Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Coupling remote sensing and eDNA to monitor environmental impact: A pilot to quantify the environmental benefits of sustainable agriculture in the Brazilian Amazon

  • Karen Dyson ,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing

    Affiliation Spatial Informatics Group, LLC, Pleasanton, California, United States of America

  • Andréa P. Nicolau,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing

    Affiliation Spatial Informatics Group, LLC, Pleasanton, California, United States of America

  • Karis Tenneson,

    Roles Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – original draft, Writing – review & editing

    Affiliation Spatial Informatics Group, LLC, Pleasanton, California, United States of America

  • Wendy Francesconi,

    Roles Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – review & editing

    Affiliation Alliance of Biodiversity International and International Center for Tropical Agriculture (CIAT), Kasarani, Nairobi

  • Amy Daniels,

    Roles Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – review & editing

    Affiliation United States Agency for International Development (USAID), Washington, DC, United States of America

  • Giulia Andrich,

    Roles Investigation, Writing – review & editing

    Affiliation Instituto de Manejo e Certificação Florestal e Agrícola (IMAFLORA), Piracicaba, Brazil

  • Bernardo Caldas,

    Roles Conceptualization, Investigation, Writing – review & editing

    Affiliation Alliance of Biodiversity International and International Center for Tropical Agriculture (CIAT), Kasarani, Nairobi

  • Silvia Castaño,

    Roles Investigation, Writing – review & editing

    Affiliation Alliance of Biodiversity International and International Center for Tropical Agriculture (CIAT), Kasarani, Nairobi

  • Nathanael de Campos,

    Roles Conceptualization, Investigation, Writing – review & editing

    Affiliation Instituto de Manejo e Certificação Florestal e Agrícola (IMAFLORA), Piracicaba, Brazil

  • John Dilger,

    Roles Formal analysis, Writing – review & editing

    Affiliation Spatial Informatics Group, LLC, Pleasanton, California, United States of America

  • Vinicius Guidotti,

    Roles Investigation, Writing – review & editing

    Affiliation Instituto de Manejo e Certificação Florestal e Agrícola (IMAFLORA), Piracicaba, Brazil

  • Iara Jaques,

    Roles Investigation, Writing – review & editing

    Affiliation Spatial Informatics Group, LLC, Pleasanton, California, United States of America

  • Ian M. McCullough,

    Roles Writing – review & editing

    Affiliation Spatial Informatics Group, LLC, Pleasanton, California, United States of America

  • Allan D. McDevitt,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – review & editing

    Affiliation Atlantic Technological University, Galway, Ireland

  • Luis Molina,

    Roles Writing – review & editing

    Affiliation Alliance of Biodiversity International and International Center for Tropical Agriculture (CIAT), Kasarani, Nairobi

  • Dawn M. Nekorchuk,

    Roles Writing – review & editing

    Affiliation Spatial Informatics Group, LLC, Pleasanton, California, United States of America

  • Tom Newberry,

    Roles Data curation, Formal analysis, Writing – review & editing

    Affiliation University of Salford, Salford, Manchester, United Kingdom

  • Cristiano Lima Pereira,

    Roles Investigation, Writing – review & editing

    Affiliation Instituto de Manejo e Certificação Florestal e Agrícola (IMAFLORA), Piracicaba, Brazil

  • Jorge Perez,

    Roles Writing – review & editing

    Affiliation Alliance of Biodiversity International and International Center for Tropical Agriculture (CIAT), Kasarani, Nairobi

  • Teal Richards-Dimitrie,

    Roles Writing – review & editing

    Affiliation Spatial Informatics Group, LLC, Pleasanton, California, United States of America

  • Ovidio Rivera,

    Roles Investigation, Writing – review & editing

    Affiliation Alliance of Biodiversity International and International Center for Tropical Agriculture (CIAT), Kasarani, Nairobi

  • Beatriz Rodriguez,

    Roles Conceptualization, Writing – review & editing

    Affiliation Alliance of Biodiversity International and International Center for Tropical Agriculture (CIAT), Kasarani, Nairobi

  • Naiara Sales,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – review & editing

    Affiliation University of Salford, Salford, Manchester, United Kingdom

  • Jhon Tello,

    Roles Investigation, Writing – review & editing

    Affiliation Alliance of Biodiversity International and International Center for Tropical Agriculture (CIAT), Kasarani, Nairobi

  • Crystal Wespestad,

    Roles Formal analysis, Investigation, Project administration, Writing – review & editing

    Affiliation Spatial Informatics Group, LLC, Pleasanton, California, United States of America

  • Brian Zutta,

    Roles Conceptualization, Investigation, Methodology, Writing – review & editing

    Affiliation Spatial Informatics Group, LLC, Pleasanton, California, United States of America

  •  [ ... ],
  • David Saah

    Roles Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing – review & editing

    Affiliation University of San Francisco, San Francisco, California, United States of America

  • [ view all ]
  • [ view less ]


Monitoring is essential to ensure that environmental goals are being achieved, including those of sustainable agriculture. Growing interest in environmental monitoring provides an opportunity to improve monitoring practices. Approaches that directly monitor land cover change and biodiversity annually by coupling the wall-to-wall coverage from remote sensing and the site-specific community composition from environmental DNA (eDNA) can provide timely, relevant results for parties interested in the success of sustainable agricultural practices. To ensure that the measured impacts are due to the environmental projects and not exogenous factors, sites where projects have been implemented should be benchmarked against counterfactuals (no project) and control (natural habitat) sites. Results can then be used to calculate diverse sets of indicators customized to monitor different projects. Here, we report on our experience developing and applying one such approach to assess the impact of shaded cocoa projects implemented by the Instituto de Manejo e Certificação Florestal e Agrícola (IMAFLORA) near São Félix do Xingu, in Pará, Brazil. We used the Continuous Degradation Detection (CODED) and LandTrendr algorithms to create a remote sensing-based assessment of forest disturbance and regeneration, estimate carbon sequestration, and changes in essential habitats. We coupled these remote sensing methods with eDNA analyses using arthropod-targeted primers by collecting soil samples from intervention and counterfactual pasture field sites and a control secondary forest. We used a custom set of indicators from the pilot application of a coupled monitoring framework called TerraBio. Our results suggest that, due to IMAFLORA’s shaded cocoa projects, over 400 acres were restored in the intervention area and the community composition of arthropods in shaded cocoa is closer to second-growth forests than that of pastures. In reviewing the coupled approach, we found multiple aspects worked well, and we conclude by presenting multiple lessons learned.


Sustainable agriculture projects, which both generate income and contribute to environmental conservation, are important to address biodiversity loss, climate change, and to improve living conditions [13]. In Brazil, sustainable agriculture approaches have been developed in response to the simultaneous pressures of combatting ongoing habitat and biodiversity loss caused by agricultural expansion and food security issues [410]. Such practices include agroforestry, sustainability certifications, and the promotion of non-timber forest products.

Environmental monitoring is essential to ensure that the promised impacts of sustainable agriculture are being achieved [11] and a growing interest in environmental monitoring provides an opportunity to improve monitoring practices [1,1216]. To improve transparency and enhance credibility, organizations need accurate, timely, and easily digestible information collected using robust methods informed by the best available science. Because the environmental impacts of projects or interventions can take many years to become evident, any monitoring system must be replicable and comparable across time. Monitoring systems should examine multiple scales to account for both individual farm and landscape-scale habitat loss or fragmentation, habitat diversity, and connectivity [7,17,18]. Similarly, to ensure that the measured impacts are due to the project and not exogenous factors, sites where projects have been implemented should be benchmarked against counterfactuals (no projects) and control (natural habitat) sites. Indicators, or predefined metrics for assessing ecosystem services, ecosystem health and biodiversity, can be used to assess relative performance rapidly and can be designed with communication to both experts and non-experts in mind.

Using remote sensing and environmental DNA (eDNA) approaches, systematic, broad-scale, multi-year monitoring efforts are financially and operationally feasible. Traditional approaches to monitoring forests and biodiversity, such as forest surveys and biodiversity transects, are expensive and require significant methodological and taxonomic expertise, particularly in megadiverse regions [19]. In contrast, products derived from remote sensing facilitate a substantial reduction in monitoring costs and simultaneously increase the timeliness of information needed to inform management [2022]. Remote sensing uses satellite and aircraft imagery and statistical approaches to detect and monitor the Earth. Existing remote sensing approaches for monitoring biodiversity allow for the evaluation of ecosystem structure and ecosystem function but are not yet extensively used in biodiversity assessment, monitoring, or conservation [23].

Similarly, eDNA monitoring has speed and cost advantages over traditional methods; thus, it is rapidly becoming a preferred method to monitor biodiversity, including in agricultural systems [2427]. eDNA refers to genetic material obtained directly from environmental samples, such as water, soil, or air, without capturing or observing the organisms themselves [28] and coupled with the metabarcoding approach allows for the simultaneous identification of multiple species from a single sample [29]. eDNA sampling approaches can monitor entire taxonomic groups at multiple spatial scales depending on whether eDNA is collected from leaves, soil, animal waste, water, or air [3035].

Coupling the two technologies allows for both wall-to-wall coverage from remote sensing and the site-specific community composition from eDNA (Fig 1). Together, these methods can be used to calculate diverse sets of indicators for monitoring sustainable agriculture projects and other sustainable interventions. These indicators can include but are not limited to measures of ecosystem function such as landscape scale and site scale habitat loss and conservation, ecosystem structure, carbon sequestration through revegetation, species richness, community composition, and relative changes in community composition over time. The approaches are cost effective, and the results provide a complete picture of activities occurring across the full project site.

Fig 1. An overview of the proposed coupled approach to environmental monitoring.

In the remote sensing component, we identify and pre-process key satellite imagery. Next, we input data into change algorithms or other models. Then, we assess the accuracy of the produced maps. In the eDNA component, we design the sampling approach by identifying locations to sample. Next, we visit sites to collect eDNA data. Then, the soil samples go through the eDNA extraction and biodiversity analysis processes. Finally, we calculate indicators from the map outputs and biodiversity results.

Here, we report on the pilot application of this approach to assess the effect of one of the sustainable agriculture initiatives of Florestas de Valor, created by the Instituto de Manejo e Certificação Florestal e Agrícola (IMAFLORA). Florestas de Valor consists of multiple initiatives including agroforestry and collection of non-timber forest products [36]. We applied the coupled monitoring approach to shaded cocoa projects implemented by IMAFLORA near São Félix do Xingu, in Pará, Brazil. Shaded cocoa in Brazil is currently being supported as an alternative to unshaded cocoa and low-yield pasturelands used for cattle ranching [37,38]. Key research questions included: 1) have the shaded cocoa projects contributed to conservation in the Brazilian Amazon? and 2) what impacts does shaded cocoa have on community structure and forest landscape patterns? We tested the relative effects of these management activities on biodiversity conservation and compared the results with counterfactual “business as usual” pastures and control second-growth forests to control for outside factors.

This implementation of coupling remote sensing and eDNA for biodiversity monitoring provided an opportunity to test multiple approaches and improve the methodology. Key areas of testing included our remote sensing mapping approach, including algorithm selection, sampling design and our data collection approach, and indicator choices. These tests contributed to lessons learned that will greatly improve coupled biodiversity monitoring methods moving forward.

Materials and methods

Study area

The study area is in the Xingu River basin, near the city of São Félix do Xingu, in the state of Pará, Brazil (06°38′30″South and 51°58′32″West). The region is warm moist equatorial, with dry months from May to October. The average annual rainfall is 2,041 mm. The average annual temperature is 25°C, with minimum and maximum temperatures of 20°C and 30°C, respectively [39]. Native tree species in the region include Attalea speciosa (Mart.) and Cedrela odorata L., for example [40]. Primary forest cover was almost entirely removed from the area in the 1960s for agriculture and was slowly replaced by secondary vegetation, including forest. Subsequently, farmers cleared secondary forests for cattle grazing [41,42]. As a result of this history, we refer to “secondary forest” as forest throughout.

