Philippa Kaur

Literature values of R
To and Q10 of leaf respirationData of RTo were read from texts, tables, and figures in all available literature (18 species; Supplementary Tables 1, 2) when measured more than once within a period of darkness in lab- and field studies where measurement temperature, To, was kept constant. The RTo-initial was defined as the initial measurement of RTo for each study/species, and further values of RTo at later points within the same night of the same study were read as well.Apparent- and inherent temperature sensitivities (Q10, Equation 1; Fig. 2b) were obtained from all available literature (ten species; Supplementary Table 2) where in the same study/species, both nocturnal values of Q10,app and of Q10,inh were obtained in response to long-term natural T-changes in the environment during the night (hours) and nocturnal values were obtained in response to short-term artificial T-changes (max 30 min), respectively.Measurements of R
To and Q10 of leaf respirationIn the field (United Kingdom, Denmark, Panama, Colombia and Brazil), RTo (µmol CO2 m−2 s−1) in 16 species (Supplementary Tables 1, 3) was measured through nocturnal periods at constant To (controlled either by block-T or leaf-T) with infra-red gas analysers (Li-Cor-6400(XT) or Li-Cor-6800, Lincoln, Nebraska, USA). Mature, attached leaves positioned in the sunlight throughout the day were chosen. Target [CO2] in the leaf cuvette was set to ambient, ranging from 390 to 410 ppm, depending on when measurements were made, and target RH = 65 ± 10%, with a flow rate of 300 µmol s–1. The RTo-initial was defined as RTo at first measurement after darkness 30 min after sunset (to conservatively avoid light-enhanced dark respiration, LEDR50,51. Leak tests were conducted prior to measurements52. The temporal resolution of measurements varied between every three minutes to once per hour for the different species. Data were subsequently binned in hourly bins.Measurements to derive Q10,inh and Q10,app were conducted in two species in a T-controlled growth cabinet and in six species in the field (Supplementary Table 2), where Q10,inh was measured in response to 10–30 min of artificial changes in T and Q10,app was calculated from measurements of RT in response to T of the environment (growth cabinet or field) at the beginning of the night and again at the end of the night (hours apart).Tree level measurements in whole-tree chambersThe night-time respiratory efflux of the entire above-ground portion (crown and bole) in large growing trees of Eucalyptus tereticornis was measured in whole-tree chambers (WTCs) in Richmond, New South Wales (Australia, (33°36ʹ40ʺS, 150°44ʹ26.5ʺE). The WTCs are large cylindrical structures topped with a cone that enclose a single tree rooted in soil (3.25 m in diameter, 9 m in height, volume of ~53 m3) and under natural sunlight, air temperature and humidity conditions. An automated system measured the net exchange of CO2 between the canopy and the atmosphere within each chamber at 15-min resolution. During the night, we used the direct measurements of CO2 evolution (measured with an infra-red gas analyser; Licor 7000, Li-Cor, Inc., Lincoln, NE)53,54 as a measure of respiration.Due to the high noise-to-signal ratio in the CO2-exchange measurements from this system when analysing the high-resolution temporal variation through each night, we chose to only analyse temporal variation in tree-RT for the nights when tree-RT-initial were amongst the top 10% of CO2-exchange signals for the entire data set. The resulting data spanned 62 nights and included hourly average measurements from three replicate chambers.Data analysis of R
To
Measurements of nocturnal leaf respiration under constant temperature conditions (RTo) were divided by the initial rate of respiration (RTo-initial) at the onset of each night. Hourly means of RTo/RTo-initial were calculated for each leaf replicate to remove measurement noise and reduce bias due to the measurement of some species at more frequent intervals throughout the night. For species with multiple leaf replicates, these hourly means of RTo/RTo-initial were then combined to create hourly averages of RTo/RTo-initial at the species level. For each species, these values were plotted as a function of time to demonstrate how RTo/RTo-initial decreases with time since the onset of darkness, from sunset until sunrise (Supplementary Fig. 1). For each species, hourly means of RTo/RTo-initial plotted as a function of time were linearised by log-transforming data and the slope of the relationship determined. To test whether the slopes of the lines differed significantly within plant functional groups (woody, non-woody), species originating from the same biome (temperate, tropical) or species measured under the same conditions (lab, field), the slopes of the lines for all species from a given functional group, biome or measurement condition were tested pairwise against each other using the slope, standard error and sample size (number of points on the x-axis) for each line and applying a 0.05 cut-off for p values after Bonferroni correction for multiple testing. 11 out of 701 comparisons came out as being significantly different, which is why within-group slope differences were considered to be overall non-significant for this analysis. t-tests were used to test whether the slopes differed between plant functional groups (tree, non-woody), species originating from different biomes (temperate, tropical) and species measured under different environmental conditions (lab, field). In these tests, the degrees of freedom varied according to the different sample sizes. Since RTo/RTo-initial plotted as a function of time always starts at 1, the intercepts do not differ between species. t-tests were performed on linearised power functions by log-transforming data in order to test potential differences between lab and field, origin of species, between woody and non-woody species and between temperate and tropical biomes. Since these functions were statistically indistinguishable in each pairing, all measurements of nocturnal leaf respiration under constant temperature conditions (n = 967 nights, 31 species) were collated into a single plot. The data were binned hourly since some studies had very few measurements on half-hourly steps. A power function was fitted with a weighting of each hourly binned value using 1/(standard error of the mean). The power function was chosen as it, better than the exponential- or linear function, can capture both sudden steep- as well as slower decrease in RTo/RTo-initial in different species. The 95% confidence interval of the power function, following the new model equation, overlaps with all the 95% confidence intervals of the hourly binned values (Fig. 1a). All data analysis, including statistical analysis and figures were performed using Python version 3.9.4.Evaluation of new equationWe performed four sets of simulations (S1-S4) using different representations of leaf and plant respiration as outlined in Supplementary Table 4. Evaluation of Equation 4 (S2; Equation 3 from Fig. 1a merged with Equation 1) in comparison with Equation 1 (S1) and Equation 5 (S4) in comparison with Equation 2 (S3), respectively, for predictions of nocturnal variation in response to natural variation in temperature, was conducted by use of independent sets of leaf level data and tree scale data. The effect of including variable nocturnal RTo is estimated as the difference between S1 and S2 and between S3 and S4, respectively.The first data set used for the evaluation consists of nine broad-leaf species for which spot measurements of leaf respiration under ambient conditions were taken at sunset and before sunrise in the field (Fig. 1b and Supplementary Fig. 2a). Of these nine species, three species (Fig. 1c) were further measured throughout the night at ambient conditions. Further, whole-tree measurements measured throughout the night at ambient conditions (Supplementary Fig. 3a–d) were also used for evaluation. Finally, comparisons of Q10,inh with Q10,app in another ten species were used to test if RTo appeared constant as assumed in Equation 1 (Supplementary Tables 2, 3 and Fig. 2b).To validate the suitability of Equation 4 and Equation 5 over equations with full temporal control, modelled respiration values were compared against observed measurements for three species at the leaf level (Supplementary Fig. 2b–d) and for Eucalyptus tereticornis at the whole-tree level using three chamber replicates and during 62 nights using hourly measurements (Supplementary Fig. 3a–d). Linear fits were applied, using ordinary least squares regressions, to plots of normalised respiration (({R}_{T}/{R}_{{T}_{0}})) predicted by the four models against the observed values. The first measurements of the night were excluded from the fits, as these were necessarily equal to unity. The standardised residuals (S) in Supplementary Figs. 2c, 3b are calculated using the equation ({S}_{i}=({R}_{{{{{{{rm{modelled}}}}}}}_{i}}/{R}_{{{{{{{rm{Modelled}}}}}}}_{0}}-{R}_{{T}_{i}}/{R}_{{T}_{0}})/sqrt{(mathop{sum }nolimits_{i}^{N}{({R}_{{{{{{{rm{modelled}}}}}}}_{i}}/{R}_{{{{{{{rm{Modelled}}}}}}}_{0}}-{R}_{{T}_{i}}/{R}_{{T}_{0}})}^{2})/{df}}), for the residual of the ith measurement, where the sum is over all measurements, df is the number of degrees of freedom, and Rmodelled are the respiration values modelled by the four equations in Supplementary Table 4.Evaluation is done by comparing observed and simulated RT/RT, initial. We evaluate the nocturnal evolution of RT/RT, initial and use (i) one-to-one line figures that include fitted regression line, R2, p value and RMSE, (ii) Taylor diagrams and (iii) use plots of standardised residuals against temperature and hours since darkness for a qualitative assessment of the simulations, to identify whether there are any model biases at specific times or temperatures. Model evaluation, statistical analysis and figures were done using python version 3.9.4.Global scale modelling of plant R and NPP
We applied the novel formulation derived in this study (Equation 4 and Equation 5) to quantify the impact of incorporating variable RTo on simulated plant R and NPP globally using the JULES land surface model32,33 following simulations outlined in Supplementary Table 4.Plant respiration in JULES and simulations for this study: The original leaf respiration representation in JULES follows either eqn 1 ({{R}_{T}={R}_{{T}_{0}}{Q}}_{10}^{(T-{T}_{0})/10}) with Q10 = 2 and To = 25 oC or Equation 1 with an additional denominator ({{R}_{T}={R}_{{T}_{0}}{Q}}_{10}^{(T-{T}_{0})/10}/leftlfloor left(1+{e}^{0.3(T-{T}_{{upp}})}right)times left(1+{e}^{0.3({T}_{{low}}-T)}right)rightrfloor) (Equation 6). For the purpose of this application, we have used Equation 1 to represent leaf respiration in standard JULES simulations. The remaining components of maintenance respiration in JULES, i.e. fine root and wood are represented as a function of leaf to root and leaf to wood nitrogen ratios and leaf respiration rates following RT (β + (Nr + Ns)/Nl) (Equation 6) with RT as leaf respiration, Nr, Ns and Nl as root, stem and leaf Nitrogen content respectively and β as a soil water factor (Equation 42 in ref. 32). This implies that any variation in leaf respiration is passed to root and wood respiration as well30,31,35. Growth respiration is estimated as a fraction (25%) of the difference between GPP and maintenance respiration (Rm) expressed as Rg = 0.25 (GPP-Rm).JULES version 5.2 was modified to simulate leaf and plant respiration using the various descriptions (Equations 1–5) outlined in the modelling protocol in Supplementary Table 4. JULES uses standard astronomical equations to calculate the times of sunrise and sunset on a given day at each grid point. We used the model leaf temperature and RT at the timestep at or immediately preceding sunset to represent Tsunset, and RT,sunset and at every timestep through the night, the time since sunset (h) was updated. We performed global simulations for the period 2000–2018 with JULES, using the global physical configuration GL8, which is an update from GL755. We used WFDEI meteorological forcing data56 available at 0.5-degree spatial resolution and 3-h temporal resolution, and disaggregated and run in JULES with a 15 min timestep. Simulations were performed using nine plant functional types (PFTs)33. To isolate the effects of the new formulation on simulated Rp and NPP from possible impacts on leaf area index (LAI) or vegetation dynamics, we prescribed vegetation phenology via seasonal LAI fields and vegetation fractional cover based on the European Space Agency’s Land Cover Climate Change Initiative (ESA LC_CCI) global vegetation distribution57, processed to the JULES nine PFTs and re-gridded to the WFDEI grid. Annual variable fields of CO2 concentrations are based on annual mean observations from Mauna Loa58. JULES was spun up using the three cycles of the 2000–2018 meteorological forcing data to equilibrate the soil moisture stores. The mean annual output of Rp and NPP over the study period (2000–2018) is computed for all simulations and the effect of the new formulation is presented as the difference between the temporal mean of simulations with and without nocturnal variation in whole plant RTo for NPP and vice versa for Rp and as percentage respect to simulations without nocturnal variation in RTo. Results are presented for grid cells where grid level NPP is >50 g m−2 yr −1 in the standard simulations to avoid excessively large % effects at very low NPP. Output from JULES was analysed and plotted using python version 2.7.16.PermitsNo permit was required in Denmark as measurements were taken in private land (of author) and public land and measurements were non-destructive. Data were collected under the Panama Department of the Environment (current name MiAmbiente) research permit under the name of Dr Kaoru Kitajima. Permit number: SE/P-16-12. Data in Brazil were collected under the minister of Environment (Ministério do Meio Ambiente—MMA), Instituto Chico Mendes de Conservação da Biodiversidade—ICMBio, Sistema de Autorização e Informação em Biodiversidade—SISBIO permit number 47080-3. No permit was required in Colombia as measurements were taken on private land, no plant samples were collected, and trees were part of an existing experiment for which one of the co-authors is the lead. No access permits were required in the UK as they were conducted on the campus of own university plus in their own private garden.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More

Nansen, C., Phillips, T. W., Parajuleeb, M. N. & Franqui, R. A. Comparison of direct and indirect sampling procedures for Plodia interpunctella in a maize storage facility. J. Stored Prod. Res. 40, 151–168 (2004).Article 

Google Scholar 
Gerken, A. R. & Campbell, J. F. Using long-term capture data to predict Trogoderma variabile Ballion and Plodia interpunctella (Hübner) population patterns. Insects 10, 93. https://doi.org/10.3390/insects10040093 (2019).Article 
PubMed Central 

Google Scholar 
Athanassiou, C. G. & Buchelos, C. T. Grain properties and insect distribution trends in silos of wheat. J. Stored Prod Res. 88, 101632 (2020).Article 

