Effective integration of drone technology for mapping and managing palm species in the Peruvian Amazon

Ximena Tagle Casapia, Rodolfo Cárdenas Vigo, Diego Marcos, Ernesto Fernández Gamarra, Harm Bartholomeus, Eurídice N. Honorio Coronado, Silvana Di Liberto Porles, Lourdes Falen, Susan Palacios, Nandin-Erdene Tsenbazar, Gordon Mitchell, Ander Dávila Díaz, Frederick C. Draper, Gerardo Flores Llampazo, Pedro E. Pérez-peña, Giovanna Chipana, Dennis Del Castillo Torres, Martin Herold, Timothy R. BakerNature Communications 2025 · DOI

Abstract

Remote sensing data could increase the value of tropical forest resources by helping to map economically important species. However, current tools lack precision over large areas, and remain inaccessible to stakeholders. Here, we work with the Protected Areas Authority of Peru to develop and implement precise, landscape-scale, species-level methods to assess the distribution and abundance of economically important arborescent Amazonian palms using field data, visible-spectrum drone imagery and deep learning. We compare the costs and time needed to inventory and develop sustainable fruit harvesting plans in two communities using traditional plot-based and our drone-based methods. Our approach detects individual palms of three species, even when densely clustered (average overall score, 74%), with high accuracy and completeness for Mauritia flexuosa (precision; 99% and recall; 81%). Compared to plot-based methods, our drone-based approach reduces costs per hectare of an inventory of Mauritia flexuosa for a management plan by 99% (USD 5 ha-1 versus USD 411 ha-1), and reduces total operational costs and personnel time to develop a management plan by 23% and 36%, respectively. These findings demonstrate how tailoring technology to the scale and precision required for management, and involvement of stakeholders at all stages, can help expand sustainable management in the tropics.

Introduction

Managing tropical forests sustainably requires precise knowledge of species distribution. In the Peruvian Amazon, arborescent palms like Mauritia flexuosa (aguaje), Oenocarpus bataua (ungurahui), and Euterpe precatoria (huasaí) are "hyperdominant" species. They are ecologically critical—storing massive amounts of carbon in peatlands—and economically essential for local fruit harvesters.

However, traditional "plot-based" inventories are logistically challenging, involving teams trekking through waterlogged, inaccessible swamps to count trees manually. This process is slow, expensive, and covers only small fractions of the landscape. While satellite imagery is available, it lacks the sub-meter resolution needed to distinguish individual palm crowns in dense jungle. This research bridges the "research-implementation gap" by creating a cost-effective, user-friendly drone-based solution for government agencies and local communities.

Methods and Datasets

The study developed a deep-learning-powered drone workflow combining low-cost hardware with a two-stage computer vision pipeline.

Visual identification of the three target palm species across different scales: fruit, ground view, and drone view with instance segmentation outlines.

Data Acquisition and Processing

Using standard commercial RGB drones (e.g., DJI Phantom 4), high-resolution imagery (~10 cm/pixel) was captured across 55 sites. The processing pipeline involved:

  1. Semantic Segmentation: DeepLab v3+ (MobileNet-v2 backbone) classified pixels into species categories. Extensive data augmentation simulated varying environmental conditions (light, wind) to ensure model robustness.
  2. Instance Segmentation: A Deep Watershed Transform solved the "clumping" problem, identifying individual palm centers and growing crowns outward to count separate trees rather than aggregate clusters.

Study Areas

The model was trained and validated using a ground-truth dataset of 5,089 georeferenced palms across 55 sites in the Loreto region. To test generalizability, the model was evaluated on new mosaics from the Madre de Dios region, hundreds of kilometers away, using different drone hardware.

Map of the study areas in Peru, showing training sites in the Loreto region (A) and test sites in the Madre de Dios region (B).

Key Results and Economic Impact

The drone-based approach demonstrated significant improvements in both accuracy and operational efficiency.

  • High Precision: For the primary species (M. flexuosa), the model achieved 99% precision and an 81% F1-score, successfully detecting individuals even in dense clusters.
  • Economic Viability: The cost of inventorying M. flexuosa dropped from USD $411 per hectare (plot-based) to just USD $5 per hectare (drone-based)—a 99% reduction. While drones involve upfront costs, the marginal cost per inventory is nearly zero; the drone approach becomes cheaper than traditional methods after just four management plans.
  • Operational Efficiency: Total operational costs for developing sustainable harvesting plans were reduced by 23%, and personnel time dropped by 36%. This allows park rangers to focus on high-level management and community engagement.
  • Scalability: The drone method enabled surveying 230 hectares in the same timeframe required for just 10 hectares on the ground.

Discussion and Limitations

While highly effective, the system has limitations. Drones primarily see the top canopy layer, leading to underestimation of sub-canopy and understory palms. Additionally, model performance was highest for M. flexuosa due to its dominance in the training data; future work requires more diverse samples for less common species like E. precatoria.

Crucially, the success of this project relied on stakeholder integration. By co-designing the tool with SERNANP (Peru’s Protected Areas Authority) and training park rangers, the researchers ensured the technology was operationally viable for real-world conservation management. Future improvements may include the use of Vertical Take-Off and Landing (VTOL) drones to cover larger areas with greater stability.

Summary updated

Abstract

Remote sensing data could increase the value of tropical forest resources by helping to map economically important species. However, current tools lack precision over large areas, and remain inaccessible to stakeholders. Here, we work with the Protected Areas Authority of Peru to develop and implement precise, landscape-scale, species-level methods to assess the distribution and abundance of economically important arborescent Amazonian palms using field data, visible-spectrum drone imagery and deep learning. We compare the costs and time needed to inventory and develop sustainable fruit harvesting plans in two communities using traditional plot-based and our drone-based methods. Our approach detects individual palms of three species, even when densely clustered (average overall score, 74%), with high accuracy and completeness for Mauritia flexuosa (precision; 99% and recall; 81%). Compared to plot-based methods, our drone-based approach reduces costs per hectare of an inventory of Mauritia flexuosa for a management plan by 99% (USD 5 ha<sup>-1</sup> versus USD 411 ha<sup>-1</sup>), and reduces total operational costs and personnel time to develop a management plan by 23% and 36%, respectively. These findings demonstrate how tailoring technology to the scale and precision required for management, and involvement of stakeholders at all stages, can help expand sustainable management in the tropics.
BibTeX
@inproceedings{casapia2025effective,
  author = {Ximena Tagle Casapia and Rodolfo Cárdenas Vigo and Diego Marcos and Ernesto Fernández Gamarra and Harm Bartholomeus and Eurídice N. Honorio Coronado and Silvana Di Liberto Porles and Lourdes Falen and Susan Palacios and Nandin-Erdene Tsenbazar and Gordon Mitchell and Ander Dávila Díaz and Frederick C. Draper and Gerardo Flores Llampazo and Pedro E. PÉREZ-PEÑA and Giovanna Chipana and Dennis Del Castillo Torres and Martin Herold and Timothy R. Baker},
  title = {{Effective integration of drone technology for mapping and managing palm species in the Peruvian Amazon}},
  booktitle = {Nature Communications},
  year = {2025},
  doi = {10.1038/s41467-025-58358-5},
}
Figure