Improving biodiversity monitoring using satellite remote sensing to provide solutions towards the 2020 conservation targets

Ensembles of data-efficient vision transformers as a new paradigm for automated classification in ecology

Monitoring biodiversity is paramount to manage and protect natural resources. Collecting images of organisms over large temporal or spatial scales is a promising practice to monitor the biodiversity of natural ecosystems, providing large amounts of data with minimal interference with the environment. Deep learning models are currently used to automate classification of organisms into taxonomic units. However, imprecision in these classifiers introduces a measurement noise that is difficult to control and can significantly hinder the analysis and interpretation of data. We overcome this limitation through ensembles of Data-efficient image Transformers (DeiTs), which not only are easy to train and implement, but also significantly outperform the previous state of the art (SOTA). We validate our results on ten ecological imaging datasets of diverse origin, ranging from plankton to birds. On all the datasets, we achieve a new SOTA, with a reduction of the error with respect to the previous SOTA ranging from 29.35% to 100.00%, and often achieving performances very close to perfect classification. Ensembles of DeiTs perform better not because of superior single-model performances but rather due to smaller overlaps in the predictions by independent models and lower top-1 probabilities. This increases the benefit of ensembling, especially when using geometric averages to combine individual learners. While we only test our approach on biodiversity image datasets, our approach is generic and can be applied to any kind of images.

The Spectral Species Concept in Living Color

Biodiversity monitoring is an almost inconceivable challenge at the scale of the entire Earth. The current (and soon to be flown) generation of spaceborne and airborne optical sensors (i.e., imaging spectrometers) can collect detailed information at unprecedented spatial, temporal, and spectral resolutions. These new data streams are preceded by a revolution in modeling and analytics that can utilize the richness of these datasets to measure a wide range of plant traits, community composition, and ecosystem functions. At the heart of this framework for monitoring plant biodiversity is the idea of remotely identifying species by making use of the 'spectral species' concept. In theory, the spectral species concept can be defined as a species characterized by a unique spectral signature and thus remotely detectable within pixel units of a spectral image. In reality, depending on spatial resolution, pixels may contain several species which renders species-specific assignment of spectral information more challenging. The aim of this paper is to review the spectral species concept and relate it to underlying ecological principles, while also discussing the complexities, challenges and opportunities to apply this concept given current and future scientific advances in remote sensing.

Mapping breeding bird species richness at management-relevant resolutions across the United States

Human activities alter ecosystems everywhere, causing rapid biodiversity loss and biotic homogenization. These losses necessitate coordinated conservation actions guided by biodiversity and species distribution spatial data that cover large areas yet have fine-enough resolution to be management-relevant (i.e., ≤5 km). However, most biodiversity products are too coarse for management or are only available for small areas. Furthermore, many maps generated for biodiversity assessment and conservation do not explicitly quantify the inherent tradeoff between resolution and accuracy when predicting biodiversity patterns. Our goals were to generate predictive models of overall breeding bird species richness and species richness of different guilds based on nine functional or life-history-based traits across the conterminous United States at three resolutions (0.5, 2.5, and 5 km) and quantify the tradeoff between resolution and accuracy and, hence, relevance for management of the resulting biodiversity maps. We summarized 18 years of North American Breeding Bird Survey data (1992-2019) and modeled species richness using random forests, including 66 predictor variables (describing climate, vegetation, geomorphology, and anthropogenic conditions), 20 of which we newly derived. Among the three spatial resolutions, the percentage variance explained ranged from 27% to 60% (median = 54%; mean = 57%) for overall species richness and 12% to 87% (median = 61%; mean = 58%) for our different guilds. Overall species richness and guild-specific species richness were best explained at 5-km resolution using ~24 predictor variables based on percentage variance explained, symmetric mean absolute percentage error, and root mean square error values. However, our 2.5-km-resolution maps were almost as accurate and provided more spatially detailed information, which is why we recommend them for most management applications. Our results represent the first consistent, occurrence-based, and nationwide maps of breeding bird richness with a thorough accuracy assessment that are also spatially detailed enough to inform local management decisions. More broadly, our findings highlight the importance of explicitly considering tradeoffs between resolution and accuracy to create management-relevant biodiversity products for large areas.