More recently, shade-grown cacao (Theobroma cacao L.) has been promoted in the region as a sustainable alternative [4244]. Shaded cocoa is thought to reduce agricultural inputs, disease susceptibility, and drought susceptibility as well as increase food security and environmental benefits [43,44]. There are two phases to cultivating shade trees in cocoa agroforestry systems. First, specific shade trees, including banana and papaya, are cultivated while native regeneration occurs, and second, these cultivated trees are thinned, and native shade trees become dominant [44; Fig 2]. During the first phase, annual crops like cassava and maize are also grown, with cassava chosen to increase the nitrogen content in the soil [44]. Native shade trees include Apuleia leiocarpa (Vogel) J.F. Macbr., Bagassa guianensis Aubl., Pouteria macrophylla (Lam.) Eyma, Erythrina verna Vell., Pouteria pariry (Ducke) Baehni, Chrysophyllum cuneifolium (Rudge) A. DC., Perebea guianensis Aubl., Spondias mombin L., Colubrina glandulosa Perkins, Cenostigma tocantinum Ducke, Annona mucosa Jacq., Handroanthus serratifolius (Vahl) S.Grose, Inga edulis Mart. and Samanea tubulosa (Benth.) Barneby and J. W. Grimes [44].

Fig 2.

Shade grown cocoa agroforestry in the Xingu River basin begins with cultivating shade trees in cocoa agroforestry systems, alongside annual crops and young native trees (left). Over time, the cultivated shade trees are thinned and the native shade trees become dominant (right). Created using components from the University of Maryland Center for Environmental Science Integration and Application Network media library; components copyright Tracey Saxby, Jane Thomas, and Dieter Tracey; under Attribution-ShareAlike 4.0 International.

IMAFLORA is a nonprofit partner in the SERVIR Amazonia consortium based in Brazil. IMAFLORA maintains a database of 150 farms participating in multiple agricultural practices and has worked with a subset of these farmers to implement shaded cocoa practices over the past 20 years [45, Fig 3]. Most of these were implemented and actively growing between 2010 to 2015. Within this context, our study boundary encapsulated the farms partnering with IMAFLORA.

Fig 3. Study area in the state of Pará, Brazil.

Yellow dots represent the farms that have partnered with IMAFLORA.

Remote sensing methods

Disturbance mapping.

Forest disturbances were mapped using a combination of two pixel-based time series mapping methods: the Continuous Degradation Detection (CODED) algorithm [46,47], and the Landsat-based detection of Trends in Disturbance and Recovery (LandTrendr) algorithm [48,49]. Both algorithms utilize Landsat collections in a time series approach (30 m spatial resolution). For this application, disturbance events were classified as deforestation and forest degradation. Deforestation was defined as a permanent conversion of forested land to non-forested land. Degradation was defined as a process that does not lead to a change in land cover but shows a reduction in tree cover canopy [50,51]. The changes were mapped for the 5-year period 2010–2015, when most of the interventions occurred. The results of the two algorithms were combined and evaluated for accuracy.

CODED is a freely available tool on Google Earth Engine (GEE), an online planetary-scale computing platform for remote sensing and satellite imagery analysis [52]. CODED uses all the Landsat imagery available from Landsat collections 5, 7, and 8 to perform a subpixel spectral mixture analysis (SMA), analyzing time series changes in the Normalized Degradation Fraction Index (NDFI; 30 m resolution; [46,50]). The spectral index of choice was the NDFI to be in accordance with CODED and since previous studies have shown that NDFI is more sensitive to disturbances in tropical forests compared to the commonly used Normalized Difference Vegetation Index (NDVI), which tends to show higher variability [4647,53]. We used 3,000 training points labeled as forest or non-forest and tested chi-squared values of both 0.9 and 0.99, which controls the width of the change-detection moving time window [51]. We defined the required number of sequential out-of-range NDFI values to flag an event as four times. In post-processing, we required the magnitude of these flagged disturbances to be above 0.4 to be considered severe enough to include in the final map. In this manner CODED can detect low-severity disturbances, which is often characteristic of the more difficult-to-detect forest degradation events, as opposed to forest loss. The final CODED output is a map with 30 m pixels labeled as non-forest, stable forest, deforestation, or degradation.

Next, we used the LandTrendr algorithm, also freely available and hosted on GEE [54]. LandTrendr is a mapping algorithm used to detect land cover changes through time series analysis of Landsat imagery. This tool aims to filter out inter-annual noise in spectral signals and generate trajectory-based time series estimates and accomplishes this through simplifying multi-year spectral trajectories into several straight-line segments that capture the progressing changes of the signal [48].

The LandTrendr implementation in GEE is connected to an image collection of all Landsat 5, 7, and 8 images. Users select an approach to convert the full time series into an annual value (one image per year) for analysis; we selected medoid composites. The medoid image compositing approach compares each pixel’s spectral band values to the median spectral values of those bands across all images within the date-constrained collection for a given year. The pixel with spectral values closest to the median value, determined by Euclidean spectral distance, is then selected [54]. In this application of LandTrendr, the algorithm was parameterized to estimate the “greatest” disturbance, and specific parameters were selected in accordance with previous studies in the Amazon [55,56]. We added a sub-classification ruleset to the disturbance events from LandTrendr using MapBiomas land cover maps from 2015 [57]. LandTrendr disturbances were classified as degradation if the land cover had returned to forest by 2015 (according to MapBiomas layers); disturbances were classified as forest loss if the land cover in 2015 was non-forest.

Lastly, the final disturbance map was generated by overlapping the CODED and LandTrendr maps. Pixels with classification disagreements between maps were reclassified according to the following rules: if at least one of the maps classified the pixel as degradation, we set the merged layer as degradation; if neither of the maps classified a pixel as degradation and at least one classified it as deforestation we set the merged layer value to deforestation; and if both maps classified the pixel as stable forest we left it as stable forest. We used this rule system because degradation was underestimated in previous studies [51,55,58].

Regeneration mapping.

We also used LandTrendr and the NDFI index to map forest regeneration [56,59]. In this application, we adjust the LandTrendr parameters to return the “greatest” gain and detect upward trends in the spectral signature of forested areas. The other parameters were adjusted to capture short-time regeneration over the 5-year time period from 2010–2015.

We made a regeneration map using MapBiomas data to compare to the map from LandTrendr. Silva Junior and colleagues [60] developed annual maps of secondary forest extent, age, increment, and loss within Brazil for 1986–2019 using maps from the Brazilian Annual Land Use and Land Cover Mapping Project [58]. Secondary forest growth is a conversion from an anthropic dominated cover classification (e.g., pasture or agriculture) to forest cover, excluding mangroves and forest plantations [60]. We used this approach to generate a map of secondary forest cover from 2010 to 2015. The final regeneration map was a combination of the LandTrendr map and the map built using MapBiomas [60]. These maps were merged using the following rules: a pixel is mapped as “regeneration” if at least one of the outputs was mapped as “regeneration.

Accuracy assessment of mapping products.

An independent validation effort was conducted leveraging high- and medium-resolution optical imagery (Planet NICFI mosaics, Google Earth Pro basemaps, Sentinel-2, and Landsat Collections) and ancillary datasets (MapBiomas products, the Global Forest Canopy Height 2019, and NDFI time series) to assess forest change dynamics on the ground [6166]. Visual interpretation was done in Collect Earth Online (CEO), a free and open-source web-based tool that facilitates data collection and validation [67,68]. The interpreters used a decision tree approach for classifying the validation samples.

For the validation of the disturbance and regeneration maps, the sample points were 30 by 30 meters square, mimicking the map output pixel sizes. The validation of both change maps (disturbance and regeneration) was done using the same set of points. We also used a stratified random sampling design. We extracted 600 points from both maps. We had 100 points for each class: degradation, stable non-forest, stable forest, deforestation, regeneration, different events (S1 Table). The two maps, the forest disturbance and forest gain maps, were merged by applying one more rule: pixels with at least two events (loss, degradation, or regeneration) were allocated to a “different events” class. For example, these areas can include regeneration events on non-forest areas followed by degradation or deforestation or disturbance events, most likely deforestation, followed by regeneration. The accuracy metrics (overall, user, and producer accuracies) and unbiased area estimates for each class were calculated through the ratio estimator approach [69] for when the strata are different from the map classes since we used the same sample points for the disturbance map and the regeneration map.

eDNA methods

eDNA field data collection.

We defined our sampling frame as the 150 farms partnering with IMAFLORA. Within this sampling frame, we verified which farms included our project sites (farms with mature shaded cocoa) by checking average vegetation height of cocoa fields (minimum average canopy height of 7.5m) and verifying that fields contained greater than 25% canopy cover in CEO [66]. We identified 49 farms containing mature shaded cocoa.

Our sampling design was chosen to reduce the influence of exogenous variables that influence biodiversity to maximize our statistical power in detecting differences between project and counterfactual sites. Sites were delineated by the cocoa field boundary for project sites, and the pasture boundary for counterfactual sites. Key exogenous variables include the amount and configuration of forest both within the farm boundaries and in a 250m buffer [7074]. To represent these exogenous influences, we selected a suite of landscape ecology metrics that best captured landscape variance using principal component analysis [PCA; 75,76]. Metrics used for clustering included within the farm boundaries: total forest area, number of forest patches, forest percentage of farm, forest contiguity, and forest aggregation index; and within the 250m buffer included: total forest area, forest percentage of landscape, and forest aggregation index [77]. We used these landscape metrics to assign each farm to a cluster using Ward’s hierarchical clustering with the Euclidean distance matrix [78].

Following cluster assignment for each farm, we used stratified random sampling to select five project sites and five counterfactual sites with the same proportion of each cluster in each group of sites. We also used the same process to select two back-up sites for each group. We avoided excluding clusters since that would have changed the sampling frame. In addition, experts from IMAFLORA identified five forest sites that represented the oldest known second growth forest areas in the study area. Permission for all sites was granted by the property owner, and no agency permits were needed for sampling on private property [79]. However, the field team experienced difficulties accessing sites, weather delays, and COVID delays. This led to only one second growth forest being sampled, reduced numbers of plots sampled in one shaded cocoa field and two pastures, and the use of both shaded cocoa backup sites and one pasture backup site.

Within each site, we randomly placed either 3 or 4 50m x 50m plots depending on the field size, with at least 25m between plots. Each plot included four sub-samples, set back from the edge of the plot by 12.5m, to sample the heterogeneity found within the plot. This plot spacing was chosen based on previous research and discussions with local experts [80,81], along with the limitations imposed by field size [82]. Approximately 30g of the topsoil (between 0–5 cm deep) was collected for each sub-sample, making sure to avoid non-soil matter including leaf litter, and sub-samples were pooled to represent one sample per plot. In total, a total of 38 plots were sampled, with 18 plots from 5 shaded-cocoa sites, 17 from 5 pasture sites and 3 from one forest site.

To avoid sample contamination and ensure consistency across samples, a protocol was followed during sampling that included the use of disposable sampling materials and gloves and samples were labeled and individually packed according to plots and sites to ensure no cross-contamination occurred [detail of methods see S2 File; 83,84]. Following collection, soil samples were stored with silica desiccant bags (minimum of 25g per sample) protected from heat and sunlight to prevent DNA degradation, and later transported to the laboratory facilities for further laboratory analysis.

eDNA extraction.

