[Paper Review] A global method to identify trees inside and outside of forests with medium-resolution satellite imagery.

John Brandt, Fred Stolle|arXiv (Cornell University)|May 13, 2020

Remote Sensing and LiDAR Applications63 references3 citations

TL;DR

This paper presents a globally consistent deep learning method using 10-meter pansharpened Sentinel-2 optical and Sentinel-1 radar imagery to detect trees both inside and outside forests at medium resolution. Trained with a fully convolutional network featuring a convolutional gated recurrent unit and feature pyramid attention layer, the model achieves over 75% user and producer accuracy in low-to-medium canopy density areas and 95% in dense forests, enabling high-accuracy global tree mapping across diverse landscapes.

ABSTRACT

Scattered trees outside of dense forests are very important for carbon sequestration, supporting livelihoods, maintaining ecosystem integrity, and climate change adaptation and mitigation. In contrast to trees inside of forests, not much is known about the spatial extent and distribution of scattered trees at a global scale. Due to the very high cost of high-resolution satellite imagery, global monitoring systems rely on medium resolution satellites to monitor land use and land use change. However, detecting and monitoring scattered trees with an open canopy using medium resolution satellites is difficult because individual trees often cover a smaller footprint than the satellites resolution. Here we present a globally consistent method to identify trees inside and outside of forests with medium resolution optical and radar imagery. Biweekly cloud-free, pansharpened 10 meter Sentinel-2 optical imagery and Sentinel-1 radar imagery are used to train a fully convolutional network, consisting of a convolutional gated recurrent unit layer and a feature pyramid attention layer. Tested across more than 215,000 Sentinel-1 and Sentinel-2 pixels distributed from -60 to +60 latitude, the proposed model exceeds 75 percent users and producers accuracy identifying trees in hectares with a low to medium density (less than 40 percent) of canopy cover, and 95 percent user's and producer's accuracy in hectares with dense (greater than 40 percent) canopy cover. When applied across large, heterogeneous landscapes, the results demonstrate potential to map trees in high detail and consistent accuracy over diverse landscapes across the globe. This information is important for understanding current land cover and can be used to detect changes in land cover such as agroforestry, buffer zones around biological hotspots, and expansion or encroachment of forests.

Motivation & Objective

• To address the lack of global-scale data on scattered trees outside dense forests, which are critical for carbon sequestration and ecosystem integrity.
• To overcome the challenge of detecting individual trees with medium-resolution satellite imagery, where tree canopies often fall below the spatial resolution.
• To develop a consistent, globally applicable method for identifying trees in heterogeneous landscapes using optical and radar satellite data.
• To improve monitoring of land use and land use change, including agroforestry, forest encroachment, and buffer zones around biodiversity hotspots.

Proposed method

• The method uses biweekly cloud-free, pansharpened 10-meter Sentinel-2 optical imagery and Sentinel-1 radar imagery as input data.
• A fully convolutional neural network architecture is employed, incorporating a convolutional gated recurrent unit layer to model temporal dynamics and spatial context.
• A feature pyramid attention layer enhances feature representation across multiple scales, improving detection of trees with varying sizes and canopy densities.
• The model is trained on over 215,000 pixels spanning latitudes from -60° to +60°, ensuring global applicability.
• The approach leverages both optical and radar data to improve robustness under varying weather and lighting conditions.
• Model performance is evaluated using user and producer accuracy across different canopy density thresholds.

Experimental results

Research questions

• RQ1 Can a deep learning model reliably detect trees in both forested and non-forested areas using medium-resolution optical and radar satellite imagery?
• RQ2 How accurate is the model in identifying trees across diverse global landscapes with varying canopy cover densities?
• RQ3 To what extent can the integration of optical and radar data improve tree detection accuracy compared to single-sensor approaches?
• RQ4 Can the model maintain consistent performance across different biomes and land cover types at a global scale?
• RQ5 How effective is the method in detecting changes in tree cover, such as agroforestry development or forest encroachment?

Key findings

• The model achieves over 75% user and producer accuracy in identifying trees within hectares that have a low to medium canopy cover density of less than 40%.
• In areas with dense canopy cover (greater than 40%), the model reaches 95% user and producer accuracy, demonstrating high reliability in forested regions.
• The method maintains consistent performance across large, heterogeneous landscapes, indicating scalability and robustness across diverse global biomes.
• The integration of optical and radar data significantly enhances detection accuracy, especially in regions with frequent cloud cover.
• The model enables high-resolution tree mapping at a global scale, supporting detailed monitoring of land cover changes such as agroforestry and forest expansion.
• The results demonstrate strong potential for operational use in tracking land use and land use change, including buffer zones and biological hotspot protections.