Remote Sensing’s Recent and Future Contributions to Landscape Ecology



The purpose of this article is to review landscape ecology research from the past 5 years to identify past and future contributions from remote sensing to landscape ecology.


Recent studies in landscape ecology have employed advances made in remote sensing. These include the use of reliable and open datasets derived from remote sensing, the availability of new sources for freely available satellite imagery, and machine-learning image classification techniques for classifying land cover types. Remote sensing data sources and methods have been used in landscape ecology to examine landscape structure. Additionally, these data sources and methods have been used to analyze landscape function including the effects of landscape structure and landscape change on biodiversity and population dynamics. Lastly, remote sensing data sources and methods have been used to analyze historical landscape changes and to simulate future landscape changes.


The ongoing integration of remote sensing analyses in landscape ecology will depend on continued accessibility of free imagery from satellite sources and open-access data-analysis software, analyses spanning multiple spatial and temporal scales, and novel land cover classification techniques that produce accurate and reliable land cover data. Continuing advances in remote sensing can help to address new landscape ecology research questions, enabling analyses that incorporate information that ranges from ground-based field samples of organisms to satellite-collected remote sensing data.



Introduction
In the last 5 years, landscape ecologists have continued their seminal focus on the relationships of pattern and process [1], addressing questions of landscape structure, landscape function, and landscape change [2]. For example, recent studies have analyzed landscape structure by examining urban green cover, rangeland distribution, wetland extent [3][4][5][6], fragmentation of forests [7], land cover and land use [8][9][10], and heterogeneity of urban and agricultural landscapes [11][12][13][14]. For relating landscape structure to ecological processes, studies have focused on habitat and resource selection by plants and animals [15][16][17][18], forest dynamics and structure [19], and pollination on agricultural lands [20,21]. For analyzing the movement across landscapes, analyses have explored movement related to corridors and connectivity [22][23][24][25][26][27] and movement of species populations related to metapopulation dynamics using genetics to track reproduction and population dispersal across generations [28,29]. For quantifying landscape change, recent studies have used landscape history to analyze disturbances such as fire and their impacts on landscape structure over time [30][31][32][33]. By analyzing prior landscape changes, other landscape ecologists have also worked towards predicting changes in landscape structure and evaluating potential impacts through system feedbacks and potential changes in land planning by using simulation models [34-53, 54 •• , 55, 56 • ].
Many of these studies in landscape ecology have relied on contributions from the field of remote sensing. Since the launch of the satellite Landsat-1 MSS in 1972, a variety of remote sensing platforms (e.g. satellite, aerial) have collected data in the form of image observations. Each sensor gathers imagery at a pre-defined spatial resolution, which denotes the ground measurement that each pixel represents in an image. Spectral resolutions vary based on the wavelength intervals that the sensors are collecting reflectance of the sun on the earth's surface. The temporal resolution of a given remote sensing platform is derived from its orbital path and speed, which determines the satellite's revisit rate for collecting a new image in the same location. Sensors currently in operation include optical sensors from NASA's Landsat program, optical and synthetic aperture radar (SAR) sensors from the European Space Agency Copernicus constellation, and many other public and privately owned airborne and spaceborne systems. Researchers are able to choose their remote sensing sources based on their research questions, whether they use sources such as unmanned aerial vehicles (UAVs), active sensors like light detection and ranging (lidar), field-based spectroscopy, cross-boundary satellites [38][39][40][41]. Users can preprocess images to correct for atmospheric interferences caused by haze, clouds, or angle of the sun [42][43][44]. By comparing imagery and ground-based measurements, users can classify land cover types (e.g., forests, wetlands, development) to analyze the landscape structure [45][46][47][48]. Freely available remote sensing data from satellite sensors with large spatial coverage has become available in the last 10 years [49][50][51][52][53]54 [49][50][51][52][53]55]. With increasing data availability for large-area coverage and medium spatial resolution sensors like Landsat and Sentinel, there has been a dramatic increase in research using satellite data in the last 5 years [52].