Soil samples were pre-processed in a Department for Environment, Food & Rural Affairs (DEFRA) licensed laboratory facility. To avoid contamination, samples were handled in a pre-PCR laboratory, using disposable tools and gloves, following standard decontamination procedures (i.e., use of bleach to clean surfaces and equipment), and personnel wore disposable full-body suits when handling the samples.

The extraction method was conducted using 2 g of mixed soil (per analyzed plot) and following the Mu-DNA soil DNA extraction protocol described by [85]. Negative controls were included, comprising DNA extraction blanks containing only the required buffers.

Following DNA extractions, DNA amplification was conducted using three sets of primers targeting two partial mitochondrial genes. First, vertebrate specific primers were used targeting ~106 bp of the 12S rRNA gene [86; forward primer 5’-TAGAACAGGCTCCTCTAG-3’ and reverse primer 5’-TTAGATACCCCACTATGC-3’]. Second, to detect arthropods, DNA extracts were amplified using two primer sets targeting different short inserts of the mtDNA COI gene. The Zeale primer set [87] was used to amplify a ~157 bp fragment, and the Gillet primers [8890] were used to amplify a ~133 bp section (Zeale: forward primer 5’-AGATATTGGAACWTTATATTTTATTTTTGG-3’ and reverse primer 5’-WACTAATCAATTWCCAAATCCTCC-3’; Gillet: forward primer 5’-CCATCTCATCCCTGCGTGTCTCCGACTCAGNNNNNNNATTCHACDAAYCAYAA RGAYATYGG-3’ and reverse primer 5’-CCTCTCTATGGGCAGTCGGTGATNNNNNNNACTATAAAARAAAATYTDAYAAA DGCRTG-3’).

PCR reactions consisted of 12.5 μl Master Mix, 7.5 μl molecular grade water, 2 μl of DNA template and 1 μl of the forward and reverse of each primer. The PCR conditions for the Riaz primer followed the [91] methodology, consisting of an incubation of 5 minutes at 95°C, then 35 cycles for 15 seconds at 95°C, 30 seconds at 57°C, ending with 30 seconds at 72°C. PCR conditions of the Gillet and Zeale primers followed the protocols set by [92]. Gillet cycles included an initial 15-minute denaturation at 95°C, then 10 cycles for 30 seconds at 94°C, 45 seconds at 49°C, 30 seconds at 72°C, 30 cycles of 30 seconds at 95°C, 45 seconds at 47°C and 30 seconds at 72°C, with a final extension following of 10 minutes at 72°C. Zeale PCR conditions began with a 15-minute denaturation at 95°C, then 40 cycles of 20 seconds each at 95°C, 30 seconds at 55°C and 1 minute at 72°C, with a final extension of 7 minutes at 72°C. PCR cycles were authenticated by electrophoresis in a 1.2% agarose gel stained with GelRed. PCRs were run in triplicates, and the success of the reactions was determined by electrophoresis on a 1.5% agarose gel. Four PCR blanks were included in each library to account for putative contaminations arising in the amplification steps. In total, 44 samples were analyzed per library, 38 eDNA soil samples, two extraction blanks and four PCR blanks. A left-sided size selection was performed using 1.2× Agencourt AMPure XP (Beckman Coulter) and the KAPA HyperPrep kit (Roche) was used to construct the Illumina libraries using the dual-indexed adapters. Libraries were quantified using the NEBNext qPCR quantification kit (New England Biolabs) and pooled in equimolar concentrations. Two Illumina MiSeq sequencing runs were conducted, one MiSeq v2 Reagent Kit (2 × 150 bp paired-end reads) and one MiSeq v3 Reagent Kit (2 × 300 bp paired-end reads).

Bioinformatic steps were conducted as described in [92]. In brief, bioinformatic analysis used the OBITools 1.2.2 metabarcoding package [93]. Read quality was assessed using FastQC, Illumina paired end aligned paired-end reads, and ngsfilter demultiplexed samples and removed primers. The obigrep command performed size selection by eliminating artifacts and ambiguous reads. Vsearch [94] clustered unique sequences and removed chimeras using uchime-denovo [95]. Sumaclust clustered sequences into Molecular Operational Taxonomic Units (MOTUs) at thresholds of 0.95-0.98. For both primer sets, analyses were processed using the 98% clustering threshold based off empirical results [96,97]. Taxonomic assignment relied on Basic Local Alignment Search Tool (BLAST, specifically blastn) against Genbank, with a minimum of 90% alignment and >80% similarity [98]. Species-level assignment required ≥98% identity, while MOTUs at 95%-98% or with multiple species were assigned at the genus level. MOTUs between 93%-95% were assigned to the family level, and MOTUs between 90%-93% were assigned to the order level [92]. Sequences were retained when they could be identified at least to the Class level. A final filtering step was conducted, including the removal of putative contaminants, tag-jumping (MOTUs represented by less than 0.01% of the total reads were removed from each sample), and non-target taxa (e.g., Human DNA). Additionally, molecular operational taxonomic units (MOTUs) were retained when the total number of reads was over 50 [99]. MOTUs were considered compositional data and treated as such [100102], except for when indicators called for species abundance measures where reads were used [103].

eDNA analysis.

The four initial biodiversity indicators and four proposed indicators were calculated based on the resulting data. For indicators using key species, we defined key species for mammals as threatened native species and excluded domesticated and invasive species from analysis more broadly [104]. We defined key species for arthropods as members of the Hymenoptera and Lepidoptera orders as important pollinator species, including for coffee crops [105,106]. These indicators require BLAST to match sequences to the Order level. As has been previously reported in other studies [e.g., 92], the Gillet primers were notably better at detecting members of Hymenoptera, and the Zeale primers were better at detecting members of Lepidoptera. Due to the limited overlap between the two primer sets, we combined the resulting datasets for further analysis. For determining if any MOTUs were associated with (indicative of) either cocoa fields or pasture and the ecological conditions found there [107,108] we used the multipatt function {indicspecies 1.7.14} with a custom wrapper [109111].

For community-based indicators, we used all sequences identified by BLAST as Arthropods, requiring identification to the Phylum level. We first accounted for zeros in the dataset using zCompositions [112], then transformed the data using compositions [113]. We calculated Aitchison distance between sites and between treatments using the Euclidean distance matrix [26,114,115]. We created PCA plots using the transformed compositional data [114,115].

Where needed, we used linear mixed models to account for the repeated sampling design to test the difference between projects and counterfactuals [116]. Significance tests were performed using ANOVA and Type II Sums of Squares [117]. All data analysis was conducted in R 4.3 [118] and all code and data are available on GitHub (

Carbon calculations

Estimates of carbon sequestration due to land cover transitions to forest and emissions from forest loss were calculated using methods modified from the New IPCC Tier-1 Global Biomass Carbon Map for the Year 2000 [119]. Tier 1 carbon estimates are defined using a look-up table which has an associated carbon value based on the classification scheme of the Global Land Cover 2000 (GLC2000) cover type, ecofloristic zone, continental region, and frontier forest designation [120]. Areas of carbon gain were determined based on areas of gain from the regeneration map product located in the shaded cocoa areas for the 150 farms. Because these farms were previously degraded pasture that were improved to shaded cocoa, we assigned the 2010 land cover to the GLC2000 land cover class of cultivated and managed lands; this was used to assign a carbon factor from the lookup tables (GLC2000; 120) Our site is in the Tropical Rainforest ecofloristic zone [121]. A 50% factor was applied for disturbed vegetation categories [122].

Morphological Spatial Pattern Analysis (MSPA)

To calculate the number of hectares of essential habitat we used Morphological Spatial Analysis (MSPA; 123). A binary image composed of the objects of interest (tree cover) and background and divides it into morphological classes that describe the spatial arrangement of tree habitat across the landscape [123]. For this application, the 8 classes we mapped include patch forest, outer edge, inner edge, core forest, secondary degradation, secondary deforestation, primary degradation, and primary deforestation (for definitions, see S2 File). The analysis consists of a customized sequence of mathematical morphological operators targeted at the description of the geometry and connectivity of the image components. The MSPA segmentation results in 25 mutually exclusive feature classes which, when merged, correspond to the initial foreground tree cover area. In this case, we created a binary image of tree canopy cover using maps from the Global Forest Change dataset [124]. These included the fractional tree cover and forest loss map from 2001 to 2019 [123].


To assess the shaded cocoa projects implemented by IMAFLORA, we conducted a pilot of TerraBio. TerraBio is a methodological framework developed to provide environmental assessment and accountability to private sector firms that (1) commercialize sustainable agriculture and forest products and/or, (2) invest in sustainable business models as profitable and conservation-driven development initiatives. TerraBio directly monitors the land cover change and biodiversity measures annually to provide timely and relevant results for investment funds, businesses, and other parties invested in the success of sustainable agricultural practices [125]. TerraBio uses a coupled approach to environmental monitoring, with the landscape component conducted with remote sensing and the biodiversity component conducted with eDNA, which work together to calculate a series of indicators. Intervention areas were shaded cocoa fields; our counterfactual areas or ‘business as usual’ areas were pasture fields, and our control areas were the areas of naturally regenerated second growth forest.

The initial set of indicators for the TerraBio pilot were derived from existing sources, including the Amazon Biodiversity Fund Brazil Key Performance Indicators (KPI) and the United States Agency for International Development/Brazil’s Partnership for the Conservation of Amazon Biodiversity standard indicators [125127]. These indicators were: Number of hectares directly restored; Number of hectares indirectly conserved; Carbon sequestration through revegetation; Number of keystone/priority species; Change in abundance of keystone/priority species; Change in species richness; Change in biodiversity index; Number of hectares showing improved biodiversity; and Number of hectares of essential habitat area conserved (Table 1).

Table 1. Overview of all indicators calculated for the TerraBio implementation.

In addition to these indicators derived from existing sources, we proposed several new indicators based on the literature [24,25,128]. These proposed biodiversity indicators included: Alpha diversity; Beta diversity; Change in beta diversity due to the intervention; and a qualitative assessment of biodiversity change due to the intervention [25,26,129,130]. Both the indicators themselves and the visualizations for all indicators were designed to be easily understood by a lay audience and for consistency in interpretation over time.


Mapping results

According to the final disturbance mapping results, 183 hectares were degraded, and 358 hectares were deforested within the 150 IMAFLORA properties (11,546 ha) between 2010–2015. Overall, producer’s and user’s accuracies (± 95% confidence interval) per class from the disturbances classified map (Fig 4) are summarized in Table 2. As expected, the stable classes (stable forest and non-forest) had higher user’s and producer’s accuracies and lower uncertainties compared to the dynamic classes (degradation and deforestation). Nevertheless, omission and commission errors happened within and across these two groups of classes (stable and dynamic classes). For example, in the dynamic classes (degradation and deforestation): most incorrectly classified pixels in the degradation class were omitted in the deforestation class and committed to the stable forest class, and both omission and commission errors in the deforestation class happened in the degradation class. The reasons behind these errors are explained in the discussion section.

Fig 4.

(A) Disturbance classification 2010–2015. Inset maps (B: B1, B2, B3) are shown with black outline and inset maps (C: C1, C2) are shown with blue outline on the main map. (B1) Inset map of disturbance classification 2010–2015 over a particular area to outline classification vs. RGB images. (B2) Inset RGB image pre-study period from Landsat 5 (July 30, 2009). (B3) Inset RGB image post-study period from Sentinel-2 (June 26, 2016). (C1, C2) Example of disturbances within one of the properties (degradation in yellow, deforestation in red, stable forest in dark green, and non-forest in gray). overlayed on the RGB Landsat 5 image (July 30, 2009).

