Temperature dependence of pollen germination in Douglas-fir (Pseudotsuga menziesii): A machine learning-based detection of pollen viability from microscopic images
- 1University of Washington
Abstract
Background and Aims: Pollen germination and tube growth are critical stages of plant reproduction that are highly sensitive to temperature but remain labor-intensive to quantify. This study aimed to develop a deep learning-based approach for pollen phenotyping and to characterize the temperature dependence of pollen germination and elongation in Douglas-fir (Pseudotsuga menziesii). Methods: A convolutional neural network (CNN) was trained to segment pollen grains and classify germination status from microscopic images. Germination percentage and pollen length were quantified across a range of incubation temperatures from 5 to 40{degrees}C. Gamma functions were fitted to temperature response curves to estimate the optimal temperatures for pollen germination and elongation among Douglas-fir populations collected across an elevational gradient. Key Results: The CNN achieved high segmentation accuracy (intersection over union = 0.846), accurately distinguishing pollen grains from the background but showing moderate accuracy in separating germinated from ungerminated pollen during early elongation. Both pollen germination and elongation exhibited bell-shaped temperature response curves with distinct thermal optima. Pollen elongation consistently reached its optimum at higher temperatures than pollen germination. Estimated optimal temperatures fell within a narrow range, approximately 19 to 23{degrees}C. No significant relationship was detected between elevation and thermal optima, although some higher-elevation populations exhibited lower optimal temperatures. Comparisons with previous analyses of three western North American conifers showed that each species occupied a distinct reproductive thermal niche corresponding to the spring temperatures of its native habitat. Conclusions: Deep learning provides an efficient approach for high-throughput quantification of pollen germination and elongation from microscopic images. The narrow thermal range for reproductive performance suggests that Douglas-fir pollen is sensitive to temperature variation and that warming climates may alter reproductive success. These findings improve our understanding of the thermal sensitivity of conifer reproduction and provide a scalable framework for assessing impacts of climate warming on forest regeneration.
Related
- Estimates of the rate of adaptive evolution in four wild populations of Nemophila menziesii
- A cellular view of seed development and germination
- Cyclic pathogen epidemics favour the evolution of delayed germination
- Phenotypic Variation Characterization and Multiple-Method Analysis of Drought Resistance in Rice at the Germination Stage
- Multi-population approach in seed germination ecology: evidence from Ranunculus ficaria
BibTeX
@misc{hsu2026temperature,
title = {{Temperature dependence of pollen germination in Douglas-fir (Pseudotsuga menziesii): A machine learning-based detection of pollen viability from microscopic images}},
author = {H.-W. Hsu and S. Zerrade and S.-H. Kim},
howpublished = {bioRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.10.750703},
url = {https://www.biorxiv.org/content/10.64898/2026.09.10.750703v1},
note = {CC BY-ND},
}
RIS
TY - UNPB
TI - Temperature dependence of pollen germination in Douglas-fir (Pseudotsuga menziesii): A machine learning-based detection of pollen viability from microscopic images
AU - H.-W. Hsu
AU - S. Zerrade
AU - S.-H. Kim
PY - 2026
DA - 2026/09/10
PB - bioRxiv
DO - 10.64898/2026.09.10.750703
UR - https://www.biorxiv.org/content/10.64898/2026.09.10.750703v1
AB - Background and Aims: Pollen germination and tube growth are critical stages of plant reproduction that are highly sensitive to temperature but remain labor-intensive to quantify. This study aimed to develop a deep learning-based approach for pollen phenotyping and to characterize the temperature dependence of pollen germination and elongation in Douglas-fir (Pseudotsuga menziesii). Methods: A convolutional neural network (CNN) was trained to segment pollen grains and classify germination status from microscopic images. Germination percentage and pollen length were quantified across a range of incubation temperatures from 5 to 40{degrees}C. Gamma functions were fitted to temperature response curves to estimate the optimal temperatures for pollen germination and elongation among Douglas-fir populations collected across an elevational gradient. Key Results: The CNN achieved high segmentation accuracy (intersection over union = 0.846), accurately distinguishing pollen grains from the background but showing moderate accuracy in separating germinated from ungerminated pollen during early elongation. Both pollen germination and elongation exhibited bell-shaped temperature response curves with distinct thermal optima. Pollen elongation consistently reached its optimum at higher temperatures than pollen germination. Estimated optimal temperatures fell within a narrow range, approximately 19 to 23{degrees}C. No significant relationship was detected between elevation and thermal optima, although some higher-elevation populations exhibited lower optimal temperatures. Comparisons with previous analyses of three western North American conifers showed that each species occupied a distinct reproductive thermal niche corresponding to the spring temperatures of its native habitat. Conclusions: Deep learning provides an efficient approach for high-throughput quantification of pollen germination and elongation from microscopic images. The narrow thermal range for reproductive performance suggests that Douglas-fir pollen is sensitive to temperature variation and that warming climates may alter reproductive success. These findings improve our understanding of the thermal sensitivity of conifer reproduction and provide a scalable framework for assessing impacts of climate warming on forest regeneration.
ER -