Despite their importance to global biodiversity and ecosystem services, tropical montane forest (TMF) disturbance regimes are understudied (Crausbay and Martin, 2016; Martin and Bellingham, 2016). Landslides, a central component of many TMF disturbance regimes, leave long-lasting legacies on montane landscapes (Restrepo et al., 2009; Walker and Shiels, 2013). They have helped shape the highly diverse TMF of the tropical Andes biodiversity hotspot (Kessler and Kluge, 2008; Richter et al., 2009), which extends from Venezuela, through Colombia, Ecuador, Peru, and Bolivia, to northeastern Chile and northwestern Argentina (Myers et al., 2000; Critical Ecosystem Partnership Fund, 2021). The hotspot is home to at least 30,000 plant species, including an estimated 15,000 endemic species, making the Andean flora the most diverse in the world (Mittermeier et al., 2011). Andean TMF is also one of Earth’s most threatened habitats, with climate change predicted to significantly alter the region’s hydrological and temperature regimes, thus changing forest compositions, disrupting ecosystem processes, and altering natural disturbance regimes (Still et al., 1999; Crausbay and Martin, 2016). Increasing scientific understanding of landslides in the tropical Andes is integral to the long-term protection and management of Andean TMF.

Landslides are widespread in the tropical Andes and contribute to large-scale ecosystem processes. For example, on average, 0.08% of Peru’s Kosñipata Valley is affected by landslides each year (Clark et al., 2016), a rate equal to the per century rate estimated for Puerto Rico’s Upper Luquillo Mountains (Guariguata, 1990) and seven times greater than estimated for eastern Puerto Rico (Larsen and Torres-Sanchez, 1992). This translates into an average hillslope turnover time of 1,320 years, more rapid than observed rates from 13 catchments in New Zealand’s western Southern Alps (Hilton et al., 2011), and 24 times faster than in the mountains of Mexico and Central America (Restrepo and Alvarez, 2006). Landsliding rates may be even higher in other parts of the Andes; a study in an area of very steep relief in the Bolivian Eastern Cordillera demonstrated landslides affect 4–6% of the landscape over 10–35 years (Blodgett and Isacks, 2007), a rate of 0.11–0.6% per year, or 11.4–17.1% per century. Landslides play a key role in landscape evolution (Korup et al., 2010) and the mobilization of geological material (Muenchow et al., 2012), including the export of sediment, rock and non-rock derived nutrients, and forest carbon to rivers (Hilton et al., 2008, 2011; Restrepo et al., 2009; Ramos Scharrón et al., 2012; Clark et al., 2016; Croissant et al., 2019; Hilton and West, 2020). By exposing underlying bedrock, landslides also make nutrients available to plants (Guariguata, 1990; Zarin and Johnson, 1995; Vitousek et al., 2003; Porder et al., 2005). Finally, populations of landslides contribute to longer-term carbon cycling in mountain landscapes (Ramos Scharrón et al., 2012; Frith et al., 2018; Hilton and West, 2020).

Here, we outline a vision for three axes of study for landslide research in Andean TMF that apply remotely-sensed data to understand how landslides contribute to TMF biodiversity, ecosystem functioning, and the potential effects of climate change on landslide regimes (Figure 1). We focus exclusively on the Andes for three reasons: (1) the vast floral diversity and high endemicity of the tropical Andes biodiversity hotspot, and its importance for global biodiversity and regional ecosystem service provisioning (Myers et al., 2000; Breuer et al., 2013; Spracklen and Righelato, 2014); (2) the broad elevational, latitudinal, and topographic gradients, longer than all other tropical mountain ranges in the world (Young et al., 2007), across which landslide dynamics play out; and (3) the existence of several long-term and thoroughly characterized plot networks across the region that provide the baseline data on mature forest structure, composition, and functioning necessary to contextualize disturbance impacts (Malhi et al., 2010; Malizia et al., 2020). Though we focus on the Andes, the themes, and research avenues we discuss are broadly applicable to TMF globally.

Technological developments in remote sensing over the past few decades have increased the efficiency and accuracy of landslide mapping, analysis, and monitoring (Lin et al., 2004; Petley, 2012; Amatya et al., 2021; Casagli et al., 2023). The relatively recent availability of medium- to high spatial and temporal resolution satellite imagery (e.g., Sentinel-2 visible bands at 10 m every 5–10 days, PlanetScope at 3 m daily), now allows scientists to observe the planet in near real-time and high detail (Finer et al., 2018). The high spatial resolution is important because previously, many freely available images (e.g., Landsat at 30 m) were not at sufficient spatial resolution to capture small landslides, an important component of landslide regimes (Clark et al., 2016; Fayne et al., 2019). Satellites with high temporal resolution largely solve the challenge of procuring cloud-free imagery from a given time period (Roy et al., 2021), which can be difficult in tropical montane locations. However, satellite imagery is just one tool of many useful for landslide research (Casagli et al., 2023). There are myriad other existing and in-development remote sensing tools, some of which are integrated in the vision outlined here, that could also be applied to these endeavors, each with their unique combination of spatial resolution, revisit time, scene extent, and accuracy (Casagli et al., 2023).

