Data For Land Cover Classification Research Articles

Exploring the Potential of the Hyperspectral Remote Sensing Data China Orbita Zhuhai-1 in Land Cover Classification

Responding to the shortcomings of China's civil remote sensing data in land cover classification, such as the difficulty of data acquisition and the low utilization rate, we used Landsat-8, China Orbita Zhuhai-1 hyperspectral remote sensing (OHS) data, and Landsat-8 + OHS data combined with band (red, green, and blue) and vegetation index features to classify land cover using maximum likelihood (ML), Mahalanobis distance (MD), and support vector machine (SVM). The results show that Landsat-8 + OHS data have the highest classification accuracy in SVM, with an overall accuracy of 83.52% and a kappa coefficient of 0.71, and this result is higher than that of Landsat-8 images and OHS images separately. In addition, the classification accuracy of OHS images was higher than that of Landsat-8 images. The results of the study provide a reference for the use of civil satellite remote sensing data in China.

Machine Learning for Global Bioclimatic Classification: Enhancing Land Cover Prediction through Random Forests

Traditional bioclimatic classification schemes have several inherent shortcomings; they do not represent anthropogenic impact, they contain a bias for global north representation, and they lack flexibility regarding novel climates that may arise due to climate change. Here we present an alternative approach, using a machine learning approach. We combine European Space Agency Land Cover Classification data with traditional bioclimate classification climate variables, and additional variables; latitude, elevation, and topography. We utilise a random forest algorithm to create a classification system that overcomes the limitations and biases of the traditional schemes. The algorithm produced is able to predict land cover classification globally at 0.5-degree resolution with 93% accuracy. The resulting classifications account for human impact, particularly via agriculture, are informed by the topography of a region, and avoids the biases that traditional bioclimatic schemes contain. The algorithm can provide insights into the drivers of land cover change, the spatial distribution of land cover change, the potential impacts on ecosystem services and human well-being. Furthermore, the random forest model serves as a novel approach to the prediction of future land cover, and can be used to identify regions at risk of a land cover transition. Our data-based machine learning approach produces larger land-cover changes due to climate change than a traditional bioclimatic scheme, especially in sensitive regions such as Amazonia. Overall, our new approach projects approximately 17.4 million square kilometre of land-cover change per degree celsius of global warming.

COMPARISON OF THE USE OF SENTINEL-1 SAR AND ALOS-2 PALSAR-2 IN MANGROVE ABOVEGROUND BIOMASS ESTIMATION IN SAN JUAN, BATANGAS, PHILIPPINES

Abstract. This study compares the potential of Sentinel-1 and ALOS-2 PALSAR-2 in estimating mangrove aboveground biomass (AGB) in San Juan, Batangas, Philippines. Mangrove forests are essential coastal ecosystems that are facing growing threats. One way of conserving them is by creating policies that can protect them. To do this effectively, information like AGB can be used as a guide. However, conventional AGB estimations are labor-intensive and ecologically disruptive. Conversely, remote sensing technologies, such as synthetic aperture radar (SAR), offer a more efficient alternative. Sentinel-1, operating in C-band, and ALOS-2 PALSAR-2, operating in L-band, are two prominent SAR platforms with global coverage, offering data for land cover classification, forest monitoring, and forest biomass estimation. This research used the backscatter values of Sentinel-1 and ALOS-2 PALSAR-2 as predictor variables in estimating mangrove AGB by correlating them to the observed AGB from a mangrove survey. The models developed using these platforms yielded limited accuracy, with low coefficient of determination (R2) (Sentinel-1 = 0.13; ALOS-2 PALSAR-2 = 0.12) and RMSE (Sentinel-1 = 8.72 Mg ha-1; ALOS-2 PALSAR-2 = 8.78 Mg ha-1). Potential sources of errors were identified, including small sample size and data noise. On the results, Sentinel-1 demonstrates a slightly better performance in terms of the R2 and RMSE in the modeling while ALOS-2 PALSAR-2 performed better in the validation, however, both still yielded suboptimal AGB estimates compared to other studies. Refining the models by incorporating additional parameters, exploring machine learning, and considering other data sources are recommended to enhance AGB estimation.

