If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
CAS Key Laboratory for Plant Diversity and Biogeography of East Asia, Kunming Institute of Botany, Chinese Academy of Sciences, 132 Lanhei Road, 650201, Kunming, Yunnan, China
CAS Key Laboratory for Plant Diversity and Biogeography of East Asia, Kunming Institute of Botany, Chinese Academy of Sciences, 132 Lanhei Road, 650201, Kunming, Yunnan, China
An under-explored area is commercial harvesting, which in animals can exert a strong selection pressure on various traits, sometimes greater even than natural selection or other human activities.
Natural populations of plants that are used by humans have likely also suffered strong pressure from harvesting, yet the potential for evolutionary change induced by humans has received surprisingly little attention.
Here, we show that the leaf coloration of a herb used in traditional Chinese medicine (Fritillaria delavayi) varies among populations, with leaves matching their local backgrounds most closely. The degree of background matching correlates with estimates of harvest pressure, with plants being more cryptic in heavily collected populations. In a human search experiment, the time it took participants to find plants was greatly influenced by target concealment. These results point to humans as driving the evolution of camouflage in populations of this species through commercial harvesting, changing the phenotype of wild plants in an unexpected and dramatic way.
In the last decade, camouflage through background matching has been verified as a defensive strategy in a number of plants, functioning to reduce herbivory,
Fritillaria delavayi is a perennial herb distributed in the alpine screes from the Hengduan mountains. It has leaves only at a young age and produces a single flower per year after the fifth year. Adult plants flower in summer (June) and die away in winter (October) annually. Leaf color of F. delavayi varies among populations from gray to brown, to green (Figures 1A–1D). Gray or brown types appear well camouflaged, while green individuals are conspicuous. Yet, after investigating all the accessible populations in NW Yunnan in the past five years, we found few herbivory marks on F. delavayi and cannot identify any natural enemies of these plants. We are also unaware of any study reporting herbivory. However, the bulb of this plant (“Lu Bei”) is an important source of Chinese traditional medicine “Chuan Bei Mu.” These wild herbs have been used for more than 2,000 years. The price of F. delavayi bulbs has increased in recent years, reaching 3,200 CNY (ca. 480 USD) per kilogram (Data S1). The mean dry weight of a single bulb is ca. 0.28 g, and over 3,500 individuals are required to harvest just 1 kg of bulbs. Thus, harvest pressure on Fritillaria is high.
Figure 1Plant Color Variation of Fritillaria delavayi among Populations
Color divergence and local adaptation for camouflage provide evidence for differential selection among populations. We measured leaf and rock colors in eight populations from SW China (locations and sample sizes are shown in Table S1). We found significant color divergence among populations in CIE L∗a∗b∗ color space, a widely used vision model designed for human color and luminance (lightness) vision (Figure S1; MANOVA on coordinates L∗, a∗, and b∗, Pillai’s trace = 1.76, F = 29.60, p < 0.001). Leaf differences among populations on color alone are significant (Figure 1A; MANOVA on coordinates a∗ and b∗, Pillai’s trace = 1.54, F7,147 = 69.53, p < 0.001), with much greater divergence in camouflaged populations (ML, PY, and PJ) than for green populations (SK, TB, and YL). Divergence was also found for luminance alone, but the effects are not as strong as those of color (Figure S2A). Populations ML, PY, and PJ are well camouflaged in either the chromatic (a∗ and b∗) or the luminance (L∗) dimensions. In these populations, leaf colors matched their native rock backgrounds better than they matched alien backgrounds (Figure S2; one-way ANOVAs, p < 0.01), showing that the current color divergence of Fritillaria delavayi is not random, but a result of population-specific selection.
We investigated the association between background matching (based on color distances between leaves and local backgrounds) and potential harvest pressure intensity. Harvest pressure was estimated with two measures: collection intensity and collection difficulty. To estimate collection intensity experienced by a population, the total collected amount of F. delavayi (dry weight from 2014 to 2019; Data S1) reported for each population was divided by the relative abundance (mean plant number in plot) of F. delavayi in the corresponding population. We found a significant negative relationship between color distance and collection intensity (Figure 2B; collection intensity was sqrt-transformed, Spearman rho correlation, r = −0.836, df = 137, p < 0.001), indicating that plants are better camouflaged in populations with heavier harvesting.
