Pawlak, C. | bioRxiv Science

bioRxiv2026

MillionTrees: A global benchmark for tree detection across geography, sensor, and annotation geometry

Airborne tree monitoring using remotely sensed data, combined with the growth of computer vision models for object detection and segmentation, can facilitate broad scale assessment of tree health, ecology and ecosystem function. However, datasets for airborne tree detection have been largely limited to a single geography, resolution, or annotation approach, making it difficult to develop general models and benchmark their utility for broad applications. We compile existing and newly published annotation datasets to create a comprehensive dataset for training and evaluating airborne tree detection in RGB imagery. The MillionTrees dataset has over a million individual trees from a broad range of forest types, tree species, and geographic locations. In addition, it has six million weakly supervised annotations from a range of sources including airborne LiDAR and photogrammetry, arborist surveys, and simulated landscapes. An evaluation standard is provided to facilitate rapid model development. We benchmark several existing tree detection algorithms and demonstrate that while current open-source models do not generalize across the diverse datasets in the benchmark, training on MillionTrees improves both within-domain and out-of-domain performance. Held-out validation datasets using ground truth derived from terrestrial LiDAR paired with overhead RGB show that overfitting to test data is common and hyperparameter tuning is an underappreciated challenge towards a global zero shot tree detection method. We discuss the limitations of the currently available data for airborne RGB tree detection, how to create more robust evaluation data and the need for better understanding of the uncertainty and utility of tree detection from airborne remote sensing.

ecology↗

bioRxiv2021

Automated satellite remote sensing of giant kelp at the Falkland Islands (Islas Malvinas)

Giant kelp populations support productive and diverse coastal ecosystems in both hemispheres at temperate and subpolar latitudes but are vulnerable to changing climate conditions as well as direct human impacts. Observations of giant kelp forests are spatially and temporally patchy, with disproportionate coverage in the northern hemisphere, despite the size and comparable density of southern hemisphere kelp forests. Satellite imagery enables the mapping of existing and historical giant kelp populations in understudied regions, but automating the detection of giant kelp in large satellite datasets requires approaches that are robust to the optical complexity of the shallow, nearshore environment. We present and compare two approaches for automating the detection of giant kelp in satellite datasets: one based on crowd sourcing of satellite imagery classifications and another based on a decision tree paired with a spectral unmixing algorithm (automated using Google Earth Engine). Both approaches are applied to satellite imagery (Landsat) of the Falkland Islands or Islas Malvinas (FLK), an archipelago in the southern Atlantic Ocean that supports expansive giant kelp ecosystems. The performance of each method is evaluated by comparing the automated classifications with a subset of expert-annotated imagery cumulatively spanning over 2,700km of coastline. Using the remote sensing approaches evaluated herein, we present the first continuous timeseries of giant kelp observations in the FLK region using Landsat imagery spanning over three decades. We do not detect evidence of long-term change in the FLK region, although we observe a recent decline in total canopy area from 2017-2021. Using a nitrate model based on nearby ocean state measurements obtained from ships and incorporating satellite sea surface temperature products, we find that the area of giant kelp forests in the FLK region is positively correlated with the nitrate content observed during the prior year. Our results indicate that giant kelp classifications using citizen science are approximately consistent with classifications based on a state-of-the-art automated spectral approach. Despite differences in accuracy and sensitivity, both approaches find high interannual variability that impedes the detection of potential long-term changes in giant kelp canopy area, although recent canopy area declines are notable and should continue to be monitored carefully.

ecology↗