Google Scholar 
Campbell, J., Mullen, M. & Dowdy, A. Monitoring stored-product pests in food processing plants with pheromone trapping, contour mapping, and mark-recapture. J. Econ. Entomol. 95, 1089–1101 (2002).CAS 
PubMed 
Article 

Google Scholar 
Arbogast, R. T., Weaver, D. K., Kendra, P. E. & Brenner, R. J. Implications of spatial distribution of insect populations in storage ecosystems. Environ. Entomol. 27, 202–216 (1998).Article 

Google Scholar 
Brenner, R. J., Focks, D. A., Arbogast, R. T., Weaver, D. K. & Shuman, D. Practical use of spatial analysis in precision targeting for integrated pest management. Am. Entomol. 44, 79–102 (1998).Article 

Google Scholar 
Arbogast, R. T., Kendra, P. E., Mankin, R. W. & McGovern, J. E. Monitoring insect pests in retail stores by trapping and spatial analysis. J. Econ. Entomol. 93, 1531–1542 (2000).CAS 
PubMed 
Article 

Google Scholar 
Arthur, F. & Phillips, T.W. Stored-product insect pest management and control. In Food Plant Sanitation; Hui, Y.H., Bruinsma, B.L., Gorham, J.R., Nip, W.-K., Tong, P.S., Ventresca, P., Eds.; Marcel Dekker, Inc, pp. 341–348(2003).Campbell, J. F., Toews, M. D., Arthur, F. H. & Arbogast, R. T. Long-term monitoring of Tribolium castaneum in two flour mills: Seasonal patterns and impact of fumigation. J. Econ. Entomol. 103, 991–1001 (2010).PubMed 
Article 

Google Scholar 
Doud, C. W. & Phillips, T. W. Activity of Plodia interpunctella (Lepidoptera: Pyralidae) in and around flour mills. J. Econ. Entomol. 93, 1842–1847 (2000).CAS 
PubMed 
Article 

Google Scholar 
Campbell, J. & Mullen, M. Distribution and dispersal behavior of Trogoderma variabile and Plodia interpunctella outside a food processing plant. J. Econ. Entomol. 97, 1455–1464 (2004).CAS 
PubMed 
Article 

Google Scholar 
Larson, Z., Subramanyam, B. & Herrman, T. Stored-product insects associated with eight feed mills in the Midwestern United States. J. Econ. Entomol. 101, 998–1005 (2008).PubMed 
Article 

Google Scholar 
Trematerra, P., Paula, M. C., Sciarretta, A. & Lazzari, S. Spatio-temporal analysis of insect pests infesting a paddy rice storage facility. Neotrop. Entomol. 33, 469–479 (2004).Article 

Google Scholar 
Arthur, F. H., Campbell, J. F. & Toews, M. D. Distribution, abundance, and seasonal patterns of Plodia interpunctella (Hübner) in a commercial food storage facility. J. Stored Prod. Res. 53, 7–14 (2013).Article 

Google Scholar 
McKay, T., White, A. L., Starkus, L. A., Arthur, F. H. & Campbell, J. F. Seasonal patterns of stored-product insects at a rice mill. J. Econ. Entomol. 110, 1366–1376 (2017).PubMed 
Article 

Google Scholar 
Roesli, R., Subramanyam, B., Fairchild, F. J. & Behnke, K. C. Trap catches of stored-product insects before and after heat treatment in a pilot feed mill. J. Stored Prod. Res. 39, 521–540 (2003).Article 

Google Scholar 
Campbell, J., Ching’oma, G.M., Toews, M.D. & Ramaswamy, S. Spatial distribution and movement patterns of stored-product insects. In Proceedings of the 9th International Working Conference on Stored Product Protection, Campinas, Sao Paulo, Brazil, 15–18 October 2006; Lorini, I., Bacaltchuk, B., Beckel, H., Deckers, D., Sundfeld, E., Santos, J.P.D., Biagi, J.D., Celaro, J.C., Faroni, L.R.D., Bortolini, L.D.F., Eds.; Brazilian Post-harvest Association—ABRAPOS: Passo Fundo, RS, Brazil, p. 18 (2006).Trematerra, P., Gentile, P., Brunetti, A., Collins, L. & Chambers, J. Spatio-temporal analysis of trap catches of Tribolium confusum du Val in a semolina-mill, with a comparison of female and male distributions. J. Stored Prod. Res. 43, 315–322 (2007).Article 

Google Scholar 
Semeao, A. A., Campbell, J. F., Whitworth, R. J. & Sloderbeck, P. E. Influence of environmental and physical factors on capture of Tribolium castaneum (Coleoptera: Tenebrionidae) in a flour mill. J. Econ. Entomol. 105, 686–702 (2012).PubMed 
Article 

Google Scholar 
Campbell, J.F., Perez-Mendoza, J. &Weier, J. Insect Pest Management Decisions in Food Processing Facilities. In Stored Product Protection; Hagstrum, D.W., Phillips, T.W., Cuperus, G., Eds.; Kansas State University, pp. 219–232 (2012).Mohandass, S., Arthur, F. H., Zhu, K. & Throne, J. E. Biology and management of Plodia interpunctella (Lepidoptera:Pyralidae) in stored products. J. Stored Prod. Res. 43, 302–311 (2007).Article 

Google Scholar 
Hamlin, J.C., Reed, W.D. & Phillips, M.E. Biology of the Indianmeal Moth on Dried Fruits in California. USDA Technical Bulletin No. 242, (1931)Hagstrum, D.W. & Subramanyam, B. Review of Stored-Product Insect Resource. AACC International (2009).Soderstrom, T., Stoica, P. & Trulsson, E. Instrumental variable methods for closed loop systems. IFAC 10th Triennial World Congress, Munich, FRG. pp. 363–368(1987).Johnson, J. A., Valero, K. A., Hannel, M. M. & Gill, R. F. Seasonal occurrence of post harvest dried fruit insects and their parasitoids in a culled fig warehouse. J. Econ. Entomol. 93, 1380–1390 (2000).CAS 
PubMed 
Article 

Google Scholar 
Nansen, C., Subramanyam, B. & Roesli, R. Characterizing spatial distribution of trap captures of beetles in retail pet stores using SADIE® software. J. Stored Prod. Res. 40, 471–483 (2004).Article 

Google Scholar 
Phillips, T.W., Berbert, R.C. &Cuperus, G.W. Post-harvest integrated pest management. In: Francis, F.J. (Ed.), Encyclopedia of Food Science and Technology. 2nd ed. Wiley Inc., pp. 2690–2701(2000).Phillips,T.W., Cogan, P.M. & Fadamiro, H.Y. Pheromones. In: Subramanyam, B., Hagstrum, D.W. (Eds.), Alternatives to Pesticides in Stored-product IPM. Kluwer Academic Publishers, pp. 273–302 (2000).Mullen, M. A. & Dowdy, A. K. A pheromone-baited trap for monitoring the Indian meal moth, Plodia interpunctella (Hübner) (Lepidoptera: Pyralidae). J. Stored Prod. Res. 37, 231–235 (2001).PubMed 
Article 

Google Scholar 
Nansen, C. & Phillips, T. W. Ovipositional responses of the Indian meal moth, Plodia interpunctella (Hübner) (Lepidoptera: Pyralidae) to oils. Ann. Entomol. Soc. Am. 96, 524–531 (2003).Article 

Google Scholar 
Hagstrum, D. W. Using five sampling methods to measure insect distribution and abundance in bins storing wheat. J. Stored Prod. Res. 36, 253–262 (2000).CAS 
PubMed 
Article 

Google Scholar 
Athanassiou, C. G., Kavallieratos, N. G., Sciarretta, A. & Trematerra, P. Mating disruption of Ephestia kuehniella (Zeller) (Lepidoptera: Pyralidae) in a storage facility: spatio-temporal distribution changed after long-term application. J. Stored Prod. Res. 67, 1–12 (2016).Article 

Google Scholar 
Lee, W. H., Jung, J. M., Kim, J., Lee, H. & Jung, S. Analysis of the spatial distribution and dispersion of Plodia interpunctella (Lepidoptera: Pyralidae) in South Korea. J. Stored Prod. Res. 86, 101577 (2020).Article 

Google Scholar 
Gerken, A.R. & Campbell, J.F. Spatial and temporal variation in stored-product insect pest distributions and implications for pest management in processing and storage facilities. Ann. Entomol. Soc. Am. saab049(2021).Athanassiou, C. G. & Buchelos, CTh. Detection of stored-wheat beetle species and estimation of population density using unbaited probe traps and grain trier samples. Ent. Exp. et Applic. 98, 67–78 (2001).Article 

Google Scholar 
Subramanyam, B. & Hagstrum, D.W. Sampling. In: Subramanyam B. & Hagstrum D.W. (eds), Integrated Management of Insects in Stored Products. Marcel Dekker Inc., pp. 135–193 (1995).Morrison, W. R. et al. Aeration to manage insects in wheat stored in the Balkan peninsula: Computer simulations using historical weather data. Agronomy 10, 1927 (2020).Article 

Google Scholar 
Toews, M. D., Campbell, J. F. & Arthur, F. H. Temporal dynamics and response to fogging or fumigation of stored-product Coleoptera in a grain processing facility. J. Stored Prod. Res. 42, 480–498 (2006).Article 

Google Scholar 
Buckman, K. A., Campbell, J. F. & Subramanyam, B. Tribolium castaneum (Coleoptera: Tenebrionidae) associated with rice mills: Fumigation efficacy and population rebound. J. Econ. Entomol. 106, 499–512 (2013).PubMed 
Article 

Google Scholar 
Campbell, J. F., Buckman, K. A., Fields, P. G. & Subramanyam, Bh. Evaluation of structural treatment efficacy against Tribolium castaneum and Tribolium confusum (Coleoptera: Tenebrionidae) using meta-analysis of multiple studies conducted in food facilities. J. Econ. Entomol. 108, 2125–2140 (2015).CAS 
PubMed 
Article 

Google Scholar 
Levene, H. Robust tests for equality of variances. In Ingram Olkin; Harold Hotelling; et al. (eds.). Contributions to Probability and Statistics: Essays in Honor of Harold Hotelling. Stanford University Press. pp. 278–292(1960).SAS Institute. SAS/STAT 9.2 User’s guide. SAS Institute (2008).Taylor, L. R. Aggregation, variance and mean. Nature 189, 732–735 (1961).ADS 
Article 

Google Scholar 
Iwao, S. A new method of sequential sampling to classify populations according to a critical density. Res. Popln. Ecol. 16, 281–288 (1975).
Google Scholar 
Green, R. H. Measurement of non-randomness in spatial distribution. Res. Popln. Ecol. 8, 1–17 (1966).
Google Scholar 
Hillhouse, T. L. & Pitre, H. N. Comparison of sampling techniques to obtain measurements of insect populations on soybeans. J. Econ. Entomol. 67, 411–414 (1974).Article 

Google Scholar 
Cassie, R. M. Frequency distribution models in the ecology of plankton and other organisms. J. Anim. Ecol. 31, 65–92 (1962).Article 

Google Scholar 
Southwood, T. R. E. Ecological Methods, with Particular Reference to the Study of Insect Population (Chapman and Hall, 1995).
Google Scholar 
Costa, M. G., Barbosa, J. C., Yamamoto, P. T. & Leal, R. M. Spatial distribution of Diaphorina citri Kuwayama (Hemiptera: Psyllidae) in citrus orchards. Scientia Agric 67, 546–554 (2010).Article 

Google Scholar 
Patil, G. P. & Stiteler, W. M. Concepts of aggregation and their quantification: A critical review with some new result and applications. Pop. Ecol. 15, 238–254 (1974).Article 

Google Scholar 
David, F. N. & Moor, P. G. Notes on contagious distribution in plant populations. Ann. Bot. 18, 47–53 (1954).Article 

Google Scholar 
Lloyd, M. Mean crowding. J. Anim. Ecol. 36, 1–30 (1967).Article 

Google Scholar 
Southwood, T. R. E. & Henderson, P. A. Ecological Methods 3rd edn. (Blackwell Sciences, 2000).
Google Scholar 
Feng, M. G. & Nowierski, R. M. Spatial distribution and sampling plans for four species of cereal aphids (Homoptera: Aphididae) infesting spring wheat in southwestern Idaho. J. Econ. Entomol. 85, 830–837 (1992).Article 