Figure 2Plant Color Variation of Fritillaria delavayi among Populations and Its Correlation with Collection Intensity, Collection Difficulty, and Human Search Time
(A) Color divergence from eight populations in human CIE L∗a∗b∗ color space (see also Figures S1 and S2).
(B) Correlation between plant camouflage (match to the background) and collection intensity (harvest amount divided by plant abundance) (p < 0.001).
(C) Correlation between plant camouflage and estimates of collection difficulty (time spent collecting a single bulb, in seconds) (p < 0.001).
(D) Capture time (milliseconds) decreasing with increasing salience value (an image-based measure of how much a target deviates from the background, without unit) in online visual selection experiment. Note trichromats spend less time than simulated dichromats finding plants (see also Table S1 and Data S1).
Collection difficulty was estimated by the time (in seconds) spent digging out a single bulb using a tool, which depends mainly on the bulb depth and the rocky substrate structure, with both factors varying among populations. Bulbs deep under tightly stacked big rocks take longer to dig out and are less heavily collected. We found a significant positive relationship between color distance and collection difficulty (Spearman rho correlation, r = 0.678, df = 138, p < 0.001). As predicted, plants in populations that are easier to collect are better camouflaged (Figure 2C). An exception is population LJ (Yulong Mt., Lijiang), where the collection is not too difficult (mean collection time 37.8 s) but the plant is green. This exception can be explained by the low collection intensity in this population.
To test the prediction that improved match to the background results in longer detection times, we developed an online citizen science experiment “spot the Fritillaria” (www.plant.sensoryecology.com). Humans have long been used to test questions related to target salience using visual displays (e.g., Farmer and Taylor
). Human subjects were asked to locate a fritillary target as quickly as possible in each of 14 randomly allocated photo slides, simulating the herb collection process by collectors. They had a free choice to play as either a trichromatic or dichromatic condition, which used images with three (red, green, and blue) or two (blue-yellow) color channels, respectively. As humans are trichromats and most other mammalian herbivores are dichromats, this set up allowed us to compare the search efficiency between human and potential natural herbivores. As expected, targets with lower salience values (better camouflage) required longer times to be located (Figure 2D; Table S1). Trichromatic players spent less time locating targets than the simulated dichromatic players (3.99 ± 0.04 versus 3.29 ± 0.06 s, mean ± SE). Given the intense commercial harvest of this species, these results show that the visual phenotype of F. delavayi may greatly influence its fitness. Furthermore, humans, being predominantly trichromatic, can exert a stronger pressure than other potential dichromatic mammalian herbivores (if any exist) on color evolution.
In principle, plant camouflage could be a result of natural selection by wild herbivores, which could have been more common in the past when the frequency of human activity was low. However, herbivores are currently very rare in the area we studied. We have not observed any animal (including free-ranging yaks, the large domestic herbivores) that feeds on either the leaf or the bulb of F. delavayi in any of the eight populations. In fact, Fritillaria species are rich in alkaloid chemical defenses, which are known to be effective in deterring herbivores, such as rodents.
Ironically, it is such alkaloid compounds that have made it a medicinal herb and induced collection. More importantly, other potential natural enemies seem unlikely to have driven the present correlation between background matching and measures of harvest intensity.
In animals, selective hunting by humans is reported to result in smaller weapon (e.g., horns or antlers) size in ungulates, but such an effect has been suggested to be limited because hunted males often reproduce before they are shot.
