Ecology

Figure 2

Schematic figure of the labeling process. Participants usually upload several images in a single report. The best photo is picked by the validator who first marks the harassing or non-appropriate photos as hidden. All the non-best photos are marked as not classified. In some rare events, two or three images are annotated from the same report. The mosquito images are classified into four different categories (Aedes albopictus, Aedes aegypti, other species or can not tell) and also the confidence of the label is marked as probable or confirmed. In this paper we excluded the not classified, the hidden and the can not tell images.

Full size image

Between 2014 and 2019, 7686 citizen-made mosquito photos were labeled through Mosquito Alert by entomology experts, with labels indicating whether Ae. albopictus appear in the photos. The photos were included in reports that Mosquito Alert participants uploaded, and each report could contain several photos, see Fig. 2. The entomology experts usually labeled the best photo of the report, but sometimes they labeled two (420 times) or three (49 times) for a single report, meaning that the dataset consisted of 7168 reports. For 6699 reports, only one image was labeled by the experts; for 420 reports two were labeled; for 49 reports three were labeled. Although these reports usually contain several photos, only the ones with expert labels were used in the analysis, as cannot be assumed that all of the photos in a report would have been given the same label.
The main goals of Mosquito Alert during this 6 year period were to monitor Ae. albopictus spreading and provide early detection of Ae. aegypti in Spain. Although people participate in Mosquito Alert all over the world, the majority of the participants and the majority of the photos are in Spain (see Fig. 1). As Ae. aegypti has not been reported in Spain in recent times, most Mosquito Alert participants lived in areas where Ae. aegypti is not present, so most of the photos are of Ae. albopictus. For the detailed yearly distribution of the photos, see Table 1.
Table 1 The collected and expert validated dataset for the period 2014–2019.
Full size table

A popular deep learning model, ResNet5026 was trained and evaluated on the collected dataset with yearly cross-validation. ResNet50 was used because of its wide popularity and its proven classification power in various datasets. As presenting infinitesimal increments of the classification power is not a goal of this paper, we do not report various ImageNet state-of-the-art model performances. Yearly cross-validation was used to rule out any possibility of information leakage (possibility of a user submitting multiple reports for the same mosquito).
The trained model is not only capable of generating highly accurate predictions, but it can also ease the human annotator workload by auto-marking the images where the neural network is confident and more accurate, leaving more uncertain cases for the entomology experts. Moreover, while visualizing the erroneous predictions a few re-occurring patterns were identified, which can serve as a proposal for how to make images that can be best processed by the model.
Several aspects of the dataset were explored as follows.
Classification
Since Mosquito Alert was centered around Ae. albopictus during the relevant time period (2014–2019), the collected dataset is biased towards this species (Table 1). We explored training classifiers on the Mosquito Alert dataset alone and also tied training on a balanced dataset, where 3896 negative samples were added from the IP10227 dataset of various non-mosquito insects as negative samples. From the IP102 dataset, images similar to mosquitoes, and images of striped insects were selected. Although the presented mosquito alert dataset is filtered to contain only mosquito images, in later use, non-mosquito images might be uploaded by the citizens. Training the CNN on a combination of mosquito and non-mosquito images can improve the model to make correct predictions, classifying non-tiger mosquitoes for those cases too. For testing, in each fold, only the Mosquito Alert dataset was used.
The trained classifiers achieved an extremely high area under the receiver operating characteristic curve (ROC AUC) score of 0.96 (see Fig. 3). The fact that the ROC AUC score for each fold was always over 0.95 proves the consistency of our classifier. Inspecting the confusion matrix shows us that the model tends to make more false positive predictions (assuming tiger mosquito is defined as the positive outcome) than false negatives, resulting in high sensitivity. The augmentation of the Mosquito Alert dataset with various insects from IP102 images to make it more balanced resulted in a slight performance boost and narrowed the gap between the number of false positive and false negative samples as expected, see Table 2.
Figure 3

Left: ROC curve calculated on the prediction of the 7686 images in the Mosquito Alert dataset with yearly cross-validation. The blue line shows the case when only the Mosquito Alert dataset was used for training, the orange when the training dataset was balanced out with the addition of non-tiger mosquito insect images from the IP102 dataset. Also a zoom into the part of the ROC curve, where the two methods differ the most is highlighted. Right: the confusion matrix was calculated on the same predictions when only the Mosquito Alert dataset was used for training. For both, a positive label means tiger mosquito is present.

Full size image

Table 2 Yearly cross-validation results with using the Mosquito Alert dataset alone and its IP102 augmented version.
Full size table

How to take a good picture?
Inspection of the weaknesses of a machine learning model is a fruitful way to gain a deeper understanding of the underlying problems and mechanisms. In our case, a careful review of the mispredicted images led us to useful insights into what makes a photo hard to classify for the deep learning model. On Fig. 8, a few selected examples are presented. Unlike humans, deep learning models rely more on textures than on shapes28. As a consequence, grid-like background patterns or striped objects may easily confuse the machine classifier. A larger rich training set can help to avoid these pitfalls, but we also have the option to advise the participants. If participants avoid confusing setups when taking photos, this can improve the accuracy of the automated classification. These guidelines can be added to the Mosquito Alert application to help participants make good images of mosquitoes.

Do not use striped structure (e.g. mosquito net or fly-flap) as a background.

Avoid complex backgrounds when possible. A few examples: patterned carpet, different nets, reflecting/shiny background, bumpy wallpaper.

Use clear, white background (e.g. a sheet of plain paper is perfect if possible) or hold the mosquito with finger pads.

Make sure that as much as possible the mosquito is in focus and covers a large area of the photo.

In general, it is desirable to have a clean white background with the mosquito centered, and with the image containing as little background as possible.
Dataset size impact on model performance
Modern deep CNNs tend to generate better predictions when trained on larger datasets. In this experiment, we trained a ResNet50 model on 10–20–(cdots )–90–100% of 6686 images and evaluated the model on the remaining 1000 images. The 1000 images were selected from the same year (2019) and all of them came from reports with only one photo. There were 709 tiger mosquitoes out of the 1000 test images. ROC AUC and accuracy were calculated with a 500 round bootstrapping of the 1000 test images.
Figure 4

Training a ResNet50 model on a subsampled training dataset. The model was tested against the same 1000 test images for all the steps and statistics of the test metric was calculated with a 500 round bootstrapping. The curve proves the diversity of the Mosquito Alert dataset and also suggests that in the future when the dataset will be even larger, the classification performance will increase.

Full size image

The mean and the standard deviation of the 500 rounds are shown in Fig. 4 for each training data size. From the figure, we can conclude that the predictive power of the model increases as more data are used. The shape of the curves also suggests that the dataset did not reach its plateau. In the upcoming years, as the dataset size increases, ROC AUC and accuracy enhancement is expected.
On measuring image quality
Through the examined period, Mosquito Alert outreach was promoting a mosquito-targeted data collection strategy. Participants were expected to report two mosquito species (Ae. aegypti and Ae. albopictus). By defining these species as positive samples and all the other potential species of mosquito as negative, the submission decision by participants becomes a binary classification problem. In the majority of cases, when participants submit an image we should expect them to think of having a positive sample. Later, based on entomological expert validation, the true label for the image was obtained.
The main goal of such a surveillance system is to keep the sensitivity of the users as high as possible while keeping their specificity at an acceptable level. Therefore, measuring the sensitivity and specificity of the users would be a plausible quality measure. Unfortunately, there is no available information regarding the non-submitted mosquitoes (the true negative and false negative ones), meaning it is impossible to measure sensitivity. The specificity can be measured only in a special case, when there are no false positive images submitted by the user, resulting in a specificity of 1. Based on the latter argument, focusing on metrics derived from the ratio of the submitted tiger mosquito images vs. all submitted images is not meaningful. Instead, the quality can be measured by the usefulness of the photos from the viewpoint of the expert validator or a CNN, as presented in the next chapter.
Quality evolution of the images through time and space
The Mosquito Alert dataset is a unique collection of mosquito images, because, among other things, it is built from 5 consecutive years (not counting 2014, where less than 100 reports were submitted) and it also provides geolocation tags. This uniqueness of the dataset provides potential identification of time and spatial evolution and dependence of the citizen-based mosquito image quality. To explore such an evolution, we performed two different experiments. Geolocation tags were converted to country, region, and city-level information via the geopy Python package. It was found, that the vast majority (95% of all) of the reports were coming from Spain so we performed the analysis only for the Spanish data.
Figure 5

Number of submitted reports and the fraction of their ratio where the entomology expert annotator could tell if tiger mosquito was presented on the photo or not. The charts are shown for the four cities, where Mosquito Alert was the most popular.

Full size image

First, we explored the fraction of the photos, where the entomology expert marked “can not tell”, because the photo was not descriptive enough to decide which species were presented. Figure 5 shows the ratio of the useful mosquito reports, when mosquito decision was possible, compared to all the mosquito reports. The chart shows the above-mentioned ratio for four Spanish cities, which have the most reports submitted (the same information is showed on Supplementary Fig. S1 as a heatmap over Spain). The Mann–Kendall test on the fraction of useful reports shows p-values of 0.09, 0.09, 0.81, 0.22 for Barcelona, Valencia, Málaga, and Girona, which does not justify the presence of a significant trend in image quality, although any conclusions drawn from five data points must be handled with a pinch of salt. It does not mean anything about the individual participants’ quality progression, because Mosquito Alert is highly open and dynamic, and active participants can constantly change. Of note, through these years, the tiger mosquitoes have widely spread from the east coast to the southern and western regions of Spain29. New (and naive) citizen scientists living in the newly colonized regions have been systematically called to action and participation, thus, limiting the overall learning rate of the Mosquito Alert participants’ population. Our results suggest, that either a dynamic balance exists between naive and experienced participants over the period of data recollection, or mosquito photographing skills are independent of the user experience level. The expectation would be that as the population in Spain became more aware of the presence of tiger mosquitoes and their associated public health risks, the system should experience an increase in the useful report ratio, at least for tiger mosquitoes, and most tiger mosquito photos maybe classified automatically.
Figure 6

1000 random samples were selected for each years data. Separated ResNet50 models were trained on each of the years and each model was tested on the rest of the years data. Metrics were calculated with a 500 round random sampling with replacement from the test data. Left: mean of the 500 round bootstrapped accuracy calculations. Right: mean of the 500 round bootstrapped ROC AUC calculations.

Full size image

Second, we subsampled randomly 1000 images from all years between 2015 and 2019. Then we trained a different ResNet50 on data from the different years and generated predictions for the rest of the data, for each year separately. This way we can explore if data from any year is a “better training material” than the others. The results see Fig. 6, shows that 2015 is the worst training material, providing 0.83–0.84 ROC AUC score for the test period, while the rest (period 2016–2019) is similar, ROC AUC varies between 0.90 and 0.93. The reason why the 2015 data found to be the least favourable for training is its class imbalance, meaning that data from 2015 is extremely biased towards tiger mosquitoes (94%), so when training on 2015 data, the model does not see enough non-tiger mosquito samples, while for the other years lower class imbalance was found (70–80%), see Table 1. In general, machine learning models for classification require a substantial amount of examples for each possible class, in our case tiger and non-tiger mosquitoes, therefore worse performance is expected when training on the 2015 data.
Other than the varying class imbalance, we can conclude that the Mosquito Alert dataset quality is consistent, we did not find any concerning difference between training and testing our model for any of the 2016–2017–2018–2019 data pairs.
Pre-filtering the images before expert validation
Generating human annotations for an image classification task is a labour-intensive and expensive part of any project especially if the annotation requires expert knowledge. Therefore, having a model that generates accurate predictions for a well-defined subset of the data saves a lot of time and cost. We assume that the trained classifier is more accurate when the prediction probability is whether high or low and more inaccurate when it is close to 0.5. With this assumption in mind one can tune the (p_{low}) and (p_{high}) probabilities, in a way that images with a prediction probability (p_{low}< p < p_{high}) are discarded and sent to human validation. Figure 7 Randomly selecting 100,000 (p_{low}) and (p_{high}) thresholds on the predictions which were created via yearly cross-validation. Each time only samples were kept where the predicted probability were out of the ([p_{low};p_{high}]) interval. Each point shows the kept data fraction and the prediction accuracy. Varying the lower and upper predicted probability almost 98% of the images are correctly predicted while keeping 80% of all the images. Full size image Varying (p_{low}) and (p_{high}) provides a trade-off between prediction accuracy and the portion of images sent to human validation. Based on Fig. 7 sending 20% of the images to human validation while having an almost 98% accurate prediction for 80% of the dataset is a fruitful way to combine human labour-power and machine learning together. More

Variation in wood density
The values of Chinese fir’s wood physical properties varied considerably among different geographic sources and Tukey-HSD testing showed that some of these differences were statistically significant (Fig. 1). The maximum value (HNYX-T) of wood all-dry density (WDD) was 62.70% higher than the minimum (FJYK-P). The WDD of each source was consistent with the classification and performance indexes of conifer trees in the timber strength grade for structural use, a standard in China’s forestry industry39: FJYK-P was at level S10 ( HNZJJ-P  > FJYK-P, for which the maximum 58.0% higher than the minimum value. According to the wood grading standards in the grain compression index, HNZJJ-P and FJYK-P were at level II (29.1–44.0 MPa) and the rest of geographic sources were at level III (44.1–59.0 MPa) (Table 3).
The compression strength perpendicular to the grain of total tensile (CPG.TT) among geographic sources was ranked as follows: HNYX-T  > JXCS-R  > HNYX-P  > HNZJJ-P  > FJYK-P (Table 4). Its maximum value (HNYX-T) was 29.3% higher than the minimum (FJYK-P). The ranking for compression strength perpendicular to the grain of total radial (CPG.TR) was slightly different: HNYX-T  > JXCS-R  > HNYX-P  > HNZJJ-P  > FJYK-P, for which the maximum was 42.1% higher than the minimum value. Compression strength perpendicular to the grain of part radial (CPG.PR) had the same rank order as CPG.TT, with a maximum value (HNYX-T) 35.0% higher than the minimum (FJYK-P). Finally, compression strength perpendicular to the grain of part tensile (CPG.PT) was ranked as HNYX-T  > JXCS-R  > HNZJJ-P  > HNYX-P  > FJYK-P for the five geographic sources of Chinese fir.
Table 4 The statistical analysis of wood mechanical properties of Chinese fir.
Full size table