Literature Review
In this review, we examine recently published manuscripts from landscape ecology that have been made possible through advances in remote sensing. We outline recent developments in remote sensing and landscape ecology, highlighting important developments from each field to illuminate their current and approaching potential. To achieve this, we employed a systematic review of highly cited literature related to landscape ecology and remote sensing for the last 5 years, from 2014 to 2019. We identified recently published manuscripts that apply remote sensing methods in landscape ecology using Web of Science. We sorted the manuscripts by overall citation count in order to identify the most prominent contributions made within this field. We terminated our exhaustive literature search after identifying all 172 manuscripts meeting the search criteria (e.g., "landscape ecology" & "remote sensing"). The 172 manuscripts were categorized by landscape ecology research themes: landscape structure, landscape change, landscape function. For each landscape ecology theme, we identified remote sensing data sources and analyses most frequently used in landscape ecology by analyzing the author keywords (Table 1). Eleven publications were removed that were unrelated to landscape ecology or remote sensing. Additional recently published manuscripts were also incorporated into the review. Once we identified recent contributions of landscape ecology, we projected future research opportunities for landscape ecology by identifying other advances in remote sensing that might also be relevant to landscape ecology, as determined from the most frequently cited manuscripts from remote sensing in Web of Science from 2014 to 2019.

Explosion of Data Diversity and Availability
The first decades of landscape ecology were characterized by a relatively data-poor setting, with only a few satellites potentially providing data and practical limits to analysis. For example, early studies in the 1970-1980s typically analyzed only one Landsat image at a time because they were expensive, had to be shipped on tapes from receiving stations, and took weeks to analyze on computers of the time period. In the last 10 years, this framework has been overturned, with hundreds of thousands of images freely available for analysis from multiple public and free remote sensing platforms [49,[51][52][53]54 •• , 55, 56 • , 57 • , 58]. These new or improved platforms include those on large satellites like Landsat-8 and Sentinel-2, on airplanes, on unmanned aerial vehicles (UAVs), via small/micro/nanosatellites, and through ground-based sensor systems. Some example sensors include passive multi-spectral optical (i.e., collecting ground reflectance along the optical light spectrum divided into three to ten segments), actively collected synthetic aperture radar (SAR) (collecting return rates of wavelengths from multiple, pulsating microwave beams), thermal sensors (i.e., heat detection), hyperspectral sensors (i.e., collecting ground reflectance of the sun at ten or more segments of the light spectrum), and actively collected lidar (i.e., collecting return rates of wavelengths from a single, pulsating laser) [59][60][61][62][63][64][65][66][67][68][69]. Meanwhile, atmospheric noise in data is decreasing. For example, "analysis-ready" imagery and data cubes are now available for Landsat imagery, which enables users to spend less time preprocessing imagery [70,71]. Whether launched by private companies or public agencies, remote sensing sources are increasing in spatial, spectral, and temporal resolutions of the observations [72,73].

Development of Algorithms for Large-scale Image Classifications
A primary focus of remote sensing research is to develop methods for converting remotely sensed data into a meaningful description or picture of what is actually on the ground. This is referred to as "classification" of the remotely sensed data. Several recent advances have greatly improved algorithms used in classification [54 • , 56 •• , 93]. For example, object-based image classifications group neighboring pixels into objects and classify the objects based on their shape, size, color, texture (spatial variation), and context (neighboring or ancillary information) [94,95]. Machine/deep learning approaches (e.g., convolutional neural networks, random forests) are automated classification algorithms that rely on minimal user interference when classifying imagery [59,63,72,[96][97][98][99][100][101]. Additionally, time-series analyses have been used to map land cover changes by stacking images from multiple sources and identifying disturbance patterns and deviations from expected values [69,74,[102][103][104][105][106][107][108][109][110][111][112][113]. This allows the rapid detection of landscape change and disturbances like forest loss and fires. Time-series analyses have created reliable globalscale landscape change datasets that are freely available for subsequent analyses [114][115][116][117][118]. For example, a regularly updated forest cover dataset including landscape changes and drivers of changes is available annually for the entire globe [119,120]. Additionally, the World Resources Institute's Global Forest Watch initiative detects forest changes globally in near real time [121]. Other recent studies have used timeseries analyses, machine learning, and object-based image analyses to analyze land surface temperatures and identify urban heat islands [122], to provide increased data to support forest inventory efforts [66], to map landscape changes related to climate change [123], to inform precision agriculture [124], to monitor air pollution [125], to quantify colored dissolved organic matter in lakes [126], to quantify aboveground biomass [127], and to track urbanization [128].