Opportunities for Monitoring Soil and Land Development to Support United Nations (UN) Sustainable Development Goals (SDGs): A Case Study of the United States of America (USA)

Land, including soil resources, makes important contributions to the United Nations (UN) Sustainable Development Goals (SDGs). However, there are challenges in identifying land/soil measurable information (e.g., indicators, metrics, etc.) to monitor the progress toward achieving these goals. This study examines the role of land/soil in selected SDGs (SDG 2: Zero Hunger; SDG 12: Responsible Consumption and Production; SDG 13: Climate Action; SDG 15: Life on Land) and provides practical examples on how to use geospatial analysis to track relevant qualitative and quantitative land/soil data using the contiguous United States of America (USA) as a case study. The innovative aspect of this study leverages geospatial technologies to track the intersection of land use/land cover (LULC) change and soil resources to quantify development trends within the overall land cover matrix to evaluate if these trends are sustainable. Classified land cover data derived from satellite-based remote sensing were used to identify the extent of developed areas in 2016 and the change in development areas since 2011. Most land development through time in the USA has caused losses (area loss of nearly 355,600 km2, with projected midpoint losses of about 5.7 × 1012 kg total soil carbon (TSC) and about $969B (where B = billion = 109, USD) in social costs of carbon dioxide emissions, SC-CO2). All ten soil orders present in the contiguous USA experienced losses from developments, which represents a loss for both biodiversity and soil diversity (pedodiversity). The contiguous USA experienced an increase in land/soil consumption between 2001 and 2016 at the expense of deciduous forest (−3.1%), evergreen forest (−3.0%), emergent herbaceous wetlands (−0.6%), and hay/pasture (−7.9%). These “new” land developments (24,292.2 km2) caused a complete projected midpoint loss of 4.0 × 1011 kg TSC, equivalent to $76.1B SC-CO2. States with the largest developed areas and the highest TSC losses with associated SC-CO2 were Texas and Florida. The proposed methodology used in this study can be applied worldwide, at various spatial scales, to help monitor SDGs over time. With improved tools to monitor SDGs, progress on these SDGs may require linking the SDGs to existing or future international and national legal frameworks.

Using PRISMA Hyperspectral Data for Land Cover Classification with Artificial Intelligence Support

Hyperspectral satellite missions, such as PRISMA of the Italian Space Agency (ASI), have opened up new research opportunities. Using PRISMA data in land cover classification has yet to be fully explored, and it is the main focus of this paper. Historically, the main purposes of remote sensing have been to identify land cover types, to detect changes, and to determine the vegetation status of forest canopies or agricultural crops. The ability to achieve these goals can be improved by increasing spectral resolution. At the same time, improved AI algorithms open up new classification possibilities. This paper compares three supervised classification techniques for agricultural crop recognition using PRISMA data: random forest (RF), artificial neural network (ANN), and convolutional neural network (CNN). The study was carried out over an area of 900 km2 in the province of Caserta, Italy. The PRISMA HDF5 file, pre-processed by the ASI at the reflectance level (L2d), was converted to GeoTiff using a custom Python script to facilitate its management in Qgis. The Qgis plugin AVHYAS was used for classification tests. The results show that CNN gives better results in terms of overall accuracy (0.973), K coefficient (0.968), and F1 score (0.842).

Trade-Offs and Synergies among 17 Ecosystem Services in Africa: A Long-Term Multi-National Analysis