Factors influencing wood physical properties
Climate factors effect on wood physical properties
The influence of precipitation on the three kinds of density was consistent. Pre in January, October, November, and December was positively related to wood density, while it was negatively correlated with density in others months, especially in May (r = − 0.39), June (r = − 0.59), and August (r = − 0.64). On a seasonal scale, Pre in summer was negatively correlated with density (r = − 0.77), but it was positively correlated with autumn (r = 0.22). MaxT was positively correlated with density during the whole year, except in May (r = − 0.34), and likewise with wood density but most strongly in summer (r = 0.75). MinT was positively correlated with density, especially in Jan (r  > 0.7), though it was not significantly so in February and October (r  0.45). Pre showed no significant correlation with TSR.LD, RSR.LD, DDS.LD, and DDS.RD, whose correlation coefficients were 0.1–0.3. But Pre was negatively correlated with VSR.LD most of the year (except July, October). AveT was negatively correlation with TSR.RD, RSR.RD, and VSR.RD in January, February, March, and winter; however, AveT showed no significant correlation with DDS.RD. AveT was negatively correlated with TSR.LD, RSR.LD, DDS.LD, and VSR.LD during the whole year. In general, MinT had a significant positive relationship to TSR.RD (r = 0.47), RSR.RD (r = 0.48), and VSR.RD (r = 0.52), except in October, and it was negatively correlated with DDS.RD. MinT was positively related to RSR.LD, VSR.LD, yet negative related to DDS.LD. MaxT was negatively correlated with TSR.RD, RSR.RD, VSR.RD in January, February, May, and December, and winter. MaxT showed no significant correlation with DDS.RD, RSR.LD, DDS.LD or VSR.LD (Fig. 2c).
Pre had significant negative correlations with all of the mechanical properties in May, June, August, and summer, as evince by Fig. 2b, which also showed positive correlations in October. As we can seen, the effects of Pre on wood density and mechanical properties have the same tendency. Pre in all other months was not significantly correlated with mechanical properties (r  0.75), while it was showed no significant correlation in Feb and Oct (r  1000. Through stepwise regression modeling, 14 variables without multicollinearity were retained (i.e., MOE, MOR, TSG, CSG, CPG.TT, CPG.TR, CPG.PT, CPG.PR, DDS.RD, WDD, DDS.LD, TSR.RD, RSR.RD, VSR.LD).
PCA was applied to the above 14 selected physical variables. These results showed that the physical properties of wood loaded strongly on the first axis of the PCA, explaining 51.8% of variation in the 14 tested properties, while the second axis explained 11.0% of it. MOE, MOR, TSR.RD, RSR.RD, and VSR.LD loaded on the positive axis of PC1 and PC2. Both DDS.LD and DDS.RD loaded on the negative axis of PC1 and PC2, while TSG, CSG, CPG.TT, CPG.TR, CPG.PT, CPG.PR, and WDD loaded on the positive axis of PC1 and the negative axis of PC2 (Fig. 3). For a comprehensive evaluation of Chinese fir’s wood physical properties, we calculated the comprehensive scores of five geographic sources via the PCA. In this respect, significant differences were detected among the five geographic sources. Among them, the comprehensive score of HNYX-T was the highest whereas that of FJYK-P was the lowest (Fig. 4).
Figure 3

Sequence diagram plot of PCA analysis showing the relationship among physical properties of wood.

Full size image

Figure 4

Mean comprehensive score of PCA plot with 95% CI. Different letters (a, b, c, d, e) mean significant difference at 0.05 level.

Full size image More

1.
BPS. Statistik Indonesia 2018 (Badan Pusat Statistik, Jakarta, 2018).
Google Scholar 
2.
KLHK. Pedoman penilaian efektivitas pengelolaan kawasan konservasi di Indonesia. (Direktorat Kawasan Konservasi dan Direktorat Jenderal Konservasi Sumber Daya Alam dan Ekosistem, Kementerian Lingkungan Hidup dan Kehutanan, Jakarta, 2015).

3.
Ditjen KSDAE. Statistik Direktorat Jenderal KSDAE 2017 (Kementerian Lingkungan Hidup dan Kehutanan Direktorat Jenderal Konservasi Sumber Daya Alam dan Ekosistem, Jakarta, 2018).

4.
Mulyana, A. et al. Kebijakan pengelolaan zona khusus. Dapatkah meretas kebuntuan dalam menata ruang Taman Nasional di Indonesia? Brief, 1 (2010).

5.
CBD. Convention on Biological Diversity https://www.cbd.int/sp/ (2019).

6.
Murninngtyas, E. et al. (eds) Indonesian Biodiversity Strategy and Action Plan (IBSAP) 2015–2020 (Ministry of the National Development Planning/ BAPPENAS, Jakarta, 2016).
Google Scholar 

7.
Wiratno. Sepuluh cara baru kelola kawasan konservasi di Indonesia: membangun “organisasi pembelajar” (Direktorat Jenderal KSDAE Kementrian Lingkungan Hidup dan Kehutanan, Jakarta, 2018).

8.
Coad, L. et al. Measuring impact of protected area management interventions: current and future use of the Global Database of Protected Area Management Effectiveness. Philos. Trans. R. Soc. B 370, 20140281. https://doi.org/10.1098/rstb.2014.0281 (2015).
Article  Google Scholar 

9.
Geldmann, J. Evaluating the effectiveness of protected areas for maintaining biodiversity, securing habitats, and reducing threats. PhD thesis. (PhD School of the Faculty of Science, University of Copenhagen, 2013).

10.
Supriatna, J., Dwiyahreni, A. A., Winarni, N., Mariati, S. & Margules, C. Deforestation of primate habitat on Sumatra and adjacent islands Indonesia. Primate Conserv. 31, 71–82 (2017).
Google Scholar 

11.
Gray, C. L. et al. Local biodiversity is higher inside than outside terrestrial protected areas worldwide. Nat. Commun. 7(12306), 1–7. https://doi.org/10.1038/ncomms12306 (2016).
CAS  Article  Google Scholar 

12.
Eklund, J. & Cabeza, M. Quality of governance and effectiveness of protected areas: crucial concepts for conservation planning. Ann. N. Y. Acad. Sci. https://doi.org/10.1111/nyas.13284 (2016).
Article  PubMed  Google Scholar 

13.
Jones, K. R. et al. One-third of global protected land is under intense human pressure. Science 360, 788–789. https://doi.org/10.1126/science.aap9565 (2018).
CAS  Article  PubMed  Google Scholar 

14.
Wiratno. Tersesat di jalan yang benar. Seribu hari mengelola Leuser (Direktorat PKPS, Jakarta, 2012).

15.
Gaveau, D. L. A. et al. Examining protected area effectiveness in Sumatra: importance of regulations governing unprotected lands. Conserv. Lett. 5, 142–148. https://doi.org/10.1111/j.1755-263X.2011.00220.x (2012).
Article  Google Scholar 

16.
Azmi, W. & Gunaryadi, D. Current status of Asian elephants in Indonesia. Gajah 35, 55–61 (2011).
Google Scholar 

17.
O’Brien, T. G. & Kinnaird, M. F. Changing populations of birds and mammals in North Sulawesi. Oryx 30, 150–156 (1996).
Article  Google Scholar 

18.
Wheeler, P. & Dwiyahreni, A. Large mammal monitoring in Lambusango. Interim Progress Report January 2007 (University of Hull, UK, 2007).

19.
Gaveau, D. L. A., Wich, S. A. & Marshall, A. J. Are protected areas conserving primate habitat in Indonesia? In An introduction to primate conservation (eds Wich, S. A. & Marshal, A. J.) 193–200 (Oxford University Press, Oxford, 2016).
Google Scholar 

20.
Wibisono, H. T. & Pusparini, W. Sumatran tiger (Panthera tigris sumatrae): a review of conservation status. Integr. Zool. 5, 313–323. https://doi.org/10.1111/j.1749-4877.2010.00219.x (2010).
Article  PubMed  Google Scholar 

21.
Dwiyahreni, A.A. et al. Forest cover changes in Indonesia’s terrestrial national parks between 2012 and 2017. Biodiversitas, 22(3), 1235–1242. https://doi.org/10.13057/biodiv/d220320 (2021).  

22.
Sanderson, E. W. et al. The human footprint and the last of the wild. Bioscience 52(10), 891–904. https://doi.org/10.1641/0006-3568(2002)052[0891:THFATL]2.0.CO;2 (2002).
Article  Google Scholar 

23.
CIESIN. United Nation’s World Population Prospects (UN WPP)-Adjusted Population Density https://sedac.ciesin.columbia.edu/data/set/gpw-v4-population-density-adjusted-to-2015-unwpp-country-totals-rev11 (2018).

24.
Wan, J. Z., Wang, C. J. & Yuc, F. H. Human footprint and climate disappearance in vulnerable ecoregions of protected areas. Global Planet. Change 170, 260–268. https://doi.org/10.1016/j.gloplacha.2018.09.002 (2018).
ADS  Article  Google Scholar 

25.
Leroux, S. J. et al. Global protected areas and IUCN designations: do the categories match the conditions?. Biol. Conserv. 143, 609–616. https://doi.org/10.1016/j.biocon.2009.11.018 (2010).
Article  Google Scholar 

26.
Ayram, C. A. C., Mendoza, M. E., Etter, A. & Salicrup, D. R. P. Anthropogenic impact on habitat connectivity: a multidimensional human footprint index evaluated in a highly biodiverse landscape of Mexico. Ecol. Ind. 72, 895–909. https://doi.org/10.1016/j.ecolind.2016.09.007 (2017).
Article  Google Scholar 

27.
Tapia-Armijos, M. F., Homeier, J. & Muntac, D. D. Spatio-temporal analysis of the human footprint in South Ecuador: influence of human pressure on ecosystems and effectiveness of protected areas. Appl. Geogr. 78, 22–32. https://doi.org/10.1016/j.apgeog.2016.10.007 (2017).
Article  Google Scholar 

28.
Li, S., Wu, J., Gong, J. & Li, S. Human footprint in Tibet: assessing the spatial layout and effectiveness of nature reserves. Sci. Total Environ. 621, 18–29. https://doi.org/10.1016/j.scitotenv.2017.11.216 (2018).
ADS  CAS  Article  PubMed  Google Scholar 

29.
Geldmann, J. et al. Effectiveness of terrestrial protected areas in reducing habitat loss and population declines. Biol. Conserv. 161, 230–238. https://doi.org/10.1016/j.biocon.2013.02.018 (2013).
Article  Google Scholar 

30.
Anderson, E. & Mammides, C. The role of protected areas in mitigating human impact in the world’s last wilderness areas. Ambio https://doi.org/10.1007/s13280-019-01213-x (2019).
Article  PubMed  PubMed Central  Google Scholar 

31.
Mulyana, A., Kosmaryandi, N., Hakim, N., Suryadi, S. & Suwito. Ruang adaptif: refleksi penataan zona/blok di kawasan konservasi (Direktorat Pemolaan dan Informasi Konservasi Alam dan Direktorat Jenderal Konservasi Sumberdaya Alam dan Ekosistem, Kementerian Lingkungan Hidup dan Kehutanan, Bogor, 2019).

32.
Landuse thematic map 2012 and 2017 (Indonesian Ministry of Environment and Forestry)

33.
Binamarga Map 2014 (Indonesian Ministry of Public Works)

34.
Rusmin, N., Edy, S., Robby, A. & Dwi, K. R. Study of the potential expansion of new rice fields in Central Maluku District to support food security in Maluku Province. IOP Conf. Ser. Earth Environ. Sci. 334, 012067. https://doi.org/10.1088/1755-1315/334/1/012067 (2019).
Article  Google Scholar 

35.
Harrison, M. E., Capilla, B. R., Thornton, S. A., Cattau, M. E., & Page, S.E. Impacts of the 2015 fire season on peat-swamp forest biodiversity in Indonesian Borneo. In 15th International Peat Congress 2016, 713–717 (2016).

36.
Laurance, W. F. et al. Averting biodiversity collapse in tropical forest protected areas. Nature 00, 1–5. https://doi.org/10.1038/nature11318 (2012).
CAS  Article  Google Scholar 

37.
Opdam, P. Exploring the role of science in sustainable landscape management: an introduction to the special issue. Sustainability 10, 1–6. https://doi.org/10.3390/su10020331 (2018).
Article  Google Scholar 

38.
Field, D. R. Symbiotic relationships between national parks and neighboring social-biological regions in National Parks and rural development. In Practice, Policy in the United States (eds Machlis, G. E. & Field, D. R.) 211–218 (Island Press, Washington, DC, 2000).
Google Scholar 

39.
IUCN. Protected area categories. Category II: National Park. https://www.iucn.org/theme/protected-areas/about/protected-areas-categories/category-ii-national-park (2017).

