ReforesTree: A Dataset for Estimating Tropical Forest Carbon Stock with Deep Learning and Aerial Imagery

Reiersen, Gyri; Dao, David; Lütjens, Björn; Klemmer, Konstantin; Amara, Kenza; Steinegger, Attila; Zhang, Ce; Zhu, Xiaoxiang

doi:10.1609/aaai.v36i11.21471

Cited by 11 publications

(9 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The average stem recall was 69.4%, with better performance in well-spaced western forests, and weaker performance in alpine conifer forests. DeepForest has been used widely outside of NEON sites (Reiersen et al 2022, Kapil et al 2022, Velasquez-Camacho et al 2023, Kwon et al 2023), with accuracies generally mirroring ∼70% for fine-tuned models from independent analysis (Gan et al 2023).…”

Section: Methodsmentioning

confidence: 99%

Individual canopy tree species maps for the National Ecological Observatory Network

Weinstein,

Marconi,

Zare

et al. 2023

Preprint

View full text Add to dashboard Cite

The ecology of forest ecosystems depends on the composition of trees. Capturing fine-grained information on individual trees at broad scales allows an unprecedented view of forest ecosystems, forest restoration and responses to disturbance. To create detailed maps of tree species, airborne remote sensing can cover areas containing millions of trees at high spatial resolution. Individual tree data at wide extents promises to increase the scale of forest analysis, biogeographic research, and ecosystem monitoring without losing details on individual species composition and abundance. Computer vision using deep neural networks can convert raw sensor data into predictions of individual tree species using ground truthed data collected by field researchers. Using over 40,000 individual tree stems as training data, we create landscape-level species predictions for over 100 million individual trees for 24 sites in the National Ecological Observatory Network. Using hierarchical multi-temporal models fine-tuned for each geographic area, we produce open-source data available as 1km^2 shapefiles with individual tree species prediction, as well as crown location, crown area and height of 81 canopy tree species. Site-specific models had an average performance of 79% accuracy covering an average of six species per site, ranging from 3 to 15 species. All predictions were uploaded to Google Earth Engine to benefit the ecology community and overlay with other remote sensing assets. These data can be used to study forest macro-ecology, functional ecology, and responses to anthropogenic change.

show abstract

Section: Methodsmentioning

confidence: 99%

Individual canopy tree species maps for the National Ecological Observatory Network

Weinstein,

Marconi,

Zare

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…To predict carbon stock, machine learning algorithms are employed to identify relevant features such as topography, climate conditions, and vegetation indices. Parameters are optimized and adjusted after creating the model, and the algorithm that achieves the highest accuracy is selected for prediction (Silatsa et al, 2020;Reiersen et al, 2022;Uniyal et al, 2022). Different machine learning algorithms produce different levels of accuracy with the same dataset.…”

Section: Introductionmentioning

confidence: 99%

Estimating carbon sequestration potential and optimizing management strategies for Moso bamboo (Phyllostachys pubescens) forests using machine learning

Lv,

Yuan,

Sun

et al. 2024

Front. For. Glob. Change

View full text Add to dashboard Cite

Estimating the carbon sequestration potential of Moso bamboo (Phyllostachys pubescens) forests and optimizing management strategies play pivotal roles in enhancing quality and promoting sustainable development. However, there is a lack of methods to simulate changes in carbon sequestration capacity in Moso bamboo forests and to screen and optimize the best management measures based on long-term time series data from fixed-sample fine surveys. Therefore, this study utilized continuous survey data and climate data from fixed sample plots in Zhejiang Province spanning from 2004 to 2019. By comparing four different algorithms, namely random forest, support vector machine, XGBoost, and BP neural network, to construct aboveground carbon stock models for Moso bamboo forests. The ultimate goal was to identify the optimal algorithmic model. Additionally, the key driving parameters for future carbon stocks were considered and future aboveground carbon stocks were predicted in Moso bamboo forests. Then formulated an optimal management strategy based on these predictions. The results indicated that the carbon stock model constructed using the XGBoost algorithm, with an R2 of 0.9895 and root mean square error of 0.1059, achieved the best performance and was considered the optimal algorithmic model. The most influential driving parameters for vegetation carbon stocks in Moso bamboo forests were found to be mean age, mean diameter at breast height, and mean culm density. Under optimal management measures, which involve no harvesting of 1–3 du bamboo, 30% harvesting of 4 du bamboo, and 80% harvesting of bamboo aged 5 du and above. Our predictions show that aboveground carbon stocks in Moso bamboo forests in Zhejiang Province will peak at 36.25 ± 8.47 Tg C in 2046 and remain stable from 2046 to 2060. Conversely, degradation is detrimental to the long-term maintenance of carbon sequestration capacity in Moso bamboo forests, resulting in a peak aboveground carbon stock of 29.50 ± 7.49 Tg C in 2033, followed by a continuous decline. This study underscores the significant influence of estimating carbon sequestration potential and optimizing management decisions on enhancing and sustaining the carbon sequestration capacity of Moso bamboo forests.

show abstract

“…Applying computer vision on remote sensing images can automate a broad range of applications, such as poverty estimation (Jean et al 2016), crop yield prediction (You et al 2017), deforestation detection (Torres et al 2021), and renewable energy mapping (Yu et al 2018;Kruitwagen et al 2021). Facing the increasing risk of climate change, remote sensing imagery and the vision models built for it can further contribute to both mitigation and adaptation by enabling the observation of the Earth surface (Helber et al 2019), detection of highpollution industry (Lee et al 2021), evaluation of carbon stock (Reiersen et al 2022), and identification of vulnerable infrastructure and populations (Huang et al 2021).…”

Section: Introductionmentioning

confidence: 99%

SkyScript: A Large and Semantically Diverse Vision-Language Dataset for Remote Sensing

Wang,

Prabha,

Huang

et al. 2024

AAAI

View full text Add to dashboard Cite

Remote sensing imagery, despite its broad applications in helping achieve Sustainable Development Goals and tackle climate change, has not yet benefited from the recent advancements of versatile, task-agnostic vision language models (VLMs). A key reason is that the large-scale, semantically diverse image-text dataset required for developing VLMs is still absent for remote sensing images. Unlike natural images, remote sensing images and their associated text descriptions cannot be efficiently collected from the public Internet at scale. In this work, we bridge this gap by using geo-coordinates to automatically connect open, unlabeled remote sensing images with rich semantics covered in OpenStreetMap, and thus construct SkyScript, a comprehensive vision-language dataset for remote sensing images, comprising 2.6 million image-text pairs covering 29K distinct semantic tags. With continual pre-training on this dataset, we obtain a VLM that surpasses baseline models with a 6.2% average accuracy gain in zero-shot scene classification across seven benchmark datasets. It also demonstrates the ability of zero-shot transfer for fine-grained object attribute classification and cross-modal retrieval. We hope this dataset can support the advancement of VLMs for various multi-modal tasks in remote sensing, such as open-vocabulary classification, retrieval, captioning, and text-to-image synthesis.

show abstract

ReforesTree: A Dataset for Estimating Tropical Forest Carbon Stock with Deep Learning and Aerial Imagery

Cited by 11 publications

References 33 publications

Individual canopy tree species maps for the National Ecological Observatory Network

Individual canopy tree species maps for the National Ecological Observatory Network

Estimating carbon sequestration potential and optimizing management strategies for Moso bamboo (Phyllostachys pubescens) forests using machine learning

SkyScript: A Large and Semantically Diverse Vision-Language Dataset for Remote Sensing

Contact Info

Product

Resources

About