Individual Tree-Crown Detection and Species Identification in Heterogeneous Forests Using Aerial RGB Imagery and Deep Learning

Remote Sensing, Journal Year: 2023, Volume and Issue: 15(5), P. 1463 - 1463

Published: March 6, 2023

Automatic identification and mapping of tree species is an essential task in forestry conservation. However, applications that can geolocate individual trees identify their heterogeneous forests on a large scale are lacking. Here, we assessed the potential Convolutional Neural Network algorithm, Faster R-CNN, which efficient end-to-end object detection approach, combined with open-source aerial RGB imagery for geolocation upper canopy layer temperate forests. We studied four species, i.e., Norway spruce (Picea abies (L.) H. Karst.), silver fir (Abies alba Mill.), Scots pine (Pinus sylvestris L.), European beech (Fagus sylvatica growing To fully explore approach identification, trained single-species multi-species models. For models, average accuracy (F1 score) was 0.76. Picea detected highest accuracy, F1 0.86, followed by A. = 0.84), F. 0.75), Pinus 0.59). Detection increased models 0.92), while it remained same or decreased slightly other species. Model performance more influenced site conditions, such as forest stand structure, less illumination. Moreover, misidentification number included increased. In conclusion, presented method accurately map location may serve basis future inventories targeted management actions to support resilient

Language: Английский

---

National tree species mapping using Sentinel-1/2 time series and German National Forest Inventory data

Remote Sensing of Environment, Journal Year: 2024, Volume and Issue: 304, P. 114069 - 114069

Published: Feb. 24, 2024

Spatially explicit and detailed information on tree species composition is critical for forest management, nature conservation the assessment of ecosystem services. In many countries, attributes are monitored regularly through sample-based inventories. combination with satellite imagery, data from such inventories have a great potential developing large-area maps. Here, high temporal resolution Sentinel-1 Sentinel-2 has been useful extracting vegetation phenology, that may also be valuable improving mapping. The objective this study was to map main in Germany using combined time series, identify address challenges related use National Forest Inventory (NFI) remote sensing applications. We generated cloud free series 5-day intervals imagery combine those monthly backscatter composites. Further, we incorporate topography, meteorology, climate account environmental gradients. To NFI training machine learning models, following challenges: 1) link pixels variable radius plots, which precise area unknown, 2) efficiently utilize mixed-species plots model validation. past, accuracies pixel-level maps were often estimated solely homogeneous pure-species stands. study, assess how well generalize mixed plot conditions. Our results show mapping large, environmentally diverse landscapes. Classification accuracy pure stands ranged between 72% 97% (F1-score) five dominant species, while less frequent remained challenging. When including assessment, decreased by 4–14 percentage points most groups. highlights importance mixed-forest when validating Based these results, discuss potentials remaining at national level. findings allow further improve national-level medium provide guidance similar approaches other countries where ground-based inventory available.

Language: Английский

---

Status, advancements and prospects of deep learning methods applied in forest studies

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2024, Volume and Issue: 131, P. 103938 - 103938

Published: June 4, 2024

Language: Английский

---

A Sentinel-2 machine learning dataset for tree species classification in Germany

Earth system science data, Journal Year: 2025, Volume and Issue: 17(2), P. 351 - 367

Published: Feb. 3, 2025

Abstract. We present a machine learning dataset for tree species classification in Sentinel-2 satellite image time series of bottom-of-atmosphere reflectance. It is geared towards training classifiers but less suitable validating the resulting maps. The based on German National Forest Inventory 2012 as well analysis-ready imagery computed using Framework Operational Radiometric Correction Environmental monitoring (FORCE) processing pipeline. From data, we extracted positions, filtered 387 775 trees upper canopy layer, and automatically corresponding reflectance from L2A images. These are labeled with species, which allows pixel-wise tasks. Furthermore, provide auxiliary information such approximate position, year possible disturbance events, or diameter at breast height. Temporally, spans years July 2015 to end October 2022, approx. 75.3 million data points 48 3 groups 13.8 observations non-tree backgrounds. Spatially, it covers whole Germany. available following DOI (Freudenberg et al., 2024): https://doi.org/10.3220/DATA20240402122351-0.