40.
Alamgir, M. et al. High-risk infrastructure projects pose imminent threats to forests in Indonesian Borneo. Sci. Rep. https://doi.org/10.1038/s41598-018-36594-8 (2019).
Article  PubMed  PubMed Central  Google Scholar 

41.
Sloan, S. et al. Transnational conservation and infrastructure development in the heart of Borneo. PLoS ONE 14(9), e0221947. https://doi.org/10.1371/journal.pone.0221947 (2019).
CAS  Article  PubMed  PubMed Central  Google Scholar 

42.
Sloan, S., Alamgir, M., Campbell, M. J., Setyawati, T. & Laurance, W. F. Development corridors and remnant-forest conservation in Sumatra Indonesia. Trop. Conserv. Sci. 12, 1–9. https://doi.org/10.1177/1940082919889509 (2019).
Article  Google Scholar 

43.
Healey, R. M. et al. Road mortality threatens endemic species in a national park in Sulawesi Indonesia. Glob. Ecol. Conserv. 24, e01281. https://doi.org/10.1016/j.gecco.2020.e01281 (2020).
Article  Google Scholar 

44.
Du, W., Penabaz-Wiley, S. M., Njeru, A. M. & Kinoshita, I. Models and approaches for integrating protected areas with their surroundings: a review of the literature. Sustainability 7, 8151–8177. https://doi.org/10.3390/su7078151 (2015).
Article  Google Scholar 

45.
Verma, M. et al. Severe human pressures in the Sundaland biodiversity hotspot. Conserv. Sci. Pract. 2(e169), 2020. https://doi.org/10.1111/csp2.169 (2020).
Article  Google Scholar 

46.
DeFries, R., Hansen, A., Newton, A. C. & Hansen, M. C. Increasing isolation of protected areas in tropical forests over the past twenty years. Ecol. Appl. 15(1), 19–26 (2005).
Article  Google Scholar 

47.
Brun, C. et al. Analysis of deforestation and protected area effectiveness in Indonesia: a comparison of Bayesian spatial models. Glob. Environ. Change 31, 285–295. https://doi.org/10.1016/j.gloenvcha.2015.02.004 (2015).
Article  Google Scholar 

48.
Mariati, S., Kusnoputranto, H., Supriatna, J. & Koestoer, R. H. Habitat Loss of Sumatran elephants (Elephas maximus sumatranus) in Tesso Nilo Forest, Riau, Indonesia. Aust. J. Basic Appl. Sci. 8(2), 248–255 (2014).
Google Scholar 

49.
Busch, J. & Ferretti-Gallon, K. What drives deforestation and what stops it? A meta-analysis. Rev. Environ. Econ. Policy 11(1), 3–23. https://doi.org/10.1093/reep/rew013 (2017).
Article  Google Scholar 

50.
Tacconi, L., Rodriguesa, R. J. & Maryudi, A. Law enforcement and deforestation: lessons for Indonesia from Brazil. For. Policy Econ. 108, 101943. https://doi.org/10.1016/j.forpol.2019.05.029 (2019).
Article  Google Scholar 

51.
Bruner, A., Gullison, G., Rice, R. E., da Fonseca, R. E. & Gustavo, A. B. Effectiveness of parks in protecting tropical biodiversity. Science 291, 125–128. https://doi.org/10.1126/science.291.5501.125 (2001).
ADS  CAS  Article  PubMed  Google Scholar 

52.
Adams, V. M., Iacona, G. D. & Possingham, H. P. Weighing the benefits of expanding protected areas versus managing existing ones. Nat. Sustain. https://doi.org/10.1038/s41893-019-0275-5 (2019).
Article  Google Scholar 

53.
Bickford, D. et al. In Indonesia’s protected areas need more protection: suggestions from island examples in biodiversity and human livelihoods in protected areas: case studies from the Malay Archipelago (eds Sodhi, N. S. et al.) 53–77 (Cambridge University Press, Cambridge, 2008).
Google Scholar 

54.
Baier, M. The Kayan Mentarang National Park: Indonesia’s new national park in North Central Borneo bordering Northern Sarawak and Sabah. Borneo Res. Bull. 40, 297 (2009).
Google Scholar 

55.
Anau, N., Hakim, A., Lekson, A. S. & Setyowati, E. Local wisdom practices of Dayak indigenous people in the management of Tana’ Ulen in the Kayan Mentarang National Park of Malinau Regency, North Kalimantan Province Indonesia. RJOAS 7(91), 156–167. https://doi.org/10.18551/rjoas.2019-07.16 (2019).
Article  Google Scholar 

56.
Susanti, R. & Zuhud, E. A. M. Traditional ecological knowledge and biodiversity conservation: the medicinal plants of the Dayak Krayan people in Kayan Mentarang National Park Indonesia. Biodiversitas 20(9), 2764–2779. https://doi.org/10.13057/biodiv/d200943 (2019).
Article  Google Scholar 

57.
Blankespoor, B., Dasgupta, S. & Wheeler, D. Protected areas and deforestation: new results from high-resolution panel data. Nat. Resour. Forum 41, 55–68. https://doi.org/10.1111/1477-8947.12118 (2017).
Article  Google Scholar 

58.
Sanderson, E. W., Walston, J. & Robinson, J. G. From bottleneck to breakthrough: urbanization and the future of biodiversity conservation. Bioscience 20, 1–15 (2018).
Google Scholar 

59.
Immanuel, G. et al. Indonesia. Pathways to sustainable land-use and food systems. FABLE Report (2019).

60.
Allan, J. R. et al. Recent increases in human pressure and forest loss threaten many natural world heritage sites. Biol. Conserv. 206, 47–55. https://doi.org/10.1016/j.biocon.2016.12.011 (2017).
Article  Google Scholar 

61.
Damayanti, E.K. Legality of National Parks and Involvement of Local People: Case Studies in Java, Indonesia and Kerala, India. Thesis. Ph.D. in Agricultural Science. (University of Tsukuba, Japan, 2008).

62.
Smiet, A. C. Forest ecology on Java: human impact and vegetation of montane forest. J. Trop. Ecol. 8, 129–152 (1992).
Article  Google Scholar 

63.
Wijaya, I. K. M. The semiotics of banyan trees spaces in Denpasar. Bali. LivaS Int. J. Livable 4(2), 48–59. https://doi.org/10.25105/livas.v4i2.5564 (2019).
Article  Google Scholar 

64.
Wijaya, M. H. & Sutrisni, K. How can the existence of customary laws protect the water preservation in the Cau Belayu (Tabanan) traditional village?. Yustisia 7(3), 600–613. https://doi.org/10.20961/yustisia.v7i3.21560 (2018).
Article  Google Scholar 

65.
Swandi, I. W. Kearifan lokal Bali untuk pelestarian alam: kajian wacana kartun-kartun majalah “Bog-Bog”. Jurnal Kajian Bali 7(2), 229–248. https://doi.org/10.24843/JKB.2017.v07.i02.p12 (2017).
Article  Google Scholar 

66.
Sobirin, S. Pranata Mangsa dan budaya kearifan lingkungan. Jurnal Budaya Nusantara 2(1), 250–264. https://doi.org/10.36456/b.nusantara.vol2.no1.a1719 (2018).
Article  Google Scholar 

67.
Apriando, T. Brubuh, kearifan masyarakat Jawa menjaga hutan. Mongabay. https://www.mongabay.co.id/2015/02/12/brubuh-kearifan-masyarakat-jawa-menjaga-hutan/ (2015).

68.
Sulfiantono, A., Hermawan, M. T. T. & Maluyi, A. Comparison of Effectiveness of the management of conservation areas of China and Indonesia. Int. J. Sci. 11, 73–82 (2013).
Google Scholar 

69.
Supriatna, J. Berwisata di Taman Nasional (Yayasan Obor, Jakarta, 2014).
Google Scholar 

70.
Venter, O. et al. Global terrestrial human footprint maps for 1993 and 2009. Sci. Data 3, 160067. https://doi.org/10.1038/sdata.2016.67 (2016).
Article  PubMed  PubMed Central  Google Scholar 

71.
Willmott, C. J. & Matsuura, K. Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance. Clim. Res. 30, 79. https://doi.org/10.3354/CR030079 (2005).
Article  Google Scholar 

72.
Viera, A. J. & Garrett, J. M. Understanding interobserver agreement: the kappa statistic. Fam. Med. 37, 360–363 (2005).
PubMed  Google Scholar 

73.
R Development Core Team. R. A Language and environment for statistical computing (R Foundation for Statistical Computing, Vienna, Austria, 2014). More

Tropical cyclone-induced bloom in NIO
We have analysed the tropical cyclones that occurred over the bay from 1997 to 2019 in accordance with the availability of satellite Chl-a measurements. Out of 51 storm events, 30 are identified as the phytoplankton bloom events (i.e. the Chl-a values greater than of 0.2 mg/m3) in BoB and 18 in AS across all seasons35,36,37. In the case of BoB, we have divided our analyses for pre-monsoon and post-monsoon, as the cyclone occurrences are rare in other seasons (e.g. winter and monsoon). Over AS, some cyclones also occur in the beginning of monsoon season. The spatio-temporal variability of cyclones is closely connected to the seasonal changes in the monsoon trough35,38. In pre-monsoon season, the trough passes over the northern BoB, but it passes through the central bay with an east-west orientation in the post-monsoon, and facilitates the formation of more number of TCs during the period39. The big seasonal change in wind shear and relative vorticity are the reasons for the lower number of cyclones in the pre-monsoon season. In general, the upwelling driven nutrient influx to the surface together with sunlight leads to the enhancement of Chl-a or phytoplankton bloom after the passage of cyclones in the open ocean40.
The Bay of Bengal cyclones
During pre-monsoon, the bay is least productive, but the western boundary current helps more production in the coastal regions41. The higher wind speed associated with TCs deepens (about 30 m) Mixed Layer Depth (MLD), and rupture the pycnocline and pumps nutrients to the surface24. In general, TCs occurring during pre-monsoon move northwards and pass north-eastern coast of India or Bangladesh42 (Fig. 1). To estimate the cyclone-induced phytoplankton bloom, we performed spatial analyses for each cyclone during the period 1997–2019 and selected cases are shown in Supplementary Fig. 1 for BOB01 in 2003, BOB01 in 2004 and Mala in 2006. The maps of Chl-a concentrations superimposed with Sea Level Anomaly (SLA) for the same period are shown in Supplementary Fig. 1. The cyclone BOB01 was a category 1 storm with a maximum sustained wind (MSW) of 39 m/s, which occurred during 10–19 May 2003. The Chl-a remained well below 0.2 mg/m3 prior to occurrence of the cyclone, which enhanced to 0.5 mg/m3 with a small area of about 1 mg/m3 at 9°–10° N, 86°–87° E, just after passage of the cyclone. This is consistent with its high Ekman Pumping Velocity (EPV) and small Translational Speed (TS) in that period. In addition, the Chl-a enhancement was higher for the Category 1 cyclone BOB01 that occurred during 14–19 May 2004 and was about 0.5 mg/m3 after passage of the cyclone. The bloom sustained for the next 5 days as also shown in Supplementary Fig. 1. On the other hand, Mala was the strongest cyclone occurred during the pre-monsoon period (24–28 April 2006) in the last two decades over BoB with a MSW speed of 61 m/s. The Chl-a increased from 0.1 to 1.0 mg/m3 after the cyclone passage in five days, with a small area of Chl-a about 1.0 mg/m3 on the immediate left of its track. The Chl-a concentration remained close to 0.5 mg/m3 in the next 5 days. Both EPV and TS were favourable for sustained upwelling and Chl-a bloom in the case of cyclone Mala. In all three cases, the closed contours of Sea surface Height Anomaly (SSHA) is negative, which indicate the presence of cold-core (cyclonic) eddies that triggered turbulent mixing and sustained Chl-a bloom. We have not used any specific eddy detection method but used the composite of SSHA to identify the presence of eddies, as done by Girishkumar et al.43.
Fig. 1: The cyclone tracks and pre-cyclone background Chl-a value.

a The study region North Indian Ocean (NIO) and the tracks for the pre- (green) and post-monsoon (yellow) cyclones for BoB and AS (red), as analysed for all cyclones occurred during the study period (1997–2019).

Full size image

We applied the same method to identify the phytoplankton bloom that occurred for each cyclone event in BoB after 1996 and estimated the corresponding EPV and TS for diagnosing the physical mechanisms that made different scales of blooms. The results are presented in Table 1 and Fig. 2. The analyses show that the bloom was comparatively larger for the cyclone BOB01 in 2004, about 3.28 mg/m3. The increase in Chl-a is negatively correlated with TS and is in agreement with the intensity of cyclone with a statistically significant correlation value of −0.30 (at the 95% confidence interval as per the P-test44). The TS is lower and bloom is larger for BOB01 in 2003, as the faster moving storms tend to intensify rapidly when compared to slower moving storms (i.e. wind speed 14 m/s). In contrast, the slow-moving storms expend more time over the ocean and thereby, increases the magnitude of upwelling to enhance the Chl-a over the region45. The estimated EPV is about 1.8 × 10−4 m/s for BOB01 in 2003 and is consistent with the observed Chl-a concentrations, whereas the EPV is about 1 × 10−4 m/s and Chl-a concentration is about 1.87 mg/m3 for the cyclone Mala. It suggests that TS has a prominent role in cyclone-induced upwelling and associated phytoplankton bloom.
Table 1 The cyclone-induced Chl-a blooms in the pre-monsoon seasons since 1997 in BoB.
Full size table

Fig. 2: The cyclone-induced Chl-a in BoB.

The observed enhancement in Chl-a (mg/m3) following the cyclone passage (5-day average) in the post and pre-monsoon seasons in BoB. The translational speed of the closest track points where the bloom occurred in (m/s), the ratio of wind speed to translational speed (WS/TS), and EPV (Ekman Pumping Velocity) as the cyclones reach their maximum intensity (m/s) are also shown in the lower panels.