Overall, producer’s and user’s accuracies (± 95% confidence interval) from the regeneration classified map (Fig 5) are summarized in Table 3. As expected, our approach performed better at identifying where regeneration did not occur compared to where regeneration did occur. The regeneration class presented low user’s and producer’s accuracies and higher levels of uncertainties. The biggest source of error was pixels being mapped as regeneration when they should not have been, leading to an overestimation of areas where regeneration occurred.

Fig 5.

(A) regeneration map output 2010–2015 overlayed on Sentinel-2 imagery (ESA). Inset maps (B: B1, B2, B3) are shown with black outline and inset maps (C: C1) are shown with blue outline on the main map. (B1) Inset map of regeneration output 2010–2015 over a particular area to outline regeneration areas vs. RGB images. (B2) Inset RGB median image pre-study period from Landsat 5 (July-August 2011). (B3) Inset RGB image post-study period from Planet NICFI mosaic (July—November 2016). (C1) Example of regeneration areas (in green) within two of the properties.

eDNA results

Overall, five cocoa sites and five pasture sites, in addition to one second growth forest site, were surveyed by the field team. While the paired sampling design was not maintained, all but one of the fields sampled came from cluster 1 so the reduction in exogenous variation should be maintained (S3 Table).

Upon the initial bioinformatics filtering steps, a yield of 14,387,489 sequencing reads was obtained from the two sequencing runs, with 10,775,915 reads for the invertebrates dataset (5,598,018 Gillet, and 5,177,897 Zeale) and 3,611,574 for vertebrates. After applying all taxonomic filtering steps and retrieving only target MOTUs, the number of reads retained were 2,665,095 reads for arthropods including both Gillet and Zeale primer sets, and 38,806 reads for vertebrates after the removal of human and domestic animal reads. For the vertebrates dataset, a very high proportion of human reads was detected (>75% of total reads, from 6.9 million mammal reads, 5.2 million belonged to humans), with 14 MOTUs from 10 unique families identified when considering only the target wild mammalian taxa. For arthropods, 290 MOTU from 27 unique identified families were retained after the removal of MOTUs that were not assigned any taxonomic information. Due to the low amount of reads and MOTU returned for vertebrates, further analyses focused on the arthropod dataset (S3 File).

Indicator results

Land Cover/ Land Use Indicator 1: Number of hectares indirectly conserved.

We found 2,871 hectares within the 150 IMAFLORA properties (excluding intervention sites) were indirectly conserved between 2010–2015.

Land Cover/ Land Use Indicator 2: Number of hectares directly restored.

We found 471 hectares within the intervention sites (areas of shaded-grown cocoa) of the 150 farms had regeneration between 2010–2015.

Land Cover / Land Use Indicator 3: Carbon sequestration through revegetation—net positive climate impact annually.

We found carbon sequestration through revegetation in shade-grown cocoa systems in the intervention farms of 44,300 Mg C (8,860 Mg C/yr) between 2010 and 2015.

Biodiversity Indicator 1: Number of key species due to intervention.

Overall, we found a total of 19 key MOTUs, including 11 in Hymenoptera and 8 in Lepidoptera. Most Hymenoptera belonged to the family Formicidae (ants), while Lepidoptera belonged to families Saturniidae and Crambidae, among others. Only one group of butterflies was detected (Hermeuptychia hermes [Fabricius], or the Hermes satyr).

One member of Hymenoptera was found in both cocoa fields and pasture. Five MOTU were found only in cocoa fields, including Labidus sp. (army ants), Solenopsis sp. (fire ants), Hileithia sp. (moth), a member of family Platygastridae and a member of Hymenoptera for which further identification was not possible. Thirteen MOTU were found only in pastures, including Crematogaster abstinens [Forel] and two other Crematogaster species, Solenopsis geminata [Fabricius] and another Solenopsis species, Argyria sp., Heliura sp., Hermeuptychia hermes, Hylesia sp. Clepsis sp., and a member of Hymenoptera for which further identification was not possible. No Hymenoptera or Lepidoptera were detected in forests. No species were consistently found to be indicator species.

Biodiversity Indicator 2: Change in abundance of keystone/ priority species due to interventions.

As this pilot only included one time period, we compared the intervention to the counterfactual for this one time period. We found that there was also no significant difference in key species abundance when comparing cocoa fields (intervention) and pasture (counterfactuals; Pr(>Chisq) = 0.2648).

Biodiversity Indicator 3: Change in species richness due to interventions.

The mean species richness in cocoa fields was 17.9 species (SD = 11.0), while on pastures mean species richness was 18.2 (SD = 9.4). When comparing total species richness between cocoa fields and pasture, we found there was no significant difference in MOTU richness (Pr(>Chisq) = 0.978; Fig 6).

Fig 6. Species richness by field type.

Each dot represents a sample point in shaded cocoa fields (red) or pasture fields (blue).

Biodiversity Indicator 4: Change in biodiversity indices due to interventions.

When comparing diversity indices between shaded cocoa fields (intervention) and pasture (counterfactuals), we found that there were no significant differences for Shannon diversity (Pr(>Chisq) = 0.3639) or for Simpson’s diversity (Pr(>Chisq) = 0.2964; Fig 7).

Fig 7.

Shannon diversity (left) and Simpson’s (right) by field type.

Biodiversity Indicator 5: Alpha diversity.

When comparing diversity indices between cocoa fields (intervention) and pasture (counterfactuals), we found that there were no significant differences for species richness (Hill’s q = 0; previous section), for effective species richness (Hill’s q = 1; Pr(>Chisq) = 0.4824), and for inverse Simpson’s (Hill’s q = 2; Pr(>Chisq) = 0.5243; Fig 8).

Fig 8. Effective species richness (L) and Inverse Simpson Diversity (R) by field type.

Each dot represents a sample point in shaded cocoa fields (red) or pasture fields (blue).

Biodiversity Indicator 6: Beta diversity.

Pairwise distances between all sampling sites ranged from 10.3 and 35.1, where 0 represents no dissimilarity, and larger distances indicate increasing dissimilarity (Fig 9). The highest dissimilarities were observed between the Cocoa 1 field and the Pasture fields.

Fig 9. Beta diversity (Aitchison distance) between each site.

Blue corresponds to lower dissimilarities between sites and yellow corresponds to higher dissimilarities between sites.

Biodiversity Indicator 7: Change in beta diversity due to the intervention.

The Aitchison distance between shaded cocoa (intervention) and pasture (BAU) was about equal to the distance between shaded cocoa (intervention) and forest (control; 16.6 and 16.8 respectively), but much smaller than the distance between pasture (BAU) and forest (control; 23.2), though this difference was not significant [131; Fig 10].

Fig 10. Aitchison distances within treatments and between treatments.

Biodiversity Indicator 8: Qualitative assessment of change in biodiversity due to the intervention

PCA graphs of plots and sites revealed that cocoa and forest sites were co-located in site-species space, while pastures were strongly separated along the first axis (Fig 11). This agrees with our findings in Biodiversity Indicator 7.

Fig 11. Arthropod PCA for arthropods found in shaded cocoa, forest, and pasture sites.

Landscape Integrity Indicator 1: No. of hectares of essential habitat area conserved

Essential habitat areas calculated using MSPA within all farm properties with shaded cocoa included 93 ha of core forest, 114 ha of patch forest, 82 ha of stable inner forest edge, and 1,000 ha of stable outer forest edge (buffer habitat; S2 Table). Thus, a total of 289 ha can be considered critical habitat within the intervention farms, and an additional 1,000 ha is stable buffer habitat.


Piloting the TerraBio framework

The pilot application of TerraBio, a coupled eDNA and remote sensing environmental monitoring approach, demonstrated the potential for such systems in monitoring sustainable agriculture such as the shaded cocoa fields implemented by IMAFLROA. Overall, our remote sensing analysis suggests that the shaded cocoa established by IMAFLORA helped revegetate over 400 hectares, and the eDNA analysis suggests that the community composition of arthropods in shaded cocoa is closer to second growth forests than that of pastures.

Based on the indicators, the broader impacts of shaded cocoa in the study area are most likely increased canopy cover, increased carbon sequestration, and more ‘forest-like’ habitat availability for arthropods. Land Cover/Land Use Indicators 1 and 2 both found support for the shaded cocoa indicators directly and indirectly restoring forest cover. Land Cover/Land Use Indicator 3 similarly found a net gain due to carbon sequestration by the shade canopy and cocoa in the intervention farms. While the results from the initially provided biodiversity indicators and vertebrates were inconclusive, the proposed biodiversity indicators found that arthropod communities in shaded cocoa fields were closer to forests than arthropod communities in pastures were to forests. This suggests that the habitat available to arthropods in shaded cocoa was more ‘forest-like’ than the pastures due to the interventions, in agreement with Landscape Integrity Indicator 1.

Importantly, our results agree with previous studies on the impacts of shaded cocoa. In Ethiopia, researchers found that shade coffee certification increased the probability of forest conservation by 19.3% [132] and that indirect forest conservation was also observed within 100m of the project areas [133]. Similarly, the contribution of shaded cocoa to biodiversity conservation viewed through the lens of retaining forest-like communities echoes decades of previous research [134137]. Contributions of shaded cocoa systems to landscape connectivity are also well supported [136,138], though systems emphasizing native trees are likely more successful than those using bananas and Erythrina fusca [139]. In Brazil’s Atlantic Forest, traditional agroforests where cacao is planted under thinned native forests called cabrucas, have greater diversity of tree species, including forest specialist tree species, and while they are not substitutes for undisturbed forest they do have a critical role in biodiversity conservation [140].

Learning from the pilot implementation

Lessons for remote sensing.

Implementing the pilot allowed us to both improve TerraBio and provide guidance for other coupled monitoring approaches for the future. In general, we found multiple aspects of our approach worked well, including robust field sampling using random samples, using the management unit as the unit of analysis, and benchmarking against a control and/or business as usual land use.

With the remote sensing component of the coupled methodology, we encountered some challenges specific to the complex agroecosystems of the Brazilian Amazon, highlighting the need for careful integration of region-specific remote sensing knowledge in developing coupled monitoring approaches. These were detected during the accuracy assessment, highlighting the importance of this step.

First, to calculate the Land Cover/Land Use and Landscape Integrity indicators, we created two map products. While the methods used to create these map products have significant support in the literature [e.g., 51,141,142], we found that applying them in this specific context had wide margins of error and some unexpected results that provided an opportunity to learn and improve upon these methods. For example, our disturbance maps, which were used for LCLU Indicator 2, showed omission errors in the stable forest class. In accordance with [143], many of the omission errors associated with this class were derived from the presence of deciduous tree species (“caducifólias’’, in Portuguese) in this region, which show a seasonal leaf color change, leaf-off pattern (Fig 12), and changes in NDFI values. Therefore, the algorithms assume a disturbance event in forested lands, classifying most of these pixels as degradation, which explains the low accuracy obtained for these classes. Both CODED and LandTrendr should be able to capture seasonal variations of forests with varying crown covers, and parametrization for local conditions can mitigate this issue [46,47,54].

Fig 12. Plot in the CEO disturbances validation project.

This pixel was misclassified as Degradation. NDFI time series show a seasonal pattern with lower NDFI values around August of each year. It is important to note that the MapBiomas classification maps these areas as “Savannic Forest Formations”. Basemap imagery is from Sentinel-2 (European Space Agency).