Advances in Landscape Ecology Using Remote Sensing
Landscape ecologists use remote sensing for three principal reasons: (1) to quantify landscape structure based on classified imagery; (2) to identify landscape change and its impact and make future predictions using statistical models; and (3) to quantify landscape function. Landscape structure is the spatial arrangement of landscape elements, such as land cover types and forest patches. Landscape change refers to the changes in the landscape structure over time and space. Landscape function is the interactions between landscape structural elements, whether through ecological processes or energy flows, such as the interactions between animal migration routes and forest connectivity.

Quantifying Landscape Structure
Remote sensing observations provide the potential to map and analyze landscape structure at a variety of spatial and temporal grains and extents. Landscape ecologists analyze both raw  [53,91,106,122,132] remote sensing data and remote sensing-derived maps to quantify landscape structure. For example, by harmonizing airborne lidar and satellite imagery, researchers were able to quantify structural connectivity and identify patches that were most important for landscape-level conservation in Alberta, Canada [129]. Landscape ecologists have extracted landscape-based information using a variety of remote sensing spectral vegetation indices (e.g., tasseled cap, leaf area index, normalized difference vegetation index (NDVI)) [130][131][132].
In Finland, researchers combined data collected by citizen scientists (e.g., landowners, students, recreationalists) with lidar-derived forest measurements to quantify landscape structure [133]. Additionally, landscape structure has frequently been quantified using open-source toolboxes designed to process remote sensing data [134][135][136][137]. The wide range of applications employing landscape-scale analyses has been made possible from the increasing availability of remote sensing sources and advances in imagery analyses (Table 3).

Future Prospects
Advances in methods for quantifying landscape structure will mirror advances made in remote sensing for image classification due to the direct relationship between a landscape's surface cover and its structure. As data diversity and availability continue to grow, information from remote sensing data seems poised to make novel advances within landscape ecology in the near future. For example, opportunities exist for increasing landscape-scale analyses focusing on biomass analyses and vegetation structure using data from the recently launched and future active sensors (e.g., NISAR, GEDI, BIOMASS, MOLI, SAOCOM1A, ICESat-2, ALOS-4, TanDEM-L, RADARSAT Constellation Mission) [138]. Additional opportunities will be created to use the finer spatial and temporal resolutions that will be provided by future optical satellites that are being built (e.g., Landsat 9, Sentinel constellation). While many landscape ecology studies take advantage of remote sensing observations collected by aerial and satellite sources, future studies can use observations from novel data sources like UAVs and microsatellites (i.e., small satellites from companies like DigitalGlobe and Planet) for very high spatial resolution observations of fine-scale landscape features [139], hyperspectral sensors for greater spectral sensitivity when using raw remote sensing values in landscape ecology models [140], and synthetic aperture radar (SAR) sensors and lidar sensors for reconstructing three-dimensional landscape structure and analyzing connectivity [141][142][143]. Landscape structure can be quantified by using feature extraction techniques and machine-learning classifiers to improve the accuracy of image classifications [80,144,145]. By quantifying landscape structure on a cloud-based processing platform like Google Earth Engine [57 • ], large-area landscape ecology structural analyses become more tenable and it will no longer be necessary to download new imagery to personal computers.

Quantifying Landscape Change
Landscape ecology studies use remote sensing images from multiple collection dates to identify landscape change, to analyze their impacts on populations, and to predict future landscape change. Satellite-based time-series data (whether from one sensor or many) provide observations spanning multiple decades of landscape change such as cumulative forest cover decline, recovery of forest species from disturbances, degradation of forest patches, and land-use change [130,132,[146][147][148][149][150][151][152]. Researchers applied a temporal trend analysis of Landsat TM time-series imagery and vegetation indices from 1987 to 2010 to map gradual and abrupt forest decline and regrowth in Québec, Canada, and inform land management policy [132]. Another study integrated multi-source imagery from NASA's Landsat MSS, TM, ETM+, the Russian KATE-200 satellite camera, and satellite Keyhole imagery to identify regions for management by evaluating the relationship between oasis changes and landscape structure in an arid region of China from 1963 to 2010 [148]. Remote sensing data has also been incorporated into existing landscape ecology simulations to model stochastic dynamics of landscape structure elements, and in turn, landscape function. For example, landscape ecologists have used remote sensing-based data to predict rates and patterns of urban expansion over time [153], to quantify landscape structure and ecosystem service changes in urban areas [154,155], and to simulate changes in soil organic carbon due to changing climate [156]. Observations from remote sensing platforms enable landscape ecologists to reconstruct landscape history for analyzing landscape changes and to inform predictive models for landscape changes.