Studies from TMF in Puerto Rico, Hawaii, Jamaica and elsewhere have identified general types of plants present during landslide succession (Guariguata, 1990; Dalling, 1994; Restrepo and Vitousek, 2001; Walker et al., 2010). However, Andean TMF are home to hyperdiverse and unique floral assemblages, with ∼50% of plant species endemic to the region (Myers et al., 2000). There remains a gap in our understanding of the types and diversity of landslide-establishing plants in the Andes. For example, the families of Asteraceae, Melastomataceae, Poaceae, and Solanaceae are both species-rich (Pérez-Escobar et al., 2022) and establish on Andean landslides during succession (Kessler, 1999; Ohl and Bussmann, 2004; Restrepo et al., 2009; Meier, 2013; Freund, 2022), meaning there may be groups of plants never or rarely observed on landslides in other regions that play important roles in Andean landslide succession. Fully cataloguing the diversity of plants found on landslides will shed light on the role of these disturbances, which provide habitat for species that do not establish, survive or grow in closed-canopy forests (Kessler, 1999), in maintaining TMF biodiversity at regional scales (Ohl and Bussmann, 2004; Richter et al., 2009).

This knowledge gap is compounded by the fact that existing information comes from piecemeal, rather than systematic, samples of a relatively small number of Andean landslides. Plant diversity on landslides is known to vary with age (a proxy for successional stage), elevation (Ohl and Bussmann, 2004; Bussmann et al., 2008), and the presence of residual forest soils (Walker et al., 1996). But in the Andes many landslides occur in steep, inaccessible parts of the landscape, making systematic field collection of vegetation data difficult or precluding it entirely. This biases the understanding of landslide vegetation to only the most readily accessible landslides (e.g., those near research stations or roads), complicating efforts to understand how biodiversity on regenerating landslides changes with time and/or across elevational or other environmental gradients. Field studies of Andean landslides have produced information from a relatively small sample of landslides of disparate (and sometimes unknown) ages between 1,400–2,800 m a.s.l. (Stern, 1995; Kessler, 1999; Ohl and Bussmann, 2004; Meier, 2013), just a portion of the expansive TMF elevational gradient found across much of the Andes. We still do not know how the types and diversity of plants occupying regenerating landslides varies with elevation, particularly below 1,400 m and above 2,800 m. Given the pool of colonizing tree species and the community of animal seed dispersers (and presumably pollinators) is known to continuously turn over across TMF elevational gradients (Patterson et al., 1998; Jankowski et al., 2013; Lough, 2017; Mena and Pacheco, 2020), the suite of species that establish on landslides, and therefore the successional process itself, likely also differs across low- and high-elevation landslides.

High resolution satellite and unoccupied aerial vehicle (UAV) imagery (Figure 2), as well as other remote sensing techniques, can help mitigate these biases by expanding the extent of TMF in which tree biodiversity can be assessed. Previous work has demonstrated that high resolution satellite imagery (Quickbird at 2.4 m and RapidEye at 5 m, respectively) alone or in combination with LiDAR can be used to measure tropical tree diversity (Fricker et al., 2015; George-Chacon et al., 2019). Very high resolution WorldView-2 satellite (Wagner et al., 2018) and UAV (Peck et al., 2012) imagery have also been used to delineate and identify tree species. On landslides specifically, high- resolution satellite and UAV imagery (at resolutions <1 m) have been used to map land cover and manually identify tree species (Furukawa et al., 2021; Freund, 2022; Saito et al., 2022). Deep learning algorithms can automate the identification of both canopy and shorter-statured species (such as those present during early stages of landslide succession), including in regenerating forests, from UAV-acquired images (Zhang et al., 2020; Moura et al., 2021; Veras et al., 2022). Using these tools, landslide researchers can now investigate ecologically relevant questions such as how plant species richness and diversity vary on landslides of similar ages across expansive elevational gradients (400–3,800 m in much of the Andes), as well as how these metrics vary across landslides of different ages within elevational bands, to better characterize the process of TMF regeneration after landslides.

Understanding how populations of landslides influence carbon uptake and storage, primary productivity, and related ecosystem processes at the landscape scale will help elucidate the resilience of montane landscapes to global change (Restrepo et al., 2009). Quantifying the ecological influence of landslides can be achieved by using air- and/or satellite-borne hyperspectral imaging, or imaging spectroscopy, to measure plant functional diversity (Jetz et al., 2016; Asner et al., 2017) of landslide-affected and undisturbed areas. This approach views plant biodiversity through the lens of species’ structural and biochemical functional traits, which correspond to their roles in ecosystem processes and services (Díaz et al., 2007). Field studies of TMF plant functional diversity have demonstrated elevational and topographic trends in plant functional diversity (Duivenvoorden and Cuello, 2012; Homeier et al., 2021; Báez et al., 2022b; Pierick et al., 2023) though these have largely focused on adult trees in mature forest (Báez et al., 2022a) covering a relatively small proportion of the wider landscape. An exception to these small-scale studies is Asner et al.’s (2014a) airborne imaging spectroscopy and LiDAR-based study, which revealed high landscape diversity and elevational turnover in functional traits and forest structure across the Amazon-Andes gradient and demonstrated the clear signature of landslides on the landscape (Asner et al., 2014a). There remains much to learn about the plant functional diversity of disturbed and regenerating Andean TMF, and how it contributes to ecosystem processes.

