Continental-Scale Land Cover Mapping at 10 m Resolution Over Europe (ELC10)

Remote Sensing, Journal Year: 2021, Volume and Issue: 13(12), P. 2301 - 2301

Published: June 11, 2021

Widely used European land cover maps such as CORINE are produced at medium spatial resolutions (100 m) and rely on diverse data with complex workflows requiring significant institutional capacity. We present a high resolution (10 map (ELC10) of Europe based satellite-driven machine learning workflow that is annually updatable. A Random Forest classification model was trained 70K ground-truth points from the LUCAS (Land Use/Cover Area frame Survey) dataset. Within Google Earth Engine cloud computing environment, ELC10 can be generated approx. 700 TB Sentinel imagery within 4 days single research user account. The achieved an overall accuracy 90% across 8 classes could account for statistical unit proportions 3.9% (R2 = 0.83) actual value. These accuracies higher than other 10-m including S2GLC FROM-GLC10. found atmospheric correction Sentinel-2 speckle filtering Sentinel-1 had minimal effect enhancing (< 1%). However, combining optical radar increased by 3% compared to alone 10% alone. conversion into homogenous polygons under Copernicus module <1%, revealing Forests robust against contaminated training data. Furthermore, requires very little achieve moderate - difference between 5K 50K only (86 vs 89%). At resolution, distinguish detailed landscape features like hedgerows gardens, therefore holds potential aerial statistics city borough level monitoring property-level environmental interventions (e.g. tree planting).

Language: Английский

---

Designing a European-Wide Crop Type Mapping Approach Based on Machine Learning Algorithms Using LUCAS Field Survey and Sentinel-2 Data

Remote Sensing, Journal Year: 2022, Volume and Issue: 14(3), P. 541 - 541

Published: Jan. 23, 2022

One of the most challenging aspects obtaining detailed and accurate land-use land-cover (LULC) maps is availability representative field data for training validation. In this manuscript, we evaluate use Eurostat Land Use Coverage Area frame Survey (LUCAS) 2018 to generate a LULC map with 19 crop type classes two broad categories woodland shrubland, grassland. The were used in combination Copernicus Sentinel-2 (S2) satellite covering Europe. First, spatially temporally consistent S2 image composites (1) spectral reflectances, (2) selection indices, (3) several bio-geophysical indicators created year 2018. From large number features, important selected classification using machine-learning algorithms (support vector machine random forest). Results indicated that could be classified an overall accuracy (OA) 77.6%, independent Our analysis three methods select optimum showed by selecting spectrally different pixels data, best OA achieved, already only 11% total data. Comparing our results similar study Sentinel-1 (S1) can achieve slightly better results, although spatial coverage was reduced due gaps Further ongoing leverage synergies between optical microwave

Language: Английский

---

UAV-based reference data for the prediction of fractional cover of standing deadwood from Sentinel time series

ISPRS Open Journal of Photogrammetry and Remote Sensing, Journal Year: 2023, Volume and Issue: 8, P. 100034 - 100034

Published: March 8, 2023

Increasing tree mortality due to climate change has been observed globally. Remote sensing is a suitable means for detecting and proven effective the assessment of abrupt large-scale stand-replacing disturbances, such as those caused by windthrow, clear-cut harvesting, or wildfire. Non-stand replacing events (e.g., drought) are more difficult detect with satellite data – especially across regions forest types. A common limitation this availability spatially explicit reference data. To address issue, we propose an automated generation using uncrewed aerial vehicles (UAV) deep learning-based pattern recognition. In study, used convolutional neural networks (CNN) semantically segment crowns standing dead trees from 176 UAV-based very high-resolution (<4 cm) RGB-orthomosaics that acquired over six in Germany Finland between 2017 2021. The local-level CNN-predictions were then extrapolated landscape-level Sentinel-1 (i.e., backscatter interferometric coherence), Sentinel-2 time series, long short term memory (LSTM) predict cover fraction deadwood per Sentinel-pixel. CNN-based segmentation UAV imagery was accurate (F1-score = 0.85) consistent different study sites years. Best results LSTM-based extrapolation fractional -2 series achieved all available --2 bands, kernel normalized difference vegetation index (kNDVI), water (NDWI) (Pearson's r 0.66, total least squares regression slope 1.58). predictions showed high spatial detail transferable Our highlight effectiveness algorithms rapid large areas imagery. Potential improving presented upscaling approach found particularly ensuring temporal consistency two sources co-registration medium resolution data). increasing publicly on sharing platforms combined mapping will further increase potential multi-scale approaches.