Both young and adult fritillary are harvested, with smaller (younger) bulbs being sold for higher prices in the retail market. Our results show that the color of Fritillaria delavayi varies among populations and closely matches their local background, with the degree of such background matching closely following the local harvest pressure for this highly valued herb in Chinese traditional medicine. Our findings are consistent with harvest pressure as the selective force driving color evolution in this plant. To further confirm this, more efforts are needed to rule out potential natural herbivores. On the other hand, experimental approaches to quantify the fitness and evolution of plant color under a relaxed harvesting pressure would be valuable in the long term. The efficacy of camouflage in these plants may also be affected by the complexity of the visual environment, as this is known to strongly affect detection of concealed targets (e.g., Dimitrova and Merilaita
This and similar factors would be valuable to explore further. In the present study, the fritillary collectors do not aim to select for color directly, but their harvest activity has influenced the adaptive evolution of plant color, intensifying phenotypic divergence. An analog is Vavilovian mimicry (weed mimicry), where the resemblance of weeds to crops has been regarded as a result of unintentional selection by humans.
Given that humans have long collected animals and plants for a variety of traits, we expect there to be many other analogous examples of humans driving changes in coloration in the wild.
Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Martin Stevens ([email protected]).
Materials Availability
This study did not generate unique reagents.
Data and Code Availability
Source data for all the figures in the paper is available as a supplementary file (Data S1). The online visual science game is available at www.plant.sensoryecology.com
Experimental Model and Subject Details
All human subjects (N = 542) that played the online game as ‘plant hunter’ and performed the visual search task were informed the intention of the game before they play. The gender of subjects is not collected, as this is not relevant to the aim of our study. All subjects gave consent to take part in the trials, and for their data to be used, and were free to leave the experiment at any time. This work was conducted with the approval of the University of Exeter Biosciences ethical committee (No. eCORN000353 v2.0). For color analysis, only leaves were collected for measurement. For collection difficulty estimation, we did not take the bulbs out, but only recorded the time spent collecting, and then filled all the substrate (rocks) back.
Method Details
Materials and study locations
Fritillaria delavayi Franch. (Liliaceae) is a perennial herb that is distributed in the Himalaya-Hengduan mountains. The habitat of this plant is the alpine screes, bare or sparsely vegetated, with elevation between 3700 to 5600 m a.s.l.
Like most plants in the alpine screes, F. delavayi grows from May to September, and its above-ground parts die away when winter comes. Young individuals in the first one to three years have only one leaf, with shape being short and needle-like (in the first year, less than 1 cm wide, often folded), to ablong-ovate (in the second year, unfolded). In the following years, it produces two or more leaves but does not flower. Plants often flower after the fifth year, sometimes producing unisexual (female sterile) flowers in the first flowering season. The adult plant usually has three to five leaves, and produces only one flower and one bulb. The leaf color of this plant varies within and among populations. To human observers, plants from some populations seem to match their substrate, and thus appear to be well camouflaged, whereas other populations are perceived as green, and not matching their substrate. Interestingly, flower color also varies among populations. In the camouflaged populations, floral colors also match the background very well.
Six Fritillaria taxa (five species, one with a variety) in China are listed as sources of Chinese Traditional Medicine, collectively named “Chuan Bei Mu”
. Among them, F. delavayi is the only species that grows in the alpine scree slopes at very high elevation. It is recognized as “Lu Bei,” as a sub-category of “Chuan Bei Mu” in pharmacopoeia and the market. These listed plants are of the most intensively harvested medicinal herb in China,
making China the biggest market for “Chuan Bei Mu” and other related traditional medicine. Although not so famous as F. cirrhosa (so called “Qing Bei” as a sub-category), bulbs of F. delavayi also have a high price and are often fraudulently traded under the same name “Bulbus Fritillariae Cirrhosae”
Our study was conducted in eight locations (Table S1) in NW Yunnan and SW Sichuan provinces, ranging in elevation from 3700 to 4800 m a.s.l. These plants are exposed to strong temperature fluctuations and high UV-radiation. The screes in these locations are formed by different rock types, either limestone or shale that vary in color.