The proper management of multiple ecosystem services (ESs) in a balanced manner is an important and challenging responsibility. However, due to infrastructural constraints, we need to understand more about the spatial interactions among ESs in most African countries. Therefore, we took 48 African countries, 5 African geopolitical regions, and the African continent as case studies to diagnose the spatial trade-offs and synergies among 17 ESs and 8 types of land use and land cover (LULC) in 2000 and 2019. The implications of our findings at the national, regional, continental, and global levels were explored. To achieve this, we mapped the spatial distributions of the 17 ESs at the continental level using classified land cover data from MODIS remotely sensed data, with a spectral band between 0.405 and 14.385 µm and a spatial resolution of 500 m. Then, we used Spearman’s rank correlation coefficient to determine the spatial interactions among the 17 ESs. The results show that regulation services showed synergies at the continental level in gas regulation (0.66), climate regulation (0.71), disturbance regulation (0.14), water regulation (0.53), water supply (0.71), and waste treatment (0.06). Moreover, we found moderate levels of interactions among most ESs in the 48 countries, with most regulating services and supporting services exhibiting trade-offs with other categories of ESs, among other findings. The results will inform scientific communities and authorities at all levels on how to deliver human well-being and quality of life, and usher in a sustainable change where we expect better ecosystem management and ecological conservation.

A polarimetric projection-based scattering characteristics extraction tool and its application to PolSAR image classification

It is necessary to adequately extract scattering characteristics from polarimetric synthetic aperture radar (PolSAR) data for land-cover classification. Current interpretation technologies, such as target decompositions, polarimetric signatures, and polarimetric coherence patterns, have been widely used to explore polarization information. However, because limited scattering models are used to characterize all ground targets, these methods are occasionally limited in the interpretation of PolSAR data. This may reduce the accuracy of the land-cover classification. In this study, we attempted to extract the scattering characteristics through another novel method. We understand the PolSAR image interpretation as the projection of the scattering model on the measured matrix. Current methods project only a limited series of scattering models onto a measured matrix. To complement current methods of PolSAR image interpretation, we propose a polarimetric projection-based scattering characteristics extraction (PPSCE) method. It projects scattering models consisting of four changing independent polarization parameters onto the measured matrix. These scattering models, associated with the geometric and physical properties of the ground targets, correspond to the entire polarization space. The PPSCE method, meanwhile, extracts polarization information sufficiently. L-band UAVSAR, C-band Radarsat-2, and P-band AIRSAR images of classical experimental areas are used to validate the PPSCE method. The results show that the proposed PPSCE method can present polarization information that is difficult to investigate by the existing methods. In addition, the scattering features derived using the PPSCE method are superior for land-cover classification. Compared to the classification results based on the coherency matrix and Wishart classifier, the proposed method improves the overall classification accuracy by 23.56% for UAVSAR data. Compared to target decompositions and polarization signatures, the overall classification accuracy of the proposed method is increased by 29.32% (RF) and 22.63% (SVM) for UAVSAR data, 7.96% (RF) and 4.36% (RF) for Radarsat-2 and AIRSAR datasets, respectively. Furthermore, the proposed method has significant potential for soil moisture inversion. https://rgdoi.net/10.13140/RG.2.2.14091.26409.

A Spatiotemporal Characterization of Water Resource Conditions and Demands as Influenced by the Hydrogeologic Framework of the Willcox Groundwater Basin, Southeastern Arizona, USA

In the Willcox Groundwater Basin (WGB), increasing rates of agricultural groundwater withdrawal have led to significant regional groundwater level decline, threatening the basin’s long-term water resource security. Updated characterization of the basin’s water resource conditions and agricultural water demand is critically important for informing groundwater resource management efforts. We developed the hydrogeologic framework of the WGB and linked groundwater level data with land cover classification data to provide a spatiotemporal assessment of water resource conditions and agricultural development in the WGB. A correlation analysis evaluated the degree of association between the basin’s mean annual depth-to-groundwater and agricultural land cover extent. Results of this study indicate that between 2008 and 2021, agricultural land cover in the WGB increased by 29%. The average rate of groundwater level change in the basin’s measured wells was calculated at −13.8 m between 2006 and 2021. We found a strong correlation between the basin’s mean annual measured depth-to-groundwater and the annual agricultural land cover extent, further reinforcing the understanding of agricultural water use in the basin as a principal driver of groundwater level decline. The methodological framework employed proved a simple and effective way to assess groundwater resources as influenced by geology and land use. The outcomes of this study provide critical information toward improved water resources management by providing an integrated understanding of local hydrogeology, groundwater level variability, and changes in agricultural land cover in arid inland basins such as those found in Arizona, USA.