Integrating remote sensing with ecology and evolution to advance biodiversity conservation

Remote sensing has transformed the monitoring of life on Earth by revealing spatial and temporal dimensions of biological diversity through structural, compositional and functional measurements of ecosystems. Yet, many aspects of Earth's biodiversity are not directly quantified by reflected or emitted photons. Inclusive integration of remote sensing with field-based ecology and evolution is needed to fully understand and preserve Earth's biodiversity. In this Perspective, we argue that multiple data types are necessary for almost all draft targets set by the Convention on Biological Diversity. We examine five key topics in biodiversity science that can be advanced by integrating remote sensing with in situ data collection from field sampling, experiments and laboratory studies to benefit conservation. Lowering the barriers for bringing these approaches together will require global-scale collaboration.

A quantitative global review of species population monitoring

Species monitoring, defined here as the repeated, systematic collection of data to detect long-term changes in the populations of wild species, is a vital component of conservation practice and policy. We created a database of nearly 1200 schemes, ranging in start date from 1800 to 2018, to review spatial, temporal, taxonomic, and methodological patterns in global species monitoring. We identified monitoring schemes through standardized web searches, an online survey of stakeholders, in-depth national searches in a sample of countries, and a review of global biodiversity databases. We estimated the total global number of monitoring schemes operating at 3300-15,000. Since 2000, there has been a sharp increase in the number of new schemes being initiated in lower- and middle-income countries and in megadiverse countries, but a decrease in high-income countries. The total number of monitoring schemes in a country and its per capita gross domestic product were strongly, positively correlated. Schemes that were active in 2018 had been running for an average of 21 years in high-income countries, compared with 13 years in middle-income countries and 10 years in low-income countries. In high-income countries, over one-half of monitoring schemes received government funding, but this was less than one-quarter in low-income countries. Data collection was undertaken partly or wholly by volunteers in 37% of schemes, and such schemes covered significantly more sites and species than those undertaken by professionals alone. Birds were by far the most widely monitored taxonomic group, accounting for around half of all schemes, but this bias declined over time. Monitoring in most taxonomic groups remains sparse and uncoordinated, and most of the data generated are elusive and unlikely to feed into wider biodiversity conservation processes. These shortcomings could be addressed by, for example, creating an open global meta-database of biodiversity monitoring schemes and enhancing capacity for species monitoring in countries with high biodiversity. Article impact statement: Species population monitoring for conservation purposes remains strongly biased toward a few vertebrate taxa in wealthier countries.

A Remote Sensing Approach to Understanding Patterns of Secondary Succession in Tropical Forest

Monitoring biodiversity on a global scale is a major challenge for biodiversity conservation. Field assessments commonly used to assess patterns of biodiversity and habitat condition are costly, challenging, and restricted to small spatial scales. As ecosystems face increasing anthropogenic pressures, it is important that we find ways to assess patterns of biodiversity more efficiently. Remote sensing has the potential to support understanding of landscape-level ecological processes. In this study, we considered cacao agroforests at different stages of secondary succession, and primary forest in the Northern Range of Trinidad, West Indies. We assessed changes in tree biodiversity over succession using both field data, and data derived from remote sensing. We then evaluated the strengths and limitations of each method, exploring the potential for expanding field data by using remote sensing techniques to investigate landscape-level patterns of forest condition and regeneration. Remote sensing and field data provided different insights into tree species compositional changes, and patterns of alpha- and beta-diversity. The results highlight the potential of remote sensing for detecting patterns of compositional change in forests, and for expanding on field data in order to better understand landscape-level patterns of forest diversity.