Further, it is important to fully understand the forest conversion process and pastureland management practices in the region, as this greatly influences the interpretation itself of disturbance samples for validation. Many of the samples analyzed represented patches of degraded forest cover or “dirty pastures”. Some of the degraded patches presented recurrent regeneration and disturbance patterns, visible in the NDFI time series due to fragmentation or border effect (Fig 13). This is sometimes not captured in the visual interpretation analysis, where just one change event is recorded, which yields omission or commission errors for the change classes (e.g., regeneration and degradation, deforestation, and degradation). For future applications, it will be important to separate the regeneration output validation from the disturbance output validation to refine confidence intervals.

Fig 13. Example plot of the CEO change validation project showing a fragmented/degraded forested patch and its variations on the NDFI time series.

Basemap imagery is from Sentinel-2 (European Space Agency).

A better understanding of the degradation process and its relationship with deforestation is vital for the decision-making process of interpreting validation samples. Many times, the disturbances happen gradually, in phases, until they reach the final clear-cut stage of deforestation. The relationship between degradation and deforestation may vary significantly across the different land tenures. The same happens with the regeneration process, which will be characterized by secondary forests with different stages depending on how many years of recovery we are seeing. The interpretation of what is happening on the ground is not always clear and straightforward, especially when the interpretation is being made through satellite imagery (Fig 14). In a span of five years, we may observe two or three different events (e.g., selective logging followed by fire and then regeneration). Without the availability of Planet NICFI data prior to 2015, the visual inspection of these events through Landsat imagery and NDFI time series interpretation was challenging, which could have yielded validation errors.

Fig 14. Example plot of the CEO change validation project showing a plot that was classified as regrowth by the algorithm, but the interpreter classified it as a single degradation event.

We note some regreening between the dates and the variation in NDFI values. Although not entirely clear, the interpreter assumed selective logging followed by a fire event in 2011. Another fire event seems to have happened in 2012. It is not clear that by the end of 2014 the area was already-established pastureland.

Finally, our estimates of carbon sequestration due to revegetation do not align well with those that have been estimated previously. For example, potential sequestration for smallholder agroforestry systems like shaded cocoa has been estimated at 1.5 to 3.5 Mg per year per hectare [144]. This rate suggests that we should expect approximately 6,000 Mg C sequestration due to revegetation over the study period based on our regeneration maps. Our overestimate is caused by the approach’s treatment of regeneration as entirely occurring within the study time period, which is unrealistic. For future applications of TerraBio, we plan to adapt the carbon calculation method by [145]. We will use forest stand age calculated by the LandTrendr algorithm to create more accurate estimates of carbon sequestration through revegetation by more realistically accounting for the rate of growth and thus rate of sequestration.

We also recognize similar limitations with the MSPA calculations by using the Global Forest Change dataset. We plan to use the LandTrendr product and simplify the MSPA classes. By applying the existing LandTrendr product, and therefore, more local-based information, to both methods we expect the results to have higher accuracy, which can be supported by confidence intervals for the case of the carbon estimates.

Lessons for monitoring with eDNA.

The success of using arthropod communities as indicators focused more on community composition points to some lessons learned for future coupled biodiversity monitoring approaches and implementations of TerraBio. The arthropod datasets did not encounter the same issues with high numbers of human reads and domesticated species, likely associated with the anthropogenic land uses [146,147], and other research has found bulk soil analysis of arthropod communities to accurately estimate diversity [148]. Combined with their ecological importance, arthropods are thus a promising taxonomic group to use as a target moving forward [105,148,149]. Both Gillet and Zeale primers succeeded in capturing different parts of the Arthropoda phylum, however most reads obtained belonged to non-target taxa, and for future studies, we recommend that both primer sets be used if both Hymenoptera and Lepidoptera groups are targeted taxa.

From the reads attributed to Arthopoda, a significant fraction of detected MOTUs could not be successfully assigned at the taxonomic rank required for the downstream analyses. The lack of resolution at short fragments associated with eDNA monitoring and sparse or incomplete reference databases is a well-known issue in understudied regions such as the Neotropics [150]. For the “key species” indicators requiring identification to species, many key arthropod species were present in only one plot. Thus, when monitoring these indicators over time, as TerraBio plans to do, researchers must either use an outdated database or constantly update previous years’ results to separate the effect of improving databases from any real biological change. Further, due to the fundamental character of eDNA data, some traditional indicators such as species abundance calculated with eDNA data can be misleading [128].

This points to one key benefit of using indicators focused on overall community composition over those focused on species specific (e.g., keystone, rare, endangered, or endemic species) used elsewhere. The initial set of indicators selected for use in the pilot implementation of TerraBio were derived from existing indicators based on traditional field ecology methods where organisms are directly observed. However, eDNA represents a fundamental shift in how ecological communities are measured. Any taxonomy-based indicator requires accurate databases linking genomic sequences with their taxonomic identity [19,25]. New approaches to ecological indicator selection can better leverage eDNA data to monitor communities using taxonomy-free approaches [24,25,128].

Our proposed indicators, chosen specifically to take advantage of eDNA data and to still be interpretable by lay people, were more successful at identifying patterns separating communities found in the intervention, even when differences were not significant due to low power. For example, Biodiversity Indicators 2–5 found no meaningful differences between the shaded cocoa and pasture sites, however, the community focused Biodiversity Indicators 7 and 8 identified that the arthropod communities in cocoa fields were more similar to forest plots than the pasture sites. Using these community ecology-based taxonomy-free approaches to fully make use of eDNA data is an important lesson learned for coupled environmental monitoring approaches, including TerraBio. However, when implementing this approach to indicators, they must be clearly communicated to stakeholders accustomed to the outputs of traditional field ecology methods. Overall, eDNA is well suited to monitoring these projects, but successful integration in monitoring, reporting, and verification (MRV) standards will require balancing the best available science with reporting requirements.

We also encountered some challenges with our sampling protocols when implemented in remote areas by field partners and packaging scientific best practices in a way that is easily accessible and actionable. During data collection, our field team ran into multiple practical issues while collecting soil samples using the plot sample design. Having specific areas to collect from proved very time consuming, which when added to the already long transit times between field locations resulted in reduced data collection. The size and shape of the fields also limited the field team’s ability to change the location of plots if necessitated by field conditions [79]. These challenges will become increasingly important when farm owners and community scientists are collecting data.

Thus, we suggest using a simplified larger volume sampling approach, and will be moving to this in future implementations of TerraBio. We are moving away from a plot-based approach with small amounts of soil collected to an approach collecting large volumes of soil from the entire site [12 liters; 151]. Recent research suggests that large soil volumes are likely needed to accurately capture community representation [e.g., 151154]. This sampling approach will also be significantly easier for farmers and community scientists to implement, and early testing is promising.

Further, during our data analysis, we noticed that fewer reads were returned for cocoa fields than for pasture fields, with multiple plots missing data entirely or exhibiting very low species richness (i.e., number of detected species, S3 Table). One potential issue was that our field sites were remote. This limited the viable approaches for sample preservation, in contrast to other studies using soil in Amazonia where either -40°C freezer storage or ethanol was available [e.g. 155], or samples could be analyzed hours after collection [156].

Additionally, differences in soil moisture may have accounted for this result, as pastures were generally drier than shaded cocoa fields and forests. If the desiccant volume was not sufficient to handle higher soil moistures, then sample degradation may have occurred unevenly between the more shaded and thus wetter cocoa and forest samples and the drier pasture samples [147,157; though see 158]. We will be testing higher volumes of desiccant and Longmire’s solution, as many of the sustainable agricultural projects occur in remote areas and tropical or sub-tropical climates [157,159]. Early results suggest this approach may help with high moisture soils, allowing for more accurate comparisons between land uses.


Approaches that directly monitor land cover change and biodiversity on an annual basis by coupling remote sensing and environmental DNA (eDNA) can provide timely and relevant results for parties interested in the success of sustainable agricultural practices. Monitoring information collected from these approaches serves two essential functions: assessing the effectiveness of project-level management actions and approaches and facilitating ongoing learning about the circumstances in which different approaches outperform others.

In this pilot, we found that shaded cocoa projects implemented by IMAFLORA contributed to forest and biodiversity conservation in the Brazilian Amazon. The broader impacts of shaded cocoa projects in the study area, as revealed through the indicators, are most likely increased canopy cover, increased carbon sequestration, and more ‘forest-like’ habitat availability for arthropods. Following the successful pilot application of TerraBio to assess the effect of shaded cocoa interventions near São Félix do Xingu, in Pará, Brazil on biodiversity conservation, the approach will be expanded to other projects created through innovative funding mechanisms.

Implementing this pilot allowed us to provide guidance for coupled monitoring approaches for the future, including TerraBio. In general, we found multiple aspects of our coupled approach worked well. These included our sampling design using randomly selected farms and using the management unit as the unit of analysis. Benchmarking the intervention against a control and business as usual land use was a unique and cost-effective way to validate results on the project level while controlling for exogenous variables. Using remote sensing and eDNA data collection in tandem to calculate indicators also provided a more holistic view than either alone.

For our remote sensing analysis, we found that detecting forest disturbances and regeneration was challenging due to regional land management practices and vegetation characteristics, suggesting better algorithm parametrization to local conditions is needed to improve future accuracy. Understanding the degradation process and its relationship with deforestation is also vital for interpreting validation samples. Additionally, more realistic approaches to forest growth for carbon calculation analysis are needed. For our eDNA analysis, we suggest moving to straightforward sampling designs using high volume sampling and replication. We also found that taxonomy-free arthropod community focused indicators were more successful at illuminating ecologically holistic differences between intervention (shaded cocoa) and business as usual (pasture) scenarios.

Supporting information

S1 Table. Sample sizes for change map validation.

Sample size per class for validation efforts. The change maps (disturbance map and regeneration map) were combined into one map and the same sample dataset was used for accuracy assessment.


S2 Table. Detailed MSPA results.

Morphological Spatial Pattern Analysis of forest change dynamics from 2001 to 2019 in the study area.


S3 Table. Biodiversity results by site.

Planned sampling and plots returning usable data after filtering.


S1 File. MPSA definitions.

A detailed description of all MPSA definitions.


S2 File. eDNA soil sampling protocol.

Detailed protocol for eDNA soil sampling.


S3 File. Vertebrate results.

Indicator results from the vertebrate primers.



We want to thank the farmers who allowed us to sample their fields for the eDNA work and Eric Bullock, for his expert advice on the use of his CODED algorithm in the Amazon.