Google Scholar  More

Structural characterization of slag samplesThe FTIR spectra of granulated blast furnace slag (Sample 1), waste slag dumped in landfill (Sample 2) and combination of both 50% granulated blast furnace slag + 50% waste slag dumped in landfill (Sample 3) are presented in Fig. 1.Figure 1FTIR spectra of slag samples.Full size imageBy analysing the spectrum (detailed figure) in the range of 700–1100 cm−1, it can be found that there are obvious absorption peaks in the spectrum of all the slag samples. The granulated blast furnace slag shows the characteristic absorption bands at 3640, 1418, 980, 944, 861, 753 and 710 cm−1. The band at 3640 cm−1 is assigned to the stretching vibration of the hydroxyl group originated from the weakly absorbed water molecules on the slag surface24. The characteristic absorption bands at 1418, 861 and 710 cm−1 are ascribed to the asymmetric stretching mode and bending mode of carbonate group, respectively and the band at 980 cm−1 are attributable to the stretching vibrations of Si–O25. The band at 944 and 752 cm−1 represent the internal vibration of [SiO4]4− and [AlO4]5− tetrahedral and comes from Si (Al)–O-antisymmetric stretching vibration26.The different vibration modes for the sample of waste slag can be observed in the FTIR spectrum. The absorption bands shown are at 1418, 873, 712, 667 and 419 cm−1. The peak at 1418 cm−1 is assigned to the asymmetric stretching mode and bending mode of carbonate group. Calcite phase is confirmed by characteristic peaks at 712 cm−1 (ʋ2 out of plane bending vibration of the CO3−2 ion) and 873 cm−1 peak (ʋ2 split in-plane bending vibrations of the CO3−2 ion27. Calcium aluminate phase is identified by characteristic peak at 419 cm−128. Peak around 667 cm−1 is described as absorption band for different M–O (metal oxide) such as Al–O, Fe–O, Mg–O etc.29.In the case of combination of both 50% granulated blast furnace slag and 50% waste slag dumped in landfill the intensity of absorption peaks is smaller in comparison with Sample 1 and Sample 2 of slag. The characteristic absorption peaks (978 and 753 cm−1) which correspond with characteristic peaks of Sample 1 are shifted compared to the Sample 1, assigned to the stretching vibrations of Si–O and to the Si (Al)–O-antisymmetric stretching vibration, respectively, can provide important evidence of chemical interaction between Sample 1 and Sample 2. The decrease of the intensity of the bands appearing at 875 and 709 cm−1 cans be attributed to overlapping the vibrations of the CO3−2 ion from calcite phase.Figure 2 presents the SEM micrographs of the slag samples (Sample 1–3). One can see the characteristic morphology- the sizes and the forms of the slag samples.Figure 2SEM images of slag samples.Full size imageAt larger magnifications it can be observed that the surface is rough and uneven, and one can notice rounded grain-like rugged formations. The slag samples display aggregated particles with average diameter of a few microns. Also, in these rounded formations it can be seen different morphologies like spheres, rods, boards specific each compound/phase from metallurgical slags.Figure 3 illustrates the EDX elemental analysis of granulated blast furnace slag (Sample 1), waste slag dumped in landfill (Sample 2) and combination of both 50% granulated blast furnace slag + 50% waste slag dumped in landfill (Sample 3).Figure 3EDX elemental map of slag samples.Full size imageOne can observe that the predominant elements in the examined area are constated in carbon, oxygen, calcium, and iron, confirming the FTIR spectra.Figure 4 shows EDX spectra of slag samples recorded on different selected punctual area, to obtain more information about the elemental composition of specific areas. For all the tested slag samples have similar elements content.Figure 4EDX spectra analysis of slag samples.Full size imageThe selected punctual areas are highlighted thus: the spheric structure are with yellow line and the structure like boards are with green line for all the analysed slag samples. In the case of Sample 1 for both structures the values of chemical elements present are similar and the silicon has a higher value at spheric structure which can be correlated with the presence of silica (SiO2). The higher content of calcium reveals that the Sample 1 is blast furnace slag dominated by calcium and silicon compositions. In the case of slag dumped in landfill (Sample 2) the content of carbon increase for both structures and some chemical elements like titanium, barium, manganese doesn`t appear in EDX spectra and the explanation for this phenomenon is that the slag was dumped in landfill for more than 30 years. One can observe for combination of both 50% granulated blast furnace slag + 50% waste slag dumped in landfill (Sample 3) that the values of all the chemical elements for both spheric and board-like structure are between the first two samples, confirming the FTIR spectra regarding chemical interaction between Sample 1.XRD patterns of the slag samples with the phases identified are shown in Fig. 5. Sample 1 show minor peaks of free CaO and MgO, which may be deleterious and cause reduction in strength. The phases and amorphous contents of the Sample 1 granulated blast furnace slag are broadly consistent with literature30. Sample 3 of slag consists of crystalline phase – Ca2Mg2SiO7, Ca2Fe2AlO5, CaCO3 and CaO as observed by the XRD analysis. In terms of the relations of phase thermal equilibrium, the compounds identified form an isomorphic series of melilites that is specific to basic metallurgical slags.Figure 5X-ray diffraction patterns of slag samples.Full size imageIn Table 1 are presented the values expressed as ppm of chemical element detected in slag samples (Sample 1, 2 and 3).Table 1 XRF analysis of the slag samples.Full size tableThe results show a large quantity of calcium in all three samples of slag. Also, the elements detected such as Fe, Al, Mg and Si are in accordance with XRD spectra.Physical–chemical characterization of soil-slag mixturesThe chemical composition of the major elements that compound the soil, soil- slag and slag samples was determined by XRF. The values expressed as ppm of chemical elements are presented in Table 2. In the case of soil sample the content of the main constituents is iron, titanium, manganese, and potentially toxic elements (PTE) such as arsenic, zinc, copper, and cobalt. For soil-slag 1 with weight ratio soil: slag (1:1) it can be observed the disappearance of the potentially toxic elements (PTE) founded in soil sample and the decrease of concentration value of zinc. When the weight ratio of slag increases at 3 (soil-slag 2 sample) the values of main component increased in accordance with values of slag sample, but in the case of soil-slag 3 sample where the weight ratio of soil is bigger (3) it can be observed the cobalt presence. Based on these XRF results we can say that take place an elimination of potentially toxic elements in contaminated soil by applying slag in a bigger proportion.Table 2 XRF analysis of the soil-slag samples.Full size tableWith the aid of a pH meter, CONSORT C 533 the important parameters of soil and slag solutions were measured as: the pH, conductivity, and the salinity, as shown in Table 3. The data presented in Table 3 suggest that the soil sampled has the pH = 5.2 corresponding to a medium acid soil, which does not sustain a high fertility and is not able to offer proper conditions for crops. Also, the pH of soil has important influence on soil fertility, decreases the availability of essential elements and the activity of soil microorganisms which can determine calcium and magnesium deficiency in plants and decreases phosphorous availability. The pH value of slag solution (12.5) corresponds as strongly basic character which is beneficial in amelioration process of acidic soils and the presence of this type of slag sustain the improving of soil characteristics, too. For the soil-slag samples the pH value increase with the increasing of the weight ratio of slag and the mixtures soil-slag obtained can be framed into the category of weakly alkaline soils.Table 3 The physical–chemical characteristics of soil and slag solutions.Full size tableThe data given in Table 4 show that the humidity of soil is bigger and decreases in soil-slag samples with adding of slag content. The values of total soil-slag porosity are between 40 and 50% and depends on the density and apparent density of the soil being influenced by the mineralogical composition, the content of organic matter and the degree of compaction and loosening of the soil, the crystalline structure of soil minerals.Table 4 The physical–chemical characteristics of soil-slag samples.Full size tableConsidering the structural and morphological characterization of the investigated slag samples we propose a recipe of blast furnace slag and of waste slag dumped in landfill in accordance with the waste directive 2008/98/EC regarding the strategic goal of EU to a complete elimination of the disposal of wastes. The slag dump of Steel Plant of Galati has an enormous quantity of unused waste slag which may be mixed with granulated blast furnace slag, to save the natural resources used as raw materials in the metallurgical technological process.The presence of Ca2+ in the composition of the slag can maintain high alkalinity in the soil for a long time in the natural environment. The alkaline pH of the soil may contribute to a decrease the available concentration of heavy metals by reducing metal mobility and bonding metals into more stable fractions. One of the objectives of this research is improving the quality of the environment by using the mixture between two different slags on agricultural lands and reintroducing them in the agricultural centre, especially in acid soils. Acidic soils are characterized by an acidic pH that has spread in recent years due to excessive fertilizers or far too aggressive work31. The production is significantly influenced, and the treatment of acid soils is usually done using a series of natural materials (lime, dolomite), the consumption being approx. 20 t/hectare depending on the acidity of the soil and the nature of the plants grown on the respective surfaces.Our research consists in improving the characteristics and qualities of the acidic soils and helping to reintroduce it into the agricultural circuit by transforming a waste into a new material friendly-environmental, the mixture of blast furnace slag and waste slag dumped in landfill. More

Countries have yet to agree to protect at least 30% of land, a crucial target proposed in the global biodiversity deal.Credit: Roberto Schmidt/AFP via Getty

A beleaguered global deal to save the environment got a financial boost last week when Germany announced that it was upping its funding for international biodiversity conservation to €1.5 billion (US$1.49 billion) a year — an increase of €0.87 billion — making it the largest national financial pledge yet to save nature. The announcement came at a 20 September meeting in New York City, where political leaders, businesses and conservation and Indigenous-rights groups came together to rally momentum and support ahead of the United Nations biodiversity summit in Montreal, Canada, in December.Conservationists welcomed the extra funding, but warned that other wealthy countries must also reach deeper into their pockets to ensure that nations agree on a new biodiversity agreement, called the Post-2020 Global Biodiversity Framework. Estimates suggest that an additional US$700 billion annually is needed to protect the environment.Concerns over insufficient financing for global biodiversity conservation have stalled negotiations and threaten to derail attempts to finalize a deal in Montreal. The forthcoming summit will be the 15th meeting of the Conference of the Parties (COP15) to the UN’s Convention on Biological Diversity.Announcing the new funds, German Chancellor Olaf Scholz said: “With this contribution, we want to send a strong signal for an ambitious outcome of the biodiversity COP-15.”Claire Blanchard, head of global advocacy at WWF, a conservation group, told Nature that the extra funding “is highly significant” and sends an important signal that rich countries are prepared to step up.But she adds: “More signals of this kind will be needed to create the environment conducive to constructive dialogue in the negotiation room.”Andrew Deutz, a specialist in biodiversity law and finance at the Nature Conservancy, a conservation group in Arlington, Virginia, says he expects further funding announcements to come in the run up to and at the COP15.Other pledgesSeveral key political leaders, including Justin Trudeau, Canada’s prime minister, echoed calls for rich nations to make urgent progress to secure the biodiversity deal. Trudeau urged countries to agree on two crucial targets proposed in the biodiversity framework, both to be met by 2030: to halt and reverse biodiversity loss, and to protect at least 30% of land and seas.The new funding was bolstered by other pledges and developments, including a promise from a partnership of some of the world’s wealthiest private philanthropic foundations and charities to add to the $5 billion they have already committed to conservation, if other countries promise more funds.The partnership — which includes the Bezos Earth Fund, an environmental fund financed by entrepreneur Jeff Bezos — has already spent around $1 billion of its promised financing over the past two years, says Cristián Samper, head of the Wildlife Conservation Society, a not-for-profit group. Samper was speaking on behalf of the partnership at the meeting in New York City.Frans Timmermans, vice-president of the European Commission, reaffirmed that Europe would double its international biodiversity funding to $1.13 billion annually — a promise originally announced in September last year. Timmermans told the meeting that the European Union would set out more details about the funding soon.Funding shortfallAlso at the meeting, a group of four countries comprising Ecuador, Gabon, the Maldives and the United Kingdom launched a joint 10-point plan to bridge the biodiversity finance gap, which is estimated at $700 billion annually.The plan sets out the financial commitments and policy reforms needed to finance biodiversity on the required scale. For example, it encourages wealthy and lower-income nations to allocate new funds for biodiversity and to quickly deliver on their existing financial pledges. It requires donor countries to ensure that funds for overseas development do no harm to biodiversity. And it asks countries to dedicate a portion of their national funding for climate change to activities that also protect and conserve nature.The plan also commits countries to ensuring that public finance is invested in ways that benefit biodiversity, and to reviewing national subsidies and redirecting those that are harmful to nature. It calls on businesses to assess and disclose commercial risks associated with biodiversity decline, and to set quantitative targets to reduce their impact on the natural world. And it encourages multilateral development banks — such as the World Bank in Washington DC — and international financial institutions to ensure that their investments benefit biodiversity, and asks that they report on their biodiversity funding in time for COP-15.So far, 15 countries, including Canada, Germany and Norway, as well as the EU have endorsed the plan.“The plan provides a clear pathway for bridging the global biodiversity finance gap. Its significance lies in the political signal it sends,” says Blanchard.António Guterres, secretary-general of the UN, urged political leaders to “act now and at scale” to secure biodiversity financing and ensure agreement on the framework. “If negotiations continue at their slow pace, we are headed to failure,” he told the meeting. More