Developing a flexible learning activity on biodiversity and spatial scale concepts using open-access vegetation datasets from the National Ecological Observatory Network

Biodiversity is a complex, yet essential, concept for undergraduate students in ecology and other natural sciences to grasp. As beginner scientists, students must learn to recognize, describe, and interpret patterns of biodiversity across various spatial scales and understand their relationships with ecological processes and human influences. It is also increasingly important for undergraduate programs in ecology and related disciplines to provide students with experiences working with large ecological datasets to develop students' data science skills and their ability to consider how ecological processes that operate at broader spatial scales (macroscale) affect local ecosystems. To support the goals of improving student understanding of macroscale ecology and biodiversity at multiple spatial scales, we formed an interdisciplinary team that included grant personnel, scientists, and faculty from ecology and spatial sciences to design a flexible learning activity to teach macroscale biodiversity concepts using large datasets from the National Ecological Observatory Network (NEON). We piloted this learning activity in six courses enrolling a total of 109 students, ranging from midlevel ecology and GIS/remote sensing courses, to upper-level conservation biology. Using our classroom experiences and a pre/postassessment framework, we evaluated whether our learning activity resulted in increased student understanding of macroscale ecology and biodiversity concepts and increased familiarity with analysis techniques, software programs, and large spatio-ecological datasets. Overall, results suggest that our learning activity improved student understanding of biological diversity, biodiversity metrics, and patterns of biodiversity across several spatial scales. Participating faculty reflected on what went well and what would benefit from changes, and we offer suggestions for implementation of the learning activity based on this feedback. This learning activity introduced students to macroscale ecology and built student skills in working with big data (i.e., large datasets) and performing basic quantitative analyses, skills that are essential for the next generation of ecologists.

PL, Poyarkov NA, Le MD, Kraus F, Agarwal I, Oliver PM, Nguyen SN, Nguyen TQ, Karunarathna S, Welton LJ, Stuart BL, Luu VQ, Bauer AM, O’Connell KA, Quah ESH, Chan KO, Ziegler T, Ngo H, Nazarov RA, Aowphol A, Chomdej S, Suwannapoom C, Siler CD, Anuar S, Tri NV, Grismer JL. Phylogenetic partitioning of the third-largest vertebrate genus in the world, Cyrtodactylus Gray, 1827 (Reptilia; Squamata; Gekkonidae) and its relevance to taxonomy and conservation

The gekkonid genus Cyrtodactylus is the third most speciose vertebrate genus in the world, containing well over 300 species that collectively range from South Asia to Melanesia across some of the most diverse landscapes and imperiled habitats on the planet. A genus-wide phylogeny of the group has never been presented because researchers working on different groups were using different genetic markers to construct phylogenies that could not be integrated. We present here Maximum likelihood and Bayesian inference mitochondrial and mito-nuclear phylogenies incorporating of 310 species that include dozens of species that had never been included in a genus-wide analysis. Based on the mitochondrial phylogeny, we partition Cyrtodactylus into 31 well-supported monophyletic species groups which, if used as recommended herein, will increase the information content of future integrative taxonomic analyses that continue to add new species to this genus at an ever-increasing annual rate. Data presented here reiterate the outcome of several previous studies indicating that Cyrtodactylus comprises an unprecedented number of narrow-range endemics restricted to single mountain tops, small islands, or karst formations that still remain unprotected. This phylogeny can provide a platform for various comparative ecological studies that can be integrated with conservation management programs across the broad diversity of landscapes and habitats occupied by this genus. Additionally, these data indicate that the true number of Cyrtodactylus remains substantially underrepresented.