Hyperspectral imaging measures electromagnetic reflectance from surfaces, in this case forest canopies, in narrow spectral bands (typically 5–10 nm) in the electromagnetic spectrum ranging from ultraviolet through medium-wave infrared (400–2,500 nm). The technique can be used to detect the chemical properties of trees, among them functional traits linked to photosynthesis, primary production, defense, biogeochemical processes, and evolutionary history. Unique combinations of these values can be used to accurately assign functional traits to canopy assemblages and even describe individual species (Clark et al., 2005), especially when paired with remote sensing techniques to measure forest structure, such as LiDAR (Lucas et al., 2008; Asner et al., 2015; Shi et al., 2018). While many studies have applied imaging spectroscopy to the detection, monitoring, and characterization of landslides (e.g., Vellico et al., 2010; Ye et al., 2019), to our knowledge this method has not yet been applied to studies of landslide regeneration in highly diverse tropical forests. However, given its power for measuring tropical plant functional diversity and linking plant communities to ecosystem-level processes (see review in Asner et al., 2017), it is a powerful option for understanding the role of landslides in Andean TMF.

Technological advances in remote sensing technology continue to expand scientific understanding of current and future landslide regimes, particularly by opening new possibilities for mapping, monitoring, and studying these natural phenomena at very large spatial scales (Guzzetti et al., 2012; Casagli et al., 2023). One additional challenge for large-scale landslide mapping not readily solved by improvements to satellite imagery alone is that the creation of landslide inventory maps has traditionally required researchers to manually examine aerial and satellite imagery to delineate affected areas (e.g., Guns and Vanacker, 2014; Clark et al., 2016; Delgado et al., 2022). However, high-resolution satellite data paired with computer vision for object and change detection (Zhong et al., 2020; Amatya et al., 2021; Lu et al., 2022) unlocks the ability to identify past landslide activity and monitor landslides in near real-time at regional and even continental scales (Yang et al., 2022). For example, this approach has recently been used to detect and map landslides in Nepal (Prakash et al., 2021; Meena et al., 2022), Taiwan, China, Japan (Ghorbanzadeh et al., 2021, 2022), and the Patagonian Andes (Morales et al., 2022). In the Andes, Morales et al. (2022) applied a convolutional neural network to Sentinel-2 images to develop a 10,000-landslide inventory covering approximately 20,000 km². Their model, the first automated landslide detection model in the region, was most accurate in areas with vegetation cover (Morales et al., 2022), suggesting this approach will work well across forested regions of the Andes.

Building on the success of Morales et al. (2022), we propose that a combination of deep learning and high-resolution satellite imagery be used to develop and maintain an automated “living landslide inventory” spanning all Andean TMF. By maximizing the spatial extent of study, and therefore range (and possible combinations) of abiotic variables represented, this effort would increase scientific understanding of how the environment and human activity (e.g., Guns and Vanacker, 2014) shapes landslide occurrences, extents, and frequencies without the limitations and biases inherent to studies at smaller spatial scales (Lobo and Dalling, 2014; Marvin et al., 2014). It will also lay the groundwork for studies of how South American landslide regimes shift in response to global climate change, a critical knowledge gap (Gariano and Guzzetti, 2016) and fill an important environmental planning and policymaking need for local, regional, and national governments in Andean countries (Hermanns et al., 2012). Here, we detail several specific scientific contributions that would be made possible by this pan-Andean landslide inventory.

Although landslides are a large and severe example of natural forest disturbances, their study is rooted in fundamental principles of ecosystem succession, plant physiology and demography, and landscape ecology. The three axes of inquiry we pose here advance important and timely questions about landslide ecology in the tropical Andes biodiversity hotspot. Answering them will require a variety of advanced remote sensing methods, including but not limited to high-resolution UAV and satellite imaging, airborne LiDAR, and satellite-borne and airborne imaging spectroscopy. This work will improve scientific understanding of the region’s biodiversity, natural disturbance dynamics, ecosystem functioning, and responses to climate change. Achieving an automated pan-Andes landslide inventory will have additional benefits for environmental policy and planning in the region, thus increasing the resilience of human and ecological communities in this critical biodiversity hotspot.

CF was responsible for the conceptualization and writing of the manuscript. MS assisted with the conceptualization and gave input on the manuscript. Both authors contributed to the article and approved the submitted version.