Study site and animalsWe studied n = 64 flocks of Japanese tits in mixed deciduous-coniferous forests in Nagano and Gumma (36°17-31’N, 138°26-39’E), Japan. Although most of the birds had not been individually colour-ringed, all the experimental trials were conducted at least 400 m apart; previous observations on colour-ringed individuals showed that this distance was enough to ensure the collection of data from different individuals30. In this site, one of the major predators of small birds is the bull-headed shrike, which is often mobbed by small birds including Japanese tits.Playback stimulusTo test whether Japanese tits recognize an alert-recruitment call sequence as a single unit, we prepared four treatments: (i) one-speaker playback of alert-recruitment call sequences, (ii) two-speaker playback of alert-recruitment call sequences with alert and recruitment calls played from different speakers, (iii) one-speaker playback of recruitment-alert call sequences, (iv) two-speaker playback of recruitment-alert call sequences with recruitment and alert calls played from different speakers (Fig. 3). We created sound files for these treatments using the software program Audacity 2.1.3 (http://www.audacityteam.org). For one-speaker treatments, we composed mono sound files where call sequences were repeated onto a single channel, whereas for two-speaker treatments, we composed stereo sound files where either alert or recruitment calls were repeated onto the right or left channels, respectively. All the files contained an equal number of alert calls (30 calls) and recruitment calls (30 calls) at the same rate (one call every 3 s), resulting in 90-s of stimuli (Fig. 3), which corresponds to the range of the natural calling rate of alert-recruitment sequences during mobbing by Japanese tits10. For all stimuli, within-call-sequence intervals between alert and recruitment calls were constant (0.1 s), which is within the range of intervals of these calls in natural call sequences11,17. In contrast, between-call-sequence intervals varied from 1.50 to 1.81 (median = 1.68) due to the difference in call length, but were constant across playback stimuli within the same “block” where the four treatments were created using the same call exemplars (see below). While alert calls are composed of three distinct note types, recruitment calls are strings of the same note type that vary in repetition number. Since the repetition number can vary depending on predator type10, we conducted predator exposure experiments to Japanese tit flocks (n = 12) and recorded call sequences towards a bull-headed shrike life-like specimen. In response to a shrike specimen, tits produced alert-recruitment call sequences with a recruitment note repetition number ranging from 5 to 15. Since the interquartile range of repetition number was 6.75 to 10, we used recruitment calls with 7–10 notes as playback stimuli in this study. In consideration for the possible influence of sound editing procedure, we created all the stimuli in the same manner; we copied alert and recruitment call parts separately from recording files, and pasted them onto background noise files to produce all four types of stimuli. Playback amplitudes were constant across treatments, 70 dB at 1.0 m measured using a sound level meter (SM-325, AS ONE Corporation). Therefore, the differences between treatments only depend on whether these calls are produced as sequences from the same source and how the calls are ordered.We carefully designed experiments to control for the possibility that individual-based acoustic features in alert and recruitment calls might influence tits’ responses. First, we prepared 16 unique sets of alert and recruitment calls using either calls from the same bird (n = 8 source individuals, n = 8 unique call sets) or from two different birds (n = 16 source individuals, n = 8 unique call sets). Then, we created the four types of treatments (i.e., alert-recruitment call sequences from the same speaker, from different speakers, and in reversed order from the same speaker and from different speakers) from each of the alert-recruitment call sets, resulting in 16 blocks of playback stimuli (Supplementary Table 3). This allows us to test the possible influence of individual-based acoustic variation on receivers’ responses.We were also careful to avoid the possible influence of population-level signatures of acoustic features: we only used Japanese tits’ call sequences that had been previously recorded from the same study population. We saved the sound files in .wav format (16-bit accuracy, 48-kHz sampling rate) onto a playback device (iPhone 8, Apple Inc.). We used the default Music app (Apple Inc.) to playback the sound files.ExperimentWe (TNS and YKM) conducted experimental trials from 26 October to 4 December 2020 and during the period of 0800 and 1600 h (Japan Standard Time). We did not conduct trials under wet and windy weather conditions, since these may influence behavioural patterns of forest birds31. First, we searched for and located a flock of Japanese tits. Upon finding a flock, we fixed a taxidermic specimen of bull-headed shrike in a perching posture on the branch at 1.8 ± 0.2 m (mean ± s.d., n = 64) above the ground. Then, we placed either one or two Bluetooth speakers (SoundLink Micro, BOSE) on tree branches at 1.6 ± 0.2 m (mean ± s.d., n = 96) above the ground, and oriented them upwards to control for the possible influence of directionality. We set the distance between the shrike specimen and the speaker(s) at 5 m. For trials with two speakers, we set the distance between speakers at 10 m, mimicking the situation in which two birds are calling (Fig. 3). The shrike specimen was first covered with a black cloth and was exposed by removing the cloth just before each trial.We began playbacks when at least two Japanese tits were present within 15 m from the shrike specimen. During 90-s of playbacks, we recorded (i) whether birds approached within 2-m of the shrike specimen during the playback and (ii) whether birds exhibited wing flicking displays12,13. We counted the number of individuals within 15 m from the shrike during 90-s of playbacks and considered it as flock size. During trials, we sat on the ground at ca. 10 m from the shrike specimen to decrease the influence of the observers’ presence on bird behaviour. To account for the inter-observer reliability32, we calculated intra-class correlation coefficient (ICC; icc function in the R package irr) between us. The lowest ICC was 0.998, indicating high degree of inter-observer reliability for the two behavioural measurements. We also video-recorded the responses of tits using a digital video camera (FDR-AX60, SONY). After completion of each trial, we checked the video recording and made an on-the-spot confirmation of the exact location at which each bird made the closest approach to the shrike specimen during the 90-s of playbacks. Then, using a tape measure, we recorded the minimum approach distance of birds to the shrike specimen. Thus, our final data set consisted of the most reliable observations confirmed by two experimenters and video evidence.The order of trials was randomized within each block (n = 16 blocks), each of which is composed of a unique alert-recruitment call set but includes four treatments differing in the number of speakers and call order. Therefore, responses to all four treatments were observed under largely similar conditions. In a few trials, the first bird to approach the shrike specimen was from a heterospecific species, such as a varied tit (n = 1) or a long-tailed tit (n = 1). To account for the possibility that these birds evoke mobbing behaviour in Japanese tits, we only used the data from instances where the first individual to approach the shrike was a Japanese tit. Otherwise, we repeated the same treatment at a different site.We used 64 unique playbacks created from 16 unique sets of alert-recruitment calls for 64 trials in order to avoid pseudoreplication33. We prepared two specimens of male bull-headed shrikes and used each of them for the equal number of trials. We did not use specimens of female shrikes since females migrate from the study site in late summer and only males were observed during the study period.Statistical analysisWe analyzed the effect of playback treatments on the mobbing behaviours of Japanese tits using generalized linear mixed models in R34,35. We used the proportions of Japanese tits in flocks that (i) approached within 2-m of the shrike specimen and (ii) exhibited wing flicking displays. For the analysis of predator approach, we prepared two vectors (i.e., the number of Japanese tits that approached the shrike specimen and the number of Japanese tits that did not approach the shrike specimen). Then, we created a single response variable by binding together these two vectors using cbind function. Similarly, for the analysis of wing flicking displays, we created a single response variable by binding two vectors (i.e., the number of tits that exhibit wing flicking and the number of tits that did not exhibit wing flicking). We fitted playback treatments as a fixed term, and flock size (maximum number of Japanese tits observed during 90-s of playback) and the way of creating playback stimuli (whether the two call types were recorded from a single individual or two individuals) as covariates. We also included identity of alert-recruitment call sets that were used for creating playback stimuli (i.e., call sets from either one or two source individuals) and identity of shrike specimens as random terms. We used a binomial error distribution and logit-link function (glmer in the R package lme4) for these models. Statistical significance was calculated by log-likelihood ratio tests using anova in the R package stats. We further conducted post-hoc pairwise comparisons between treatments by using estimated marginal means (emmeans in the R package emmeans). When making pairwise comparisons, we adjusted p-values by applying a false discovery rate control for multiple testing36. All tests were two-sided and the significance level was set at α = 0.05. Exact p-values are reported when p ≥ 0.0001.Ethics statementAll protocols were approved by the ethics committee of Kyoto University, the Ministry of the Environment, and the Forestry Agency of Japan, and adhered to Guidelines for the Use of Animals of the Association for the Study of Animal Behaviour/Animal Behavior Society37.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More