PL, Poyarkov NA, Le MD, Kraus F, Agarwal I, Oliver PM, Nguyen SN, Nguyen TQ, Karunarathna S, Welton LJ, Stuart BL, Luu VQ, Bauer AM, O’Connell KA, Quah ESH, Chan KO, Ziegler T, Ngo H, Nazarov RA, Aowphol A, Chomdej S, Suwannapoom C, Siler CD, Anuar S, Tri NV, Grismer JL. Phylogenetic partitioning of the third-largest vertebrate genus in the world, Cyrtodactylus Gray, 1827 (Reptilia; Squamata; Gekkonidae) and its relevance to taxonomy and conservation

The gekkonid genus Cyrtodactylus is the third most speciose vertebrate genus in the world, containing well over 300 species that collectively range from South Asia to Melanesia across some of the most diverse landscapes and imperiled habitats on the planet. A genus-wide phylogeny of the group has never been presented because researchers working on different groups were using different genetic markers to construct phylogenies that could not be integrated. We present here Maximum likelihood and Bayesian inference mitochondrial and mito-nuclear phylogenies incorporating of 310 species that include dozens of species that had never been included in a genus-wide analysis. Based on the mitochondrial phylogeny, we partition Cyrtodactylus into 31 well-supported monophyletic species groups which, if used as recommended herein, will increase the information content of future integrative taxonomic analyses that continue to add new species to this genus at an ever-increasing annual rate. Data presented here reiterate the outcome of several previous studies indicating that Cyrtodactylus comprises an unprecedented number of narrow-range endemics restricted to single mountain tops, small islands, or karst formations that still remain unprotected. This phylogeny can provide a platform for various comparative ecological studies that can be integrated with conservation management programs across the broad diversity of landscapes and habitats occupied by this genus. Additionally, these data indicate that the true number of Cyrtodactylus remains substantially underrepresented.

Combining Satellite Remote Sensing and Climate Data in Species Distribution Models to Improve the Conservation of Iberian White Oaks (Quercus L.)

The Iberian Peninsula hosts a high diversity of oak species, being a hot-spot for the conservation of European White Oaks (Quercus) due to their environmental heterogeneity and its critical role as a phylogeographic refugium. Identifying and ranking the drivers that shape the distribution of White Oaks in Iberia requires that environmental variables operating at distinct scales are considered. These include climate, but also ecosystem functioning attributes (EFAs) related to energy–matter exchanges that characterize land cover types under various environmental settings, at finer scales. Here, we used satellite-based EFAs and climate variables in species distribution models (SDMs) to assess how variables related to ecosystem functioning improve our understanding of current distributions and the identification of suitable areas for White Oak species in Iberia. We developed consensus ensemble SDMs targeting a set of thirteen oaks, including both narrow endemic and widespread taxa. Models combining EFAs and climate variables obtained a higher performance and predictive ability (true-skill statistic (TSS): 0.88, sensitivity: 99.6, specificity: 96.3), in comparison to the climate-only models (TSS: 0.86, sens.: 96.1, spec.: 90.3) and EFA-only models (TSS: 0.73, sens.: 91.2, spec.: 82.1). Overall, narrow endemic species obtained higher predictive performance using combined models (TSS: 0.96, sens.: 99.6, spec.: 96.3) in comparison to widespread oaks (TSS: 0.80, sens.: 92.6, spec.: 87.7). The Iberian White Oaks show a high dependence on precipitation and the inter-quartile range of Normalized Difference Water Index (NDWI) (i.e., seasonal water availability) which appears to be the most important EFA variable. Spatial projections of climate–EFA combined models contribute to identify the major diversity hotspots for White Oaks in Iberia, holding higher values of cumulative habitat suitability and species richness. We discuss the implications of these findings for guiding the long-term conservation of Iberian White Oaks and provide spatially explicit geospatial information about each oak species (or set of species) relevant for developing biogeographic conservation frameworks.