Effect of the Synergetic Use of Sentinel-1, Sentinel-2, LiDAR and Derived Data in Land Cover Classification of a Semiarid Mediterranean Area Using Machine Learning Algorithms

Land cover classification in semiarid areas is a difficult task that has been tackled using different strategies, such as the use of normalized indices, texture metrics, and the combination of images from different dates or different sensors. In this paper we present the results of an experiment using three sensors (Sentinel-1 SAR, Sentinel-2 MSI and LiDAR), four dates and different normalized indices and texture metrics to classify a semiarid area. Three machine learning algorithms were used: Random Forest, Support Vector Machines and Multilayer Perceptron; Maximum Likelihood was used as a baseline classifier. The synergetic use of all these sources resulted in a significant increase in accuracy, Random Forest being the model reaching the highest accuracy. However, the large amount of features (126) advises the use of feature selection to reduce this figure. After using Variance Inflation Factor and Random Forest feature importance, the amount of features was reduced to 62. The final overall accuracy obtained was 0.91 ± 0.005 (α = 0.05) and kappa index 0.898 ± 0.006 (α = 0.05). Most of the observed confusions are easily explicable and do not represent a significant difference in agronomic terms.

NATIONAL LAND COVER MAPPING USING VARIOUS REMOTE SENSING DATASETS IN GEE

National land-cover maps are essential for the development of the countries as land-use patterns have shifted dramatically throughout the world in the previous decades. With the latest development in the remote sensing community, the power and ease of web-mapping and web-based map and GIS services have increased. This work investigates several datasets for land cover categorization on a national scale across North Macedonia (25,713 km2) using Sentinel images within Google Earth Engine (GEE), a cloud computing platform designed to store and process huge data sets for analysis and ultimate decision making. Both single and synergetic use of Sentinel-1 and Sentinel-2 satellite images have been investigated. The basic land-cover components are generated, upon which the more detailed final land-cover/land-use data were derived and defined. Comparing the results of the datasets indicate the influence of the radar data over the optical data in land cover classification. Also, the influence of the investigated data over every class is calculated. The results showed that the various datasets lead to different overall accuracy. Also, the different datasets performed differently over single classes, even though their overall accuracy was the same. As a result, a high accuracy national-level land cover classification has been created. The results have been compared to the latest Corine data. The results can be critical to making informed policy, development, planning, and resource management decisions. This provided the standardized references from which landscape changes could be determined and quantified.

Fusion of hyperspectral and LiDAR data using sparse stacked autoencoder for land cover classification with 3D-2D convolutional neural network

Fusion of complementary information from multisensor data is of great importance for identifying the land covers. However, integration of multisource information is a challenging task. A framework is developed to integrate hyperspectral and LiDAR data for land cover classification. In the proposed method, sparse stacked autoencoders are used to represent the spectral and spatial information in a compact form. The spatial information is extracted both from hyperspectral and LiDAR data using morphological operators. The encoded spectral and spatial features are combined with elevation information to form a joint feature vector. The joint features are fed to a convolutional neural network (CNN) classifier to classify the land covers. The CNN classifier is a hybrid three- and two-dimensional (3D)–(2D) model having three 3D convolutional layers and one 2D convolutional layer. The experiments are carried out on two datasets Houston and Samford to evaluate the performance of the proposed method. The results have demonstrated the effectiveness of the method with global κ = 0.9285 and global naive accuracy (OA) of 93.44% for Houston data. For Samford data, it achieves κ = 0.9811 and OA = 98.93 % .

