Digital Image Analysis: a Substitute for Destructive Measures in Lettuce Production Research?

Assessments of plant growth and biomass are ubiquitous in commercial and non-commercial crop production. Such assessments are followed by major decisions (e.g., harvest timing) that channel resources and influence commercial profit potential. In research, the materials and methods required for assessment affect other aspects of experimentation and, therefore, discovery. Destructive harvests are important because they influence treatment selection, replicate number and size, and the resources required for data collection. Destructive sampling also diminishes the opportunities for truly repeated measures. Our familiarity with pre-harvest, in-field cabbage yield prediction and associations between direct, instrumented and human assessments of lettuce leaves led us to ask if digital imagery could be employed to greater benefit in lettuce production and research. Specifically, could image acquisition and analysis complement or substitute for standard, destructive measures of fresh biomass (yield)? If yes, under what conditions? Field and greenhouse plantings of red- and green-leaved lettuce direct-seeded (2700 seed/m²) into 0.61m x 0.61m x 0.15m raised beds and 0.3m x 0.3m x 0.05m plastic trays were established at the OARDC in Wooster, OH over four seasons in 2008 and 2009. Seedling images (.jpg format) were captured on days 7-30 after sowing at approximately 0.61m above the crop canopies at variable times during the photoperiod using hand-held and tripod-mounted digital cameras focused on a 0.3m x 0.3m grid within the plot. Live plant samples representative of the digital image were also collected by hand and processed, recording tissue weight and leaf area. A total of 384 field and 168 greenhouse images were taken Oct. 2008-Nov. 2009 and analyzed using preset and user-defined settings in WinCam^TM (Regent Instruments, Canada) software. These settings and the reference grid in the image allowed the software to calculate the area of the image covered by crop canopy. Differentiating leaves and rooting medium background was least reliable in images containing red leaves and providing sufficient contrast between the background and the leaf material was a consistent hurdle. Nevertheless, Pearson correlation coefficients (r) between computer-generated and direct measures of leaf area in the greenhouse study were 0.93-0.96 10-14 days after sowing. And, in field studies, r values for correlations between direct measures of plant biomass and WinCam^TM-derived estimates of leaf area were 0.76-0.94 (P<0.0001). We conclude that digital image analysis may be useful in real-time, non-destructive assessments of early-stage lettuce canopy development, particularly in settings dominated by green leaves.