  1. 1. Bhagwat SA, Willis KJ, Birks HJ, Whittaker RJ. Agroforestry: a refuge for tropical biodiversity? Trends in ecology & evolution. 2008 May 1;23(5):261–7. pmid:18359125
  2. 2. Gomiero T, Pimentel D, Paoletti MG. Is there a need for a more sustainable agriculture?. Critical reviews in plant sciences. 2011 Jan 1;30(1–2):6–23.
  3. 3. Ruggerio CA. Sustainability and sustainable development: A review of principles and definitions. Science of the Total Environment. 2021 Sep 10;786:147481. pmid:33965820
  4. 4. Perfecto I, Armbrecht I, Philpott SM, Soto-Pinto L, Dietsch TV. Shaded coffee and the stability of rainforest margins in northern Latin America. Stability of tropical rainforest margins: Linking ecological, economic and social constraints of land use and conservation. 2007:225–61.
  5. 5. Martinelli LA, Naylor R, Vitousek PM, Moutinho P. Agriculture in Brazil: impacts, costs, and opportunities for a sustainable future. Current Opinion in Environmental Sustainability. 2010 Dec 1;2(5–6):431–8.
  6. 6. Sano EE, Rosa R, Brito JL, Ferreira LG. Land cover mapping of the tropical savanna region in Brazil. Environmental monitoring and assessment. 2010 Jul;166:113–24. pmid:19504057
  7. 7. Ferreira J, Pardini R, Metzger JP, Fonseca CR, Pompeu PS, Sparovek G, et al. Towards environmentally sustainable agriculture in Brazil: challenges and opportunities for applied ecological research. Journal of applied Ecology. 2012 Jun;49(3):535–41.
  8. 8. Lewis SL, Edwards DP, Galbraith D. Increasing human dominance of tropical forests. Science. 2015 Aug 21;349(6250):827–32. pmid:26293955
  9. 9. Jezeer RE, Verweij PA, Santos MJ, Boot RG. Shaded coffee and cocoa–double dividend for biodiversity and small-scale farmers. Ecological economics. 2017 Oct 1;140:136–45.
  10. 10. Curtis PG, Slay CM, Harris NL, Tyukavina A, Hansen MC. Classifying drivers of global forest loss. Science. 2018 Sep 14;361(6407):1108–11. pmid:30213911
  11. 11. Credit Suisse and McKinsey. Conservation finance: From niche to mainstream: The building of an institutional asset class. Credit Suisse, Zurich, 2016.
  12. 12. Thompson BS. Impact investing in biodiversity conservation with bonds: An analysis of financial and environmental risk. Business Strategy and the Environment. 2023 Jan;32(1):353–68.
  13. 13. Tropek R, Sedláček O, Beck J, Keil P, Musilová Z, Šímová I, et al. Comment on “High-resolution global maps of 21st-century forest cover change”. Science. 2014 May 30;344(6187):981–.
  14. 14. Milder JC, Arbuthnot M, Blackman A, Brooks SE, Giovannucci D, Gross L, et al. An agenda for assessing and improving conservation impacts of sustainability standards in tropical agriculture. Conservation biology. 2015 Apr;29(2):309–20. pmid:25363833
  15. 15. Hill SL, Arnell A, Maney C, Butchart SH, Hilton-Taylor C, Ciciarelli C, et al. Measuring forest biodiversity status and changes globally. Frontiers in Forests and Global Change. 2019 Nov 29;2:70.
  16. 16. Threlfall R, King A, Shulman J, Bartels W. The time has come: The KPMG Survey of sustainability reporting 2020. KMPG IMPACT: Singapore. 2020:63. Available online at Accessed 6/22/23.
  17. 17. Edwards DP, Fisher B, Wilcove DS. High Conservation Value or high confusion value? Sustainable agriculture and biodiversity conservation in the tropics. Conservation Letters. 2012 Jan;5(1):20–7.
  18. 18. Tscharntke T, Milder JC, Schroth G, Clough Y, DeClerck F, Waldron A, et al. Conserving biodiversity through certification of tropical agroforestry crops at local and landscape scales. Conservation Letters. 2015 Jan;8(1):14–23.
  19. 19. Beng KC, Corlett RT. Applications of environmental DNA (eDNA) in ecology and conservation: opportunities, challenges and prospects. Biodiversity and Conservation. 2020 Jun;29:2089–121.
  20. 20. Slough T, Kopas J, Urpelainen J. Satellite-based deforestation alerts with training and incentives for patrolling facilitate community monitoring in the Peruvian Amazon. Proceedings of the National Academy of Sciences. 2021 Jul 20;118(29):e2015171118. pmid:34253599
  21. 21. Townsend PA, Clare JD, Liu N, Stenglein JL, Anhalt‐Depies C, Van Deelen TR, et al. Snapshot Wisconsin: networking community scientists and remote sensing to improve ecological monitoring and management. Ecological Applications. 2021 Dec;31(8):e02436. pmid:34374154
  22. 22. Zhang T, Zhang W, Yang R, Liu Y, Jafari M. CO2 capture and storage monitoring based on remote sensing techniques: A review. Journal of Cleaner Production. 2021 Jan 25;281:124409.
  23. 23. Reddy C.S. Remote sensing of biodiversity: what to measure and monitor from space to species? Biodiversity and Conservation, 30(10), 2021. pp.2617–2631.
  24. 24. Leese F, Bouchez A, Abarenkov K, Altermatt F, Borja Á, Bruce K, et al. Why we need sustainable networks bridging countries, disciplines, cultures and generations for aquatic biomonitoring 2.0: a perspective derived from the DNAqua-Net COST action. In advances in ecological research 2018 Jan 1 (Vol. 58, pp. 63–99). Academic Press.
  25. 25. Cordier T, Alonso‐Sáez L, Apothéloz‐Perret‐Gentil L, Aylagas E, Bohan DA, Bouchez A, et al. Ecosystems monitoring powered by environmental genomics: a review of current strategies with an implementation roadmap. Molecular Ecology. 2021 Jul;30(13):2937–58. pmid:32416615
  26. 26. Ladin ZS, Ferrell B, Dums JT, Moore RM, Levia DF, Shriver WG, et al. Assessing the efficacy of eDNA metabarcoding for measuring microbial biodiversity within forest ecosystems. Scientific Reports. 2021 Jan 15;11(1):1629. pmid:33452291
  27. 27. Kestel J.H., Field D.L., Bateman P.W., White N.E., Allentoft M.E., Hopkins A.J., et al. Applications of environmental DNA (eDNA) in agricultural systems: Current uses, limitations and future prospects. Science of the Total Environment, 847, 2022. p.157556. pmid:35882340
  28. 28. Taberlet P, Prud’homme SM, Campione E, Roy J, Miquel C, Shehzad W, et al. Soil sampling and isolation of extracellular DNA from large amount of starting material suitable for metabarcoding studies. Molecular ecology. 2012 Apr;21(8):1816–20. pmid:22300434
  29. 29. Deiner K, Bik HM, Mächler E, Seymour M, Lacoursière‐Roussel A, Altermatt F, et al. Environmental DNA metabarcoding: Transforming how we survey animal and plant communities. Molecular ecology. 2017 Nov;26(21):5872–95. pmid:28921802
  30. 30. Thomsen P. F. & Willerslev E. Environmental DNA—An emerging tool in conservation for monitoring past and present biodiversity. Biol. Conserv. 2015 183, 4–18.
  31. 31. Sales N.G., McKenzie M.B., Drake J., Harper L.R., Browett S.S., Coscia I., et al. Fishing for mammals: Landscape‐level monitoring of terrestrial and semi‐aquatic communities using eDNA from riverine systems. Journal of Applied Ecology. 2020 57(4), pp.707–716.
  32. 32. Nørgaard L., Olesen C.R., Trøjelsgaard K., Pertoldi C., Nielsen J.L., Taberlet P., et al. eDNA metabarcoding for biodiversity assessment, generalist predators as sampling assistants. Scientific reports, 2021 11(1), pp.1–12.
  33. 33. Banerjee P., Stewart K.A., Dey G., Antognazza C.M., Sharma R.K., Maity J.P., et al. Environmental DNA analysis as an emerging non-destructive method for plant biodiversity monitoring: a review. AoB Plants. 2022 14(4), p.plac031. pmid:35990516
  34. 34. Altermatt F, Carraro L, Antonetti M, Albouy C, Zhang Y, Lyet A, et al. Quantifying biodiversity using eDNA from water bodies: General principles and recommendations for sampling designs. Environmental DNA. 2023 May 16.
  35. 35. Aucone E, Kirchgeorg S, Valentini A, Pellissier L, Deiner K, Mintchev S. Drone-assisted collection of environmental DNA from tree branches for biodiversity monitoring. Science Robotics. 2023 Jan 18;8(74):eadd5762. pmid:36652506
  36. 36. Imaflora. Florestas de Valor: the bioeconomy that generates income and preserves a standing forest. 2019. Accessed 7/11/2023. Available online at:
  37. 37. Johns ND. Conservation in Brazil’s chocolate forest: the unlikely persistence of the traditional cocoa agroecosystem. Environmental Management. 1999;23(1):31. pmid:9817770
  38. 38. Gramacho I. C. P., Magno A. E. S., Mandarino E. P., and Matos A. (1992). Cultivo e beneficiamento do cacau. Ilhéus, CEPLAC–MAPA, 66–73.
  39. 39. Instituto Nacional de Meteorologia (INMET). Banco de dados meteorológicos. Accessed on April 27, 2022.
  40. 40. Carneiro V., Sablayrolles P., Oliveira C., & das Graças Sablayrolles M. Composição e Diversidade Florística de Remanescentes Florestais Presentes em Lotes de Agricultores na Área de Proteção Ambiental Triunfo do Xingu, Pará, Brasil. Enciclopédia Biosfera, 2012 Nov 30;8(15).
  41. 41. Mertens B, Poccard‐Chapuis R, Piketty MG, Lacques AE, Venturieri A. Crossing spatial analyses and livestock economics to understand deforestation processes in the Brazilian Amazon: the case of Sao Felix do Xingu in South Para. Agricultural economics. 2002 Nov;27(3):269–94.
  42. 42. Schmink M, Hoelle J, Gomes CV, Thaler GM. From contested to ‘green’frontiers in the Amazon? A long-term analysis of São Félix do Xingu, Brazil. The Journal of Peasant Studies. 2019 Feb 23;46(2):377–99.
  43. 43. Schroth G, Garcia E, Griscom BW, Teixeira WG, Barros LP. Commodity production as restoration driver in the Brazilian Amazon? Pasture re-agro-forestation with cocoa (Theobroma cacao) in southern Pará. Sustainability Science. 2016 Mar;11:277–93.
  44. 44. Braga DP, Domene F, Gandara FB. Shade trees composition and diversity in cacao agroforestry systems of southern Pará, Brazilian Amazon. Agroforestry Systems. 2019 Aug 15;93:1409–21.
  45. 45. Neves, Andressa, Nachtergaele, Marcos Froes, and Gonçalves, Eduardo Trevisan. São Félix do Xingu: farmers lead exchange of experiences in cocoa management workshops. Available online at: 2015.Accessed 9/7/2022.
  46. 46. Bullock EL, Nolte C, Reboredo Segovia AL, Woodcock CE. Ongoing forest disturbance in Guatemala’s protected areas. Remote sensing in ecology and conservation. 2020 Jun;6(2):141–52. pmid:32617175
  47. 47. Bullock EL, Woodcock CE, Olofsson P. Monitoring tropical forest degradation using spectral unmixing and Landsat time series analysis. Remote sensing of Environment. 2020 Mar 1;238:110968.
  48. 48. Kennedy RE, Yang Z, Cohen WB. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1. LandTrendr—Temporal segmentation algorithms. Remote Sensing of Environment. 2010 Dec 15;114(12):2897–910.
  49. 49. Kennedy RE, Cohen WB, Schroeder TA. Trajectory-based change detection for automated characterization of forest disturbance dynamics. Remote Sensing of Environment. 2007 Oct 15;110(3):370–86.