Full size image

Post-monsoon is the active storm season over BoB, and about 25 cyclones with significant enhancement in Chl-a concentration are identified during the 1997–2019 period. As the haline stratification is stronger in BoB due to the monsoon rain and river water influx, the presence of BL increases SST, which fuel the storms over the bay46. Presence of BL weakens the impact of cooling in the mixed layer driven by cyclones and favours the intensification of post-monsoon cyclones2. We have also analysed the variability in BL, MLD, isothermal layer depth (ILD) and Chl-a for selected storms passed over the Argo Floats (see next section). The cyclones either form or develop further over the southeast BoB, but some move west northwest and cross the peninsular coast. Some cyclones recurve towards the west central bay and pass the central and northeast coast of India, but some hit Bangladesh and upper Burma coast47, as illustrated in Fig. 1.
Figure 3 presents a closer look at the bloom and its spatial distribution for selected cyclones during the post-monsoon season; e.g. the cyclones Sidr, Madi and Vardah overlaid with SSHA contours. Sidr, a category 4 cyclone occurred during 11–16 November 2007 with a MSW of about 44 m/s. The Chl-a is about 0.5 mg/m3 during the cyclone period at the right side of the track, but the bloom has spread to a wider area with values close to 0.5 mg/m3 just after the passage of cyclone. The analyses of SSHA further provide evidence for the eddy-mediated phytoplankton bloom. The phytoplankton bloom also sustained for another 5 days. This is also in agreement with that reported in other analyses, although bloom values were estimated for 19 November in the other studies48,49. The cyclone Madi, occurred during 6–13 December 2013, showed an enhancement of about 0.5 mg/m3 during the cyclone period with a region of 1 mg/m3 in the left side of the track. Some regions with 2–3 mg/m3 are also observed at the right and left sides of cyclone track, and the bloom sustained for the next 5 days with values of about 1 mg/m3 in the adjacent areas. The closed contours of negative SSHA suggest the presence of cyclonic eddies there. The phytoplankton bloom during this particular period is also contributed by the cyclone Lehar that occurred a week before, in 23–28 November; demonstrating the impact of occurrences of consecutive storms over the same oceanic region. Nevertheless, the cyclone Vardah showed an enhancement of about 1.92 mg/m3, which is higher than that of Sidr due to the higher EPV of the former. As for Lehar and Madi, there was another cyclone Nada that appeared during the period 29 November–2 December 2016, just before the appearance of Vardah, and that storm might have also contributed to the Chl-a bloom during the period of Vardah.
Fig. 3: The spatial extension of Chl-a bloom for selected post-monsoon cyclones in BoB.

The Chl-a averaged, overlaid with SSHA contours (solid—positive and dashed—negative), for 5 days before cyclone, during the entire cyclone period, five days just after the passage of cyclone and next 5 days for a Sidr (2007), b Madi (2013), and c Vardah (2016). The tracks of the respective cyclones are also shown.

Full size image

Figure 4 shows the time evolution of physical and biological observations during the period of TCs Phailin, Hudhud and Vardah. The biogeochemical Argo float WMO ID 2902086 was closer to the track of TC Phailin, and the float WMO ID 2902114 was near the tracks of Hudhud and Vardah. Supplementary Table 1 shows the name of cyclones, Argo IDs, and distance between the float and nearest track point of respective cyclones. Figure 4 (right) represents the subsurface temperature up to 200 m with MLD, ILD, BLT and D23 (23° isotherm) for selected cyclones occurred over BoB. Figure 4 (left) represents the subsurface Chl-a concentration up to 200 m driven by the same cyclones. Prior to the passage of cyclones, the profiles represent typical hydrographic state of the oceans with warm waters near the surface and cold waters in the subsurface. The MLD was shallower about 20 m and the ocean was warm from September to December in 2013 during the passage of cyclone Phailin. The other cyclones occurred in 2013 were Helen, Lehar and Madi. Similar situation was observed in September–December of 2014 for Hudhud, but a colder and deeper MLD is observed in September–December of 2016 for Vardah. These are consistent with the climatological oceanic characteristics observed in BoB during the post-monsoon seasons. The time-depth cross-section of Chl-a reveals that the Chl-a concentration remains small in the surface, but about 0.8–1.0 mg/m3 at 40–60 m for the cyclone Phailin. The Chl-a concentration is about 3 mg/m3 for the cyclone Hudhud and about 1.5 mg/m3 for Vardah.
Fig. 4: Bio-Argo measurements.

Temporal evolution (right panel) of depth-time section up to 200 m of temperature of some selected cyclones in the BoB. The MLD (cyan), ILD (blue), BLT (green), and D23 (black) are also indicated in the figure. The vertical black lines indicate the cyclone period. The subsurface Chl-a concentration (left panel) up to 200 m for some selected cyclones in BoB. The vertical black lines indicate the cyclone period.

Full size image

To identify the differential oceanic response of the cyclones at the Argo float locations, we further examined the presence of eddies that play a major role in regulating the physical and biogeochemical processes. The analysis of 7-day SSHA composite before and during the cyclone period at the location of Argo float shows the presence of cold-core (negative SSHA) eddies before the passage of cyclone Phailin, Madi and Hudhud, whereas a warm-core (positive SSHA) eddy prior to the passage of Vardah (Fig. 5). The lower TS and a cold-core eddy during the cyclone Hudhud, and higher TS and a warm-core eddy during the cyclone Vardah produce contrasting oceanic response43. The temperature measurements during the periods of Hudhud and Vardah exhibit comparable response to cold and warm-core eddies, respectively. Another feature of cold-core eddies is trapping the near inertial oscillations in the mixed layer50, which accelerates the entrainment at the bottom of mixed layer and vertical shear as observed during the period of Hudhud. Conversely, the proximity of warm-core eddies triggers rapid vertical dispersion of near inertial energy, which suppresses the mixing and shear as for Vardah50.
Fig. 5: Eddies and primary productivity.

The 7-day composite of sea level anomaly (m) before and during the cyclones in BoB. The black solid lines represent the cyclone tracks, the stars represent the genesis location of the cyclone and the box represents the Argo float location.

Full size image

Vardah was a category 1 cyclone that occurred during 6–13 December 2016. The Chl-a amount before the passage of cyclone was about 0.23 mg/m3, but it escalated to 1.92 mg/m3 in 5 days after the passage of cyclone. The bloom continued to exist for the next 5 days as shown in Fig. 3. Unlike the pre-monsoon cases, for which the Chl-a is restored back to open ocean values in 10 days after the passage of cyclones, the bloom continued to persist even longer periods for the post-monsoon cases. The changes in Chl-a concentrations before and after the passage of cyclones in all three cases are greater than 0.2 mg/m3 and are higher for the lower category tropical storms. These analyses are consistent with the frequent occurrence of cyclones over the south east BoB during this season, as shown in Supplementary Table 2. It is also compelling to note that higher intensity cyclones occur over the north as compared to south BoB, which may be due to the presence of BL in the northern BoB as BL does not exist or insignificantly shallow in the south BoB in any season. The barrier layer in turn favours intensification of tropical cyclones whereas the absence of BL favours the storm-induced upwelling that eventually makes the Chl-a blooms45. Note that the stratification is very strong in northern BoB due to the river water input there51.
Supplementary Table 2 and Fig. 2 also show the results of Chl–bloom events in the post-monsoon seasons in 1997–2019. For instance, the Chl-a increased from 0.53 to 1.13 mg/m3 in the northern and southwestern BoB after the super cyclone of 1999 (25 October–3 November), as also shown by Madhu et al.52. Similar enhancements in Chl-a are estimated for BOB08 (1997), BOB05 (2000) and Madi (2013), about 0.6–3 mg/m3, depending on the cyclones. Although the cyclones BOB06 (1999), Sidr (2007), Giri (2010) and Phailin (2013) were category 4 or 5 cyclones, the high TS and lower EPV did not magnify the Chl-a concentrations to the level of bloom initiated by other cyclones. The enhancement of Chl-a estimated in our study during the cyclone Phailin is in agreement with the reported value of 0.9 mg/m3 for the period 16–24 October 2013 by Vidya et al.21. They also computed the bloom associated with Thane, about 0.7 mg/m3 in the post-cyclone period (1–8 January 2012) at 10°–13° N and 82°–86° E. Nevertheless, we have estimated about 1.8 mg/m3 for the post-cyclone period for Thane. The cyclone Hudhud produced a Chl-a bloom of up to 2.8 mg/m3 in 8–15 October 2014 along the track, as analysed by Chacko27 using the MODIS data, which is very close to our estimate of 2.8 mg/m3. We find similar enhancements in Chl-a that reported by Rao et al.26 for BOB05 in 16–23 November 2000, about 1.2 mg/m3 as deduced from the MODIS data. The other cyclones show moderate bloom values, below 1 mg/m3 as listed in Supplementary Table 2. Nevertheless, the low intensity cyclones such as BOB08 (1997) and Thane (2011) exhibit notable increment in Chl-a following the passage of cyclone, about 1.25–1.8 mg/m3, which is in agreement with their comparatively lower TS and higher EPV during the cyclone period. It also attests the impact and significance of TS in deciding the amplitude of phytoplankton bloom; suggesting sustained low intensity winds trigger strong upwelling to cause intense bloom events.
The Arabian Sea cyclones
In Arabian Sea, about 18 out of 33 cyclones are identified as phytoplankton bloom events (55%) during the study period 1997–2019, in which one occurred in pre-monsoon, three in monsoon and nine in post-monsoon seasons. Since the frequency of occurrences is very small, we have not separated the analyses into seasons or regions of landfall, but a selected case is presented in Supplementary Fig 2. For a better understanding of the behaviour of cyclones, we have selected three cyclones, ARB01 (2001), Mukda (2006) and Megh (2015), one in each season for this discussion. Supplementary Fig 3 illustrates the spatial distribution of Chl-a superimposed with SSHA for the selected cyclones. The ARB01 (2001) was a category 3 cyclone that occurred during 21–28 May 2001 with a MSW of about 60 m/s. The surface Chl-a concentration before the cyclone appearance was below 0.2 mg/m3 due to the profound heating in May53,54. It was one of the strongest cyclones appeared over AS, but measurements were sparse during the cyclone period, and thus, only a small area of about 1 mg/m3 is observed at 16°–17° N, 69°–72° E after passage of the cyclone. Analysis of IRS–P4 measurements by Subramanyam et al.29 found a very large bloom of about 5–8 mg/m3 at 17° N, from 67° E to 71° E. However, our analyses show the bloom of about 2.07 mg/m3 in the region 67°–68° E, 16°–17° N. The difference in bloom values could be due to the difference in datasets, region and period of analyses. As found in the case of BoB, there are closed contours of negative SSHA, which strengthens the observed cyclone-induced and eddy-mediated phytoplankton bloom.
Mukda was a tropical storm that occurred during 21–24 September 2006 with a MSW of 28 m/s. The Chl-a along the right side of the track was above 0.5 mg/m3 even before the cyclone period. The cyclone Megh was considered as the worst to hit Yemen and it occurred just after the passage of another cyclone Chapala over the same region. Megh was a category 3 cyclone with a MSW of 57 m/s. The Chl-a was about 0.7 mg/m3 in the post-cyclone stage, but a small region of about 1.0 mg/m3 was also observed at the right end of the track. Table 2 and Fig. 6 show the analysis of phytoplankton bloom occurrences in AS during the study period (1997–2019). It shows higher Chl-a concentrations in connection with the cyclones ARB01 and ARB02 in 2001, Mukda in 2006 and Megh in 2015, and are consistent with their lower TS. The situation in 2015 was also similar, in which the Category 4 cyclone Chapala showed higher bloom than that of the category 3 cyclone Megh. Similarly, the Category 4 cyclone Kyaar triggered higher bloom than that of the Category 3 cyclone Maha in 2019.
Table 2 The cyclone-induced Chl-a blooms in the post-monsoon seasons since 1997 in AS.
Full size table

Fig. 6: The Chl-a blooms associated with cyclones in Arabian Sea.

The observed enhancement in Chl-a (mg/m3) following the cyclone passage (5-day average) over AS. The translational speed of the closest track points where the bloom occurred in (m/s), the ratio of wind speed to translational speed (WS/TS) and EPV when cyclone reached its maximum intensity (m/s) are also shown.

Full size image

Supplementary Fig. 3 shows the time evolution of biological and physical observations from the Argo float (WMO ID 2902120) in the period of TCs Nilofar, Chapala and Megh. The distance between float location and nearest track point is provided in Supplementary Table 1. The temperature profiles show warm waters near the surface and cold waters in the subsurface before the cyclone passage, as for a typical oceanic state (Supplementary Fig. 4). The MLD is shallower about 20 m and the ocean is warm throughout September–December 2014 and October–December 2015 during the passage of cyclones Nilofar, Megh and Chapala. The time-depth cross-section of chlorophyll shows about 0.8–1 mg/m3 at 40–60 m. However, the Chl-a values remain about 1 mg/m3 for cyclones Megh and Chapala close to the surface; supporting the satellite measurements. The analysis of 7-day SSHA composite shows a cold-core eddy before the passage of cyclone Nilofar whereas warm-core eddies before the passage of Chapala and Megh at the buoy location (Supplementary Fig. 5). The warm-core eddies before the passage of Chapala and Megh could also be the reason for their rapid intensification.
Tropical storms and category 1 cyclones
Although a number of cyclones occurred over NIO during the 1997–2019 period, the phytoplankton bloom happened mostly for the storms and lower category cyclones. For instance, there were five cyclones that appeared over BoB in the pre-monsoon seasons that triggered phytoplankton bloom, four (80%) of them were either tropical storms or category 1 cyclones (Table 1). Similarly, out of 25 cyclones that made Chl-a blooms in BoB during the post-monsoon seasons, 20 of them were (80%) either tropical storms or category 1 cyclones. An analogues occurrence of cyclone-induced phytoplankton bloom is observed for the lower category cyclones in AS, where eight cyclones out of 18 (44.4%) were either tropical storms or category 1 cyclones. These analyses suggest that the slow-moving storms stay more time over the oceans and impart high momentum to upwell the subsurface nutrient-rich water, leading to the phytoplankton blooms in the open oceans with a time lag of 4–12 days, as illustrated in Fig. 7 (blue coloured bar chart). The bloom is as higher as about 20–500% with respect to the pre-cyclone Chl-a levels, and is even up to 1385% as for the case of BoB01 in 2003 and 3758% for the cyclone Gonu in AS (Supplementary Fig. 6); demonstrating the impact and scale of cyclone-induced primary productivity in the open oceans. This slow-moving cyclone-induced primary productivity is very important in the context of climate change, as there is a global slowdown in the translational speed of tropical cyclones.
Fig. 7: The change in cyclone-induced bloom and time lag.

Left: with respect to pre-cyclone Chl-a values (blue), 0.5 mg/m3 (dark blue) and 0.2 mg/m3 (magenta) for the post-monsoon cyclones in Bay of Bengal (BoB). Right: The time lag in days in cyclone-induced Chl-a bloom with respect to the pre-cyclone Chl-a (blue histograms, left) for the post-monsoon cyclones of BoB.