New techniques for old fires: Using deep learning to augment fire maps from the early satellite era

Disturbances such as fire play a critical role in forest ecosystems. However, anthropogenic fires can profoundly impact forests to the point of destabilizing ecosystems. In addition, fires have legacy effects on environments which may be observed in forests for decades after the fire is extinguished. Thus, understanding the extent of historic fires in a landscape is vital to understanding current forest structure and ecological processes (e.g., carbon sequestration capacity and provision of habitat) and, therefore, essential for informing land-management decisions. However, little work has been done to map forest fires pre 1980s due to the challenges of interpreting imagery from the 1970s-era Landsat Multispectral Scanner (MSS) platform. MSS imagery is distinguished from recent satellite missions through lower temporal, spatial, and spectral resolutions. Recent advances in image processing have brought the goal of high-quality MSS classifications within reach. In this study, we use deep learning, specifically UNet (a fully convolutional neural network (CNN)), to detect historic forest fires in MSS imagery for the forest-dominated regions of Quebec, Canada. While other studies have applied deep learning to present-day satellite data for land cover classification, hardly any work has specifically applied deep learning to MSS data for fire detection. We trained our UNet model on 206 MSS images that were labelled by applying thresholds to the Burned Area Index inside polygons drawn by the authors around burned areas. We then used the trained model to label burns in 5104 MSS images that were compiled to generate annual burned area maps. Our results identified (with a 95% confidence interval) 3503.95 ± 484.90 km2 of burns not previously reported in any database; this represents a 35.30 ± 3.94% increase in the total known burned area across the forest-dominated regions of Quebec between 1973 and 1982.

Delimitation of Urban Hot Spots and Rural Cold Air Formation Areas for Nocturnal Ventilation Studies Using Urban Climate Simulations

Due to global warming, the conservation or enhancement of urban ventilation during synoptically calm and hot weather conditions is receiving increasing attention in climate resilient urban and regional planning. The transport of cool air from rural surroundings into the city by local winds during nighttime is important for the alleviation of the urban heat island intensity and heat load in particular. A simple statistical method, which objectively identifies urban thermal hot spots and areas of rural cold air formation from thermodynamic urban climate model simulations is described and applied to Aschaffenburg, a medium-sized town located in hilly terrain in south-central Germany. The delimitated hot spots and nocturnal cold air formation areas are influenced by local land cover, and also by the surrounding landscape heterogeneity, surface energy exchange and atmospheric mixing processes. The results illustrate limitations of hot spot or cool spot estimation methods based purely on the analysis of classified land cover data. Nocturnal backward airflow trajectories from thermal hot spots in the city and forward trajectories from rural areas with substantial cold air formation are calculated to determine which cold air formation areas are contributing to ventilation and advective cooling of thermal hot spots. It is found that nocturnal ventilation mechanisms are not bound to municipal boundaries, which highlights the need for regional cooperation in urban climate adaptation. The described method provides guidance to urban and regional planners in order to protect important cold air formation areas, e.g., from urban sprawl, and it can be applied to study impacts of planning scenarios. Options for improvement or extension of the method are discussed.

A dual-stream high resolution network: Deep fusion of GF-2 and GF-3 data for land cover classification

Land cover classification (LCC) is an important application in remote sensing data interpretation and invariably faces big intra-class variance and sample imbalance in remote sensing images. The optical image is obtained by satellites capturing the spectral information of the Earth’s surface, and the synthetic aperture radar (SAR) image is produced by the satellite actively transmitting and receiving the electromagnetic wave signals reflected from land covers. Because of the limitations of the optical image, a single modality (optical image) might be disturbed by external conditions, especially complex weather. Using heterogeneous SAR and optical images for LCC can reduce the negative impact caused by single-modal data damage, and multi-modal data can also be used as supplementary information to enhance classification accuracy. However, general LCC methods mainly focus on remote sensing data of a single modality without fully considering the multi-modalities of land covers. Therefore, we propose a dual-stream deep high-resolution network (DDHRNet) to deeply integrate SAR and optical data at the feature level in every branch. The network can effectively exploit the complementary information in heterogeneous images. It improves classification performance and achieves significant improvements in the classification of clouded images. A multi-modal squeeze-and-excitation (MSE) module is also utilized to fuse the features. Compared with the ordinary methods, MSE modules can lead to an improvement of about 1% to 5% in overall accuracy (OA), Kappa coefficients, and mean intersection over union (mIoU). Besides, in order to evaluate our method, we describe in detail the preprocessing process of Gaofen-2 (GF2) and Gaofen-3 (GF3) data before they are used in the LCC task. The experiments show that the proposed method performs well compared with other excellent segmentation methods and obtains the best performance on heterogeneous images from GF2 and GF3. The code and datasets are available at: https://github.com/XD-MG/DDHRNet.