Effective ecosystem monitoring requires a multi-scaled approach

Ecosystem monitoring is fundamental to our understanding of how ecosystem change is impacting our natural resources and is vital for developing evidence-based policy and management. However, the different types of ecosystem monitoring, along with their recommended applications, are often poorly understood and contentious. Varying definitions and strict adherence to a specific monitoring type can inhibit effective ecosystem monitoring, leading to poor program development, implementation and outcomes. In an effort to develop a more consistent and clear understanding of ecosystem monitoring programs, we here review the main types of monitoring and recommend the widespread adoption of three classifications of monitoring, namely, targeted, surveillance and landscape monitoring. Landscape monitoring is conducted over large areas, provides spatial data, and enables questions relating to where and when ecosystem change is occurring to be addressed. Surveillance monitoring uses standardised field methods to inform on what is changing in our environments and the direction and magnitude of that change, whilst targeted monitoring is designed around testable hypotheses over defined areas and is the best approach for determining the causes of ecosystem change. The classification system is flexible and can incorporate different interests, objectives, targets and characteristics as well as different spatial scales and temporal frequencies, while also providing valuable structure and consistency across distinct ecosystem monitoring programs. To support our argument, we examine the ability of each monitoring type to inform on six key types of questions that are routinely posed for ecosystem monitoring programs, such as where and when change is occurring, what is the magnitude of change, and how can the change be managed? As we demonstrate, each type of ecosystem monitoring has its own strengths and weaknesses, which should be carefully considered relative to the desired results. Using this scheme, scientists and land managers can design programs best suited to their needs. Finally, we assert that for our most serious environmental challenges, it is essential that we include information from each of these monitoring scales to inform on all facets of ecosystem change, and this is best achieved through close collaboration between the scales. With a renewed understanding of the importance of each monitoring type, along with greater commitment to monitor cooperatively, we will be well placed to address some of our greatest environmental challenges.

Multi-temporal assessment of grassland α- and β-diversity using hyperspectral imaging

While more and more studies are exploring the application of remote sensing in assessing biodiversity for different ecosystems, most consider biodiversity at one point in time. Using several remote-sensing-based metrics, we asked how well remote sensing can detect biodiversity (both α- and β-diversity) in a prairie grassland across time using airborne hyperspectral data collected in two successive years (2017 and 2018) and at different periods in the growing season (2018). The ability to detect biodiversity using "spectral diversity" and "spectral species" types indeed varied significantly over a 2-yr timespan. Toward the end of the growing season in 2018, the relationship between field- and remote-sensing-based α- and β-diversity weakened compared to data collected from the same season in the previous year. This contrasting pattern between the two years was likely influenced by prescribed fire, altered weather, and the resulting shifting species composition and phenology. These findings indicate that direct detection of α- and β-diversity in grasslands should be multi-temporal when possible and should consider the effect of disturbances, climate variables, and phenology. We demonstrate an essential role for airborne platforms in developing a global biodiversity monitoring system involving forthcoming space-borne hyperspectral sensors.

Improving on MODIS MCD64A1 Burned Area Estimates in Grassland Systems: A Case Study in Kansas Flint Hills Tall Grass Prairie

Uncertainty in satellite-derived burned area estimates are especially high in grassland systems, which are some of the most frequently burned ecosystems in the world. In this study, we compare differences in predicted burned area estimates for a region with the highest fire activity in North America, the Flint Hills of Kansas, USA, using the moderate resolution imaging spectroradiometer (MODIS) MCD64A1 burned area product and a customization of the MODIS MCD64A1 product using a major ground-truthing effort by the Kansas Department of Health and Environment (KDHE-MODIS customization). Local-scale ground-truthing and the KDHE-MODIS product suggests MODIS burned area estimates under predicted fire occurrence by 28% over a 19-year period in the Flint Hills ecoregion. Between 2001 and 2019, MODIS product indicated <1 million acres burned on average, which was far below the KDHE-MODIS customization (mean = 2.6 million acres). MODIS also showed that <1% of the Flint Hills burned 5 times from 2001–2019 (2001, 2002, 2007, 2012 and 2013), whereas KDHE-MODIS customization showed this never happened in any single year. KDHE-MODIS also captured some areas of the Flint Hills that burned every year (19 times out of 19 years), which is well-known with field inventory data, whereas the maximum fire occurrence in MODIS was 14 times in 19 years. Finally, MODIS never captured >8% burned area for any given year in the Flint Hills, even in years when fire activity was highest (2008, 2009, 2011, 2014). Based on these results, coupling MODIS burned area computations with local scale ground-truth efforts has the potential to significantly improve fire occurrence estimates and reduce uncertainty in other grassland and savanna regions.