Full size image

To test robustness of the estimates of cyclone-induced change in Chl-a (i.e. Fig. 7), we also considered two other background Chl-a values (i.e. 0.2 and 0.5 mg/m3), which were also taken as the background Chl-a of the ocean basins and the Chl-a threshold for bloom detection. Since the value extracted from the 1° × 1° latitude-longitude region at the track (i.e. blue diagrams) is different from the basin average and bloom threshold values, there are significant differences in the amplitude of blooms, as displayed in Fig. 7. It shows that the pre-cyclone Chl-a values are between 0.5 and 0.2 mg/m3. Therefore, the change in Chl-a is higher with the estimates based on 0.2 mg/m3 (magenta histogram) and about 10 cyclones show a change in Chl-a of about 400%. The highest bloom of about 800–850% is found for Madi (2013), Hudhud (2014) and BOB05 (1999). On the other hand, the change in Chl-a with respect to 0.5 mg/m3 (dark blue histogram) is lower than that with the pre-cyclone estimates, and the change is mostly within 250%, although few cyclones show around 400%. The highest bloom is observed for the cyclone Madi (2013), about 300%. In AS, the Chl-a bloom is mostly between 300 and 1000%, except for Gonu in 2007. The change in Chl-a is about 6000% with respect to the basin average of 0.2 mg/m3, and about 3500% based on the bloom threshold for the cyclone Gonu. The assessment confirm that the cyclone-induced bloom (change in percent) in AS is about five times higher than that of BoB.
The impact of ENSO and IOD on Chl-a blooms
Several studies have examined the relationship between El Niño and Southern Oscillation (ENSO) and cyclone activity across different oceanic basins12,55,56. The influence of ENSO on tropical cyclone activity in BoB during the period 1997–2010 is also investigated by Girishkumar et al.19. We have chosen the dates after the cyclone passage, and considered the Niño and IOD indices to classify the cyclones occurred in the El Niño, La Niña, normal, positive IOD (PIOD) and negative IOD (NIOD) years, as listed in Table 1, Table 2 and Supplementary Table 2. In addition, we have prepared the composites of Chl-a and SSHA for 10 days before and after the passage of each cyclone to assess the inter-annual variability in Chl-a and SSHA with respect to the El Niño, La Niña, normal, PIOD and NIOD years (Fig. 8, for BoB). Out of the 25 cyclones, three of them occurred in El Niño, fourteen in La Niña, nine in normal, four in PIOD and five in NIOD years. The cyclones those occurred in PIOD or NIOD years also happened to be in the El Niño/La Niña years and therefore included in both analyses, and are shown in the figure. The number of cyclones are more in the La Niña years, which were mostly followed by the normal, PIOD and NIOD years. The magnitude of phytoplankton bloom is higher in the PIOD years than that in the NIOD years. In the El Niño years, the magnitude of bloom is comparatively smaller and the bloom in normal years is around 0.5 mg/m3.
Fig. 8: The differences in Chl-a during El Niño, La Niña, normal, PIOD and NIOD years.

The SSHA a 10-day before and b during the passage of cyclone) and Chl-a composite maps with respect to El Niño, La Niña, normal, PIOD and NIOD years in the BoB. The respective cyclone tracks are also shown.

Full size image

Supplementary Fig. 9 shows the composite of SSHA and Chl-a with respect to El Niño, La Niña, normal, PIOD and NIOD years in AS. Here, more number of cyclones occurred in the El Niño years as compared to that in the La Niña and normal years. In contrast, there are more number of cyclones in the PIOD years than the NIOD years in AS, but the amplitude of bloom is higher for the NIOD years. These are also the reasons for the differences in phytoplankton bloom in AS and BoB, as the impact of ENSO and IOD events is different in both basins. The response of cyclones in IOD years are similar to those in the La Niña years. Although the spatial extent of bloom is larger in the El Niño years owing to the higher number of cyclone occurrences, the magnitude of bloom is higher for the cyclones occurred in La Niña and NIOD years. The normal years exhibit bloom similar to that of the El Niño years. In BoB, the analysis of SSHA composite for the El Niño, La Niña, IOD and normal years is dominated by negative SSHA (suggesting the presence of cold-core eddies), but the normal years are more influenced by warm-core eddies (e.g. Fig. 8). In AS, on the other hand, the normal, La Niña and IOD years are dominated by cold-core eddies, whereas the El Niño years are overwhelmed by warm-core eddies (e.g. Supplementary Fig. 9). The influence of IOD is higher than that of ENSO, which is one of the reasons for the inter-annual variability of phytoplankton blooms. The characteristics of phytoplankton blooms in AS and BoB are in contrast with the differences in SST in IOD years in both basins, and this feature is also found with the cyclone-induced blooms. There are noticeable difference in Chl-a concentrations among the normal and El Niño, La Niña, PIOD or NIOD years, and are exhibited in Supplementary Figs. 7 and 8. More

1.
Ceballos, G. et al. Accelerated modern human–induced species losses: Entering the sixth mass extinction. Sci. Adv. 1, e1400253 (2015).
ADS  PubMed  PubMed Central  Article  Google Scholar 
2.
Cardillo, M. et al. Human population density and extinction risk in the world ’ s carnivores. PLoS Biol. 2, 909–914 (2004).
CAS  Article  Google Scholar 

3.
Pimm, S. L. et al. Can we defy nature’s end ?. Science 293, 2207–2208 (2001).
CAS  PubMed  Article  Google Scholar 

4.
Ripple, W. J. et al. Bushmeat hunting and extinction risk to the world’s mammals. R. Soc. Open Sci. 3, 160498 (2016).
ADS  PubMed  PubMed Central  Article  Google Scholar 

5.
Chapron, G. et al. Recovery of large carnivores in Europe’s modern human-dominated landscapes. Science 346, 1517–1519 (2014).
ADS  CAS  PubMed  Article  Google Scholar 

6.
Owen-Smith, N. Pleistocene extinctions: the pivotal role of megaherbivores. Paleobiology 13, 351–362 (1987).
Article  Google Scholar 

7.
Pradhan, N. M. B. & Wegge, P. Dry season habitat selection by a recolonizing population of Asian elephants Elephas maximus in lowland Nepal. Acta Theriol. (Warsz) 52, 205–214 (2007).
Article  Google Scholar 

8.
Hayward, M. W. et al. The reintroduction of large carnivores to the Eastern Cape South Africa: an assessment. Oryx 41(205), 214 (2007).
Google Scholar 

9.
Owen-Smith, N. Megaherbivores: the influence of very large body size on ecology. Trends Ecol. Evol. https://doi.org/10.1111/j.1523-1739.1989.tb00246.x (1989).
Article  Google Scholar 

10.
Karki, J. B., Jhala, Y. V. & Khanna, P. P. Grazing lawns in Terai Grasslands, Royal Bardia National Park, Nepal1. Biotropica 32, 423–429 (2000).
Article  Google Scholar 

11.
McNaughton, S. J. Serengeti migratory wildebeest: facilitation of energy flow by grazing. Science 191, 92–94 (1976).
ADS  CAS  PubMed  Article  Google Scholar 

12.
Skarpe, C. et al. The return of the giants: ecological effects of an increasing elephant population. Ambio 33, 276–282 (2004).
PubMed  Article  Google Scholar 

13.
Foose, T. J. & Van Strien, N. Asian Rhinos—Status Survey and Conservation Action Plan. Vol. 32 (1997).

14.
Subedi, N. et al. Population status, structure and distribution of the greater one-horned rhinoceros Rhinoceros unicornis in Nepal. Oryx 47, 352–360 (2013).
Article  Google Scholar 

15.
Talukdar, B. R. R. C. Asian Rhino specialist group report. Pachyderm 53, 25–27 (2013).
Google Scholar 

16.
Hedges, S., Sagar Baral, H., Timmins, R. & Duckworth, J. Bubalus arnee. IUCNRed List Threat. Species 2008 (2008).

17.
Leader-Williams, N. Fate riding on their horns—and genes?. ORYX 47, 311–312 (2013).
Article  Google Scholar 

18.
Amin, R., Thomas, K., Emslie, R. H., Foose, T. J. & VanStrien, N. An overview of the conservation status of and threats to rhinoceros species in the wild. Int. Zoo Yearb. 40, 96–117 (2006).
Article  Google Scholar 

19.
IUCN Red List Categories and Criteria, Version 3.1, second edition | IUCN Library System. IUCN (2012).

20.
Phillips, S. J. & Dudík, M. Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography 31, 161–175 (2008).
Article  Google Scholar 

21.
Lacy, R. C. Vortex: A computer simulation model for population viability analysis. Wildl. Res. 20, 1–13 (1993).
Article  Google Scholar 

22.
IUCN/SSC. Guidelines for Reintroductions and Other Conservation Translocations. Version 1.0. Gland, Switzerland: IUCN Species Survival Commission. Ecologial Applications (2013).

23.
Guillera-Arroita, G. et al. Is my species distribution model fit for purpose? Matching data and models to applications. Glob. Ecol. Biogeogr. 24, 276–292 (2015).
Article  Google Scholar 

24.
Elith, J. & Leathwick, J. R. Species distribution models: ecological explanation and prediction across space and time. Annu. Rev. Ecol. Evol. Syst. 40, 677–697 (2009).
Article  Google Scholar 

25.
Veloz, S. D. Spatially autocorrelated sampling falsely inflates measures of accuracy for presence-only niche models. J. Biogeogr. 36, 2290–2299 (2009).
Article  Google Scholar 

26.
Kramer-Schadt, S. et al. The importance of correcting for sampling bias in MaxEnt species distribution models. Divers. Distrib. 19, 1366–1379 (2013).
Article  Google Scholar 

27.
Jhala, Y., Qureshi, Q. & Gopal, R. Status of Tigers, Copredators and Prey in India, 2014 (2015).

28.
Graham-Rowe, D. Biodiversity: endangered and in demand. Nature 480, S101–S103 (2011).
ADS  CAS  PubMed  Article  Google Scholar 

29.
Frankham, R. Genetic considerations in reintroduction programmes for top-order, terrestrial predators. In Reintroduction of Top-Order Predators 371–387 (Wiley-Blackwell, 2009). https://doi.org/10.1002/9781444312034.ch17

30.
Martin, E., Kumar Talukdar, B. & Vigne, L. Rhino poaching in Assam: challenges and opportunities. Pachyderm (2009).

31.
Rawat, G. S., Goyal, S. P. & Johnsingh, A. J. T. Ecological observations on the grasslands of Corbett Tiger Reserve India. Indian Forester 123(958), 963 (1997).
Google Scholar 

32.
Dinerstein, E. & Schaller, G. The Return of the Unicorns: Natural History and Conservation of Greater—One Horned Rhinoceros (2003). https://doi.org/10.7312/dine08450.

33.
N Subedi 2012 Effect of Mikania micrantha on thrdemography, Habitat use, and Nutrition of Greater one horned rhinoceros in Chitwan National Park Dr. Philos Thesis

34.
Mathur, V.B., Gopal, R., Yadav, S.P., P. R. S. Management Effectiveness Evaluation (MEE) of Tiger Reserves in India: Process and Outcomes 97 (2011).

35.
Amin, R., Thomas, K., Emslie, R. H., Foose, T. J. & Van Strien, N. V. An overview of the conservation status of and threats to rhinoceros species in the wild. Int. Zoo Yearb. 40, 96–117 (2006).
Article  Google Scholar 

36.
Singh, S. P., Sharma, A. & Talukdar, B. K. Translocation of Rhinos within Assam : a successful third round of the second phase of translocations under Indian Rhino Vision (IRV) 2020, 1–6 (2012).
Google Scholar 

37.
Jhala, Y. ., Qureshi, Q., Gopal, R. & Sinha, P. . Status of the Tigers, Co-predators, and Prey in India, 2010. (2011).

38.
Rai, S. After 250 years, rhinos set to make comeback in west UP|Meerut News—Times of India. Times of India (2016).

39.
Subedi, N., Lamichhane, B. R., Amin, R., Jnawali, S. R. & Jhala, Y. V. Demography and viability of the largest population of greater one-horned rhinoceros in Nepal. Glob. Ecol. Conserv. 12, 241–252 (2017).
Article  Google Scholar 

40.
IUCN Standards and Petitions Committee. IUCN Standards and Petitions Committee. Stand. Petitions Comm. 1, 1–60 (2019).
Google Scholar 

41.
Cockrill, W. R. The water buffalo: a review. Br. Vet. J. 137, 8–10 (1981).
CAS  PubMed  Article  Google Scholar 

42.
Kumar, S. et al. Mitochondrial DNA analyses of Indian water buffalo support a distinct genetic origin of river and swamp buffalo. Anim. Genet. 38, 227–232 (2007).
CAS  PubMed  Article  Google Scholar 

43.
Aiyadurai, A. Wildlife hunting and conservation in Northeast India—a need for an interdisciplinary understanding.pdf. Int. J. Gall. Conserv. 2, 61–73 (2011).
Google Scholar 

44.
Jhala, H. Y., Pokheral, C. P. & Subedi, N. Well being and conservation awareness of communities around Chitwan National Park, Nepal|Jhala|Indian Forester. Indian For. 145, 114–120 (2019).
Google Scholar 

45.
Heinen, J. T. & Kandel, R. Threats to a small population: a census and conservation recommendations for wild buffalo Bubalus arnee in Nepal. Oryx 40, 324–330 (2006).
Article  Google Scholar 

46.
Choudhury, A. The decline of the wild water buffalo in north-east India. Oryx 28, 70–73 (1994).
Article  Google Scholar 

47.
Magioncalda, W. A modern insurgency: India’s evolving naxalite problem. S. Asia Monitor 140 (2010).

48.
Melles, S. J., Fortin, M.-J., Lindsay, K. & Badzinki, D. Expanding northward: influence of climate change, forest connectivity, and population processes on a threatened species’ range shift. Glob. Chang. Biol. 17, 17–31 (2011).
ADS  Article  Google Scholar 

49.
Challender, D. W. S. & MacMillan, D. C. Poaching is more than an enforcement problem. Conserv. Lett. 7, 484–494 (2014).
Article  Google Scholar 

