Use of PhenoCam measurements and image analysis to inform the ALMANAC process-based simulation model

Authors: JACOT, JACQUELINE - Oakland Research Institute; Kiniry, James; Williams, Amber; CORONEL, ADDISON - University Of Texas; SU, JIANZHONG - University Of Texas; MILLER, GRETCHEN - Texas A&M University; MOHANTY, BINAYAK - University Of Texas; SAHA, AMARTYA - Archbold Biological Station; GOMEZ-CASANOVAS, NURIA - University Of Illinois; Johnson, Jane; Browning, Dawn

Submitted to: Journal of Experimental Agriculture International
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/18/2021
Publication Date: 7/5/2021
Citation: Jacot, J., Kiniry, J.R., Williams, A.S., Coronel, A., Su, J., Miller, G.R., Mohanty, B., Saha, A., Gomez-Casanovas, N., Johnson, J.M., Browning, D.M. 2021. Use of PhenoCam measurements and image analysis to inform the ALMANAC process-based simulation model. Journal of Experimental Agriculture International. 43(4):120-144. https://doi.org/10.9734/jeai/2021/v43i430684.
DOI: https://doi.org/10.9734/jeai/2021/v43i430684

Interpretive Summary: Remote sensing measured near the ground has been used to document seasonal growth patterns for diverse plant communities. This study quantified greenness from PhenoCam images measured over growing seasons, and applied a simple equation (Beer’s law) to compare leaf area index (LAI) development with the ALMANAC process-based simulation model. We examined patterns and modeled LAI at eight diverse sites as part of the Long-Term Agroecosystem Research (LTAR) network. Sections of the images at each location were used to derived a greenness value. After photos were analyzed to determine greenness, each site was simulated in ALMANAC. Finally, the greenness data were converted to LAI and compared with modeled LAI. Results indicate that PhenoCam imagery can be used to improve parameters for leaf area development in ALMANAC by adjusting parameter values to better match LAI derived values in diverse environments. Soybeans, mesquite, and maize produced the most successful match between the model simulations and PhenoCam data out of the eight species simulated. This study represents the first independent evaluation of the ALMANAC process-based plant growth model with imagery from the PhenoCam network. The results show how PhenoCam data can make a valuable contribution to parameter determination in process-based models, making these models much more realistic. The impact of these conclusions can be increased as the methods are further refined.

Technical Abstract: Near-surface remote sensing has been used to document seasonal growth patterns (i.e., phenology) for plant communities in diverse habitats. This study quantified greenness from PhenoCam image time series, and applied Beer’s law with established extinction coefficients to compare leaf area index (LAI) development with the ALMANAC model. We examined patterns and modeled LAI at eight agroecosystem sites as part of the Long-Term Agroecosystem Research (LTAR) network selected for their diversity and amount of and quality of available data. Regions of interest (ROIs) in PhenoCam images were used to derived Green Chromatic Coordinate (GCC) at each site. Next, photos were analyzed in the L*A*B* color space in ImageJ to determine greenness and then the site was simulated in ALMANAC. Finally, the GCC and greenness data were converted to LAI and compared with modeled LAI. Results indicate that PhenoCam time series imagery can be used to improve parameters for leaf area development in ALMANAC by adjusting parameter values to better match LAI derived values in diverse environments. Soybeans, mesquite, and maize produced the most successful match between the model simulations and PhenoCam data out of the eight species simulated. This study represents, to our knowledge, the first independent evaluation of the ALMANAC process-based plant growth model with imagery in agroecosystems available from the PhenoCam network. The results show how PhenoCam data can make a valuable contribution to parameter determination in process-based models, making these models much more realistic. The impact of these conclusions can be increased as the methods are further refined.