Ethical statementThis research project was ethically reviewed and approved by the Ethics Committee of the Instituto Gulbenkian de Ciência (license reference: A009.2018), and by the Portuguese National Entity that regulates the use of laboratory animals (DGAV – Direção Geral de Alimentação e Veterinária (license reference: 008958). All experiments conducted on animals followed the Portuguese (Decreto-Lei n° 113/2013) and European (Directive 2010/63/EU) legislations, concerning housing, husbandry and animal welfare.Escherichia coli clonesThe ancestral invader E. coli strain expresses a Yellow Fluorescent Protein (YFP), and carries streptomycin and ampicillin resistance markers for easiness of isolation from the mouse feces [galK::amp (pZ12)::PLlacO−1-YFP, strR (rpsl150), ΔlacIZYA::scar]. An E. coli strain used for the in vivo competition experiments is isogenic to the ancestral invader but expresses a Cyan Fluorescent Protein (CFP) and carries streptomycin and chloramphenicol resistance markers [galK::chlor (pZ12)::PLlacO−1-CFP, strR (rpsl150), ΔlacIZYA::scar]. The resident E. coli lineage was isolated from the feces along time using McConkey + 0.4% lactose medium, as previously described9. All the resident clones sampled from each mouse belong to E.coli phylogenetic group B9. The invader E. coli strains (YFP and CFP) derive from the K-12 MG1655 strain (DM08) and exhibit a gat negative phenotype, gatZ::IS112. The resident E. coli clone used for the competition experiments in the mouse gut expresses a mCherry fluorescent protein and a chloramphenicol resistance marker, allowing to distinguish the invader and resident strains in the mice feces.E. coli clones were grown at 37 °C under aeration in liquid media Luria broth (LB) from SIGMA — or McConkey and LB agar plates. Media were supplemented with antibiotics streptomycin (100 µg/mL), ampicillin (100 µg/mL) or chloramphenicol (30 µg/mL) when specified.Serial plating of 1X PBS dilutions of feces in LB agar plates supplemented with the appropriate antibiotics were incubated overnight and YFP, CFP or mCherry-labeled bacterial numbers were assessed by counting the fluorescent colonies using a fluorescent stereoscope (SteREO Lumar, Carl Zeiss). The detection limit for bacterial plating was ~300 CFU/g of feces9.In vivo evolution and competition experimentsAll mice (Mus musculus) used in this study were supplied by the Rodent Facility at Instituto Gulbenkian de Ciência (IGC) and were given ad libitum access to food (Rat and Mouse No.3 Breeding (Special Diets Services) and water. Mice were kept at 20-24 °C and 40-60% humidity with a 12-h light-dark cycle. For the in vivo evolution experiment we used the gut colonization model previously established9. Briefly, mice drank water with streptomycin (5 g/L) only for 24 h before a 4 h starvation period of food and water. The animals were then inoculated by gavage with 100 µL of an E. coli bacterial suspension of ~108 colony-forming units (CFUs). Mice A2, B2, D2, E2, G2, H2 and I2 were successfully colonized with the invader E. coli, while mice C2 and F2 failed to be colonized. Six- to eight-week-old C57BL/6 J non-littermate female mice were kept in individually ventilated cages under specified pathogen-free (SPF) barrier conditions at the IGC animal facility. Fecal pellets were collected during more than one year ( >400 days) and stored in 15% glycerol at −80 °C for later analysis. In the competition experiments between the invader ancestral E. coli and evolved populations, we colonized the mice using a 1:1 ratio of each genotype, with bacterial loads being assessed and frozen on a daily basis after gavage.In vivo competition experiments in which the two modes of selection (directional and diversifying) were acting for a longer time period were performed using evolved invader E. coli populations colonizing mice D2, B2 and A2, H2. Here we used both male (n = 8) and female (n = 8) C57BL/6 J mice aged six- to eight-week-old treated with streptomycin during 3 days before gavage. E. coli populations evolving for short time periods do not allow for strong conclusions on which mode of selection is taking place. Evolved invader populations such as I2 or G2 were therefore not used for in vivo fitness assays. To assess the impact of the mouse resident E. coli in the competitive fitness of dgoR we performed one-to-one competitions between the invader ancestral and dgoR KO clones. We first homogenized the mice microbiotas by co-housing the animals during seven days. The animals (n = 6, female C57BL/6 J mice aged six- to eight-week-old) were then maintained under co-housing and given streptomycin-supplemented (5 g/L) water during seven days to break colonization resistance and eradicate their resident E. coli. At this point, the co-housed mice were removed from the antibiotic-supplemented water for two days. The following day, one group of mice was gavaged with an mCherry-expressing resident E. coli (n = 3 mice) while the other group (n = 3) was not, with all animals being individually caged from this point on and receiving normal water without antibiotic. The day after gavage, all mice were colonized with a mix (1:1) of the invader ancestral and the dgoR KO clones, and the bacterial loads were assessed and frozen on a daily basis.Microbiota analysisFecal DNA was extracted with a QIAamp DNA Stool MiniKit (Qiagen), according to the manufacturer’s instructions and with an additional step of mechanical disruption32. 16 S rRNA gene amplification and sequencing was carried out at the Gene Expression Unit from Instituto Gulbenkian de Ciência, following the service protocol. For each sample, the V4 region of the 16 S rRNA gene was amplified in triplicate, using the primer pair F515/R806, under the following PCR cycling conditions: 94 °C for 3 min, 35 cycles of 94 °C for 60 s, 50 °C for 60 s, and 72 °C for 105 s, with an extension step of 72 °C for 10 min. Samples were then pair-end sequenced on an Illumina MiSeq Benchtop Sequencer, following Illumina recommendations. Sampling for microbiota analysis was performed until the microbiota composition stabilized (~1 year after the antibiotic perturbation).QIIME2 version 2017.1133 was used to analyze the 16 S rRNA sequences by following the authors’ online tutorials (https://docs.qiime2.org/2017.11/tutorials/). Briefly, the demultiplexed sequences were filtered using the “denoise-single” command of DADA2 version 1.1434, and forward and reverse sequences were trimmed in the position in which the 25th percentile’s quality score got below 20. Diversity analysis was performed following the QIIME2 tutorial35. Beta diversity distances were calculated through Unweighted Unifrac36. For taxonomic analysis, OTU were picked by assigning operational taxonomic units at 97% similarity against the Greengenes database version 13 (Greengenes 13_8 99% OTUs (250 bp, V4 region 515 F/806 R))37.Whole-genome sequencing and analysis pipelineDNA was extracted38 from E. coli populations (mixture of  > 1000 clones) or a single clone growing in LB plates supplemented with antibiotic to avoid contamination. DNA concentration and purity were quantified using Qubit and NanoDrop, respectively. The DNA library construction and sequencing were carried out by the IGC genomics facility using the Illumina Miseq platform. Processing of raw reads and variants analysis was based on the previous work39. Briefly, sequencing adapters were removed using fastp version 0.20.040 and raw reads were trimmed bidirectionally by 4 bp window sizes across which an average base quality of 20 was required to be retained. Further retention of reads required a minimum length of 100 bps per read containing at least 50% base pairs with phred scores at or above 20. BBsplit (part of BBMap version 38.9)41 was used to remove likely contaminating reads as explained previously39. Separate reference genomes were used for the alignment of invader (K-12 (substrain MG1655; Accession Number: NC_000913.2)) and resident (Accession Number: SAMN15163749) E. coli genomes. Alignments were performed via three alignment approaches: BWA-sampe version 0.7.1742, MOSAIK version 2.743, and Breseq version 0.35.144,45. Final average alignment depths for invader and resident populations across time points equalled 302 (median = 236) and 253 (median = 235), respectively. While Breseq provides variant analysis in addition to alignment, other variant calling approaches were used to identify putative variation in the sequenced genomes, and to verify data from Breseq. A naïve pipeline39 using the mpileup utility within SAMtools version 1.946 and a custom script written in python was employed. Only reads with a minimum mapping quality of 20 were considered for analysis, and variant calling was limited to bases with call qualities of at least 30. At these positions, a minimum of 5 quality reads had to support a putative variant on both strands (with strand bias, pos. strand / neg. strand, above 0.2 or below 5) for further consideration. Finally, mutations were retained if detected in more than one of the alignment approaches, and if they reached a minimum frequency of 5% at a minimum of one time point sampled. Further simple and complex small variants were considered from freebayes version 0.9.2147 with similar thresholds, while insertion sequence movements and other mobile element activity was inferred via is mapper version 248 and panISa version 0.1.649, as well as Breseq, as previously described39. All putative variants were verified manually in IGV version 2.750,51. Raw sequencing reads were deposited in the sequence read archive under bioproject PRJNA666769. Population dynamics of lineage-specific dynamics and the resulting Muller plots were inferred manually and are meant strictly as a means of presenting the data. In order to generate these plots, mutations were sorted by frequency (descending for each time point at which the population was sampled). The largest frequency mutations were considered major lineages within which minor frequency mutations occurred. Assuming that a mutation, which arises subsequent to a preexisting mutation (an already differentiated lineage) cannot exceed the frequency of that preexisting mutation at any point, and will fluctuate in frequency with the preexisting one, we assigned mutations to the lineages within each population. While this resolved the majority of high frequency and medium frequency mutations, low-frequency mutations within the Muller plots cannot be placed with high confidence, and are only included for completeness.Prophage induction rateTo calculate the maximum prophage induction rate we grew E. coli lysogenic clones, starting with the same initial OD600 values: ~0.1 (Bioscreen C system, Oy Growth Curves Ab Ltd), with agitation at 37 °C in LB medium in the presence or absence of mitomycin C along time (5 µg/mL)9. The OD600 values were normalized by dividing the ones in the presence of mitomycin C by the ones in the absence of mitomycin C (sampling interval: 30 min). The LN of this ratios along time originates a lysis curve, where the maximum slope corresponds to the maximal prophage induction rate for each clone analyzed. We tested evolved clones from mouse A2, H2 and G2 against the ancestral clone which only carries the Nef and the KingRac prophages. We also tested clones of the resident strain that had evolved in the presence of the invader for more than 400 days (these clones were sampled from mouse A2).
E. coli growth rate, growth curves, cell aggregation, biofilm and motility capacityTo calculate the maximum bacterial growth rate, we grew E. coli lysogenic clones, starting with the same initial OD600 values: ~0.1 (Bioscreen C system, Oy Growth Curves Ab Ltd), with agitation at 37 °C in LB medium along time using reading intervals of 30 min. The LN of the OD600 values along time originates a growth curve, where the maximum slope corresponds to the maximum bacterial growth rate for each clone analyzed.To test for metabolic differences of the psuK/fruA mutation, growth curves of evolved lysogenic E. coli clones, bearing the Nef and KingRac prophages, with or without the psuK/fruA mutation were performed with the same initial OD600 value (~0.03) for each clone. The clones were grown in glucose (0.4%) minimal medium (MM9-SIGMA) with or without pseudouridine (80 μM) and absorbance values were obtained using the Bioscreen C apparatus during 12 h.Frozen stocks of E. coli clones were used to seed tubes with 5 mL of liquid LB. These were incubated overnight at 37 °C under static conditions to assess the formation of cell flocks/clumps, observable to the naked eye, in order to evaluate the formation of cell aggregates. Biofilm was tested according a previously published protocol52 and to evaluate the motility capacity we adapted the protocol from Croze and colleagues53. Briefly, overnight E. coli clonal cultures grown with agitation at 37 °C in 5 mL LB medium supplemented with streptomycin (100 ug/mL) were adjusted to the same absorbance and a 3uL volume was dropped on top of soft agar (0.25%). Plates were incubated at 37 °C and photos were taken at day 1, 2 and 5 post-inoculation to assess swarming motility phenotype.Number of E. coli generations during mouse gut colonizationTo estimate the number of generations of E. coli in the mouse gut, we used a previously described protocol to measure the fluorescent intensity of a probe specific to E. coli 23 S rRNA (as a measure of ribosomal content) that correlates with the growth rate of the bacterial cells54. We measured the number of generations of the ancestral E. coli clone while colonizing the gut of 2 mice, treated during 24 h with streptomycin (5 g/L) before gavage, during 25 days.Plasmid DNA extraction and PCR detection of ~69Kb (repA) and ~109Kb (repB) plasmidsPlasmid DNA was extracted from overnight cultures using a Plasmid Mini Kit (Qiagen), according to the manufacturer’s guidelines. Specific primers for the amplification of repA and repB genes, were used to determine the frequency of the 68935 bp (~69 Kb) and 108557 bp (~109 Kb) plasmids, respectively, in the invader E. coli population.The primers used for repA gene were:repA-Forward: 5’-CAGTCCCCTAAAGAATCGCCCC-3’ and repA-Reverse: 5’-TGACCAGGAGCGGCACAATCGC-3’.For repB the primer sequences were:repB-Forward: 5’-GTGGATAAGTCGTCCGGTGAGC-3’ and repB-Reverse: 5’-GTTCAAACAGGCGGGGATCGGC3’.PCR amplification of plasmid-specific genes was performed in 12 isolated random clones from mouse A2 at days 104 and 493. PCR reactions were performed in a total volume of 25 μL, containing 1 μL of plasmid DNA, 1X Taq polymerase buffer, 200 μM dNTPs, 0.2 μM of each primer and 1.25 U Taq polymerase. PCR reaction conditions: 95 °C for 3 min, followed by 35 cycles of 95 °C for 30 s, 65 °C for 30 s and 72 °C for 30 s, finalizing with 5 min at 72 °C. DNA was visualized on a 2% agarose gel stained with GelRed and run at 160 V for 60 min.Construction of the dgoR KO mutantP1 transduction was used to construct a ΔdgoR mutant (dgoR KO). This KO strain was created by replacing the wild-type dgoR in the invader ancestral YFP-expressing genetic background by the respective knock-out from the KEIO collection, strain JW562755, in which the dgoR sequence is replaced by a kanamycin resistance cassette. The presence of the cassette was confirmed by PCR using primers dgoK-F: GCGATGTAGCGAGCTGTC, and yidX-R: GGGAATAAACCGGCAGCC. PCR reactions were performed in a total volume of 25 μL, containing 1 μL of DNA, 1X Taq polymerase buffer, 200 μM dNTPs, 0.2 μM of each primer and 1.25 U Taq polymerase. PCR reaction conditions: 95 °C for 3 min, followed by 35 cycles of 95 °C for 30 s, 65 °C for 30 s and 72 °C for 30 s, finalizing with 5 min at 72 °C. DNA was visualized in a 2% agarose gel stained with GelRed and run at 160 V for 60 min.RNA extraction, DNAse treatment, RT-PCR and qPCRThe Qiagen RNeasy Mini Kit was used for RNA extraction. RNA concentration and quality were evaluated in the Nanodrop 2000 and by gel-electrophoresis. DNase treatment was performed with the RQ1 DNase (Promega) by adding 0.5 μl of DNase to 1 μg of RNA and 1 μl buffer in a final volume of 15 ul, followed by incubation 30 min at 37 °C. Afterwards, 1 ul of stop solution was added and incubation for 15 min at 65 °C was performed to inactivate the DNase. As a control for complete DNA digest a PCR was performed on the reactions including positive controls. Reverse transcription was performed with M-MLV RT[-H] (Promega) by mixing 1 μg of RNA with 0.5 μl random primers (Promega) and nuclease free water to a volume of 15 μl, incubation at 70 °C for 5 min and a quick cool down on ice. Afterwards the reverse transcription was accomplished by adding 5 μl of RT buffer, 0.5 μl RT enzyme and 2 μl dNTP mix, followed by incubation for 10 min at 25 °C, 50 min at 50 °C and 10 min at 70 °C. The resulting cDNA was diluted 100-fold in nuclease free water before changes in gene expression were detected using the The QuantStudio 7Flex (Applied Biosystems) with iTaq Universal SYBR Green Supermix (BioRad) and the following cycling protocol: Hold stage: 2 min at 50 °C, 10 min at 95 °C. PCR stage (40 cycles): 15 s at 95 °C, 30 s at 58 °C, 30 s at 60 °C. Melt curve stage: 15 s at 95 °C, 1 min at 50 °C then increments of 0.05 °C/s until 95 °C. Melt curve analysis was performed to verify product homogeneity. All reactions included six biological and three technical replicates for each sample. A relative quantification method of analysis with normalization against the endogenous control rrsA and employing the primer specific efficiencies was used according to the Pfaffl method (add reference). The primers used were designed with PrimerQuest (idt). The used primer sequences were: psuK – TGCGTTAGCAGCGATTGA, AATTTACGCCTGGTGGAGTAG; arcA – GATTCATGGTACGGGACAGTAG, CCGTGACAACGAAGTCGATAA; yjtD – CGCACATGGATCTGGTGATA, GGCGTGGCGTAGTAATGATA and rrsR – GTCAGCTCGTGTTGTGAAATG, CCCACCTTCCTCCAGTTTATC.Statistics and reproducibilityCorrelation between microbiota diversity measures and E. coli loads (CFU) or persistence (1-presence or 0-absence) was performed in R using the statistical package rmcorr (version 0.5.2)56 and lme4 (version 1.1-10)57, respectively. The rate of accumulation of new ISs in vivo was compared using Wilcoxon paired signed ranked test for expected and observed insertions, while the rate of selective sweeps correlation was performed using the Spearman Correlation test. Selective sweeps were taken to be mutations or HGT events that reached  > 95% frequency in the population and kept high frequency until the end of the colonization. Statistical analysis of prophage induction as well as biofilm levels was performed using the Mann-Whitney test in GraphPad Prism (version 8.4.3). A single sample T-Test was used test if the growth rate of evolved invader clones deviates from the mean of the ancestral. A Wilcoxon rank sum test with continuity correction was used to compare the relative expression levels of the evolved clones with the ancestral. P values of x0 are of order of their inverse selection coefficient (up to logarithmic corrections):$${{{{{rm{G}}}}}}left(xright)approx 1,{{{{{rm{T}}}}}}left(xright)sim frac{1}{s}.$$
(1)