  50. 50. Souza CM Jr, Roberts DA, Cochrane MA. Combining spectral and spatial information to map canopy damage from selective logging and forest fires. Remote Sensing of Environment. 2005 Oct 15;98(2–3):329–43.
  51. 51. Aryal RR, Wespestad C, Kennedy R, Dilger J, Dyson K, Bullock E, et al. Lessons Learned While Implementing a Time-Series Approach to Forest Canopy Disturbance Detection in Nepal. Remote Sensing. 2021 Jul 7;13(14):2666.
  52. 52. Gorelick N, Hancher M, Dixon M, Ilyushchenko S, Thau D, Moore R. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote sensing of Environment. 2017 Dec 1;202:18–27.
  53. 53. Schultz Michael, et al. "Performance of vegetation indices from Landsat time series in deforestation monitoring." International journal of applied earth observation and geoinformation 52 (2016): 318–327.
  54. 54. Kennedy RE, Yang Z, Gorelick N, Braaten J, Cavalcante L, Cohen WB, et al. Implementation of the LandTrendr algorithm on google earth engine. Remote Sensing. 2018 May 1;10(5):691.
  55. 55. Reygadas Y, Spera S, Galati V, Salisbury DS, Silva S, Novoa S. Mapping forest disturbances across the Southwestern Amazon: tradeoffs between open-source, Landsat-based algorithms. Environmental Research Communications. 2021 Sep 16;3(9):091001.
  56. 56. Fragal Everton Hafemann, Thiago Sanna Freire Silva, and Evlyn Márcia Leão de Moraes Novo. "Reconstructing historical forest cover change in the Lower Amazon floodplains using the LandTrendr algorithm." Acta Amazonica 46 (2016): 13–24.
  57. 57. Souza CM Z. Jr Shimbo J, Rosa MR, Parente LL, AAlencar A, Rudorff BF, et al. Reconstructing three decades of land use and land cover changes in brazilian biomes with landsat archive and earth engine. Remote Sensing. 2020 Aug 25;12(17):2735.
  58. 58. Souza J.; Siqueira J.V.; Sales M.H.; Fonseca A.V.; Ribeiro J.G.; Numata I.; et al. Ten-Year Landsat classification of deforestation and forest degradation in the Brazilian Amazon. Remote Sens. 2013, 5, 5493–5513.
  59. 59. Hua Jianwen, et al. "Improved Mapping of Long-Term Forest Disturbance and Recovery Dynamics in the Subtropical China Using All Available Landsat Time-Series Imagery on Google Earth Engine Platform." IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 14 (2021): 2754–2768.
  60. 60. Silva Junior CH, Heinrich VH, Freire AT, Broggio IS, Rosan TM, Doblas J, et al. Benchmark maps of 33 years of secondary forest age for Brazil. Scientific data. 2020 Aug 14;7(1):269. pmid:32796858
  61. 61. Farr TG, Rosen PA, Caro E, Crippen R, Duren R, Hensley S, et al. The shuttle radar topography mission. Reviews of geophysics. 2007 Jun;45(2).
  62. 62. Olofsson Pontus, et al. "Good practices for estimating area and assessing accuracy of land change." Remote Sensing of Environment 148 (2014): 42–57.
  63. 63. ESA. SENTINEL-2 User Handbook, 2015, vol. 1, pp. 64.
  64. 64. Buchhorn M., Lesiv M., Tsendbazar N. E., Herold M., Bertels L., and Smets B. Copernicus global land cover layers—collection 2. Remote Sensing, 2020. 12(6), 1044.
  65. 65. Planet. Planet and NICFI partnership data. 2021.
  66. 66. Potapov P., Li X., Hernandez-Serna A., Tyukavina A., Hansen M. C., Kommareddy A., et al. Mapping global forest canopy height through integration of GEDI and Landsat data. Remote Sensing of Environment. 2021. 253, 112165.
  67. 67. Bey A., Sánchez-Paus Díaz A., Maniatis D., Marchi G., Mollicone D., Ricci S., et al. Collect earth: Land use and land cover assessment through augmented visual interpretation. Remote Sensing. 2016. 8(10), 807.
  68. 68. Saah D, Johnson G, Ashmall B, Tondapu G, Tenneson K, Patterson M, et al. Collect Earth: An online tool for systematic reference data collection in land cover and use applications. Environmental Modelling & Software. 2019 Aug 1;118:166–71.
  69. 69. Stehman SV. Estimating area and map accuracy for stratified random sampling when the strata are different from the map classes. International Journal of Remote Sensing. 2014 Jul 3;35(13):4923–39.
  70. 70. Jordan G., Baskent E. Z., & Whittaker G. A. Spatial forest modelling and landscape management. (1995).
  71. 71. Gustafson EJ. Quantifying landscape spatial pattern: what is the state of the art?. Ecosystems. 1998 Mar;1(2):143–56.
  72. 72. Turner MG, Gardner RH, O’neill RV, O’Neill RV. Landscape ecology in theory and practice. Springer New York; 2001.
  73. 73. Brudvig LA. The restoration of biodiversity: where has research been and where does it need to go?. American journal of botany. 2011 Mar;98(3):549–58. pmid:21613146
  74. 74. de Souza Leite M, Tambosi LR, Romitelli I, Metzger JP. Landscape ecology perspective in restoration projects for biodiversity conservation: a review. Natureza & Conservação. 2013;11:108–18.
  75. 75. McCabe G. P. Principal variables. Technometrics. (1984). 26(2), 137–144.
  76. 76. Pacheco J, Casado S, Porras S. Exact methods for variable selection in principal component analysis: Guide functions and pre-selection. Computational statistics & data analysis. 2013 Jan 1;57(1):95–111.
  77. 77. Hesselbarth MH, Sciaini M, With KA, Wiegand K, Nowosad J. landscapemetrics: an open‐source R tool to calculate landscape metrics. Ecography. 2019 Oct;42(10):1648–57.
  78. 78. Maechler M., Rousseeuw P., Struyf A., Hubert M., Hornik K. (2019). cluster: Cluster Analysis Basics and Extensions. R package version 2.1.0.
  79. 79. Dyson K, Ziter C, Fuentes TL, Patterson MS. Conducting urban ecology research on private property: Advice for new urban ecologists. Journal of Urban Ecology. 2019;5(1):juz001.
  80. 80. Guerrieri A, Bonin A, Münkemüller T, Gielly L, Thuiller W, Francesco Ficetola G. Effects of soil preservation for biodiversity monitoring using environmental DNA. Molecular Ecology. 2021 Jul;30(13):3313–25. pmid:33034070
  81. 81. Dickie I.A., Boyer S., Buckley H.L., Duncan R.P., Gardner P.P., Hogg I.D., et al, 2018. Towards robust and repeatable sampling methods in eDNA‐based studies. Molecular Ecology Resources, 18(5), pp.940–952. pmid:29802793
  82. 82. Dyson K, Dawwas E, Poulton Kamakura R, Alberti M, Fuentes TL. Say where you sample: Increasing site selection transparency in urban ecology. Ecosphere. 2023 Mar;14(3):e4466.
  83. 83. Foucher A, Evrard O, Ficetola GF, Gielly L, Poulain J, Giguet-Covex C, et al. Persistence of environmental DNA in cultivated soils: implication of this memory effect for reconstructing the dynamics of land use and cover changes. Scientific Reports. 2020 Jun 29;10(1):10502. pmid:32601368
  84. 84. Echeverría-Beirute F, Varela-Benavides I, P Jiménez-Madrigal J, Carvajal-Chacon M, Guzmán-Hernández T. eDNA extraction protocol for metagenomic studies in tropical soils. Biotechniques. 2021 Sep;71(6):580–6. pmid:34636654
  85. 85. Sellers GS, Di Muri C, Gómez A, Hänfling B. Mu-DNA: a modular universal DNA extraction method adaptable for a wide range of sample types. Metabarcoding and Metagenomics. 2018 May 7;2:e24556.
  86. 86. Riaz T, Shehzad W, Viari A, Pompanon F, Taberlet P, Coissac E. ecoPrimers: inference of new DNA barcode markers from whole genome sequence analysis. Nucleic acids research. 2011 Nov 1;39(21):e145. pmid:21930509
  87. 87. Zeale M. R., Butlin R. K., Barker G. L., Lees D. C., & Jones G. Taxon‐specific PCR for DNA barcoding arthropod prey in bat faeces. Molecular ecology resources. (2011). 11(2), 236–244. pmid:21429129
  88. 88. Gillet F, Tiouchichine ML, Galan M, Blan F, Némoz M, Aulagnier S, et al. A new method to identify the endangered Pyrenean desman (Galemys pyrenaicus) and to study its diet, using next generation sequencing from faeces. Mammalian Biology. 2015 Nov;80:505–9.
  89. 89. Hebert PD, Cywinska A, Ball SL, DeWaard JR. Biological identifications through DNA barcodes. Proceedings of the Royal Society of London. Series B: Biological Sciences. 2003 Feb 7;270(1512):313–21. pmid:12614582
  90. 90. Hajibabaei M, Shokralla S, Zhou X, Singer GA, Baird DJ. Environmental barcoding: a next-generation sequencing approach for biomonitoring applications using river benthos. PLoS one. 2011 Apr 13;6(4):e17497. pmid:21533287
  91. 91. Kelly RP, Port JA, Yamahara KM, Crowder LB. Using environmental DNA to census marine fishes in a large mesocosm. PloS one. 2014 Jan 15;9(1):e86175. pmid:24454960
  92. 92. Browett S.S., Curran T.G., O’Meara D.B. et al. Primer biases in the molecular assessment of diet in multiple insectivorous mammals. Mamm Biol 101, 293–304 (2021).
  93. 93. Boyer F, Mercier C, Bonin A, Le Bras Y, Taberlet P, Coissac E. obitools: A unix‐inspired software package for DNA metabarcoding. Molecular ecology resources. 2016 Jan;16(1):176–82. pmid:25959493
  94. 94. Rognes T, Flouri T, Nichols B, Quince C, Mahé F. VSEARCH: a versatile open source tool for metagenomics. PeerJ. 2016 Oct 18;4:e2584. pmid:27781170
  95. 95. Edgar RC, Haas BJ, Clemente JC, Quince C, Knight R. UCHIME improves sensitivity and speed of chimera detection. Bioinformatics. 2011 Aug 15;27(16):2194–200. pmid:21700674
  96. 96. Clarke LJ, Beard JM, Swadling KM, Deagle BE. Effect of marker choice and thermal cycling protocol on zooplankton DNA metabarcoding studies. Ecology and evolution. 2017 Feb;7(3):873–83. pmid:28168024
  97. 97. Razgour O, Clare EL, Zeale MR, Hanmer J, Schnell IB, Rasmussen M, et al. High‐throughput sequencing offers insight into mechanisms of resource partitioning in cryptic bat species. Ecology and evolution. 2011 Dec;1(4):556–70. pmid:22393522
  98. 98. Benson D. A., Cavanaugh M., Clark K., Karsch-Mizrachi I., Lipman D. J., Ostell J., et al. (2013). GenBank. Nucleic acids research, 41 (D1), D36–D42.
  99. 99. Valsecchi E, Bylemans J, Goodman SJ, Lombardi R, Carr I, Castellano L, et al. Novel universal primers for metabarcoding environmental DNA surveys of marine mammals and other marine vertebrates. Environmental DNA. 2020 Oct;2(4):460–76.
  100. 100. Gloor GB, Macklaim JM, Pawlowsky-Glahn V, Egozcue JJ. Microbiome datasets are compositional: and this is not optional. Frontiers in microbiology. 2017 Nov 15;8:2224. pmid:29187837
  101. 101. Calle ML. Statistical analysis of metagenomics data. Genomics & informatics. 2019 Mar;17(1).
  102. 102. McKnight DT, Huerlimann R, Bower DS, Schwarzkopf L, Alford RA, Zenger KR. Methods for normalizing microbiome data: an ecological perspective. Methods in Ecology and Evolution. 2019 Mar;10(3):389–400.
  103. 103. Di Muri C, Handley LL, Bean CW, Li J, Peirson G, Sellers GS, et al. Read counts from environmental DNA (eDNA) metabarcoding reflect fish abundance and biomass in drained ponds. bioRxiv. 2020 Jul 30:2020–07.
