Dataset of Ground-Based Vineyard Thermal Images, Stem Water Potential, Meteorological, and Soil Water Probe Measurements.
Description
This dataset includes ground thermal images with corresponding masks (manually segmented and automatically segmented using a pretrained Unet++ fine-tuned on manual masks), stem water potential measurements, meteorological data, and soil water probe readings from the 2024 vineyard growing season in Lisbon, Portugal. It supports research on water status estimation methods, thermal image processing and segmentation, canopy temperature analysis, CWSI evaluation, and the effects of solar panels, grape variety, time of day, and row orientation on thermal imaging.
Files
Steps to reproduce
Study site: Adult drip-irrigated vineyard plot (1.6 ha) at Tapada da Ajuda, Lisbon (38.70679822, −9.18477230). Planting density: 4000 plants/ha (1 m vine spacing, 2.5 m row spacing). White Vitis vinifera varieties: ‘Moscatel de Setúbal’, ‘Arinto’, ‘Encruzado’ (on 1103P rootstock) and ‘Viosinho’ (on 110R rootstock). Meteorological data: Obtained from Instituto Português do Mar e da Atmosfera (IPMA) stations at Tapada da Ajuda (300 m), Geofísico (3.3 km), and Gago Coutinho (8.3 km). Parameters: relative humidity, wind speed, air temperature, radiation. Recorded at 10-minute intervals. Soil water content: Measured with seven EnviroSCAN capacitance probes (Sentek Technologies, Australia). Depths: 20, 40, 60, and 80 cm. Recorded at 15-minute intervals. Stem water potential (SWP): Measured with a Scholander pressure chamber. Procedure: enclose leaves in foil and plastic bags at 11–12 h, remove after ~1.5 h, then measure. One undamaged, fully mature leaf per randomly selected vine across the four varieties. Thermal imaging: Camera: FLIR A35 (320 × 256 px, 60 fps, 13° × 10° FOV, 7.5–13 µm, NETD <0.05 °C @ 30 °C, emissivity 0.96). Acquisition: handheld at 35–45° angle, ~0.8 m from vine, from both east and west canopy sides. Times of day: Morning (9:00–11:30), Midday (11:30–14:00), Afternoon (14:00–17:00). Wet reference targets included on some days. Image segmentation: Thermal image masks created manually and via automated segmentation. Automated segmentation used Unet++ with ResNet34 backbone pretrained on ImageNet and fine-tuned on manual masks.
Institutions
- Universidade de Lisboa Instituto Superior de Agronomia
- Universidade de Lisboa Instituto de Sistemas e Robotica
- Universidade de Lisboa Instituto Superior Tecnico