Clonal interference under uniform directional selection. This mode occurs in asexual populations when adaptive mutations become frequent enough to interfere with one another59,60,61. Only a fraction of the established adaptive mutations reaches fixation; sojourn times to intermediate frequencies are set by a global coalescence rate (widetilde{sigma }) that is higher than the typical selection coefficient of individual mutations62:

$${{{{{rm{G}}}}}}left(xright) < 1,{{{{{rm{T}}}}}}left(xright)sim frac{1}{widetilde{sigma}}.$$ (2) Details of these dynamics depend on the spectrum of selection coefficients and on the overall mutation rate, which set the strength of clonal interference. For moderate interference, where a few concurrent beneficial mutations compete for fixation, we expect a roughly exponential drop of the frequency propagator, (Gleft(xright)sim {{exp }}left(-lambda xright)), reflecting the probability that a trajectory reaches frequency x without interference by a stronger competing clade. Moderate interference generates an effective neutrality for weaker beneficial mutations and at higher frequencies63. This regime has been mapped for influenza64. In the asymptotic regime of a travelling fitness wave, where many beneficial mutations are simultaneously present, the fate of a mutation is settled in the range of small frequencies; that is, at the tip of the wave65. In this regime, emergent neutrality affects the vast majority of beneficial mutations and most of the frequency regime66. Hence, the frequency propagator rapidly drops to its asymptotic value (Gleft(x=1right)ll 1.) Adaptation under diversifying selection. More complex selection scenarios involve selection within and between ecotypes, i.e., subpopulations occupying distinct ecological niches67,68. An important factor generating niches and ecotypes is the differential use of food and other environmental resources. In this mode, ecotype-specific, conditionally beneficial mutations reach intermediate frequencies after a time given by their within-ecotype selection coefficients, but fixation can be slowed down or suppressed by diversifying (negative frequency-dependent) cross-ecotype selection18, $${{{{{rm{G}}}}}}left(xright)approx 1,{{{{{rm{T}}}}}}left(xright)sim frac{1}{s},left(xlesssim ,frac{1}{2}right)$$ (3) $${{{{{rm{G}}}}}}left(xright) < 1,{{{{{rm{T}}}}}}left(xright)gg frac{1}{s}left(xto 1right).$$ (4) The details depend on the details of the eco-evolutionary model (synergistic vs. antagonistic interactions, carrying capacities, amount of resource competition vs. explicitly frequency-dependent selection). In a model with directional selection within ecotypes, conditionally beneficial mutations rapidly fix within ecotypes, but lead only to finite shifts of the ecotype frequencies. In the simplest case, the resulting dynamics of ecotype frequencies is diffusive, resulting in an effectively neutral turnover of ecotypes18. Given negative frequency-dependent selection between ecotypes, fixations become even rarer and can be completely suppressed; that is, ecotypes can become stable on the time scales of observation. The separation of time and selection scales between intra- and cross-ecotype frequency changes is expected to be a robust feature of ecotype-dependent selection: sojourn of adaptive alleles to intermediate frequencies is fast, fixation is slower and rarer. In other words, ecotype-dependent selection is characterized by two regimes of coalescence times T(x).Frequency propagators and the coalescence time spectra expected under these evolutionary modes are qualitatively sketched in Supplementary Fig. 11. For periodic sweeps under directional selection (dark green, left column), G(x) depends weakly on x and T(x) is set by rapid sweeps for all x. For clonal interference under directional selection (green, center column), G(x) decreases substantially with increasing x and T(x) becomes uniformly shorter. Under negative frequency-dependent selection (brown, right column), G(x) decreases substantially with increasing x, while T(x) substantially increases for large x and diverges in case of strong frequency-dependent selection generating stable ecotypes (dashed lines). (see Supplementary Fig. 11 for the results of simulations assuming a model of direction selection or assuming a resource competition model where ecotype formation occurs31.The ({{{{{boldsymbol{p}}}}}})-({{{{{boldsymbol{tau }}}}}}) selection testThis test is based on qualitative characteristics of the functions G(x), T(x) and does not depend on details of the evolutionary process. We evaluate G(x) and T(x) for host-specific families of frequency trajectories; sojourn times are counted from an initial frequency x0=0.01. Origination times at this frequency are inferred by backward extrapolation of the first observed trajectory segment; the reported results are robust under variations of the threshold x0 and the extrapolation procedure. We then compute two summary statistics: the probability (p) that a mutation established at an intermediate frequency xm reaches near-fixation at a frequency xf,$$p=frac{{{{{{rm{G}}}}}}({x}_{f})}{{{{{{rm{G}}}}}}({x}_{m})},$$ (5) and the corresponding fraction of sojourn times,$$tau=,frac{{{{{{rm{T}}}}}}({x}_{f})}{{{{{{rm{T}}}}}}({x}_{m})}.$$ (6) Here we use xm=0.3 and xf=0.95 to limit the uncertainties of empirical trajectories at low and high frequency; however, the selection test is robust under variation of these frequencies. We find evidence for different modes of evolution: The long-term frequency trajectories of mice B2, D2 and E2 are consistent with predominantly frequency-dependent selection (Fig. 2, Fig. 4a–c). The propagator G(x) is a strongly decreasing function of x, resulting in fixation probabilities (p) 0.6, as measured by time ratios τ  > 3.

The trajectories of mice A2, G2, and I2 show a signature of recurrent selective sweeps and clonal interference under uniform directional selection (Fig. 4a–c). The propagator G(x) is a decreasing function of x, resulting in fixation probabilities (p=0.2-0.8), depending on the strength of clonal interference. Fixation times are short, giving time ratios (tau lesssim 2).

The shorter trajectory of mouse H2 signals periodic sweeps under uniform directional selection (Fig. 3, Fig. 4a–c). The origination rate of mutations is lower than in the longer trajectories, and G(x) shows a weak decrease with (p=1.) Sojourn times T(x) are short and grow uniformly with x, resulting in a time ratio τ=2.25. This pattern is expected under directional selection in the low mutation regime: (Tleft(xright)={{log }}left[x/(1-x)right]/{s}) for individual mutations with a uniform selection coefficient s, leading to τ=2.0 for xm=0.3 and xf=0.95 (this value is marked as a dashed line in Fig. 4c).

The trajectories of non-mutator lines in the long-term in vitro evolution experiment of Good et al1, evaluated over the first 7500 generations, show an overall signal of clonal interference under uniform directional selection (Fig. 4c, Supplementary Fig. 12). The frequency propagators G(x) are strongly decreasing functions of x and sojourn times T(x) grow uniformly with x. We find (p=0.2-0.8) and (tau lesssim 2), similar to the pattern in mice A2, G2, and I2.

Code for Selection testsThe code for selection tests from the mutation frequency trajectories can be found in the Supplementary Information file.Reporting summaryFurther information on research design is available in the Nature Research Reporting Summary linked to this article. More

Gardner, T. A., Côté, I. M., Gill, J. A., Grant, A. & Watkinson, A. R. Long-term region-wide declines in Caribbean corals. Science 301, 958–960 (2003).ADS 
PubMed 
Article 

Google Scholar 
Bruno, J. F. & Selig, E. R. Regional decline of coral cover in the Indo-Pacific: Timing, extent, and subregional comparisons. PLoS ONE 2, e711 (2007).ADS 
PubMed 
PubMed Central 
Article 

Google Scholar 
De’Ath, G., Fabricius, K. E., Sweatman, H. & Puotinen, M. The 27-year decline of coral cover on the Great Barrier Reef and its causes. Proc. Natl. Acad. Sci. U. S. A. 109, 17995–17999 (2012).ADS 
PubMed 
PubMed Central 
Article 

Google Scholar 
Hughes, T. P., Graham, N. A. J., Jackson, J. B. C., Mumby, P. J. & Steneck, R. S. Rising to the challenge of sustaining coral reef resilience. Trends Ecol. Evol. 25, 633–642 (2010).PubMed 
Article 

Google Scholar 
Pandolfi, J. M., Connolly, S. R., Marshall, D. J. & Cohen, A. L. Projecting coral reef futures under global warming and ocean acidification. Science 333, 418–422 (2011).ADS 
PubMed 
Article 

Google Scholar 
Hughes, T. P. et al. Global warming and recurrent mass bleaching of corals. Nature 543, 373–377 (2017).ADS 
PubMed 
Article 

Google Scholar 
Bindoff, N. L. et al. Chapter 5: Changing Ocean, Marine Ecosystems, and Dependent Communities. IPCC Special Report on the Ocean and Cryosphere in a Changing Climate (2019).Richmond, R. H. Reproduction and recruitment in corals: Critical links in the persistence of reefs. In Life and Death of Coral Reefs (ed. Birkeland, C. E.) 175–197 (Springer, 1997).Chapter 

Google Scholar 
Trapon, M. L., Pratchett, M. S., Hoey, A. S. & Baird, A. H. Influence of fish grazing and sedimentation on the early post-settlement survival of the tabular coral Acropora cytherea. Coral Reefs 32, 1051–1059 (2013).ADS 
Article 

Google Scholar 
Gallagher, C. & Doropoulos, C. Spatial refugia mediate juvenile coral survival during coral–predator interactions. Coral Reefs 36, 51–61 (2017).ADS 
Article 

Google Scholar 
Vermeij, M. J. A. & Sandin, S. A. Density-dependent settlement and mortality structure the earliest life phases of a coral population. Ecology 89, 1994–2004 (2008).PubMed 
Article 

Google Scholar 
Vermeij, M. J. A., Smith, J. E., Smith, C. M., Vega Thurber, R. & Sandin, S. A. Survival and settlement success of coral planulae: Independent and synergistic effects of macroalgae and microbes. Oecologia 159, 325–336 (2009).ADS 
PubMed 
Article 

Google Scholar 
Ricardo, G. F., Jones, R. J., Nordborg, M. & Negri, A. P. Settlement patterns of the coral Acropora millepora on sediment-laden surfaces. Sci. Total Environ. 609, 277–288 (2017).ADS 
PubMed 
Article 

Google Scholar 
Brunner, C. A., Uthicke, S., Ricardo, G. F., Hoogenboom, M. O. & Negri, A. P. Climate change doubles sedimentation-induced coral recruit mortality. Sci. Total Environ. 768, 143897 (2021).ADS 
PubMed 
Article 

Google Scholar 
Birrell, C. L., McCook, L. J., Willis, B. L. & Diaz-Pulido, G. A. Effects of benthic algae on the replenishment of corals and the implications for the resilience of coral reefs. In Oceanography and Marine Biology: An Annual Review 25–63 (CRC Press, 2008).Chapter 

Google Scholar 
Karcher, D. B. et al. Nitrogen eutrophication particularly promotes turf algae in coral reefs of the central Red Sea. PeerJ 2020, 1–25 (2020).
Google Scholar 
Kirschner, C. M. & Brennan, A. B. Bio-inspired antifouling strategies. Annu. Rev. Mater. Res. 42, 211–229 (2012).ADS 
Article 

Google Scholar 
Webster, N. S. et al. Metamorphosis of a scleractinian coral in response to microbial biofilms. Appl. Environ. Microbiol. 70, 1213–1221 (2004).ADS 
PubMed 
PubMed Central 
Article 

Google Scholar 
Heyward, A. J. & Negri, A. P. Natural inducers for coral larval metamorphosis. Coral Reefs 18, 273–279 (1999).Article 

Google Scholar 
Negri, A. P., Webster, N. S., Hill, R. T. & Heyward, A. J. Metamorphosis of broadcast spawning corals in response to bacteria isolated from crustose algae. Mar. Ecol. Prog. Ser. 223, 121–131 (2001).ADS 
Article 

Google Scholar 
Tebben, J. et al. Induction of larval metamorphosis of the coral Acropora millepora by tetrabromopyrrole isolated from a Pseudoalteromonas bacterium. PLoS ONE 6, 1–8 (2011).Article 

Google Scholar 
Sneed, J. M., Sharp, K. H., Ritchie, K. B. & Paul, V. J. The chemical cue tetrabromopyrrole from a biofilm bacterium induces settlement of multiple Caribbean corals. Proc. R. Soc. B Biol. Sci. 281, 1–9 (2014).
Google Scholar 
Tebben, J. et al. Chemical mediation of coral larval settlement by crustose coralline algae. Sci. Rep. 5, 1–11 (2015).Article 

Google Scholar 
Carpenter, R. C. & Edmunds, P. J. Local and regional scale recovery of Diadema promotes recruitment of scleractinian corals. Ecol. Lett. 9, 268–277 (2006).Article 

Google Scholar 
Box, S. J. & Mumby, P. J. Effect of macroalgal competition on growth and survival of juvenile Caribbean corals. Mar. Ecol. Prog. Ser. 342, 139–149 (2007).ADS 
Article 

Google Scholar 
Linares, C., Cebrian, E. & Coma, R. Effects of turf algae on recruitment and juvenile survival of gorgonian corals. Mar. Ecol. Prog. Ser. 452, 81–88 (2012).ADS 
Article 

Google Scholar 
McCook, L. J., Jompa, J. & Diaz-Pulido, G. Competition between corals and algae on coral reefs: A review of evidence and mechanisms. Coral Reefs 19, 400–417 (2001).ADS 
Article 

Google Scholar 
Nugues, M. M., Smith, G. W., Van Hooidonk, R. J., Seabra, M. I. & Bak, R. P. M. Algal contact as a trigger for coral disease. Ecol. Lett. 7, 919–923 (2004).Article 

Google Scholar 
Fong, J. et al. Allelopathic effects of macroalgae on Pocillopora acuta coral larvae. Mar. Environ. Res. 151, 104745. https://doi.org/10.1016/j.marenvres.2019.06.007 (2019).Article 
PubMed 

Google Scholar 
Hauri, C., Fabricius, K. E., Schaffelke, B. & Humphrey, C. Chemical and physical environmental conditions underneath mat- and canopy-forming macroalgae, and their effects on understorey corals. PLoS ONE 5, 1–9 (2010).Article 

Google Scholar 
Bay, L. K. et al. Reef Restoration and Adaptation Program : Intervention Technical Summary. A report provided to the Australian Government by the Reef Restoration and Adaptation Program. (2019).Anthony, K. R. N. et al. Interventions to help coral reefs under global change—A complex decision challenge. PLoS ONE 15, 1–14 (2020).Article 

Google Scholar 
Vardi, T. et al. Six priorities to advance the science and practice of coral reef restoration worldwide. Restor. Ecol. 29, 1–7 (2021).Article 

Google Scholar 
Heyward, A. J., Rees, M. & Smith, L. D. Coral spawning slicks harnessed for large-scale coral culture. Progr. Abstr. Int. Conf. Sci. Asp. Coral Reef Assess. Monit. Restor. 104, 188–189 (1999).
Google Scholar 
Harrison, P., Villanueva, R. & De la Cruz, D. Coral Reef Restoration using Mass Coral Larval Reseeding (Southern Cross University, 2016).
Google Scholar 
de la Cruz, D. W. & Harrison, P. L. Enhancing coral recruitment through assisted mass settlement of cultured coral larvae. PLoS ONE 15, e0242847. https://doi.org/10.1371/journal.pone.0242847 (2020).Article 