Future Prospects
As the temporal revisit rate of satellite image observations gets shorter, landscape ecologists will be able to see landscape changes as they happen in near real time, whether they are persistent (e.g., fire), ephemeral (e.g., floods), or gradual (e.g., forest degradation) [157][158][159]. Increased data frequency will be useful for analyzing landscape changes at daily or monthly resolutions rather than only annual resolutions. Additionally, by accessing publicly available near-real-time global datasets that map land cover changes using cloud-based platforms like Google Earth Engine, landscape ecologists will be able to perform their own analyses more rapidly without developing their own image classification protocols. Multi-temporal landscape analyses of the same landscape or analyses comparing different landscapes will become increasingly accessible by employing data fusion methods to combine observations from multiple sensors and weighing the evidence from each classification [54 •• , 129, 160-162]. Such analyses have been previously difficult for landscape ecology due to data collection limitations and financial costs of imagery. However, open-access satellites provide multi-scale views for free [37,163]. Robust predictive models that are able to include remote sensing classifications derived from multiple sources or classifications with continuous values (e.g., forest quality on a continuous scale rather than discrete classes) will be useful for incorporating future data sources more readily into existing landscape ecology models.

Understanding Landscape Function
Landscape ecologists can analyze landscape function of the study area by combining information derived from remote sensing with information from other sources into landscape ecology models. For example, satellite-derived ecosystem service indicators (e.g., water quality, soil moisture, and soil erosion) can be analyzed in combination with land cover information (e.g., wetland area) to estimate ecosystem service provisioning [164,165]. Habitat classifications identifying  [129] Multiple data sources with varying spatial and thematic resolution Predicted seasonal land surface temperatures Determined strong predictors of land surface temperatures to include percent of impervious surfaces, percent of tree canopy from spring to fall, and vegetative-based indices from summer to fall [131] Multiple data sources from ground observations and airborne lidar Quantified aboveground forest biomass and vegetation structure Identified spatially explicit biodiversity indicators for bird habitats for 41 different species in boreal forest regions [133] Refining spatial resolution from remote sensing sources Examined landscape surface metrics at a higher spatial resolution to assess scale-dependent relationships Found that map accuracy for data aggregation of sub-pixel remote sensing classifications was dependent on spatial heterogeneity of the landscape [136] Synthetic aperture radar (SAR) data sources

Calculated resistance maps for habitat connectivity
Found that SAR-based maps explained more of the species abundance for forest beetles than aerial photograph-based maps [141] Active (lidar) and passive (AVIRIS) aerial sensors Multiple data sources from ground samples and MODIS time-series data Quantified the temporal and spatial patterns of land-use regime shifts Identified that regime shifts were caused by livestock/pasture privatization and installment iron fences in Qinghai-Tibetan Plateau using oral histories and GIS data [150] Multi-temporal satellite analysis Quantified land cover change and calculated landscape composition and configuration Found that an abandoned mine landscape was less heterogeneous with fewer dense conifer patches and more bare patches than a comparison protected site over 20 years in the Northwest Territories, Canada [160] High-resolution radar and lidar data Mapped vegetative structure and modeled dispersal barriers of population structure using Circuitscape connectivity software Determined that higher elevations in the Amazon rainforest were related to larger genetic distances between macaw individuals due to mountain ridges limiting gene flows [182] Multiple data sources from ground observations and remote sensing-derived land cover data Identified relationships between land cover classes and reptile roadkills using a hotspot analysis Identified hotspots of amphibian/reptile deaths near arable land, suburban areas, and vineyards using freely available remote sensing data in eastern Austria [184] Satellite time-series data Analyzed multi-temporal landscape mosaic patterns Identified changes in coverage and pattern of six dominant tree species in Saskatchewan, Canada [185] Long-term remote sensing time-series dataset Compared models for calculating landscape connectivity between scenario models and a 25-year surface water time series Found that scenario-derived connectivity models were less effective than remote sensing-derived time series for identifying important areas for connectivity in south-eastern Australia [186] population preferences and vulnerability related to landscape change, and primary productivity related to spatial distributions of species have been assessed using object-based classifications and random forest machine-learning algorithms of satellite data, vegetation structural observations provided by lidar data, and gross primary productivity values derived from the enhanced vegetation index [139,143,145,166,167]. By fusing spectral indices like NDVI with vegetative structural information provided by lidar and topographical information derived from SAR observations, human impacts on vegetation patterns and environmental gradients can be analyzed [142].
Research focusing on urban landscape ecology has analyzed remote sensing data like land surface temperature products to examine the relationship between land surface temperature and land cover/use [168][169][170]. By incorporating remote sensing data like the National Land Cover Database, NDVI, and Landsat 7 ETM+ observations with land surface temperatures, the urban heat island effect can be analyzed and used to predict future land surface temperatures [131].