Color measurements
Color was measured based on reflectance spectra from eight populations. Twelve to 27 leaf samples (specific sample sizes were shown in Table S1) from different individuals were collected in the field for reflectance measurements. Flower color was not included in the analysis as a large proportion of plants were collected without flowers, either because they were still of young age or because the harvest season is often after flowering time. For the background, we collected 20 to 40 rocks (specific sample sizes were shown in Table S1) within a radius of 5 cm of each focal plant. All samples were kept in plastic bags and brought back to our guesthouse (where AC power was available) within two h and measured using spectrometer FLAME equipped with HD2000 UV-VIS-NIR light source (OceanOptics, FL, USA). A PTFE standard was used as a white standard. For both plants and rocks, the upper surface of the sample, which was seen by the observers, was used in measurement. Each sample was measured three times and then averaged before further analysis. Reflectance between 390 and 750 nm was used in the calculation, as the expected observers are human and other mammals.
Quantifying background matching
Color matching of plants against rock backgrounds was measured by color similarity between each target and rock, as measured by human vision models. To do this, the spectrum of each sample was converted to photon catch values equivalent to L, M and S cone responses,
CIE L∗a∗b∗ is a scaled opponent model that is specifically designed for human color vision, with L∗ correlating to perceived lightness, a∗ and b∗ approximately describing green to red and blue to yellow variations, respectively. It uses CIE XYZ tristimulus values as the input data, which can be calculated from reflectance and irradiance spectra. It also includes chromatic adaptation (“von Kries transformation”) to account for color constancy under different light conditions. Color similarity was calculated as the Euclidean distance between color loci (the locations of the color points) in a color space. A closer distance between plant and background color loci indicates better background matching. Luminance (perceived lightness) is indicated by L∗ and was analyzed separately. Cone sensitivity curves (color matching functions) were obtained from http://www.cvrl.org/cmfs.htm. D65 (daylight) was used as the irradiance condition. For the sake of comparison with other widely used metrics, we also calculated color distances in the form of Just Noticeable Differences (JNDs). The Euclidean color distance in CIE L∗a∗b∗ space is thought to be roughly equivalent to JNDs, with 2.3 L∗a∗b∗ unit being approximately equal to 1 JND.
the Excel spreadsheet implementation, available at http://www2.ece.rochester.edu/∼gsharma/ciede2000/), which uses CIEL∗a∗b∗ color coordinates as the input, with the output being JND color differences to humans. The results can be seen in Figure S2C, and are directly comparable to the results obtained in CIE lab space (cf.Figure S2B).
Estimation of harvest pressure
An ideal way to estimate harvest pressure is to investigate the proportion of harvested plants in each population over time for a long enough period that evolution could act. However, such population-specific historical data cannot be obtained. Therefore, we used two metrics that allowed us to estimate the harvest pressure acting upon a population, the collection intensity and collection difficulty. Collection intensity was calculated as:
For each population, CI is the collection intensity (a ratio without unit), which is represented by the number of collected individuals divided by the estimated number of total individuals. Wc is the total dry weight (g) of plant bulbs collected in the most recent six years (from 2014 to 2019). Ws, is the mean dry weight (g) per single bulb (individual), estimated from the bulbs purchased from these populations. N is the mean number of plants per m2 counted in the plots. S is the area (in m2) that F. delavayi distributed. Larger collection amounts (Wc / Ws) occurring in lower abundance populations (N × S) would yield higher collection intensity values. Dry weights of bulbs collected (WC) in the last six years were obtained from the local herb dealer in the village close to the population we studied. There was only one head herb dealer in the village, who gathered fritillary bulbs and other herbs from private collectors. At least in the area we studied, each location (and the adjacent regions) is managed, collected and utilized by a separate village to avoid any interest conflict. The trading record of F. delavayi was written under the local plant name “Zhimu.” Records from seven out of the eight populations were obtained. We could not obtain data on the collection amount in population in ML (Saganai Mt., Muli, SW Sichuan province), although recently dug collection potholes were seen. These data may systemically underestimate the real collection amount, as the collectors may sell the bulbs to tourists or use the bulbs for themselves as well. However, it should reflect a general variation of collection among populations. The weight of a single bulb (Ws) was measured from 100 to 200 bulbs in each population using a balance. To estimate the abundance of F. delavayi, we counted the plant number in ten 2 m × 2 m plots in about population and calculated the mean plant number per m2 (N). The plots were chosen in the very habitat that F. delavayi can be found. In an area about 300 m x 300 m, when we found an individual, the surrounding 2 m × 2 m area was set as a plot, and checked carefully. We tried to count every individual in the plot, although the very small needle-like first-year individuals in rock crevices could sometimes be overlooked. Next, we estimated the distribution area (S) occupied by F. delavayi in each population using high resolution satellite images accessed from GoogleEarth, based on previous surveys in these regions. Specifically, we used the highest resolution images in Google Earth Pro (V7.3.3.7786, 2020CNES/Airbus), and used a polygon tool to select the habitats carefully and obtained the area in m2. We did not use an automatic approach to calculate the habitat area because it is not accurate enough. The areas were selected based on the habitat of F. delavayi and accessibility. Like other alpine scree plants, the distribution of F. delavayi greatly depends on the microhabitat; i.e., it grows on screes composed of rocks that are not too large, and never on meadows, shrubs and cliffs (which can be distinguished in the satellite images).