Google Scholar 
Chamberland, V. F., Snowden, S., Marhaver, K. L., Petersen, D. & Vermeij, M. J. A. The reproductive biology and early life ecology of a common Caribbean brain coral, Diploria labyrinthiformis (Scleractinia: Faviinae). Coral Reefs 36, 83–94 (2017).ADS 
Article 

Google Scholar 
Randall, C. J. et al. Sexual production of corals for reef restoration in the Anthropocene. Mar. Ecol. Prog. Ser. 635, 203–232 (2020).ADS 
Article 

Google Scholar 
Miller, M. W. et al. Settlement yields in large-scale in situ culture of Caribbean coral larvae for restoration. Restor. Ecol. https://doi.org/10.1111/rec.13512 (2021).Article 

Google Scholar 
Baria-Rodriguez, M. V., de la Cruz, D. W., Dizon, R. M., Yap, H. T. & Villanueva, R. D. Performance and cost-effectiveness of sexually produced Acropora granulosa juveniles compared with asexually generated coral fragments in restoring degraded reef areas. Aquat. Conserv. Mar. Freshw. Ecosyst. 29, 891–900 (2019).Article 

Google Scholar 
Doropoulos, C., Elzinga, J., ter Hofstede, R., van Koningsveld, M. & Babcock, R. C. Optimizing industrial-scale coral reef restoration: Comparing harvesting wild coral spawn slicks and transplanting gravid adult colonies. Restor. Ecol. 27, 758–767 (2019).Article 

Google Scholar 
Kuffner, I. B., Andersson, A. J., Jokiel, P. L., Rodgers, K. S. & MacKenzie, F. T. Decreased abundance of crustose coralline algae due to ocean acidification. Nat. Geosci. 1, 114–117 (2008).ADS 
Article 

Google Scholar 
Webster, N. S., Uthicke, S., Botté, E. S., Flores, F. & Negri, A. P. Ocean acidification reduces induction of coral settlement by crustose coralline algae. Glob. Change Biol. 19, 303–315 (2013).ADS 
Article 

Google Scholar 
Randall, C. J., Giuliano, C., Heyward, A. J. & Negri, A. P. Enhancing coral survival on deployment devices with microrefugia. Front. Mar. Sci. 8, 662263. https://doi.org/10.3389/fmars.2021.662263 (2021).Article 

Google Scholar 
Kuffner, I. B. et al. Inhibition of coral recruitment by macroalgae and cyanobacteria. Mar. Ecol. Prog. Ser. 323, 107–117 (2006).ADS 
Article 

Google Scholar 
Arnold, S. N., Steneck, R. S. & Mumby, P. J. Running the gauntlet: Inhibitory effects of algal turfs on the processes of coral recruitment. Mar. Ecol. Prog. Ser. 414, 91–105 (2010).ADS 
Article 

Google Scholar 
Speare, K. E., Duran, A., Miller, M. W. & Burkepile, D. E. Sediment associated with algal turfs inhibits the settlement of two endangered coral species. Mar. Pollut. Bull. 144, 189–195 (2019).PubMed 
Article 

Google Scholar 
Tebben, J., Guest, J. R., Sin, T. M., Steinberg, P. D. & Harder, T. Corals like it waxed: Paraffin-based antifouling technology enhances coral spat survival. PLoS ONE 9, 1–8 (2014).Article 

Google Scholar 
Almeida, E., Diamantino, T. C. & de Sousa, O. Marine paints: The particular case of antifouling paints. Prog. Org. Coat. 59, 2–20 (2007).Article 

Google Scholar 
Negri, A. P., Smith, L. D., Webster, N. S. & Heyward, A. J. Understanding ship-grounding impacts on a coral reef: Potential effects of anti-foulant paint contamination on coral recruitment. Mar. Pollut. Bull. 44, 111–117 (2002).PubMed 
Article 

Google Scholar 
Smith, L. D., Negri, A. P., Philipp, E., Webster, N. S. & Heyward, A. J. The effects of antifoulant-paint-contaminated sediments on coral recruits and branchlets. Mar. Biol. 143, 651–657 (2003).Article 

Google Scholar 
Jacobson, A. H. & Willingham, G. L. Sea-nine antifoulant: An environmentally acceptable alternative to organotin antifoulants. Sci. Total Environ. 258, 103–110 (2000).ADS 
PubMed 
Article 

Google Scholar 
Silva, V. et al. Isothiazolinone biocides: Chemistry, biological, and toxicity profiles. Molecules 25, 991. https://doi.org/10.3390/molecules25040991 (2020).Article 
PubMed Central 

Google Scholar 
da Silva, A. R., da Guerreiro, A. S., Martins, S. E. & Sandrini, J. Z. DCOIT unbalances the antioxidant defense system in juvenile and adults of the marine bivalve Amarilladesma mactroides (Mollusca: Bivalvia). Comp. Biochem. Physiol. Part C 250, 109169 (2021).
Google Scholar 
Cima, F. et al. Preliminary evaluation of the toxic effects of the antifouling biocide Sea-Nine 211TM in the soft coral Sarcophyton cf. glaucum (Octocorallia, Alcyonacea) based on PAM fluorometry and biomarkers. Mar. Environ. Res. 83, 16–22 (2013).PubMed 
Article 

Google Scholar 
Wendt, I., Backhaus, T., Blanck, H. & Arrhenius, Å. The toxicity of the three antifouling biocides DCOIT, TPBP and medetomidine to the marine pelagic copepod Acartia tonsa. Ecotoxicology 25, 871–879 (2016).PubMed 
Article 

Google Scholar 
Chen, L. et al. Identification of molecular targets for 4,5-dichloro-2-n-octyl-4-isothiazolin-3-one (DCOIT) in teleosts: New insight into mechanism of toxicity. Environ. Sci. Technol. 51, 1840–1847 (2017).ADS 
PubMed 
Article 

Google Scholar 
Martins, S. E., Fillmann, G., Lillicrap, A. & Thomas, K. V. Review: Ecotoxicity of organic and organo-metallic antifouling co-biocides and implications for environmental hazard and risk assessments in aquatic ecosystems. Biofouling 34, 34–52 (2018).PubMed 
Article 

Google Scholar 
Moon, Y. S., Kim, M., Hong, C. P., Kang, J. H. & Jung, J. H. Overlapping and unique toxic effects of three alternative antifouling biocides (Diuron, Irgarol 1051 ®, Sea-Nine 211 ® ) on non-target marine fish. Ecotoxicol. Environ. Saf. 180, 23–32 (2019).PubMed 
Article 

Google Scholar 
Su, Y. et al. Toxicity of 4,5-dichloro-2-n-octyl-4-isothiazolin-3-one (DCOIT) in the marine decapod Litopenaeus vannamei. Environ. Pollut. 251, 708–716 (2019).PubMed 
Article 

Google Scholar 
Fonseca, V. B., da Guerreiro, A. S., Vargas, M. A. & Sandrini, J. Z. Effects of DCOIT (4,5-dichloro-2-octyl-4-isothiazolin-3-one) to the haemocytes of mussels Perna perna. Comp. Biochem. Physiol Part C 232, 108737. https://doi.org/10.1016/j.cbpc.2020.108737 (2020).Article 

Google Scholar 
Ferreira, V. et al. Effects of nanostructure antifouling biocides towards a coral species in the context of global changes. Sci. Total Environ. 799, 149324 (2021).ADS 
PubMed 
Article 

Google Scholar 
de Campos, B. G. et al. A preliminary study on multi-level biomarkers response of the tropical oyster Crassostrea brasiliana to exposure to the antifouling biocide DCOIT. Mar. Pollut. Bull. 174, 112141 (2022).Article 

Google Scholar 
Maia, F. et al. Incorporation of biocides in nanocapsules for protective coatings used in maritime applications. Chem. Eng. J. 270, 150–157 (2015).Article 

Google Scholar 
Santos, J. V. N. et al. Can encapsulation of the biocide DCOIT affect the anti-fouling efficacy and toxicity on tropical bivalves?. Appl. Sci. 10, 1–12 (2020).Article 

Google Scholar 
Detty, M. R., Ciriminna, R., Bright, F. V. & Pagliaro, M. Environmentally benign sol-gel antifouling and foul-releasing coatings. Acc. Chem. Res. 47, 678–687 (2014).PubMed 
Article 

Google Scholar 
Korschelt, K., Tahir, M. N. & Tremel, W. A Step into the future: Applications of nanoparticle enzyme mimics. Chemistry 24, 9703–9713 (2018).PubMed 
Article 

Google Scholar 
Herget, K. et al. Haloperoxidase mimicry by CeO2-x nanorods combats biofouling. Adv. Mater. 29, 1–8 (2017).Article 

Google Scholar 
Korschelt, K. et al. CeO2-: X nanorods with intrinsic urease-like activity. Nanoscale 10, 13074–13082 (2018).PubMed 
Article 

Google Scholar 
Herget, K., Frerichs, H., Pfitzner, F., Tahir, M. N. & Tremel, W. Functional enzyme mimics for oxidative halogenation reactions that combat biofilm formation. Adv. Mater. 30, 1–28 (2018).Article 

Google Scholar 
Doropoulos, C., Ward, S., Marshell, A., Diaz-Pulido, G. & Mumby, P. J. Interactions among chronic and acute impacts on coral recruits: The importance of size-escape thresholds. Ecology 93, 2131–2138 (2012).PubMed 
Article 

Google Scholar 
Ji, Z. et al. Designed synthesis of CeO2 nanorods and nanowires for studying toxicological effects of high aspect ratio nanomaterials. ACS Nano 6, 5366–5380 (2012).PubMed 
PubMed Central 
Article 

Google Scholar 
Herget, K. et al. Supporting Information: Haloperoxidase mimicry by CeO2-x nanorods combats biofouling. Adv. Mater. 29, 1603823 (2017).Article 

Google Scholar 
Sokolova, A. et al. Spontaneous multiscale phase separation within fluorinated xerogel coatings for fouling-release surfaces. Biofouling 28, 143–157 (2012).PubMed 
Article 

Google Scholar 
Schneider, C. A., Rasband, W. S. & Eliceiri, K. W. NIH Image to ImageJ: 25 years of image analysis. Nat. Methods 9, 671–675 (2012).PubMed 
PubMed Central 
Article 

Google Scholar 
ImageJ Release Notes. https://imagej.nih.gov/ij/notes.html.Arganda-Carreras, I. et al. Trainable Weka Segmentation: A machine learning tool for microscopy pixel classification. Bioinformatics 33, 2424–2426 (2017).PubMed 
Article 

Google Scholar 
Arganda-Carreras, I. et al. Supplementary Data: Trainable Weka Segmentation: A Machine Learning Tool for Microscopy Pixel Classification: Trainable Weka Segmentation User Manualhttps://doi.org/10.1093/bioinformatics/btx180 (2017).Vyas, N., Sammons, R. L., Addison, O., Dehghani, H. & Walmsley, A. D. A quantitative method to measure biofilm removal efficiency from complex biomaterial surfaces using SEM and image analysis. Sci. Rep. 6, 2–11 (2016).Article 

Google Scholar 
Carbone, D. A., Gargano, I., Pinto, G., De Natale, A. & Pollio, A. Evaluating microalgae attachment to surfaces: A first approach towards a laboratory integrated assessment. Chem. Eng. Trans. 57, 73–78 (2017).
Google Scholar 
Moreno Osorio, J. H. et al. Early colonization stages of fabric carriers by two Chlorella strains. J. Appl. Phycol. 32, 3631–3644 (2020).Article 

Google Scholar 
Ricardo, G. F. et al. Impacts of water quality on Acropora coral settlement: The relative importance of substrate quality and light. Sci. Total Environ. 777, 146079. https://doi.org/10.1016/j.scitotenv.2021.146079 (2021).ADS 
Article 
PubMed 

Google Scholar 
Macadam, A., Nowell, C. J. & Quigley, K. Machine learning for the fast and accurate assessment of fitness in coral early life history. Remote Sens. 13, 1–17 (2021).Article 

Google Scholar 
Negri, A. P. & Heyward, A. J. Inhibition of Fertilization and Larval Metamorphosis of the Coral Acropora millepora (Ehrenberg, 1834) by Petroleum Products. Mar. Pollut. Bull. 41, 420–427 (2000).Article 

Google Scholar 
Nordborg, F. M., Flores, F., Brinkman, D. L., Agustí, S. & Negri, A. P. Phototoxic effects of two common marine fuels on the settlement success of the coral Acropora tenuis. Sci. Rep. 8, 1–12 (2018).Article 

Google Scholar 
R Core Team. R: A Language and Environment for Statistical Computing. (2021).Wickham, H. et al. Welcome to the Tidyverse. J. Open Source Softw. 4, 1686. https://doi.org/10.21105/joss.01686 (2019).ADS 
Article 

Google Scholar 
Pinheiro, J., Bates, D., Debroy, S., Sarkar, D. & R Core Team. Linear and nonlinear mixed effects models contact. Linear nonlinear Mix. Eff. Model. 3, 103–135 (2021).
Google Scholar 
Fox, J. & Weisberg, S. An R Companion to Applied Regression (Sage Publications, 2019).
Google Scholar 
Lenth, R. V. Emmeans: Estimated Marginal Means. https://cran.r-project.org/package=emmeans (2021).Brooks, M. E. et al. glmmTMB balances speed and flexibility among packages for zero-inflated generalized linear mixed modeling. R J. 9, 378–400 (2017).Article 