50.
Hirzel, A. H., Hausser, J., Chessel, D. & Perrin, N. Ecological-niche factor analysis: How to compute habitat-suitability maps without absence data?. Ecology 83, 2027–2036 (2002).
Article  Google Scholar 

51.
Stockman, A. K., Beamer, D. A. & Bond, J. E. An evaluation of a GARP model as an approach to predicting the spatial distribution of non-vagile invertebrate species. Divers. Distrib. 12, 81–89 (2006).
Article  Google Scholar 

52.
Merow, C., Smith, M. J. & Silander, J. A. A practical guide to MaxEnt for modeling species’ distributions: what it does, and why inputs and settings matter. Ecography (Cop.) 36, 1058–1069 (2013).
Article  Google Scholar 

53.
Phillips, S. J. & Dudı, M. Modeling of species distributions with Maxent : new extensions and a comprehensive evaluation. Ecograohy 31, 161–175 (2008).
Article  Google Scholar 

54.
Elith, J. et al. A statistical explanation of MaxEnt for ecologists. Divers. Distrib. 17, 43–57 (2011).
Article  Google Scholar 

55.
Phillips, S. J., Anderson, R. P., Dudík, M., Schapire, R. E. & Blair, M. E. Opening the black box: an open-source release of Maxent. Ecography (Cop.) 40, 887–893 (2017).
Article  Google Scholar 

56.
Congalton, R. G. A review of assessing the accuracy of classifications of remotely sensed data. Remote Sens. Environ. 37, 35–46 (1991).
ADS  Article  Google Scholar 

57.
Roy, D. P. et al. Landsat-8: Science and product vision for terrestrial global change research. Remote Sens. Environ. 145, 154–172 (2014).
ADS  Article  Google Scholar 

58.
Dinerstein, E. & Price, L. Demography and habitat use by greater one-horned rhinoceros in Nepal. J. Wildl. Manage. 55, 401 (1991).
Article  Google Scholar 

59.
Bontemps, S. et al. GLOBCOVER 2009 Products Description and Validation Report.

60.
Barsi, J., Lee, K., Kvaran, G., Markham, B. & Pedelty, J. The spectral response of the landsat-8 operational land imager. Remote Sens. 6, 10232–10251 (2014).
ADS  Article  Google Scholar 

61.
Laurie, A. Behavioural ecology of the Greater one-horned rhinoceros (Rhinoceros unicornis). J. Zool. 196, 307–341 (2009).
Article  Google Scholar 

62.
Dnerstein, E. & Price, L. Demography and habitat use by greater one-horned rhinoceros in Nepal author(s): Eric Dinerstein and Lori Price Source. J. Wildl. Manage. 55, 401–411 (1991).
Article  Google Scholar 

63.
Pettorelli, N. et al. The Normalized Difference Vegetation Index (NDVI): unforeseen successes in animal ecology. Climate Res. 46, 15–27 (2011).
ADS  Article  Google Scholar 

64.
Raffini, F. et al. From nucleotides to satellite imagery: approaches to identify and manage the invasive pathogen Xylella fastidiosa and its insect vectors in Europe. Sustainability 12, 4508 (2020).
CAS  Article  Google Scholar 

65.
Norberg, A. et al. A comprehensive evaluation of predictive performance of 33 species distribution models at species and community levels. Ecol. Monogr. https://doi.org/10.1002/ecm.1370 (2019).
Article  Google Scholar 

66.
Peterson, A. T. & Nakazawa, Y. Environmental data sets matter in ecological niche modelling: an example with Solenopsis invicta and Solenopsis richteri. Glob. Ecol. Biogeogr. https://doi.org/10.1111/j.1466-8238.2007.00347.x (2007).
Article  Google Scholar 

67.
Laurie, A. Behavioural ecology of the greater one-horned rhinoceros Rhinoceros unicornis. J. Zool. 196, 307–341 (1982).
Article  Google Scholar 

68.
Kriticos, D. J. et al. CliMond: Global high-resolution historical and future scenario climate surfaces for bioclimatic modelling. Methods Ecol. Evol. 3, 53–64 (2012).
Article  Google Scholar 

69.
Rodríguez, E. et al. An Assessment of the SRTM Topographic Products.

70.
Watson, J. E. M., Whittaker, R. J. & Dawson, T. P. Habitat structure and proximity to forest edge affect the abundance and distribution of forest-dependent birds in tropical coastal forests of southeastern Madagascar. Biol. Conserv. 120, 311–327 (2004).
Article  Google Scholar 

71.
Society WC. Last of the Wild Project, Version 2, 2005 (LWP-2): Global Human Influence Index (HII) Dataset (Geographic) Society. Conserv Wildl https://doi.org/10.7927/H4BP00QC (2005).
Article  Google Scholar 

72.
Phillips, S. J. et al. Sample selection bias and presence-only distribution models: Implications for background and pseudo-absence data. Ecol. Appl. 19, 181–197 (2009).
PubMed  Article  Google Scholar 

73.
Jiménez-Valverde, A. Insights into the area under the receiver operating characteristic curve (AUC) as a discrimination measure in species distribution modelling. Glob. Ecol. Biogeogr. 21, 498–507 (2011).
Article  Google Scholar 

74.
Allouche, O., Tsoar, A. & Kadmon, R. Assessing the accuracy of species distribution models: prevalence, kappa and the true skill statistic (TSS). J. Appl. Ecol. 43, 1223–1232 (2006).
Article  Google Scholar 

75.
Shabani, F., Kumar, L. & Ahmadi, M. Assessing accuracy methods of species distribution models: AUC, specificity, sensitivity and the true skill statistic. 18, (2018).

76.
Warren, D. L. & Seifert, S. Ecological niche modeling in Maxent: the importance of model complexity and the performance of model selection criteria. Ecol. Soc. Am. 21, 335–342 (2011).
Google Scholar 

77.
Wiltshire, K. H. & Tanner, J. E. Comparing maximum entropy modelling methods to inform aquaculture site selection for novel seaweed species. Ecol. Modell. 429, 109071 (2020).
Article  Google Scholar 

78.
Smeraldo, S. et al. Modelling risks posed by wind turbines and power lines to soaring birds: the black stork (Ciconia nigra) in Italy as a case study. Biodivers. Conserv. 29, 1959–1976 (2020).
Article  Google Scholar 

79.
Puri, K. & Joshi, R. A case study of greater one horned rhinoceros (Rhinoceros unicornis) in Kaziranga National Park of Assam India. NeBIO 9, 307–309 (2018).
Google Scholar 

80.
DNPWC 2015. Rhino Count Report 2015 DNPWC.

81.
Mukherjee, T., Sharma, L. K., Saha, G. K., Thakur, M. & Chandra, K. Past, present and future: combining habitat suitability and future landcover simulation for long-term conservation management of Indian rhino. Sci. Rep. https://doi.org/10.1038/s41598-020-57547-0 (2020).
Article  PubMed  PubMed Central  Google Scholar 

82.
CCF West Bengal. Estimation of Indian Rhinoceros (Rhinoceros unicornis) 2019 West Bengal (2019)

83.
Boyce, M. Population viability analysis: adaptive management for threatened and endangered species. 226–238 (1997).

84.
Khatri, T. B., Shah, D. N. & Mishra, N. Wild Water Buffalo Bubalus arnee in Koshi Tappu Wildlife Reserve, Nepal: status, population and conservation importance. J. Threat. Taxa 04, 3294–3301 (2012).
Article  Google Scholar 

85.
Phillips, S., Anderson, R., Dudík, M., Schapire, R. & Blair, M. Opening the black box: an open-source release of Maxent. Ecography 40, 887–893 (2017). More

1.
Martin, E. A. et al. The interplay of landscape composition and configuration: New pathways to manage functional biodiversity and agroecosystem services across Europe. Ecol. Lett. 22, 1083–1094. https://doi.org/10.1111/ele.13265 (2019).
Article  PubMed  Google Scholar 
2.
Daane, K. M., Vincent, C., Isaacs, R. & Ioriatti, C. Entomological opportunities and challenges for sustainable viticulture in a global market. Annu. Rev. Entomol. 63, 193–214. https://doi.org/10.1146/annurev-ento-010715-023547 (2018).
CAS  Article  PubMed  Google Scholar 

3.
Viers, J. H. et al. Vinecology: Pairing wine with nature. Conserv. Lett. 6, 287–299. https://doi.org/10.1111/conl.12011 (2013).
Article  Google Scholar 

4.
Pertot, I. et al. A critical review of plant protection tools for reducing pesticide use on grapevine and new perspectives for the implementation of IPM in viticulture. Crop Prot. 97, 70–84. https://doi.org/10.1016/j.cropro.2016.11.025 (2017).
ADS  CAS  Article  Google Scholar 

5.
Roehrich, R. & Boller, E. Tortricids in Vineyards. In Tortricid Pests. Their Biology, Natural Enemies and Control (eds Van der Geest, L. P. S. & Evenhuis, H. H.) 507–514 (Elsevier Science Publishers, 1991).

6.
Ioriatti, C., Lucchi, A. & Varela, L. G. Grape berry moths in Western European vineyards and their recent movement into the New World. In Arthropod Management in Vineyards: Pests, Approaches, and Future Directions (eds Bostanian, N. J., Vincent, C. & Isaacs, R.) 339–359 (Springer, 2012).

7.
Stellwaag, F. Die Weinbauinsekten der Kulturländer (Paul Parey, Berlin, 1928).
Google Scholar 

8.
Pavan, F., Girolami, V. & Sacilotto, G. Second generation of grape berry moths, Lobesia botrana (Den. & Schiff.) (Lep., Tortricidae) and Eupoecilia ambiguella (Hb.) (Lep., Cochylidae): Spatial and frequency distributions of larvae, weight loss and economic injury level. J. Appl. Entomol. 122, 361–368. https://doi.org/10.1111/j.1439-0418.1998.tb01513.x (1998).
Article  Google Scholar 

9.
Svobodova, E. et al. Determination of areas with the most significant shift in persistence of pests in Europe under climate change. Pest Manag. Sci. 70, 708–715. https://doi.org/10.1002/ps.3622 (2014).
CAS  Article  PubMed  Google Scholar 

10.
Reineke, A. & Thiéry, D. Grapevine insect pests and their natural enemies in the age of global warming. J. Pest Sci. 89, 313–328. https://doi.org/10.1007/s10340-016-0761-8 (2016).
Article  Google Scholar 

11.
Gutierrez, A. P., Ponti, L., Gilioli, G. & Baumgartner, J. Climate warming effects on grape and grapevine moth (Lobesia botrana) in the Palearctic region. Agric. For. Entomol. 20, 255–271. https://doi.org/10.1111/afe.12256 (2018).
Article  Google Scholar 

12.
Schartel, T. E. et al. Reconstructing the European Grapevine Moth (Lepidoptera: Tortricidae), Invasion in California: Insights from a successful eradication. Ann. Entomol. Soc. Am. 112, 107–117. https://doi.org/10.1093/aesa/say056 (2019).
Article  Google Scholar 

13.
Gutierrez, A. P. et al. Prospective analysis of the invasive potential of the European grapevine moth Lobesia botrana (Den. & Schiff.) in California. Agric. For. Entomol. 14, 225–238. https://doi.org/10.1111/j.1461-9563.2011.00566.x (2012).
Article  Google Scholar 

14.
Gilligan, T. M. et al. Discovery of Lobesia botrana ([Denis and Schiffermüller]) in California: An invasive species new to North America (Lepidoptera: Tortricidae). Proc. Entomol. Soc. Wash. 113, 14–30. https://doi.org/10.4289/0013-8797.113.1.14 (2011).
Article  Google Scholar 

15.
Selkoe, K. A. & Toonen, R. J. Microsatellites for ecologists: A practical guide to using and evaluating microsatellite markers. Ecol. Lett. 9, 615–629 (2006).
Article  Google Scholar 

16.
Reineke, A., Karlovsky, P. & Zebitz, C. P. W. Preparation and purification of DNA from insects for AFLP-analysis. Insect Mol. Biol. 7, 95–99. https://doi.org/10.1046/j.1365-2583.1998.71048.x (1998).
CAS  Article  PubMed  Google Scholar 

17.
Reineke, A. et al. A novel set of microsatellite markers for the European Grapevine Moth Lobesia botrana isolated using next-generation sequencing and their utility for genetic characterization of populations from Europe and the Middle East. Bull. Entomol. Res. 105, 408–416. https://doi.org/10.1017/S0007485315000267 (2015).
CAS  Article  PubMed  Google Scholar 

18.
Schuelke, M. An economic method for the fluorescent labeling of PCR fragments. Nat. Biotechnol. 18, 233–234 (2000).
CAS  Article  Google Scholar 

19.
Kalinowski, S. T., Taper, M. L. & Marshall, T. C. Revising how the computer program CERVUS accommodates genotyping error increases success in paternity assignment. Mol Ecol 16, 1099–1106. https://doi.org/10.1111/j.1365-294X.2007.03089.x (2007).
Article  PubMed  PubMed Central  Google Scholar 

20.
Excoffier, L. & Lischer, H. E. L. Arlequin suite ver. 3.5: A new series of programs to perform population genetics analyses under Linux and Windows. Mol. Ecol. Resour. 10, 564–567. https://doi.org/10.1111/j.1755-0998.2010.02847.x (2010).
Article  PubMed  PubMed Central  Google Scholar 

21.
Raymond, M. & Rousset, F. GENEPOP (version 1.2): Population genetics software for exact tests and ecumenicism. J. Hered. 86, 248–249 (1995).
Article  Google Scholar 

22.
Goudet, J. FSTAT (Version 1.2): A computer program to calculate F-statistics. J. Hered. 86, 485–486. https://doi.org/10.1093/oxfordjournals.jhered.a111627 (1995).
Article  Google Scholar 

23.
Rice, W. R. Analyzing tables of statistical tests. Evolution 43, 223–225 (1989).
Article  Google Scholar 

24.
Chapuis, M.-P. & Estoup, A. Microsatellite null alleles and estimation of population differentiation. Mol. Biol. Evol. 24, 621–631. https://doi.org/10.1093/molbev/msl191 (2007).
CAS  Article  PubMed  Google Scholar 