Language: Английский

---

Tree canopy extent and height change in Europe, 2001–2021, quantified using Landsat data archive

Remote Sensing of Environment, Journal Year: 2023, Volume and Issue: 298, P. 113797 - 113797

Published: Sept. 7, 2023

European forests are among the most extensively studied ecosystems in world, yet there still debates about their recent dynamics. We modeled changes tree canopy height across Europe from 2001 to 2021 using multidecadal spectral data Landsat archive and calibration Airborne Laser Scanning (ALS) spaceborne Global Ecosystem Dynamics Investigation (GEDI) lidars. Annual was regression ensembles integrated with annual removal maps produce harmonized map time series. From these series, we derived extent a ≥ 5 m threshold. The root-mean-square error (RMSE) for both ALS-calibrated GEDI-calibrated ≤4 m. user's producer's accuracies estimated reference sample ≥94% 80% maps. Analyzing found that area increased by nearly 1% overall during past two decades, largest increase observed Eastern Europe, Southern British Isles. However, after year 2016, declined. Some regions reduced between 2021, highest reduction Fennoscandia (3.5% net decrease). continental of tall (≥ 15 height) decreased 3% 2021. decline agrees FAO statistics on timber harvesting intensification increasing severity natural disturbances. decreasing indicates forest carbon storage capacity Europe.

Language: Английский

---

The importance of blue and green landscape connectivity for biodiversity in urban ponds

Basic and Applied Ecology, Journal Year: 2021, Volume and Issue: 57, P. 129 - 145

Published: Oct. 29, 2021

The negative impact of urbanization on biodiversity can be buffered by blue (e.g., rivers, ponds) and green parks, forests) spaces. However, to prevent loss reduce the risk local extinctions, spaces need connected corridors, so that organisms may disperse between sites. Landscape connectivity affects community composition metacommunity dynamics facilitating dispersal. goal this study was test relative roles pond environmental properties, spatial structure, functional landscape differentiation invertebrate metacommunities in urban ponds city Stockholm, Sweden. We characterized as (distance water bodies), (land use), combined blue-green connectivity. estimated using electrical circuit theory identify dispersal corridors. Interestingly, while is often used single-taxon studies, method has rarely been applied multiple taxa forming a metacommunity, we have done study. Indeed, our contributes toward an increased focus role at level. determined most important factor explaining differentiation, with environment contributing comparatively little, structure least. Combined had major influence structuring communities, 7.8% variance across ponds. Furthermore, found associated increase number species. In summary, results suggest preserve ponds, it enhance connectivity, open could augment corridors maintaining metacommunities. To generalize these findings, future studies should compare how cities.

Language: Английский

---

Spatially Stratified and Multi-Stage Approach for National Land Cover Mapping Based on Sentinel-2 Data and Expert Knowledge

Remote Sensing, Journal Year: 2022, Volume and Issue: 14(8), P. 1865 - 1865

Published: April 13, 2022

Portugal is building a land cover monitoring system to deliver products annually for its mainland territory. This paper presents the methodology developed produce prototype relative 2018 as first map of future annual series (COSsim). A total thirteen classes are represented, including most important tree species in Portugal. The mapping approach includes two levels spatial stratification based on landscape dynamics. Strata analysed independently at higher level, while nested sublevels can share data and procedures. Multiple stages analysis implemented which subsequent improve outputs precedent stages. goal adjust local tackle specific problems or divide complex tasks several parts. Supervised classification Sentinel-2 time post-classification with expert knowledge were performed throughout four overall accuracy estimated 81.3% (±2.1) 95% confidence level. Higher thematic was achieved southern Portugal, significantly improved quality map.

Language: Английский

---

ELULC-10, a 10 m European Land Use and Land Cover Map Using Sentinel and Landsat Data in Google Earth Engine

Remote Sensing, Journal Year: 2022, Volume and Issue: 14(13), P. 3041 - 3041

Published: June 24, 2022

Land Use/Land Cover (LULC) maps can be effectively produced by cost-effective and frequent satellite observations. Powerful cloud computing platforms are emerging as a growing trend in the high utilization of freely accessible remotely sensed data for LULC mapping over large-scale regions using big geodata. This study proposes workflow to generate 10 m map Europe with nine classes, ELULC-10, European Sentinel-1/-2 Landsat-8 images, well LUCAS reference samples. More than 200 K 300 situ surveys respectively, were employed inputs Google Earth Engine (GEE) platform perform classification an object-based segmentation algorithm Artificial Neural Network (ANN). A novel ANN-based preparation was also presented remove noisy samples from dataset. Additionally, improved several rule-based post-processing steps. The overall accuracy kappa coefficient 2021 ELULC-10 95.38% 0.94, respectively. detailed report accuracies provided, demonstrating accurate different such Woodland Cropland. Furthermore, post processing class identifications when compared current studies. could supply seasonal, yearly, change considering proposed integration complex machine learning algorithms large survey data.