Future Prospects
For analyzing landscape function, advances will be made in landscape ecology by using new remote sensing data sources and analyses to quantify interactions between landscape structure and ecological processes (e.g., land cover type and population movement). Calls have been made to shift habitat assessments from categorical indices (e.g., low, medium, and high) to continuous values (e.g., 0-100) to better evaluate impacts of landscape change on biodiversity and incorporate error quantification into landscape ecology models [115, 171 • , 172]. This shift towards continuous values would capitalize on advances made in remote sensing for classifying gradients of sub-pixel land cover and forest quality, per-pixel confidences in classification, and data uncertainty measurements [147, 171 • , 172-174]. New sensors like GEDI and continuous data such as forest quality can provide more functional information about the landscape in terms of species distribution, resource distribution, and three-dimensional habitat connectivity [175]. Landscape ecology studies that incorporate remote sensing images can also incorporate data from non-remote sensing sources like crowdsourcing, participatory research, and other existing geospatial datasets [54 •• , 56 • , [176][177][178]. For example, landscape ecologists can incorporate geolocations of bird sightings collected by citizen scientists in eBird (eBird. org) in combination with vertical vegetative structure data from lidar to improve models analyzing species distribution or biodiversity. Additionally, the fusion of data and imagery from multiple sources can increase spatial, temporal, and spectral resolutions by updating the data cube with the finest resolution data available to better analyze landscape processes [54 • , 112, 179-181]. For example, often genetic and metapopulation studies examine landscape changes that occur at scales finer than landscape changes captured by medium-resolution satellites like Landsat. Therefore, there is an opportunity to assimilate very high spatial resolution remote sensing data from microsatellites and UAVs or temporally fine-scale satellite time series to analyze metapopulation dynamics [182].

Conclusions
The advances that have been made in landscape ecology using remote sensing can inform future opportunities for integrating remote sensing in landscape ecology studies. Landscape ecology has made advances in quantifying landscape connectivity, using genetics to analyze metapopulation dynamics, examining multi-functional and social-ecological systems, simulating future landscape changes, and establishing landscape histories to inform and model future landscape changes. These advances have been made possible in part due to remote sensing including the production of reliable land cover datasets that use new data sources, time series of remotely sensed data and three-dimensional data, machine-learning classification techniques, and free data accessibility. Within landscape ecology, remote sensing images and analyses have been applied to construct multi-scale, multi-temporal, and multi-source landscape-scale analyses.
Upcoming data sources will be used to estimate functional attributes of a landscape such as interactions between landscape elements and ecological processes, which can then be integrated into existing landscape ecology models that relate landscape structure or landscape change to ecological responses like species diversity. Remote sensing-derived data can either inform the landscape structure and landscape change or the ecological responses, depending on research objectives and data availability. The fusion of remote sensing observations from multiple sources into data cubes can increase temporal and spatial resolutions without trading off spatial extent coverage. Near-real-time monitoring provided by open-access satellite sensors can provide landscapes pre and post-change at the time steps necessary to evaluate impacts on ecosystem processes. Ultimately, these advances in data sources at varying scales and resolutions from very highresolution to large-area coverage enable landscape ecology analyses that can be produced more rapidly, for larger study regions, and for longer study periods.

Compliance with Ethical Standards
Conflict of Interest Morgan Crowley and Jeffrey Cardille declare that they have no conflict of interest.
Human and Animal Rights and Informed Consent This article does not contain any studies with human or animal subjects performed by any of the authors.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.