Language: Английский

---

Towards operational UAV-based forest health monitoring: Species identification and crown condition assessment by means of deep learning

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 219, P. 108785 - 108785

Published: March 6, 2024

Uncrewed Aerial Vehicles (UAVs) have emerged as a promising tool for complementing terrestrial surveys, offering unique advantages forest health monitoring (FHM). UAVs the potential to improve or even replace core tasks such crown condition assessment, bridging gap between ground-based surveys and traditional remote sensing platforms. However, present approaches not yet fully exploited very high temporal resolution flexible convenient utilization that offer under cloudy skies. In this paper, we provide standardized data pipeline semi-automatically generate reference by merging UAV-based related species-specific health. Furthermore, investigated of Convolutional Neural Networks (CNNs) classify main tree species their conditions based on data. Therefore, acquired multispectral drone imagery 235 different ICP large scale plots (Level-I plots) distributed across Bavaria three consecutive years (2020–2022). Using highly heterogeneous time-series dataset, encompassing diverse weather lighting conditions, stand characteristics, spatial distribution study areas, successfully classified five species, genus level classes dead trees, including status occurring in Germany. This way managed 14 distinct with an average macro F1-score 0.61 using EfficientNet CNN architecture. The highest class-specific apart from class trees (0.97) was achieved Picea abies healthy (0.80). If participating countries Forests program adopt our approach harmonize monitoring, many could be reduced replaced, leading significant time cost savings. We open-source analysis strategies can potentially extended throughout Europe. Our findings demonstrate UAV deep learning modernize management efficiency sustainability. recommend integrating drones ground systems take advantage benefits.

Language: Английский

---

Common Practices and Taxonomy in Deep Multiview Fusion for Remote Sensing Applications

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Journal Year: 2024, Volume and Issue: 17, P. 4797 - 4818

Published: Jan. 1, 2024

The advances in remote sensing technologies have boosted applications for Earth observation. These provide multiple observations or views with different levels of information. They might contain static temporary resolution, addition to having types and amounts noise due sensor calibration deterioration. A great variety deep learning models been applied fuse the information from these views, known as multi-view multi-modal fusion learning. However, approaches literature vary greatly since terminology is used refer similar concepts illustrations are given techniques. This article gathers works on observation by focusing common practices literature. We summarize structure insights several publications concentrating unifying points ideas. In this manuscript, we a harmonized while at same time mentioning various alternative terms that topics covered reviewed focus supervised use neural network models. hope review, long list recent references, can support future research lead unified advance area.

Language: Английский

---

Map of forest tree species for Poland based on Sentinel-2 data

Earth system science data, Journal Year: 2024, Volume and Issue: 16(6), P. 2877 - 2891

Published: June 20, 2024

Abstract. Accurate information on forest tree species composition is vital for various scientific applications, as well inventory and management purposes. Country-wide, detailed maps are a valuable resource environmental management, conservation, research, planning. Here, we performed the classification of 16 dominant genera in Poland using time series Sentinel-2 imagery. To generate comprehensive spectral–temporal information, created seasonal aggregations known metrics (STMs) within Google Earth Engine (GEE). STMs were computed short periods 15–30 d during spring, summer, autumn, covering multi-annual observations from 2018 to 2021. The Polish Forest Data Bank served reference data, and, obtain robust samples with pure stands only, data validated through automated visual inspection based very-high-resolution orthoimagery, resulting 4500 polygons serving training test data. mask was derived available land cover datasets GEE, namely ESA WorldCover Dynamic World dataset. Additionally, incorporated topographic climatic variables GEE enhance accuracy. random algorithm employed process, an area-adjusted accuracy assessment conducted cross-validation datasets. results demonstrate that country-wide stand mapping achieved exceeding 80 %; however, this varies greatly depending species, region, observation frequency. We provide freely accessible resources, including map data: https://doi.org/10.5281/zenodo.10180469 (Grabska-Szwagrzyk, 2023a).

