1. In both basic and applied studies, quantification of herbivory on foliage is a key metric in characterizing plant-herbivore interactions, which underpin many ecological, evolutionary, and agricultural processes. Current methods of quantifying herbivory are slow or inaccurate. We present LeafByte, a free iOS application for measuring leaf area and herbivory. LeafByte can save data automatically, read and record barcodes, handle both light and dark colored plant tissue, and be used non-destructively. 
2. We evaluate its accuracy and efficiency relative to existing herbivory assessment tools.
3. LeafByte has the same accuracy as ImageJ, the field standard, but is 50% faster. Other tools, such as BioLeaf and grid quantification, are quick and accurate, but limited in the information they can provide. Visual estimation is quickest, but it only provides a coarse measure of leaf damage and tends to overestimate herbivory.
4. LeafByte is a quick and accurate means of measuring leaf area and herbivory, making it a useful tool for research in fields such as ecology, entomology, agronomy, and plant science.

Methods for Testing LeafByte

Accuracy

Accuracy_Masterfile_.csv

To confirm the accuracy of ImageJ and LeafByte, we used both methods to measure artificial "leaves" of known area. We printed out 16 black rectangles of known area with white "holes" of known size and analyzed them with both LeafByte and ImageJ, compared their results to the known area.

Precision

Leafbyte_compare_users.csv

We tested if different researchers analyzing the same leaves got the same results. To create herbivory, we excised thirty Solanum tuberosum leaves and allowed a single first instar Leptinotarsa decemlineata larvaeto feed for 24 hours. Three independent researchers measured leaf area and herbivory using LeafByte.

Comparisons of different methods

LeafByte Test.csv

We collected 67 leaves from 14 plant species (Supporting Information 3) from the Cornell Botanical Garden and grounds. Leaves were selected to represent a range of morphologies and were categorized by shape and margin type. If the leaf was undamaged, we created simple and complex artificial herbivory using hole punches and razor blades to remove 0-50% of the leaf. We recorded whether the leaf was damaged on the margin (n=36) or only internally (n=22). Herbivory was estimated visually and using grid quantification with 2mm2 grid paper (Coley 1983). For visual estimation, herbivory was estimated to the nearest 5%. Leaves with 0-2.5% herbivory were rounded to 5%. The leaves were then flattened between a sheet of printer paper with the scale printed on it and a Premium Matte Film Shield Screen Protector (J&D, Middleton, MA) and photographed. Each photograph was analyzed using LeafByte, BioLeaf, and ImageJ by at least two different researchers per method. LeafByte and ImageJ provided total leaf area, absolute herbivory, and percent herbivory. BioLeaf and visual quantification provided only percent herbivory, and the grid method provided only total herbivory. We also recorded the time it took to analyze each leaf and record the data. For ImageJ, we did not include the time it took to photograph and upload the pictures. 

Statistics

All statistics were performed using R, Version 3.5.2 (R Core Team, 2018). We built global mixed effects models using the nlme package (Pinheiro et al., 2018). We dropped non-significant predictors from the models in a backwards stepwise fashion, assessed pairwise differences between the methods using emmeans (Lenth, R., 2019), and adjusted for multiple comparisons using false discovery rate. 

Accuracy 

To test for differences in measurement accuracy between ImageJ and LeafByte, we ran linear mixed effects models with area and herbivory as response variables. Method was included as a fixed effect, and the known size of each artificial leaf was set as the reference value. Additionally, we used an equivalency test (TOSTER, Lakens 2017) to evaluate whether the methods produced the same results (as opposed to linear models that test for differences). We used ¼ of the standard deviation as upper and lower bounds of the model.

Precision

Because data were non-normally distributed, we used a Kruskul Wallis test to assess the effect of individual users on estimates of leaf area and leaf area consumed.

Comparisons of different methods

To analyze the effect of method on leaf area, we ran a linear mixed effects model with leaf area as the response variable and the interaction between method and leaf shape as predictor variables. Species and leaf ID were included as random effects in all models. Leaf areas were log transformed to meet assumptions of homoscedasticity. 

To analyze the effect of method on herbivory, we ran a linear mixed effects model with herbivory as the response variable and the interaction between method and number of holes, and between method and presence of leaf margin herbivory as predictor variables. To analyze the effect of method on percent area consumed data, we ran a binomial generalized linear mixed effects model with herbivory as a response variable and the interaction between method and number of holes and the interaction between method and presence of leaf margin herbivory as fixed effects. Because low levels of herbivory (0-2.5%) were rounded to 5% rather than 0% when using visual quantification, we analyzed both the full data set and data where percent herbivory was greater than 5% to ensure that rounding did not skew our results.

Comparing different LeafByte scale types

leafbyte_Line_v-dot_(2).csv

Initially, we used a straight line of known length as the scale (Fig. 4A), similar to how ImageJ is often used. Testing and observations found that holding a phone at an angle (which happens commonly) leads to skewed results. To fix this problem, we designed a system with 4 dots in a square acting as the scale. This allows LeafByte to identify and correct skew. To compare the line and dots scale methods and validate that our skew-correction resulted in more accurate output, we photographed 46 leaves (Table 1) at 0, 15, and 30 degree angles as determined by the mobile level application ‘Bubble level for iPhone’ (version 3.04). The leaves were photographed and analyzed using LeafByte (version 0.0.7) using a line as a scale, or LeafByte (version 1.0.0) using 4 dots in a square as a scale.  All statistics were performed using R, Version 3.5.2 (R Core Team, 2018). We built linear mixed effects models using the nlme package (Pinheiro et al., 2018). We dropped non-significant predictors from the models in a backwards stepwise fashion, assessed pairwise differences between the methods using emmeans (Lenth, R., 2019), and adjusted for multiple comparisons using false discovery rate.

See read me file for details.