf area index (LAI). Such relations often make assumptions that make it hard to link the derived LAI to realistic trees and forest canopies. In previous work we developed a set of analytical expressions to describe LiDAR waveforms with only a limited number of assumptions based on radiative transfer. These expressions were a function of crown macro-structure and LAI. The expressions were successfully tested when applied on crown archetypes, but showed significant error when applied to more realistic crowns. In this study, we analyse the effect of clumping on inferring LAI from realistic trees. Despite the potential of the expressions to detect subtle changes in LAI, absolute inferred LAI values can be significantly off. However, the strong correlation between true and inferred LAI (R2>; 0.97) for the two test cases in this study, allows for calibration of the inferred LAI values.

Here, we test and optimise an approach to monitor and detect tropical deforestation in near-real time by comparing it with a seasonal-trend model fitted onto the historical time series. The method detects disturbances in near-real time by comparing newly acquired satellite data with a modelled stable period based on historic satellite time series data. The method allows the differentiation between normal and abnormal change in near-real time and is based on the Break For Additive Seasonal Trend (BFAST) concept [5, 6, 7]. Testing is done by analysis of 16-day MODIS satellite image time series (MOD13Q1) for areas in the Brazilian Amazon region for which optimised deforestation systems (e.g. PRODESDETER, IMAZON) are available.

Reference levels (RLs) are key points that determine the starting point of carbon credits payment within REDD+ framework. This work approaches the RLs using remote sensing data and analysis. Objectives of this study are: 1) mapping of forest cover change and 2) predicting carbon emissions/removals as a basis of RLs assessment. Forest cover change was estimated using national land cover map from the Ministry of Forestry in Indonesia. Combined with social economic parameters, deforestation drivers were analyzed to predict current/future deforestation. We elaborate available biomass data for predicting carbon density of each forest type and carbon biomass dynamics between 2000–2009. Hence, Riau Province was selected to conduct more detailed study. Relationship between deforestation, forest degradation, and carbon emissions/removals in this province were discussed.

The Globcover initiative comprises the development and demonstration of a service that in first instance produces a global land cover map for year 2005/2006. Globcover uses MERIS fine resolution (300 m) mode data acquired between mid 2005 and mid 2006 and, for maximum user benefit, the thematic legend is compatible with the UN land cover classification system (LCCS). This new product updates and complements the other existing comparable global products, such as the global land cover map at 1 km resolution for the year 2000 (GLC2000) produced by JRC. It is expected to improve such previous global product, in particular because of the finer spatial resolution. The Globcover project is an initiative of ESA in cooperation with an international network of partner including EEA, FAO, GOFC-GOLD, IGBP, JRC and UNEP.

An international initiative aimed at the harmonization and validation of existing and future land cover datasets is needed to support operational earth observation of land. The goal is to overcome current limitations of land cover datasets with respect to their compatibility and comparability and unknown accuracy. These limitations significantly hinder a variety of applications. Key entities in this effort are the Land Cover Implementation Team of Global Observation of Forest Cover/Global Observation of Land Dynamics, the Global Land Cover Network, and the CEOS Group on Calibration and Validation. In their recent efforts, they have explored and provided the methodological and organizational resources to foster such an international cooperation. The approaches described in this paper include an introduction of the UN Land Cover Classification System as a common land cover language and a basis for legend translation. All actors involved in land cover mapping are invited to participate in this initiative.

The Joint Research Centre of the European Commission (JRC), in partnership with 30 institutions, has produced a global land cover map for the year 2000, the GLC 2000 map. The validation of the GLC2000 product has now been completed. The accuracy assessment relied on two methods: a confidence-building method (quality control based on a comparison with ancillary data) and a quantitative accuracy assessment based on a stratified random sampling of reference data. The sample site stratification used an underlying grid of Landsat data and was based on the proportion of priority land cover classes and on the landscape complexity. A total of 1265 sample sites have been interpreted. The first results indicate an overall accuracy of 68.6%. The GLC2000 validation exercise has provided important experiences. The design-based inference conforms to the CEOS Cal-Val recommendations and has proven to be successful. Both the GLC2000 legend development and reference data interpretations used the FAO Land Cover Classification System (LCCS). Problems in the validation process were identified for areas with heterogeneous land cover. This issue appears in both in the GLC2000 (neighborhood pixel variations) and in the reference data (cartographic and thematic mixed units). Another interesting outcome of the GLC2000 validation is the accuracy reporting. Error statistics are provided from both the producer and user perspective and incorporates measures of thematic similarity between land cover classes derived from LCCS

This paper explores an addition to theory in urban geography pertaining to spatio‐temporal dynamics. Remotely sensed data on the historical extent of urban areas were used in a spatial metrics analysis of geographical form of towns and cities in the Central Valley of California (USA). Regularities in the spatio‐temporal pattern of urban growth were detected and characterized over a hundred year period. To test hypotheses about variation over geographical scale, multiple spatial extents were used in examining a set of spatial metric values including an index of contagion, the mean nearest neighbor distance, urban patch density and edge density. Through changes in these values a general temporal oscillation between phases of diffusion and coalescence in urban growth was revealed. Analysis of historical datasets revealed preliminary evidence supporting an addition to the theory of urban growth dynamics, one alluded to in some previous research, but not well developed. The empirical results and findings provide a lead for future research into the dynamics of urban growth and further development of existing urban theory.

This study evaluated how spectral resolution of high-spatial resolution optical remote sensing data influences detailed mapping of urban land cover. A comprehensive regional spectral library and low altitude data from the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) were used to characterize the spectral properties of urban land cover. The Bhattacharyya distance was applied as a measure of spectral separability to determine a most suitable subset of 14 AVIRIS bands for urban mapping. We evaluated the performance of this spectral setting versus common multispectral sensors such as Ikonos by assessing classification accuracy for 26 urban land cover classes. Significant limitations for current multispectral sensors were identified, where the location and broadband character of the spectral bands only marginally resolved the complex spectral characteristics of the urban environment, especially for built surface types. However, the AVIRIS classification accuracy did not exceed 66.6% for 22 urban cover types, primarily due to spectral similarities of specific urban materials and high within-class variability.