Language: Английский

---

GloUTCI-M: a global monthly 1 km Universal Thermal Climate Index dataset from 2000 to 2022

Earth system science data, Journal Year: 2024, Volume and Issue: 16(5), P. 2407 - 2424

Published: May 22, 2024

Abstract. Climate change has precipitated recurrent extreme events and emerged as an imposing global challenge, exerting profound far-reaching impacts on both the environment human existence. The Universal Thermal Index (UTCI), serving important approach to comfort assessment, plays a pivotal role in gauging how humans adapt meteorological conditions copes with thermal cold stress. However, existing UTCI datasets still grapple limitations terms of data availability, hindering their effective application across diverse domains. We have produced GloUTCI-M, monthly dataset boasting coverage extensive time series spanning March 2000 October 2022, high spatial resolution 1 km. This is product comprehensive leveraging multiple sources advanced machine learning models. Our findings underscored superior predictive capabilities CatBoost forecasting (mean absolute error, MAE = 0.747 °C; root mean square RMSE 0.943 coefficient determination, R2=0.994) when compared models such XGBoost LightGBM. Utilizing geographical boundaries stress areas at scale were effectively delineated. Spanning 2001–2021, annual was recorded 17.24 °C, pronounced upward trend. Countries like Russia Brazil key contributors increasing, while countries China India exerted more inhibitory influence this Furthermore, contrast datasets, GloUTCI-M excelled portraying distribution finer resolutions, augmenting accuracy. can enhance our capacity evaluate experienced by humans, offering substantial prospects wide array applications. publicly available https://doi.org/10.5281/zenodo.8310513 (Yang et al., 2023).

Language: Английский

---

A Review: Tree Species Classification Based on Remote Sensing Data and Classic Deep Learning-based Methods

Published: April 8, 2024

Timely and accurate information on tree species is crucial for the sustainable management of natural resources, forest inventory, biodiversity detection, carbon stock calculation. The advancement remote sensing technology artificial intelligence has facilitated acquisition analysis data, resulting in more precise effective classification species. Multimodal data deep learning seem to be current research mainstream, whether or not. review methods perspectives analyze unimodal multimodal this realm missing. To bridge gap, we search major trends methods, provide a detailed overview classic learning-based classification, discuss limitations.

Language: Английский

---

Spectral-temporal traits in Sentinel-1 C-band SAR and Sentinel-2 multispectral remote sensing time series for 61 tree species in Central Europe

Remote Sensing of Environment, Journal Year: 2024, Volume and Issue: 307, P. 114162 - 114162

Published: April 17, 2024

Tree species maps derived from satellite imagery increasingly support forest administrations and nature conservation authorities with large-scale up-to-date information. However, many are often excluded or aggregated in classification tasks due to a limited knowledge of the most suitable predictors. Our study aims gain better understanding optical polarimetric traits for tree mapping by examining Sentinel-1 Sentinel-2 time series 61 temperate Europe. For selection 32 optical, structural variables, principal component analysis revealed that variables mainly explain variance data contributing "seasonality" "foliage color" components. contribute "texture" component. The Normalized Difference Vegetation Index (NDVI), Tasseled Cap Greenness (TCG) Radar (RVI) were chosen as key further analysis. Seasonality was found be dominant aspect all vegetation indices. Furthermore, TCG useful distinguish between early late budding species. RVI showed large potential discriminate conifers, which is attributed crown volume effect C-band SAR. Using exploratory analysis, we examined influence management, biogeographical meteorological factors on Fagus sylvatica, Pinus sylvestris, Picea abies. NDVI relatively robust different conditions. two conifer however, strong spatial variations presumably caused conditions across area. could therefore lead uncertainties gradients. This contributes improvement based dual-polarimetric thus benefits other stakeholders their monitoring decision-making.

Language: Английский

---