25.
Pritchard, J. K., Stephens, M. & Donnelly, P. Inference of population structure using multilocus genotype data. Genetics 155, 945–959 (2000).
CAS  PubMed  PubMed Central  Google Scholar 

26.
Falush, D., Stephens, M. & Pritchard, J. K. Inference of population structure using multilocus genotype data: Linked loci and correlated allele frequencies. Genetics 164, 1567–1587 (2003).
CAS  PubMed  PubMed Central  Google Scholar 

27.
Evanno, G., Regnaut, S. & Goudet, J. Detecting the number of clusters of individuals using the software STRUCTURE: A simulation study. Mol. Ecol. 14, 2611–2620. https://doi.org/10.1111/j.1365-294X.2005.02553.x (2005).
CAS  Article  PubMed  PubMed Central  Google Scholar 

28.
Earl, D. A. & vonHoldt, B. M. STRUCTURE HARVESTER: A website and program for visualizing STRUCTURE output and implementing the Evanno method. Conserv. Genet. Resour. 4, 359–361. https://doi.org/10.1007/s12686-011-9548-7 (2012).
Article  Google Scholar 

29.
Littell, R. C., Milliken, G. A., Stroup, W. W., Wolfinger, R. D. & Schabenberger, O. SAS for Mixed Models (SAS Institute, Cary, 2006).
Google Scholar 

30.
Saour, G. Flight ability and dispersal of European grapevine moth gamma-irradiated males (Lepidoptera: Tortricidae). Fla. Entomol. 99, 73–78. https://doi.org/10.1653/024.099.sp110 (2016).
Article  Google Scholar 

31.
Schmitz, V., Roehrich, R. & Stockel, J. Dispersal of marked and released Lobesia botrana in a small isolated vineyard and the effect of synthetic sex pheromone on moth movements. J. Int. Sci. Vigne Vin 30, 67–72 (1996).
CAS  Google Scholar 

32.
Sciarretta, A., Zinni, A., Mazzocchetti, A. & Trematerra, P. Spatial analysis of Lobesia botrana (Lepidoptera: Tortricidae) male population in a Mediterranean agricultural landscape in Central Italy. Environ. Entomol. 37, 382–390. https://doi.org/10.1093/ee/37.2.382 (2008).
CAS  Article  PubMed  Google Scholar 

33.
De Meeûs, T. Revisiting FIS, FST, wahlund effects, and null alleles. J. Hered. 109, 446–456. https://doi.org/10.1093/jhered/esx106 (2018).
Article  PubMed  Google Scholar 

34.
Delbac, L. & Thiéry, D. Damage to grape flowers and berries by Lobesia botrana larvae (Denis & Schiffernüller) (Lepidoptera: Tortricidae), and relation to larval age. Aust. J. Grape Wine R. 22, 256–261. https://doi.org/10.1111/ajgw.12204 (2016).
Article  Google Scholar 

35.
Fuentes-Contreras, E. et al. Measuring local genetic variability in populations of codling moth (Lepidoptera: Tortricidae) across an unmanaged and commercial orchard interface. Environ. Entomol. 43, 520–527. https://doi.org/10.1603/en13131 (2014).
Article  PubMed  Google Scholar 

36.
Dong, Z. K., Li, Y. F. & Zhang, Z. Y. Genetic diversity of melon aphids Aphis gossypii associated with landscape features. Ecol. Evol. 8, 6308–6316. https://doi.org/10.1002/ece3.4181 (2018).
Article  PubMed  PubMed Central  Google Scholar 

37.
Vialatte, A., Dedryver, C. A., Simon, J. C., Galman, M. & Plantegenest, M. Limited genetic exchanges between populations of an insect pest living on uncultivated and related cultivated host plants. P. Roy. Soc. B-Biol. Sci. 272, 1075–1082. https://doi.org/10.1098/rspb.2004.3033 (2005).
Article  Google Scholar 

38.
Manel, S. & Holderegger, R. T. years of landscape genetics. Trends Ecol. Evol. 28, 614–621. https://doi.org/10.1016/j.tree.2013.05.012 (2013).
Article  PubMed  Google Scholar 

39.
Tscharntke, T. et al. When natural habitat fails to enhance biological pest control—Five hypotheses. Biol. Conserv. 204(Part B), 449–458. https://doi.org/10.1016/j.biocon.2016.10.001 (2016).
Article  Google Scholar  More

Overview of methods
First, we modeled and mapped investible forest carbon, and its climate mitigation potential across the tropics at 1-km resolution. Second, we compared our estimates of investible forest carbon with actual volumes of VCUs reported by 25 real-world VCS forest protection projects. Third, we modeled the relative profitability of investible forest carbon sites to produce a global forest carbon return-on-investment map based on their NPV.
All calculations were based on data dated between 2012 and 2017 and at a resolution of 0.00833 degrees (~1 km). To ensure data standardization, we resampled (bilinear) finer-scaled data where necessary, for example, for data sourced from the European Space Agency – Climate Change Initiative -Land Cover27. We only considered tropical forests between ~23.44°N and 23.44°S, and excluded all land cover types that would preclude forests, for example, savannas, bare ground, water, agriculture and urban areas27.
Investible forest carbon
We first estimated the total volume of CO2 associated with three carbon pools in tropical forests: aboveground carbon, belowground carbon, and soil organic carbon. Next, we applied key VCS criteria, including additionality, to model and map investible forest carbon across the tropics.
Mapping total volume of CO2 associated with tropical forests
Aboveground carbon
We applied a stoichiometric factor of 0.475 to recent spatial data on aboveground carbon biomass12 (i.e., for period 2012–2016), to convert it from biomass to carbon stock values, based on established carbon accounting methodology3,28,29. We performed an uncertainty analysis to account for potential variability in this stoichiometric factor (see ‘Uncertainty analysis’ section below). We applied a conversion factor of 3.67 to derive the volume of CO2 associated with this carbon pool3.
Belowground carbon
We derived belowground carbon biomass by applying two different allometric equations relating root to shoot biomass30 to the most recent spatial dataset on aboveground carbon biomass12, following established carbon accounting methodology3,28,29. The two equations are: belowground biomass = 0.489 × aboveground biomass^0.89; and belowground biomass = 0.26 × aboveground biomass. We then applied a stoichiometric factor of 0.475 to the estimated belowground carbon biomass to convert it from biomass to carbon stock values. Next, we calculated the mean, minimum and maximum values for belowground carbon based on an uncertainty analysis (see ‘Uncertainty analysis’ section below). We applied a conversion factor of 3.67 to derive the volume of CO2 associated with this carbon pool3.
Soil organic carbon
We also considered soil carbon due to its potentially significant contributions to carbon storage31 and despite potential uncertainties and variability surrounding its measurements32. Specifically, we utilized the organic carbon density of the topsoil layer (0–30 cm) obtained from the European Soil Data Centre33 as it represented the best data available of soil organic carbon. We applied a conversion factor of 3.67 to derive the volume of CO2 associated with this carbon pool3.
Applying VCS criteria to map investible forest carbon
The criterion of additionality is a pre-condition for certifying all carbon credits under the VCS. This implies that only the volume of forest carbon that are under imminent threat of decline or loss if left unprotected by a conservation intervention can be certified under the VCS. We derived the volume of forest carbon under threat of loss based on best available proxy data on projected future deforestation rates across the tropics13 (through to the year 2029), and annualized over the prediction period (15 years). We applied this estimated annual deforestation rate to the total volume of CO2 associated with tropical forests as estimated above, to derive the volume of CO2 that would be certifiable and therefore investible under the VCS.
We also assumed a conservative 10-year decay estimate for the belowground carbon pool9.
Additionally, we excluded lands that will likely not be certifiable for other reasons9, including recently deforested areas34 (i.e., for the period 2010–2017), as well as human settlements located within these forests35.
Lastly, we accounted for the VCS requirement to set aside buffer credits of 20% to account for the risk of non-permanence associated with Agriculture, Forestry and Other Land Use projects (AFOLU)9.
Comparing estimates of investible forest carbon to verified carbon units
We compared our estimates of investible forest carbon with actual volumes of VCUs reported by real-world VCS forest protection projects (https://verra.org/).
We identified a set of 25 VCS forest protection projects from across 16 countries that met the following criteria: ii) includes spatial data on project boundary in their project documentation; ii) the project extent is located entirely within the tropics; and 3) has been verified (i.e., either “verified, under verification” or “verification approve”) (Table S2).
We extracted the shapefiles (i.e., geometric polygons) of these VCS projects, and overlay them on our map of investible forest carbon to extract the volume of investible forest carbon (CO2) from our analysis that corresponds to each of the 25 VCS forest project.
We then compared our estimates of investible forest carbon to the volume of VCUs issued between 2005 and 2018 for each VCS project. The number of data points reported per year for each project ranged from 1–10, and generated a total of 111 data points for comparison. We then assessed the degree of correlation (i.e., Pearson’s correlation), relative accuracy (i.e., Root Mean Square Error; RMSE), and statistical difference (i.e., paired t-test) between the two datasets.
Estimating return-on-investment
Based on our map of investible forest carbon, we modeled the relative profitability of investible forest carbon sites to produce a global forest carbon return-on-investment map based on their NPV. We calculated NPV of these returns based on several simplifying assumptions following established values from previous studies19.
First, we estimated the cost of project establishment at $25 ha−1. This was based on a wide range of costs that are key to the development of a project, including but not limited to project design, governance and planning, enforcement, zonation, land tenure and acquisition, surveying and research19,36,37.
Second, we estimated an annual maintenance cost of $10 ha−1, which included aspects such as education and communication, monitoring, sustainable livelihoods, marketing, finance and administration19,36,37.
Third, we assumed a constant carbon price of $5.8 t−1CO2 for the first five years. This price was based on an average price of carbon for avoided deforestation projects recently reported by Forest Trends’ Ecosystem Marketplace6 (i.e., for the period 2006–2018). After the first five years, we assumed a 5% price appreciation for subsequent years over a project timeframe of 30 years19.
Based on these criteria, we calculated NPV of annual and accumulated profits over the 30 years, based on a 10% risk-adjusted discount rate.
Separately, we repeated the analysis using a range of starting carbon prices, including $1, $5, $10, $15, $25, $50, $100 t−1CO2, based on cost effectiveness thresholds from previous studies1. In these analyses, other assumptions remain unchanged, including the project establishment and annual maintenance cost, price appreciation, discount rates and timeframe. Based on these criteria and excluding sites that would be unable to breakeven (i.e., yielding net negative NPVs), we calculated the potential profitable forest areas, as a percentage of the total investible forest areas, associated with these different starting carbon prices.
All values of investible carbon and return-on-investment (based on NPV) were summarized to global, regional, and country level estimates (see Table 1). For countries that extend beyond tropical latitudes, we only analyze and present data for their tropical extents. These values were rounded to the nearest 1000 values.
Uncertainty analyses
Stoichiometric factor
Previous studies utilized a range of stoichiometric factors, typically ranging between 0.45 and 0.503,28,29. We account for this variability by first using a stoichiometric factor of 0.475, which was based on the median value across these reference studies3,28,29. We then repeated the analyses with stoichiometric factors of 0.45 and 0.50 to calculate the respective minimum and maximum values of above and belowground carbon per cell.
Root to shoot biomass allometric equations
Many site-specific factors can influence the ratio of root to shoot biomass, resulting in variability of the best-fit allometric equations30. Here, we account for this variability by utilizing the two allometric equations that best matches global data30. This produced two sets of spatially explicit estimates of belowground biomass, from which we calculated the average, minimum and maximum values per cell.
Aboveground biomass
We incorporated uncertainties, reported at standard deviations, which were inherent to the aboveground biomass dataset12.
Leakage effects
We considered three scenarios of leakage, where the protection of an area of forest results in deforestation beyond its borders to the amounts of 10%, 20%, and 30% of the areas’ carbon volume. This reduces the total investible carbon within each cell, thereby causing a decrease in return-on-investment and the climate mitigation potential within profitable areas to 81.9 ± 51.1, 64.6 ± 40.4 and 48.3 ± 30.3%, or 909.1 ± 567.4, 716.7 ± 448.4, 535.8 ± 336.3 MtCO2 yr−1, respectively (Table S1).
Establishment and maintenance costs
We also considered two scenarios of establishment and maintenance cost, where the overall direct cost of protecting areas from deforestation increases by 50% and 100%. We find that this reduces the climate mitigation potential in profitable areas to 80.4 ± 50.4 and 65.7 ± 41.6% or 892.1 ± 559.2 and 728.8 ± 462.2 MtCO2 yr−1 respectively (Table S1).
Opportunity costs
We also considered the potential for alternative land-use such as agriculture or timber extraction to outcompete the value of protecting forests through carbon financing means. Utilizing agricultural rents (based on 18 crops) and timber value as a proxy for opportunity cost38, we excluded areas where opportunity cost exceeds projected net present values. This results in a large decrease in overall climate mitigation potential, almost comparable to the 30% leakage scenario, to 52.3 ± 33.2% or 580.6 ± 368.6 MtCO2 yr−1 within remaining areas.
All analyses were performed in R version 3.6.039, utilizing the package “raster” for processing and calculations of raster layers40. Map visualizations were formed in QGIS41.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article. More

1.
Moran, N. A., Ochman, H. & Hammer, T. J. Evolutionary and ecological consequences of gut microbial communities. Annu. Rev. Ecol. Syst. 50, 451–475 (2019).
Article  Google Scholar 
2.
Engel, P. & Moran, N. A. The gut microbiota of insects—Diversity in structure and function. FEMS Microbiol. Rev. 37, 699–735 (2013).
CAS  PubMed  PubMed Central  Article  Google Scholar 

3.
Douglas, A. E. Multiorganismal insects: Diversity and function of resident microorganisms. Annu. Rev. Entomol. 60, 17–34 (2015).
CAS  PubMed  Article  PubMed Central  Google Scholar 