Vermont Global Warming Solutions Act: The Costs of Inaction from Land Conversions

The Vermont (VT) Global Warming Solutions Act (GWSA, 2020) sets greenhouse gas (GHG) emissions reduction targets at 26% below 2005 by 2025, 40% below 1990 by 2030 and 80% below 1990 by 2050 for energy-related emissions only. Vermont’s omission of GHG emissions from land conversions could result in significant costs of inaction (COI), which could hinder the state’s mitigation and adaptation plans and result in climate crisis-related risks (e.g., credit downgrade). Science-based spatio-temporal data of GHG emissions from soils because of land conversions can be integrated into the conceptual framework of “action” versus “inaction” to prevent GHG emissions. The application of soil information data and remote sensing analysis can identify the GHG emissions from land conversions, which can be expressed as “realized” social costs of “inaction”. This study demonstrates the rapid assessment of the value of regulating ecosystems services (ES) from soil organic carbon (SOC), soil inorganic carbon (SIC), and total soil carbon (TSC) stocks, based on the concept of the avoided social cost of carbon dioxide (CO2) emissions for VT by soil order and county using remote sensing and information from the State Soil Geographic (STATSGO) and Soil Survey Geographic Database (SSURGO) databases. Classified land cover data for 2001 and 2016 were downloaded from the Multi-Resolution Land Characteristics Consortium (MRLC) website. These results provide accurate and quantitative spatio-temporal information about likely GHG emissions, which can be linked to VT’s climate action plan. A failure to considerably reduce emissions from land conversions would increase climate change costs and potential legal consequences for VT and beyond its borders.

FUZZY COMPREHENSIVE EVALUATION OF LAND COVER CLASSIFICATION DATA

Abstract. The demarcation of the evaluation grade of surveying and mapping products is too rigid, and the evaluation results are often distorted due to the abnormality of an evaluation index. This paper takes the land cover classification data quality evaluation as an example and analyses the quality model, which includes quality factors and weight coefficients. In this paper, a multilevel fuzzy comprehensive evaluation method is introduced, a reasonable and efficient membership function is designed, and the evaluation results are evaluated by using the relative index of effectiveness and the probability of a single value distribution. The results show the fuzzy comprehensive evaluation method based on the premise of suitability weighting can make the evaluation results retain rich information, and the evaluation results are more reasonable and the improved fuzzy comprehensive evaluation method can be applied to the quality evaluation of other natural resource surveys and monitoring products.

MULTI-FREQUENCY POLINSAR DATA ARE ADVANTAGEOUS FOR LAND COVER CLASSIFICATION – A VISUAL AND QUANTITATIVE ANALYSIS

Abstract. This paper investigates the enhanced potential of using multi-frequency PolInSAR data for land cover classification. In order to enable a descriptive analysis that goes beyond the mere comparison of classification accuracies, a two-step classification process is applied. First, polarimetric and interferometric features are extracted and projected into a 3-dimensional feature space by using the supervised dimension reduction algorithm Uniform Manifold Approximation and Projection (UMAP). Subsequently, based on the expressive 3-dimensional representation a simple yet sufficient k-nearest neighbors (KNN) classifier is applied to assign a land cover class to each pixel. In this way, besides the simplified classification, the visualization of the underlying data structure is possible and contributes to a better explanation and analysis of classification results. The data analyzed in this way are airborne L- and S-band PolInSAR data acquired by the F-SAR system. The visual analysis of reduced feature spaces as well as the quantitative analysis of classification results reveal the benefits of combining both frequencies with regard to class separability.

