Summary
A bioRxiv preprint describes a convolutional neural network that measured Douglas-fir pollen germination and elongation across temperatures from 5 to 40°C. The study found narrow, distinct thermal optima for the two reproductive processes, mostly between 19 and 23°C.
A deep-learning system has been used to measure how temperature affects pollen germination and tube elongation in Douglas-fir (Pseudotsuga menziesii), according to a bioRxiv preprint from researchers at the University of Washington. The analysis found that the two processes have distinct but narrow temperature optima, estimated at roughly 19 to 23°C across the Douglas-fir populations examined.
The study combines automated image analysis with controlled temperature experiments. Its results suggest that temperature-sensitive pollen development could be an important factor when assessing how warming affects conifer reproduction, although the experiment measured pollen-stage responses rather than reproduction or regeneration in the field.
Measuring pollen development with a neural network
Pollen germination and the subsequent growth of the pollen tube are important stages in plant reproduction. They are also time-consuming to quantify manually because researchers must inspect microscopic images and classify many individual pollen grains.
The researchers trained a convolutional neural network, or CNN, to identify pollen grains in microscope images and classify whether they had germinated. A CNN is a type of machine-learning model commonly used for image recognition: it learns visual patterns from labelled examples and then applies those patterns to new images.
The model achieved an intersection-over-union score of 0.846 for segmenting pollen grains from the background. Intersection over union compares the area identified by the model with the area in the reference annotation; a higher score indicates closer overlap. The network was less consistent when distinguishing germinated from ungerminated pollen during the early stages of tube elongation, which the authors describe as moderate accuracy.
The automated measurements were used to calculate germination percentage and pollen length at incubation temperatures ranging from 5 to 40°C.
Germination and elongation preferred different temperatures
Both germination and elongation followed bell-shaped temperature-response curves. In practical terms, performance increased as temperature approached a favourable range and declined when conditions moved away from it.
The optimum for pollen elongation was consistently higher than the optimum for pollen germination. Across Douglas-fir populations collected along an elevational gradient, the estimated optimal temperatures remained within an approximately 19–23°C range.
The analysis found no significant relationship between elevation and the thermal optima. Some populations from higher elevations nevertheless showed lower optimal temperatures, a pattern that may be relevant to how reproductive traits vary across environments.
The authors also compared the results with earlier analyses of three western North American conifers. They report that the species occupied distinct reproductive thermal niches, with those niches corresponding to the spring temperatures of their native habitats.
Why the temperature range matters
A narrow optimum means that modest changes in temperature can shift pollen conditions away from the range associated with the highest measured germination or elongation. For Douglas-fir, this provides a measurable link between temperature and a reproductive stage that precedes successful forest regeneration.
The machine-learning method is also significant as a measurement tool. Faster image-based scoring could allow researchers to examine more pollen samples, temperatures and populations than would be practical through entirely manual counts. The preprint presents the approach as a scalable framework for studying how climate warming may affect conifer reproduction.
The findings are reported in a preprint, and the model’s weaker performance during early elongation is an important qualification for interpreting the automated classifications. The experiment also focused on pollen germination and tube growth under controlled incubation conditions. How those responses translate into seed production and regeneration under changing field conditions remains a separate ecological question.