4.
Paniagua Voirol, L. R., Frago, E., Kaltenpoth, M., Hilker, M. & Fatouros, N. E. Bacterial symbionts in Lepidoptera: Their diversity, transmission, and impact on the host. Front. Microbiol. 9, 556 (2018).
PubMed  PubMed Central  Article  Google Scholar 

5.
Mason, C. J. Complex relationships at the intersection of insect gut microbiomes and plant defenses. J. Chem. Ecol. 46, 793–807 (2020).
CAS  PubMed  Article  PubMed Central  Google Scholar 

6.
Hammer, T. J., Sanders, J. G. & Fierer, N. Not all animals need a microbiome. FEMS Microbiol. Lett. 366, fnz117 (2019).
CAS  PubMed  Article  PubMed Central  Google Scholar 

7.
Jones, A., Mason, C., Felton, G. & Hoover, K. Host plant and population source drive diversity of microbial gut communities in two polyphagous insects. Sci. Rep. 9, 2792 (2019).
ADS  PubMed  PubMed Central  Article  CAS  Google Scholar 

8.
Hammer, T. J., Janzen, D. H., Hallwachs, W., Jaffe, S. P. & Fierer, N. Caterpillars lack a resident gut microbiome. Proc. Natl. Acad. Sci. 114, 9641–9646 (2017).
CAS  PubMed  Article  PubMed Central  Google Scholar 

9.
Broderick, N. A., Raffa, K. F., Goodman, R. M. & Handelsman, J. Census of the bacterial community of the gypsy moth larval midgut by using culturing and culture-independent methods. Appl. Environ. Microbiol. 70, 293–300 (2004).
CAS  PubMed  PubMed Central  Article  Google Scholar 

10.
Shao, Y., Arias-Cordero, E., Guo, H., Bartram, S. & Boland, W. In vivo Pyro-SIP assessing active gut microbiota of the cotton leafworm, Spodoptera littoralis. PLoS ONE 9, e85948 (2014).
ADS  PubMed  PubMed Central  Article  CAS  Google Scholar 

11.
Priya, N. G., Ojha, A., Kajla, M. K., Raj, A. & Rajagopal, R. Host plant induced variation in gut bacteria of Helicoverpa armigera. PLoS ONE 7, e30768 (2012).
PubMed  Article  CAS  PubMed Central  Google Scholar 

12.
Mason, C. J. & Raffa, K. F. Acquisition and structuring of midgut bacterial communities in gypsy moth (Lepidoptera: Erebidae) larvae. Environ. Entomol. 43, 595–604 (2014).
PubMed  Article  PubMed Central  Google Scholar 

13.
Martemyanov, V. V. et al. Phenological asynchrony between host plant and gypsy moth reduces insect gut microbiota and susceptibility to Bacillus thuringiensis. Ecol. Evol. 6, 7298–7310 (2016).
PubMed  PubMed Central  Article  Google Scholar 

14.
Chen, B. et al. Gut microbiota metabolic potential correlates with body size between mulberry-feeding lepidopteran pest species. Pest Manag. Sci. 76, 1313–1323 (2020).
CAS  PubMed  Article  PubMed Central  Google Scholar 

15.
Su’ad, A. Y. et al. Host plant-dependent effects of microbes and phytochemistry on the insect immune response. Oecologia 191, 141–152 (2019).
Article  Google Scholar 

16.
Mason, C. J. et al. Plant defenses interact with insect enteric bacteria by initiating a leaky gut syndrome. Proc. Natl. Acad. Sci. 116, 15991–15996 (2019).
CAS  PubMed  Article  PubMed Central  Google Scholar 

17.
Staudacher, H. et al. Variability of bacterial communities in the moth Heliothis virescens indicates transient association with the host. PLoS ONE 11, e0154514 (2016).
PubMed  PubMed Central  Article  CAS  Google Scholar 

18.
Ericsson, A. C., Personett, A. R., Turner, G., Dorfmeyer, R. A. & Franklin, C. L. Variable colonization after reciprocal fecal microbiota transfer between mice with low and high richness microbiota. Front. Microbiol. 8, 196 (2017).
PubMed  PubMed Central  Article  Google Scholar 

19.
Kreisinger, J. et al. Temporal stability and the effect of transgenerational transfer on fecal microbiota structure in a long distance migratory bird. Front. Microbiol. 8, 50 (2017).
PubMed  PubMed Central  Article  Google Scholar 

20.
Stappenbeck, T. S. & Virgin, H. W. Accounting for reciprocal host-microbiome interactions in experimental science. Nature 534, 191–199 (2016).
ADS  CAS  PubMed  Article  PubMed Central  Google Scholar 

21.
Mikaelyan, A., Thompson, C. L., Hofer, M. J. & Brune, A. Deterministic assembly of complex bacterial communities in guts of germ-free cockroaches. Appl. Environ. Microbiol. 82, 1256–1263 (2016).
CAS  PubMed  PubMed Central  Article  Google Scholar 

22.
Salem, H., Florez, L., Gerardo, N. & Kaltenpoth, M. An out-of-body experience: The extracellular dimension for the transmission of mutualistic bacteria in insects. Proc. R. Soc. Lond. B Biol. Sci. 282, 20142957 (2015).
Google Scholar 

23.
Powell, J. E., Martinson, V. G., Urban-Mead, K. & Moran, N. A. Routes of acquisition of the gut microbiota of Apis mellifera. Appl. Environ. Microbiol. 80, 7378–7387 (2014).
PubMed  PubMed Central  Article  CAS  Google Scholar 

24.
Brune, A. Symbiotic digestion of lignocellulose in termite guts. Nat. Rev. Microbiol. 12, 168–180 (2014).
ADS  CAS  Article  Google Scholar 

25.
Chen, B. et al. Gut bacterial and fungal communities of the domesticated silkworm (Bombyx mori) and wild mulberry-feeding relatives. ISME J. 12, 2252–2262 (2018).
CAS  PubMed  PubMed Central  Article  Google Scholar 

26.
Hannula, S., Zhu, F., Heinen, R. & Bezemer, T. Foliar-feeding insects acquire microbiomes from the soil rather than the host plant. Nat. Commun. 10, 1254 (2019).
ADS  PubMed  PubMed Central  Article  CAS  Google Scholar 

27.
Montezano, D. G. et al. Host plants of Spodoptera frugiperda (Lepidoptera: Noctuidae) in the Americas. Afr. Entomol. 26, 286–300 (2018).
Article  Google Scholar 

28.
Day, R. et al. Fall armyworm: Impacts and implications for Africa. Outlooks Pest Manag. 28, 196–201 (2017).
Article  Google Scholar 

29.
Visôtto, L. E., Oliveira, M. G. A., Guedes, R. N. C., Ribon, A. O. B. & Good-God, P. I. V. Contribution of gut bacteria to digestion and development of the velvetbean caterpillar, Anticarsia gemmatalis. J. Insect Physiol. 55, 185–191 (2009).
PubMed  Article  CAS  PubMed Central  Google Scholar 

30.
Xiang, H. et al. Microbial communities in the larval midgut of laboratory and field populations of cotton bollworm (Helicoverpa armigera). Can. J. Microbiol. 1092, 1085–1092 (2006).
Article  Google Scholar 

31.
Tang, X. et al. Complexity and variability of gut commensal microbiota in polyphagous lepidopteran larvae. PLoS ONE 7, e36978 (2012).
ADS  CAS  PubMed  PubMed Central  Article  Google Scholar 

32.
Gomes, A. F. F., Omoto, C. & Cônsoli, F. L. Gut bacteria of field-collected larvae of Spodoptera frugiperda undergo selection and are more diverse and active in metabolizing multiple insecticides than laboratory-selected resistant strains. J. Pest Sci. 93, 833–851 (2020).
Article  Google Scholar 

33.
Acevedo, F. E. et al. Fall armyworm-associated gut bacteria modulate plant defense responses. Mol. Plant-Microbe Interact. 30, 127–137 (2017).
CAS  PubMed  Article  PubMed Central  Google Scholar 

34.
Gichuhi, J. et al. Diversity of fall armyworm, Spodoptera fugiperda and their bacterial community in Kenya. PeerJ 8, e8701 (2020).
PubMed  PubMed Central  Article  Google Scholar 

35.
Wagner, M. R., Busby, P. E. & Balint-Kurti, P. Analysis of leaf microbiome composition of near-isogenic maize lines differing in broad-spectrum disease resistance. New Phytol. 225, 2152–2165 (2020).
CAS  PubMed  Article  PubMed Central  Google Scholar 

36.
Naveed, M., Mitter, B., Reichenauer, T. G., Wieczorek, K. & Sessitsch, A. Increased drought stress resilience of maize through endophytic colonization by Burkholderia phytofirmans PsJN and Enterobacter sp. FD17. Environ. Exp. Bot. 97, 30–39 (2014).
CAS  Article  Google Scholar 

37.
Keshri, J. et al. Microbiome dynamics during ensiling of corn with and without Lactobacillus plantarum inoculant. Appl. Microbiol. Biotechnol. 102, 4025–4037 (2018).
CAS  PubMed  Article  Google Scholar 

38.
Mason, C. J. et al. Diet influences proliferation and stability of gut bacterial populations in herbivorous lepidopteran larvae. PLoS ONE 15, e0229848 (2020).
CAS  PubMed  PubMed Central  Article  Google Scholar 

39.
Chan, Q. W. T., Melathopoulos, A. P., Pernal, S. F. & Foster, L. J. The innate immune and systemic response in honey bees to a bacterial pathogen, Paenibacillus larvae. BMC Genomics 10, 387 (2009).
PubMed  PubMed Central  Article  CAS  Google Scholar 

40.
Mazumdar, T. et al. Survival strategies of Enterococcus mundtii in the gut of Spodoptera littoralis: A live report. bioRxiv. https://doi.org/10.1101/2020.02.03.932053 (2020).

41.
Mason, C. J., Jones, A. G. & Felton, G. W. Co-option of microbial associates by insects and their impact on plant–folivore interactions. Plant Cell Environ. https://doi.org/10.1111/pce.13430 (2018).
Article  PubMed  PubMed Central  Google Scholar 

42.
Mason, C. J., Rubert-Nason, K. F., Lindroth, R. L. & Raffa, K. F. Aspen defense chemicals influence midgut bacterial community composition of gypsy moth. J. Chem. Ecol. 41, 75–84 (2014).
PubMed  Article  CAS  PubMed Central  Google Scholar 

43.
Chaturvedi, S., Rego, A., Lucas, L. K. & Gompert, Z. Sources of variation in the gut microbial community of Lycaeides melissa caterpillars. Sci. Rep. https://doi.org/10.1038/s41598-017-11781-1 (2017).
Article  PubMed  PubMed Central  Google Scholar 

44.
Fescemyer, H. W. et al. Maize toxin degrades peritrophic matrix proteins and stimulates compensatory transcriptome responses in fall armyworm midgut. Insect Biochem. Mol. Biol. 43, 280–291 (2013).
CAS  PubMed  Article  PubMed Central  Google Scholar 

45.
Hammer, T. J., McMillan, W. O. & Fierer, N. Metamorphosis of a butterfly-associated bacterial community. PLoS ONE 9, e86995 (2014).
ADS  PubMed  PubMed Central  Article  CAS  Google Scholar 

46.
Chippendale, G. M. Metamorphic changes in haemolymph and midgut proteins of the southwestern corn borer, Diatraea grandiosella. J. Insect Physiol. 16, 1909–1920 (1970).
CAS  Article  Google Scholar 

47.
Pechan, T., Cohen, A., Williams, W. P. & Luthe, D. S. Insect feeding mobilizes a unique plant defense protease that disrupts the peritrophic matrix of caterpillars. Proc. Natl. Acad. Sci. 99, 13319–13323 (2002).
ADS  CAS  PubMed  Article  PubMed Central  Google Scholar 

48.
Mohan, S. et al. Degradation of the S. frugiperda peritrophic matrix by an inducible maize cysteine protease. J. Insect Physiol. 52, 21–28 (2006).
CAS  PubMed  Article  PubMed Central  Google Scholar 

49.
Tsuji, G. Y., Hoogenboom, G. & Thornton, P. K. Understanding Options for Agricultural Production Vol. 7 (Springer Science & Business Media, Berlin, 2013).
Google Scholar 

50.
Apprill, A., McNally, S., Parsons, R. & Weber, L. Minor revision to V4 region SSU rRNA 806R gene primer greatly increases detection of SAR11 bacterioplankton. Aquat. Microb. Ecol. 75, 129–137 (2015).
Article  Google Scholar 

51.
Parada, A. E., Needham, D. M. & Fuhrman, J. A. Every base matters: Assessing small subunit rRNA primers for marine microbiomes with mock communities, time series and global field samples. Environ. Microbiol. 18, 1403–1414 (2016).
CAS  PubMed  Article  PubMed Central  Google Scholar 

52.
Schloss, P. D. et al. Introducing mothur: Open-source, platform-independent, community-supported software for describing and comparing microbial communities. Appl. Environ. Microbiol. 75, 7537–7541 (2009).
CAS  PubMed  PubMed Central  Article  Google Scholar 

53.
R Core Team. R: A Language and Environment for Statistical Computing. (2020).

54.
Kolde, R. pheatmap: Pretty Heatmaps. R package version 1.0.12. https://CRAN.R-project.org/package=pheatmap (2018).

55.
Kay, M. & Wobbrock, J. ARTool: Aligned Rank Transform for Nonparametric Factorial ANOVAs. https://doi.org/10.5281/zenodo.594511, R package version 0.10.7, https://github.com/mjskay/ARTool (2020).

56.
Wobbrock, J., Findlater, L., Gergle, D., & Higgins, J.. The Aligned Rank Transform for Nonparametric Factorial Analyses Using Only ANOVA Procedures. In Proceedings of the ACM Conference on Human Factors in Computing Systems (CHI ’11), 143–146.(2011).

57.
Raubenheimer, D. & Simpson, S. L. Analysis of covariance: An alternative to nutritional indices. Entomol. Exp. Appl. 62, 221–231 (1992).
Article  Google Scholar  More