Although collection intensity mentioned above reflects the harvest pressure at least in the recent period, this intensity might have changed over a long history. As a supplementary metric, we also estimated the collection difficulty, which may significantly influence the intention of collection and harvest pressure. Despite their commercial value, plant bulbs in alpine screes are often hard to collect. The collection difficulty varies among populations, depending on the depth of the bulb underground and the rock structure where the bulb grows, which varies among populations. For example, bulbs deep under tightly stacked big rocks are very difficult to collect. As far as we know, these factors per se do not influence the plant color phenotype. And importantly, this parameter remains unchanged through years. Collection difficulty was measured as the time spent (in seconds) collecting a single bulb by the authors, using a stop clock on a smart phone. It is possible that the experienced local collectors may spend less time to dig out the bulbs, but our estimate should reflect the general pattern of inter-population variation. Blubs from nine to 18 individuals were dug out to estimate the mean time spent for each population, and then were backfilled. Seven populations were included in estimation except population YG (Yagong Mt., Nixi), as we were prevented from revisiting this location by debris flow.
Visual selection assessments
To simulate the selection process and test the prediction that improved match to the background results in longer detection times, we developed an online citizen science visual experiment based on photographs. A field survey in a collection area might be a more straightforward approach, but we cannot control factors such as weather (which may influence the background color), light conditions, and the experience of the collectors. More importantly, we cannot quantify the background matching of collected plants without disturbing the collection process. Furthermore, to obtain enough trials in the field, we would need to encourage collection, which is not our intention for this plant, close to being endangered. In contrast, a visual online experiment is powerful enough to obtain a general conclusion of this process, one with which we can quantify the camouflage efficacy of each plant in the experiment and obtain corresponding capture times. Similar experiments using online citizen science games have proven to be a powerful way of assessing camouflage efficacy and the role of color vision in focal animals such as birds and crabs.
Photographs Images of Fritillaria plants used in the game were taken from several locations in Shangri-La (Tianbao, Shika, Hongshan) and Deqin (Pujin) NW Yunnan province during June-July 2017. A Nikon D7100 with Tamron 90 mm lens were used for taking photos at ca. 5 m away from the plant target on a non-raining day. Aperture was kept at F/10. A photo always included both the target plants and the surrounding rock backgrounds. To control for light conditions, a standard color checker (color passport, X-rite) was set in a second photograph that shared the same light conditions. Each photo only included a single focal plant, but other accompanying plants may inevitably co-occur in some photos (as a distraction, which was considered in the model). All plants were photographed from their visible viewing angle to make sure they can be seen without obstacles. The location of the target was made random, in different parts, but not on the very edge of the photo. As both young and adult plants are harvested, photos included both of them. Photos of various difficulty to locate (estimated by experience) were used in the game. In camouflage search tasks, trichromats do not always perform better than dichromats. Dichromatic vision has long been thought to have advantages in detecting camouflage objects,
therefore it is worth comparing it with trichromatic vision in the visual experiment. To simulate the scene that viewed by a general mammalian dichromat with long and shortwave cone types, all trichromatic photographs were also converted into blue-yellow dichromatic images by combing red and green channels (Y = (R + G)/2, in ImageJ, as used by Troscianko et al.