EFFICIENT TRAINING DATA GENERATION BY CLUSTERING-BASED CLASSIFICATION

Abstract. Insufficient amount or complete absence of reference data for the training of classifiers is a general topic. Especially the state-of-the-art deep learning approaches have to deal with the availability or adaption of this reference data to produce the reliable results they are designed for. This paper will pursue different approaches according to the absence of training data for land cover classification from aerial images. First, we will analyze the performance of traditional classification in the absence of reference data using clustering techniques and salient features for the assignment of semantic labels. Second, we will transfer the results as training data to a DeepLabv3+ CNN with pre-trained weights to demonstrate the usability of the generated training data. Third, we expand the clustering approaches and combine them with a Random Forest classifier. Finally, if user interaction and manual annotation of training data are still necessary, we also introduce our labeling GUI that enables a simple, fast, and comfortable training data generation with only a few clicks. To evaluate our procedure, we used two datasets, including the Vaihingen benchmark, for which ground truth is available. Without any interactive steps except setting a few algorithm paremeters, we achieved an overall accuracy of 75% using the Deeplab method with image data only.

WEAKLY SUPERVISED LEARNING FOR TREELINE ECOTONE CLASSIFICATION BASED ON AERIAL ORTHOIMAGES AND AN ANCILLARY DSM

Abstract. Convolutional neural networks (CNNs) effectively classify standard datasets in remote sensing (RS). Yet, real-world data are more difficult to classify using CNNs because these networks require relatively large amounts of training data. To reduce training data requirements, two approaches can be followed – either pretraining models on larger datasets or augmenting the available training data. However, these commonly used strategies do not fully resolve the lack of training data for land cover classification in RS. Our goal is to classify trees and shrubs from aerial orthoimages in the treeline ecotone of the Krkonoše Mountains, Czechia. Instead of training a model on a smaller, human-labelled dataset, we semiautomatically created training data using an ancillary normalised Digital Surface Model (nDSM) and image spectral information. This approach can complement existing techniques, trading accuracy for a larger labelled dataset while assuming that the classifier can handle the training data noise. Weakly supervised learning on a CNN led to 68.99% mean Intersection over Union (IoU) and 81.65% mean F1-score for U-Net and 72.94% IoU and 84.35% mean F1-score for our modified U-Net on a test set comprising over 1000 manually labelled points. Notwithstanding the bias resulting from the noise in training data (especially in the least occurring tree class), our data show that standard semantic segmentation networks can be used for weakly supervised learning for local-scale land cover mapping.

Sensitivity of biomass burning emissions estimates to land surface information

Abstract. Emissions from biomass burning (BB) are a key source of atmospheric tracer gases that affect the atmospheric carbon cycle. We developed four sets of global BB emissions estimates (named GlcGlob, GlcGeoc, McdGlob, and McdGeoc) using a bottom-up approach and by combining the remote sensing products related to fire distribution with two aboveground biomass (AGB) and two land cover classification (LCC) distributions. The sensitivity of the estimates of BB emissions to the AGB and LCC data was evaluated using the carbon monoxide (CO) emissions associated with each BB estimate. Using the AGB and/or LCC data led to substantially different spatial estimates of CO emissions, with a large (factor of approximately 3) spread of estimates for the mean annual CO emissions: 526±53, 219±35, 624±57, and 293±44 Tg CO yr−1 for GlcGlob, GlcGeoc, McdGlob, and McdGeoc, respectively, and 415±47 Tg CO yr−1 for their ensemble average (EsmAve). We simulated atmospheric CO variability at an approximately 2.5∘ grid using an atmospheric tracer transport model and the BB emissions estimates and compared it with ground-based and satellite observations. At ground-based observation sites during fire seasons, the impact of intermittent fire events was poorly defined in our simulations due to the coarse resolution, which obscured temporal and spatial variability in the simulated atmospheric CO concentration. However, when compared at the regional and global scales, the distribution of atmospheric CO concentrations in the simulations shows substantial differences among the estimates of BB emissions. These results indicate that the estimates of BB emissions are highly sensitive to the AGB and LCC data.