Integrating intraseasonal grassland dynamics in cross-scale distribution modeling to support waterbird recovery plans

Despite much discussion about the utility of remote sensing for effective conservation, the inclusion of these technologies in species recovery plans remains largely anecdotal. We developed a modeling approach for the integration of local, spatially measured ecosystem functional dynamics into a species distribution modeling (SDM) framework in which other ecologically relevant factors are modeled separately at broad scales. To illustrate the approach, we incorporated intraseasonal water-vegetation dynamics into a cross-scale SDM for the Common Snipe (Gallinago gallinago), which is highly dependent on water and vegetation dynamics. The Common Snipe is an Iberian grassland waterbird characteristic of European agricultural meadows and a member of one of the most threatened bird guilds. The intraseasonal dynamics of water content of vegetation were measured using the standard deviation of the normalized difference water index time series computed from bimonthly images of the Sentinel-2 satellite. The recovery plan for the Common Snipe in Galicia (northwestern Iberian Peninsula) provided an opportunity to apply our modeling framework. Model accuracy in predicting the species' distribution at a regional scale (resulting from integration of downscaled climate projections with regional habitat-topographic suitability models) was very high (area under the curve [AUC] of 0.981 and Boyce's index of 0.971). Local water-vegetation dynamic models, based exclusively on Sentinel-2 imagery, were good predictors (AUC of 0.849 and Boyce's index of 0.976). The predictive power improved (AUC of 0.92 and Boyce's index of 0.98) when local model predictions were restricted to areas identified by the continental and regional models as priorities for conservation. Our models also performed well (AUC of 0.90 and Boyce's index of 0.93) when projected to updated water-vegetation conditions. Our modeling framework enabled incorporation of key ecosystem processes closely related to water and carbon cycles while accounting for other factors ecologically relevant to endangered grassland waterbirds across different scales, allowed identification of priority areas for conservation, and provided an opportunity for cost-effective recovery planning by monitoring management effectiveness from space.

Geospatial modeling of land cover change in the Chocó-Darien global ecoregion of South America; One of most biodiverse and rainy areas in the world

The tropical rain forests of northwest South America fall within the Chocó-Darien Global Ecoregion (CGE). The CGE is one of 25 global biodiversity hotspots prioritized for conservation due to its high biodiversity and endemism as well as threats due to deforestation. The analysis of land-use and land-cover (LULC) change within the CGE using remotely sensed imagery is challenging because this area is considered to be one of the rainiest places on the planet (hence high frequency of cloud cover). Furthermore, the availability of high-resolution remotely sensed data is low for developing countries before 2015. Using the Random Forest ensemble learning classification tree system, we developed annual LULC maps in the CGE from 2002 to 2015 using a time series of cloud-free MODIS vegetation index products. The MODIS imagery was processed through a Gaussian weighted filter to further correct for cloud pollution and matched to visual interpretations of land cover and land use from available high spatial resolution imagery (WorldView-2, Quick Bird, Ikonos and GeoEye-1). Validation of LULC maps resulted in a Kappa of 0.87 (Sd = 0.008). We detected a gradual replacement of forested areas with agriculture (mainly grassland planted to support livestock grazing), and secondary vegetation (agriculture reverting to forest) across the CGE. Forest loss was higher between 2010–2015 when compared to 2002–2010. LULC change trends, deforestation drivers, and reforestation transitions varied according to administrative organization (countries: Panamanian CGE, Colombian CGE, and Ecuadorian CGE).