This image-based parameter estimates target camouflage (or more strictly, lack of camouflage) by taking both predator perception and the visual background into account. It combines different visual attention-relevant visual features, i.e., luminance, color, and orientation contrasts, into a single value to give holistic target conspicuousness taking into account the weight of each feature. For the focal plant in each photo, the salience value was calculated using a MATLAB code modified from Pike
(www.plant.sensoryecology.com). Subjects were expected to locate the fritillary target as soon as possible in each of the 14 photo slides. A few plant samples were shown to inform the players of what the target looked like and what the target did not look like. Subjects were asked whether they had played this game before or not (experience), and whether they would like to play as a human (trichromatic) or a yak (dichromatic). Then 14 random (out of 48) slides of fritillaria photos were shown in a random order. The subjects had up to 15 s to point out (by clicking the target as soon as they saw it) the location of the target. Click coordinates were recorded. When the target region was clicked successively, a green circle with a sound response was shown on the target location, and the capture time was recorded (in milliseconds) and displayed, before moving on to the next slide. Although various aspects of the technical set up of online games could potentially limit the level of precision with regards to measuring timing differences, most if not all of these would run at speeds substantially beyond the reaction times recorded here; any possible limitations should be limited to within or less than 10-20 ms. Even if such limitations exist, this would still afford high precision with regards to human reactions, and any lag in the various technical systems should be consistent across participants, and so could not explain the results obtained. If the player failed to find the focal plant before the time was up, a red circle was shown on the target location with a different sound, before moving on to the next slide. An average capture time was shown after the game finished. All data collected were anonymously, and no ID information was used to identify individuals.
Online experiments to test theories of camouflage and other forms of adaptive coloration have become increasingly popular in recent years and proven to be a powerful method of testing the efficacy of anti-predator defenses (e.g.,
). Our study is particularly well suited to this, given that the hypothesized observer is humans themselves. Nonetheless, there are potential limitations with online experiments,
including that the monitors of participants are uncalibrated and will vary. Nonetheless, we are confident that the approach is robust and accurate for our key questions here. Most crucially, there is no reason to expect that display variation would add any systematic bias in any direction, but instead should simply add noise. In fact, the great advantage of online experiments is in the number of participants that can be used, which work to greatly overcome any such noise effects. The other potential issue, beyond color reproduction, is that in online games it is not possible to control the environment where participants undertake the games (e.g., they may be distracted by other things happening around them), but this should also simply add noise.
Quantification and Statistical Analysis
Color divergence and background matching
Collectively, we measured reflectance spectra of 155 leaves and 240 rocks from eight populations. Each sample was measured three times to obtain an average value. To examine phenotype divergence, we used MANOVA (multivariate analysis of variance) to analyze the color divergence among populations, with coordinates of color loci (L∗, a∗ and b∗) as the dependent variables (with Pillai’s trace to generate F-test). As color and luminance are often used separately in visual tasks, we also analyzed the chromatic (using a∗ and b∗ as dependent variables, MANOVA) and luminance (L∗, one-way ANOVA) dimensions, separately.
We used the color distance between the focal plant and its background to estimate the degree of background matching. Chromatic color similarity and luminance were analyzed separately. As the real background is complex in the field, and light conditions fluctuate, the chromatic dimension should be more important for target detection. For chromatic similarity, distances between the loci (with a and b coordinates in CIEL∗a∗b∗ space) of each of the N plant and M rocks were calculated, generating N × M distance values (see specific sample sizes of each population in Table S1). These values were then averaged by plant individuals, creating the N distances for each population. For luminance (lightness), the difference in L∗ value between color loci was analyzed using the same method (in the one-dimensional condition). One-way ANOVAs were used to analyze the difference of these distances among populations. All color distance values were square-root-transformed to improve homogeneity in analyses, the original values were shown in Figures S2.
Camouflage and harvest pressure correlation
We used a Spearman rho method to test the correlation between background matching and the intensity of harvest pressure (in terms of collection intensity and collection difficulty).