Language: Английский

---

Landscape context and farm characteristics are key to farmers' adoption of agri-environmental schemes

Land Use Policy, Journal Year: 2022, Volume and Issue: 121, P. 106320 - 106320

Published: Aug. 20, 2022

Agri-environmental schemes (AES) belong to the main instruments of European Union's Common Agricultural Policy (CAP) foster sustainable farming practices that contribute conservation biodiversity, ecosystem services, climate change mitigation and adaptation. Farmers' attitudes towards these voluntary measures socio-economic factors influencing their decisions have been widely studied through interviews or surveys. However, it remains unclear whether spatial patterns AES adoption can be predicted based on farm structural environmental variables. In this study, we combine biophysical maps with information structure landscape context model influence variables implementation at both field level. We fit a set regression models using characteristics (e.g. size specialization, size) as well elevation, soil fertility, presence protected areas) predictors Mulde River Basin in Germany case study. Our analysis reveals distribution explained by factors: tend implemented larger farms specialized permanent grassland cultivation are typically located areas lower fertility. At level, preferably allocated fields close water bodies small woody features. The effect different farm-related varies across AES-schemes indicating complex farmers take into consideration when allocating scheme field. As our study shows quantifiable tendency place unproductive and/or areas, supports previous evidence criticizing global allocate protection regions low agricultural value, which results goals not being met. presented here support development future AES, e.g. developing tailored currently unlikely adopt thus improving effectiveness environmentally friendly practices.

Language: Английский

---

A spatiotemporal ensemble machine learning framework for generating land use/land cover time-series maps for Europe (2000–2019) based on LUCAS, CORINE and GLAD Landsat

PeerJ, Journal Year: 2022, Volume and Issue: 10, P. e13573 - e13573

Published: July 21, 2022

A spatiotemporal machine learning framework for automated prediction and analysis of long-term Land Use/Land Cover dynamics is presented. The includes: (1) harmonization preprocessing spatial input datasets (GLAD Landsat, NPP/VIIRS) including five million harmonized LUCAS CORINE Cover-derived training samples, (2) model building based on k-fold cross-validation hyper-parameter optimization, (3) the most probable class, class probabilities variance predicted per pixel, (4) LULC change time-series produced maps. ensemble consists a random forest, gradient boosted tree classifier, an artificial neural network, with logistic regressor as meta-learner. results show that important variables mapping in Europe are: seasonal aggregates Landsat green near-infrared bands, multiple Landsat-derived spectral indices, surface water probability, elevation. Spatial indicates consistent performance across years overall accuracy (a weighted F1-score) 0.49, 0.63, 0.83 when predicting 43 (level-3), 14 (level-2), classes (level-1). Additional experiments models generalize better to unknown years, outperforming single-year known-year classification by 2.7% unknown-year 3.5%. Results assessment using 48,365 independent test samples shows 87% match validation points. (time-series NDVI images) suggest forest loss large parts Sweden, Alps, Scotland. Positive negative trends general land degradation restoration classes, “urbanization” showing trend. An advantage ML fitted can be used predict were not included its dataset, allowing generalization past future periods, e.g. prior 2000 beyond 2020. generated data stack (ODSE-LULC), points, publicly available via ODSE Viewer. Functions prepare run modeling are eumap library Python.

Language: Английский

---

Thematic Comparison between ESA WorldCover 2020 Land Cover Product and a National Land Use Land Cover Map

Land, Journal Year: 2023, Volume and Issue: 12(2), P. 490 - 490

Published: Feb. 16, 2023

This work presents a comparison between global and national land cover map, namely the ESA WorldCover 2020 (WC20) Portuguese use/land map (Carta de Uso e Ocupação do Solo 2018) (COS18). Such is relevant given current amount of publicly available LULC products (either or global) where such comparative studies enable better understanding regarding different sets information their production, focus characteristics, especially when comparing authoritative maps built by mapping agencies focused products. Moreover, this also aimed at complementing validation report released with WC20 product, which on continental level accuracy assessments, no additional for specific countries. The were compared following framework composed four steps: (1) class nomenclature harmonization, (2) computing cross-tabulation matrices (3) determining area occupied each harmonized in data source, (4) visual to illustrate differences focusing landscape details. Some due minimum unit ofCOS18 WC20, nomenclatures focuses either use cover. Overall, results show that while detail able distinguish small occurrences artificial surfaces grasslands within an urban environment, often not sparse/individual trees from neighboring cover, common occurrence landscape. While selecting users should be aware can have range causes, as scale, temporal reference, errors.

Language: Английский

---