  104. 104. IUCN. The IUCN Red List of Threatened Species. Version 2021–3. 2021. Accessed on 29 April 2022.
  105. 105. Latini AO, Silva DP, Souza FM, Ferreira MC, de Moura MS, Suarez NF. Reconciling coffee productivity and natural vegetation conservation in an agroecosystem landscape in Brazil. Journal for Nature Conservation. 2020 Oct 1;57:125902.
  106. 106. Toledo-Hernández M, Wanger TC, Tscharntke T. Neglected pollinators: Can enhanced pollination services improve cocoa yields? A review. Agriculture, ecosystems & environment. 2017 Sep 1;247:137–48.
  107. 107. Bakker JD. Increasing the utility of indicator species analysis. Journal of Applied Ecology. 2008 Dec;45(6):1829–35.
  108. 108. De Cáceres M, Legendre P, Wiser SK, Brotons L. Using species combinations in indicator value analyses. Methods in Ecology and Evolution. 2012 Dec;3(6):973–82.
  109. 109. De Cáceres M, Legendre P. Associations between species and groups of sites: indices and statistical inference. Ecology. 2009 Dec;90(12):3566–74/. pmid:20120823
  110. 110. Dyson K. Conserving native trees increases native bird diversity and community composition on commercial office developments. Journal of Urban Ecology. 2020 Dec 15;6(1):juaa033.
  111. 111. Dyson K. Custom Community Ecology Helper R Scripts. (2023). <> accessed 6 June 2023.
  112. 112. Palarea-Albaladejo J, Martín-Fernández JA. zCompositions—R package for multivariate imputation of left-censored data under a compositional approach. Chemometrics and Intelligent Laboratory Systems. 2015 Apr 15;143:85–96.
  113. 113. Van den Boogaart KG, Tolosana-Delgado R. “Compositions”: a unified R package to analyze compositional data. Computers & Geosciences. 2008 Apr 1;34(4):320–38.
  114. 114. Aitchison J. Principal component analysis of compositional data. Biometrika. 1983 Apr 1;70(1):57–65.
  115. 115. Gloor GB, Wu JR, Pawlowsky-Glahn V, Egozcue JJ. It’s all relative: analyzing microbiome data as compositions. Annals of epidemiology. 2016 May 1;26(5):322–9.
  116. 116. Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects models using lme4. arXiv preprint arXiv:1406.5823. 2014 Jun 23.
  117. 117. Fox J, Weisberg S. An R companion to applied regression. Sage publications; 2018 Sep 27. URL:
  118. 118. R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. 2022. URL
  119. 119. Gibbs HK, Ruesch A. New IPCC tier-1 global biomass carbon map for the year 2000. Environmental System Science Data Infrastructure for a Virtual Ecosystem (ESS-DIVE)(United States); 2008 Nov 6.
  120. 120. Bartholome E, Belward AS. GLC2000: a new approach to global land cover mapping from Earth observation data. International Journal of Remote Sensing. 2005 May 1;26(9):1959–77.
  121. 121. FAO. Global ecological zones for FAO forest reporting: 2010 update. FAO: Rome, Italy. 2012.
  122. 122. Ruesch Aaron, and Gibbs Holly K. New IPCC Tier-1 Global Biomass Carbon Map For the Year 2000. Available online from the Carbon Dioxide Information Analysis Center [], Oak Ridge National Laboratory, Oak Ridge, Tennessee. 2008.
  123. 123. Soille P, Vogt P. Morphological segmentation of binary patterns. Pattern recognition letters. 2009 Mar 1;30(4):456–9.
  124. 124. Hansen MC, Potapov PV, Moore R, Hancher M, Turubanova SA, Tyukavina A, et al. High-resolution global maps of 21st-century forest cover change. science. 2013 Nov 15;342(6160):850–3. pmid:24233722
  125. 125. USAID. Private sector engagement: Mobilizing finance for Amazon biodiversity conservation. 2020. Accessed November 8th, 2022. Available online at:
  126. 126. Althelia Climate Fund. Impact Report 2021. 2021. Accessed April 21st, 2022. Available online at:
  127. 127. USAID Partnership for the Conservation for Amazon Biodiversity. Indicators Template. 2019. Accessed April 21st, 2022. Available online at:
  128. 128. Pawlowski J, Bonin A, Boyer F, Cordier T, Taberlet P. Environmental DNA for biomonitoring. Molecular Ecology. 2021 Jul;30(13):2931. pmid:34176165
  129. 129. Drinkwater R, Jucker T, Potter JH, Swinfield T, Coomes DA, Slade EM, et al. Leech blood‐meal invertebrate‐derived DNA reveals differences in Bornean mammal diversity across habitats. Molecular Ecology. 2021 Jul;30(13):3299–312. pmid:33171014
  130. 130. Jost L. Entropy and diversity. Oikos. 2006 May;113(2):363–75.
  131. 131. Pocock SJ. The simplest statistical test: how to check for a difference between treatments. Bmj. 2006 May 25;332(7552):1256. pmid:16735336
  132. 132. Takahashi R, Todo Y. The impact of a shade coffee certification program on forest conservation using remote sensing and household data. Environmental Impact Assessment Review. 2014 Jan 1;44:76–81.
  133. 133. Takahashi R, Todo Y. Coffee certification and forest quality: evidence from a wild coffee forest in Ethiopia. World Development. 2017 Apr 1;92:158–66.
  134. 134. Francesconi W, Nair PK, Levey DJ, Daniels J, Cullen L. Butterfly distribution in fragmented landscapes containing agroforestry practices in Southeastern Brazil. Agroforestry systems. 2013 Dec;87:1321–38.
  135. 135. Jha S, Bacon CM, Philpott SM, Ernesto Méndez V, Läderach P, Rice RA. Shade coffee: update on a disappearing refuge for biodiversity. BioScience. 2014 May 1;64(5):416–28.
  136. 136. Schroth G, Harvey CA. Biodiversity conservation in cocoa production landscapes: an overview. Biodiversity and Conservation. 2007 Jul;16:2237–44.
  137. 137. Maney C, Sassen M, Hill SL. Modelling biodiversity responses to land use in areas of cocoa cultivation. Agriculture, Ecosystems & Environment. 2022 Feb 1;324:107712.
  138. 138. de Jesús Cervantes-López M, Andresen E, Hernández-Ordónez O, Mora F, Reynoso VH, Arroyo-Rodríguez V. Lightly-harvested rustic cocoa is a valuable land cover for amphibian and reptile conservation in human-modified rainforest landscapes. Journal of Tropical Ecology. 2022 Sep;38(5):312–21.
  139. 139. Cassano CR, Schroth G, Faria D, Delabie JH, Bede L. Landscape and farm scale management to enhance biodiversity conservation in the cocoa producing region of southern Bahia, Brazil. Biodiversity and Conservation. 2009 Mar;18:577–603.
  140. 140. Sambuichi RH, Vidal DB, Piasentin FB, Jardim JG, Viana TG, Menezes AA, et al. Cabruca agroforests in southern Bahia, Brazil: tree component, management practices and tree species conservation. Biodiversity and Conservation. 2012 Apr;21:1055–77.
  141. 141. Hunt DA, Tabor K, Hewson JH, Wood MA, Reymondin L, Koenig K, et al. Review of remote sensing methods to map coffee production systems. Remote Sensing. 2020 Jun 25;12(12):2041.
  142. 142. Pasquarella VJ, Arévalo P, Bratley KH, Bullock EL, Gorelick N, Yang Z, et al. Demystifying LandTrendr and CCDC temporal segmentation. International Journal of Applied Earth Observation and Geoinformation. 2022 Jun 1;110:102806.
  143. 143. Chen S, Woodcock CE, Bullock EL, Arévalo P, Torchinava P, Peng S, et al. Monitoring temperate forest degradation on Google Earth Engine using Landsat time series analysis. Remote Sensing of Environment. 2021 Nov 1;265:112648.
  144. 144. Montagnini F, Nair PR. Carbon sequestration: an underexploited environmental benefit of agroforestry systems. InNew Vistas in Agroforestry: A Compendium for 1st World Congress of Agroforestry, 2004 2004 (pp. 281–295). Springer Netherlands.
  145. 145. Bernal B, Murray LT, Pearson TR. Global carbon dioxide removal rates from forest landscape restoration activities. Carbon balance and management. 2018 Dec;13(1):1–3.
  146. 146. Sales NG, Kaizer MD, Coscia I, Perkins JC, Highlands A, Boubli JP, et al. Assessing the potential of environmental DNA metabarcoding for monitoring Neotropical mammals: a case study in the Amazon and Atlantic Forest, Brazil. Mammal Review. 2020 Jul;50(3):221–5.
  147. 147. Strickler KM, Fremier AK, Goldberg CS. Quantifying effects of UV-B, temperature, and pH on eDNA degradation in aquatic microcosms. Biological Conservation. 2015 Mar 1;183:85–92.
  148. 148. Oliverio AM, Gan H, Wickings K, Fierer N. A DNA metabarcoding approach to characterize soil arthropod communities. Soil Biology and Biochemistry. 2018 Oct 1;125:37–43.
  149. 149. Arribas P, Andújar C, Bidartondo MI, Bohmann K, Coissac É, Creer S, et al. Connecting high‐throughput biodiversity inventories: Opportunities for a site‐based genomic framework for global integration and synthesis. Molecular Ecology. 2021 Mar;30(5):1120–35. pmid:33432777
  150. 150. Jackman J.M., Benvenuto C., Coscia I., Oliveira Carvalho C., Ready J.S., Boubli J.P., et al. eDNA in a bottleneck: Obstacles to fish metabarcoding studies in megadiverse freshwater systems. Environmental DNA, 2021. 3(4), pp.837–849.
  151. 151. Leempoel K, Hebert T, Hadly EA. A comparison of eDNA to camera trapping for assessment of terrestrial mammal diversity. Proceedings of the Royal Society B. 2020 Jan 15;287(1918):20192353. pmid:31937227
  152. 152. Cantera I, Cilleros K, Valentini A, Cerdan A, Dejean T, Iribar A, et al. Optimizing environmental DNA sampling effort for fish inventories in tropical streams and rivers. Scientific reports. 2019 Feb 28;9(1):3085. pmid:30816174
  153. 153. Holmes V. Environmental DNA Monitoring of Non-Native Mudpuppy (Necturus Maculosus) and Transient Rainbow Smelt (Osmerus mordax). The University of Maine; 2021.
  154. 154. Sakata MK, Watanabe T, Maki N, Ikeda K, Kosuge T, Okada H, et al. Determining an effective sampling method for eDNA metabarcoding: a case study for fish biodiversity monitoring in a small, natural river. Limnology. 2021 Apr;22:221–35.
  155. 155. Echeverría-Beirute F, Varela-Benavides I, P Jiménez-Madrigal J, Carvajal-Chacon M, Guzmán-Hernández T. eDNA extraction protocol for metagenomic studies in tropical soils. Biotechniques. 2021 Sep;71(6):580–6. pmid:34636654
  156. 156. Zinger L, Taberlet P, Schimann H, Bonin A, Boyer F, De Barba M, et al. Body size determines soil community assembly in a tropical forest. Molecular Ecology. 2019 Feb;28(3):528–43. pmid:30375061
  157. 157. Guerrieri A, Bonin A, Münkemüller T, Gielly L, Thuiller W, Francesco Ficetola G. Effects of soil preservation for biodiversity monitoring using environmental DNA. Molecular Ecology. 2021 Jul;30(13):3313–25. pmid:33034070
  158. 158. Donald J, Murienne J, Chave J, Iribar A, Louisanna E, Manzi S, et al. Multi-taxa environmental DNA inventories reveal distinct taxonomic and functional diversity in urban tropical forest fragments. Global Ecology and Conservation. 2021 Sep 1;29:e01724.
  159. 159. Williams KE, Huyvaert KP, Piaggio AJ. No filters, no fridges: a method for preservation of water samples for eDNA analysis. BMC research notes. 2016 Dec;9(1):1–5. pmid:27278936