Factors influencing focal plant detection using an online experiment
It is possible that the subjects failed to find the target because they were distracted from the game temporarily. Therefore, trials that failed to find the target were not included in the analysis (see also
). We noticed that sometimes players could have being using a random scatter-gun strategy to locate the target successfully instead of looking for the target visually. Such data were discarded by excluding data for slides that received clicks more than three times. The final sample size of records is 6,849 from 542 subjects (during Aug. 2018 to Nov. 2019).
We used a linear mixed model to examine whether the capture time was influenced by the following fixed factors: target conspicuousness (salience value), visual condition (trichromatic or dichromatic images), player experience (whether played the game before), screen size, distance of target from edge, distraction (whether there is a distraction, the flowers of non-focal species), and trial number (the number of slide shown to the player), with ID as a random effect. Capture time and EdgeDistance were log-transformed to improve normality. The full model contains all terms and their 2-way interactions, specified using function lmer (package lme4
The model was then simplified by AIC (Akaike information criterion) in a stepwise algorithm with the backward direction, using step function. The simplified model contained all seven fixed terms and seven 2-way interactions.
Acknowledgments
We thank Zhe Chen, Hong-Liang Chen, Ze-Min Guo, Zi-Jue Ren, and Xiang-Guang Ma for field assistant; De-You Li for advice on harvest pressure intensity estimation; Anna Hughes for help on the game and MATLAB code; and all the players that participated in the visual selection imitation game. We thank Adrian G. Dyer and two anonymous referees for valuable comments. This work was supported by the Second Tibetan Plateau Scientific Expedition and Research (STEP) program (2019QZKK0502 to H.S.), the Strategic Priority Research Program of the Chinese Academy of Sciences (XDA 20050203 to H.S.), NSFC (31971569 and 31670214 to Y.N.), Yunnan Ten Thousand Talents Plan Young & Elite Talents Project (YNWR-QNBJ-2018-183 to Y.N.) and Youth Innovation Promotion Association, CAS (2018427 to Y.N.), the Major Program of the National Natural Science Foundation of China (31590823 to H.S.), the Key Projects of the Joint Fund of the National Natural Science Foundation of China (U1802232 to H.S.), and the National Key R&D Program of China (2017YFC0505200 to H.S.).
Author Contributions
All authors conceived and designed the study and experiments. H.S. designed the strategies for plant sample collection and harvest pressure estimation. M.S. designed the strategies for color analyses and harvest pressure estimation, and N.Y. and M.S. the computer experiment. N.Y. and M.S. processed data and analyzed results. N.Y. and H.S. prepared display items. Y.N., H.S., and M.S. wrote the manuscript.
This file includes eight data sheets. The sheet ‘Spectra’ includes reflectance spectra of leaf (LF) and rock (RK) samples from eight populations. Sheet ‘ColorModelRef’ contains parameters used in color analyses. Sheet ‘CoordinatesInCIELAB’ contains coordinates of color loci in CIEL∗a∗b∗ space, calculated from spectral data. Sheet ‘Distance-ab’ contains mean chromatic color distance (between leaves and rocks) in CIEL∗a∗b∗ space in different populations. Sheet ‘Differ-L’ contains mean luminance difference (between leaves and rocks) in CIEL∗a∗b∗ space in different populations. Sheet ‘Price’ contains market price of Fritillaria delavayi bulbs from 2016 to 2020. Sheet ‘Abundance’ contains plant abundance estimated by ten plots in each population. Sheet ‘Harvest’ contains total havest weight, average weight of a single bulb and distributed area in each population. Sheet ’ Difficulty ’ contains time spend for digging out a single bulb in each population. Sheet ‘Game’ contains data used in the visual selection experiment.
31655-9&id=fx1.jpg&_x_tr_sl=en&_x_tr_tl=es&_x_tr_hl=es&_x_tr_pto=wapp){kind=link}
31655-9&id=gr1.jpg&_x_tr_sl=en&_x_tr_tl=es&_x_tr_hl=es&_x_tr_pto=wapp){kind=link}
31655-9&id=gr2.jpg&_x_tr_sl=en&_x_tr_tl=es&_x_tr_hl=es&_x_tr_pto=wapp){kind=link}