Google Scholar 
Dafforn, K. A., Lewis, J. A. & Johnston, E. L. Antifouling strategies: History and regulation, ecological impacts and mitigation. Mar. Pollut. Bull. 62, 453–465 (2011).PubMed 
Article 

Google Scholar 
Wu, R. et al. Room temperature synthesis of defective cerium oxide for efficient marine anti-biofouling. Adv. Compos. Hybrid Mater. https://doi.org/10.1007/s42114-021-00256-7 (2021).Article 

Google Scholar 
Hu, M. et al. Nanozymes in nanofibrous mats with haloperoxidase-like activity to combat biofouling. ACS Appl. Mater. Interfaces 10, 44722–44730 (2018).PubMed 
Article 

Google Scholar 
He, X. et al. Haloperoxidase mimicry by CeO2-x nanorods of different aspect ratios for antibacterial performance. ACS Sustain. Chem. Eng. 8, 6744–6752 (2020).Article 

Google Scholar 
Saxena, P. & Harish,. Nanoecotoxicological reports of engineered metal oxide nanoparticles on algae. Curr. Pollut. Rep. 4, 128–142 (2018).Article 

Google Scholar 
Xu, Y. et al. Effects of cerium oxide nanoparticles on bacterial growth and behaviors: Induction of biofilm formation and stress response. Environ. Sci. Pollut. Res. 26, 9293–9304 (2019).Article 

Google Scholar 
Xu, Y. et al. Mechanistic understanding of cerium oxide nanoparticle-mediated biofilm formation in Pseudomonas aeruginosa. Environ. Sci. Pollut. Res. 25, 34765–34776 (2018).Article 

Google Scholar 
Tang, Y. et al. Hybrid xerogel films as novel coatings for antifouling and fouling release. Biofouling 21, 59–71 (2005).PubMed 
Article 

Google Scholar 
Gunari, N. et al. The control of marine biofouling on xerogel surfaces with nanometer-scale topography. Biofouling 27, 137–149 (2011).PubMed 
Article 

Google Scholar 
Maia, F. et al. Silica nanocontainers for active corrosion protection. Nanoscale 4, 1287–1298 (2012).ADS 
PubMed 
Article 

Google Scholar 
Martins, R. et al. Effects of a novel anticorrosion engineered nanomaterial on the bivalve: Ruditapes philippinarum. Environ. Sci. Nano 4, 1064–1076 (2017).Article 

Google Scholar 
Gutner-Hoch, E. et al. Antimacrofouling efficacy of innovative inorganic nanomaterials loaded with booster biocides. J. Mar. Sci. Eng. 6, 15. https://doi.org/10.3390/jmse6010006 (2018).Article 

Google Scholar 
Negri, A. P. & Heyward, A. J. Inhibition of coral fertilisation and larval metamorphosis by tributyltin and copper. Mar. Environ. Res. 51, 17–27 (2001).PubMed 
Article 

Google Scholar 
Morse, D. E., Hooker, N., Morse, A. N. C. & Jensen, R. A. Control of larval metamorphosis and recruitment in sympatric agariciid corals. J. Exp. Mar. Biol. Ecol. 116, 193–217 (1988).Article 

Google Scholar 
Harrington, L., Fabricius, K., De’ath, G. & Negri, A. Recognition and selection of settlement substrata determine post-settlement survival in corals. Ecology 85, 3428–3437 (2004).Article 

Google Scholar 
Jorissen, H., Baumgartner, C., Steneck, R. S. & Nugues, M. M. Contrasting effects of crustose coralline algae from exposed and subcryptic habitats on coral recruits. Coral Reefs 39, 1767–1778 (2020).Article 

Google Scholar 
Figueiredo, J. et al. Toxicity of innovative anti-fouling nano-based solutions to marine species. Environ. Sci. Nano 6, 1418–1429 (2019).Article 

Google Scholar 
Shafir, S., Abady, S. & Rinkevich, B. Improved sustainable maintenance for mid-water coral nursery by the application of an anti-fouling agent. J. Exp. Mar. Biol. Ecol. 368, 124–128 (2009).Article 

Google Scholar  More

Experimental designA 2-year field experiment was conducted at the Modern Agricultural Research and Development Base of Henan Province (113° 35′–114° 15′ E, 34° 53′–35° 11′ N). In order to enhance the diversity of LAI data, a split-plot design with a variety of field management measures and three replications was selected for the experiment (Fig. 1). The size of each experiment plot was 40 m2, the soil texture was predominantly sandy loam and sandy clay loam, as determined by textural analysis of soil samples collected before planting. Maize cultivar Dedan-5 was used in the experiment, which was planted on June 12, 2019, and June 20, 2020, with a row spacing of 42 cm and a planting density of 7 seedlings·m−2. The soil and cultivar in field experiments were representatives of those in the region. The irrigation, pesticide, and herbicide control practices followed local management for maize production.Figure 1The experimental design.Full size imageLAI measurements and UAV-based image acquisitionThe measurements of LAI were conducted at four growth stages including the tasseling stage (TS), flowering stage (FS), grain-filling stage (GS), and milk-ripe stage (MS) of maize in 2019 and 2020, a total of 264 LAI data of maize were collected during the 2-year field trial (Table 1). In order to reduce the impact of plant variability, the random sampling method was used to collect LAI samples. For each plot, three plants were randomly selected to measure the total green leaf area with the non-destructive portable leaf area meter (Laser Area Meter CI-203; CID Inc.). And the average leaf area of selected plants represented the single plant leaf area in each experiment plot. The LAI of each plot wasTable 1 Description of samplings.Full size table$$mathrm{LAI}=mathrm{LA}*mathrm{D}$$
(1)
where (mathrm{LA}) is the leaf area of a single plant in each plot; (mathrm{D}) is the planting density in one square meter.PHANTOM 4 PRO (DJI-Innovations Inc., Shenzhen, China) is a multi-rotor UAV equipped with a 20-megapixel visible-light camera that was employed to capture digital images. Aerial observations were conducted on the same dates as the LAI measurements, which was between 10:30 a.m. and 2:00 p.m. local time when the solar zenith angle was minimal. The UAV was flown automatically based on preset flight parameters and waypoints, with a forward overlap of 80% and a side overlap of 60%. A three-axis gimbal integrated with the inertial navigation system stabilized the camera, the automatic camera mode with fixed ISO (100) and a fixed exposure was used during the flight. Altogether, 4192 images were taken in eight flights from a flight height of 29.36 m above ground, with a spatial resolution of 0.008 m.The measurements of maize LAI were carried out with permission from the Modern Agricultural Research and Development Base of Henan Province. All experiments were carried out in accordance with relevant institutional, national, and international guidelines and legislation.Image pre-processingDJI Terra (version 2.3.3) was used to generate ortho-rectified images based on the structure from motion algorithms and a mosaic blending model. The main procedures are as follows: (1) extract feature points and match features according to the longitude, latitude, elevation, roll angle, pitch angle, and heading angle of each image; (2) build dense 3D point clouds by using dense multi-view stereo matching algorithm; (3) build a 3D polygonal mesh based on the vector relationship between each point in the dense cloud; (4) establish a 3D model with both external image and internal structure by merging the mosaic image into the 3D model; (5) generate digital orthophoto map (DOM).Vegetation indices (VIs) derived from the UAV-based digital imageryDigital imagery records the intensity of visible red (R), green (G), and blue (B) bands in individual pixels24. In order to enhance the vegetation parameters contained in the digital image, fourteen commonly used RGB-based VIs were collected, and their correlation with the LAI of maize at different growth stages was evaluated. Table 2 shows the detailed information of the selected RGB-based VIs.Table 2 RGB-based VIs for LAI estimation.Full size tableCentered on the point where LAI was measured, regions of interests (ROIs) with a size of 100*100 were clipped from the digital image. Python 3.7.3 was used for extracting the R, G, B information of maize and computing the RGB-based VIs from ROIs. In order to reduce the effects of light and shadow, the R, G, B color space of the image was normalized according to the followings:$$mathrm{r}=frac{R}{R+G+B}$$
(2)
$$g=frac{G}{R+G+B}$$
(3)
$$b=frac{B}{R+G+B}$$
(4)
where r, g, and b are the normalized values. R, G, B are the pixel values from the digital images based on each band.Pearson correlation analysisBefore regression analysis, the Pearson correlation analysis was performed to determine the relationship between maize LAI and different RGB-based VIs extracted from the digital image. Pearson correlation coefficient ((mathrm{r})) reflects the degree of linear correlation between two variables, which is between − 1 and 1. The calculation formula of Pearson correlation coefficient was expressed as follows:$$mathrm{r}= frac{sum_{i=1}^{n}left({X}_{i}-overline{X }right)left({Y}_{i}-overline{Y }right)}{sqrt{sum_{i=1}^{n}{left({X}_{i}-overline{X }right)}^{2}}sqrt{sum_{i=1}^{n}{left({Y}_{i}-overline{Y }right)}^{2}}}$$
(5)
where (X), (mathrm{Y}) are variables, (n) is the number of variables.Regression methodsLinear regression (LR)Linear regression is an approach for modelling the relationship between dependent and independent variables. The case of one independent variable is called unary linear regression (ULR), the expressions can be expressed as follows:$$mathrm{y}={beta }_{0}+{beta }_{1}x+varepsilon $$
(6)
where (varepsilon ) is deviation, which satisfies the normal distribution. (x), (mathrm{y}) are variables. ({beta }_{0}), ({beta }_{1}) are the intercept and slope of the regression line, respectively.For more than one independent variable, the regression process is called multiple linear regression (MLR), the expressions can be expressed as:$$mathrm{y}={beta }_{0}+{beta }_{1}{x}_{1}+{beta }_{2}{x}_{2}+dots +{beta }_{n}{x}_{n}$$
(7)
where ({x}_{1}),( {x}_{2}), …, ({x}_{n}), (mathrm{y}) are variables, ({beta }_{0}), ({beta }_{1}), ({beta }_{2}), …, ({beta }_{n}) are coefficients that determined by least square method and gradient descent method38.The RGB-based VIs with the highest Pearson correlation coefficient was used to establish the ULR model, and VIs with a correlation coefficient higher than 0.7 were used to establish the MLR model. In each growth stage, 70% of observation data were randomly selected for establishing models, and the remaining 30% of data were used as the testing dataset to assess the model performance.Back propagation neural networks (BPNN)In this study, a three-layer BPNN model was established for LAI estimation (Fig. 2). RGB-based VIs with a correlation coefficient higher than 0.7 were selected as the input variables. Tan-Sigmoid activation function was used in the hidden layer, and the Levenberg–Marquardt algorithm was selected as the training function. The maximum epoch of BPNN training was set to 1000, the learning rate was set to 0.005, and the MSE was set to 0.001. The observation data set was split into the training set and the testing dataset with a ratio of 7:3. The training dataset was used to fit the weights and bias of the BPNN model, the testing dataset was used to evaluate the model performance. Before training, data normalization was conducted for the input and output variables, and the denormalization was required to convent the output variable back into the original units after training.Figure 2Three-layer BPNN model.Full size imageRandom forest (RF)RF is a non-parametric ensemble ML method that operates by constructing a multitude of decision trees at training time and outputting the average prediction of the individual trees (Fig. 3). The bootstrapping approach was used to collect different sub-training data from the input training dataset to construct individual decision trees.Figure 3Random forest model.Full size imageThe construction process of RF regression model is as follows:

(1)

The value of (mathrm{n}_mathrm{estimators}) was tested from 50 to 1000 in increments of 50, and the value of 500 was finally selected according to higher R2 and lower RMSE.

(2)

At each node per tree, (mathrm{m}_mathrm{try}) RGB-based VIs was randomly selected from all 14 vegetation indices, and the best split was chosen according the lowest Gini Index. (mathrm{m}_mathrm{try}) was tested from 3 to 10, and the final value was 6.

(3)

The other parameters in the RF model were kept as default values according to the (mathrm{RandomForestRegressor}) function in (mathrm{Scikit}-mathrm{learn library}).

(4)

For each tree, the data splitting process in each internal node was repeated from the root node until a pre-defined stop condition was reached.

(5)

Similar with LR and BPNN model, the RGB-based VIs with a correlation coefficient higher than 0.7 were selected as the input variables, and the output variable is LAI.

Data analysis and performance evaluationThe repeated random sampling validation method was used to evaluate the generalization performance of different models. The training and testing dataset were randomly split 500 times. For each split, the LR, BPNN, and RF models were fitted to the training dataset, and the estimation accuracy was evaluated using the testing dataset. The coefficient of determination (R2), root mean square error (RMSE), and Akaike information criterion (AIC) of the training dataset were used for the assessment of models39, and the estimation accuracy was evaluated by R2 and RMSE of the testing dataset. Mathematically, a higher R2 corresponds to a smaller RMSE, and thus represents better model performance. The procedures of LAI inversion using UAV-based digital imagery and ML methods were shown in Fig. 4.Figure 4Flowchart of LAI inversion using UAV-based remote sensing and ML methods.Full size imageThe construction and evaluation of models was performed using Python 3.7.3 in Windows 10 operating system with Intel Core i7-9700 processor, 3.00 GHz CPU, and 32 GB RAM. The processing software is Spyder. The statistical analysis and figure plotting were performed in R × 64 4.0.3. More