Demand for Ecosystem Services Drive Large-Scale Shifts in Land-Use in Tropical Mountainous Watersheds Prone to Landslides

Abstract

An increasing frequency of extreme atmospheric events is challenging our basic knowledge about the resilience mechanisms that mediate the response of small mountainous watersheds (SMW) to landslides, including production of water-derived ecosystem services (WES). We hypothesized that the demand for WES increases the connectivity between lowland and upland regions, and decreases the heterogeneity of SMW. Focusing on four watersheds in the Central Andes of Colombia and combining “site-specific knowledge”, historic land cover maps (1970s and 1980s), and open, analysis-ready remotely sensed data (GLAD Landsat ARD; 1990–2000), we addressed three questions. Over roughly 120 years, the site-specific data revealed an increasing demand for diverse WES, as well as variation among the watersheds in the supply of WES. At watershed-scales, variation in the water balances—a surrogate for water-derived ES flows—exhibited complex relationships with forest cover. Fractional forest cover (p_i) and forest aggregation (AI_i) varied between the historic and current data sets, but in general showed non-linear relationships with elevation and slope. In the current data set (1990–2000), differences in the number of significant, linear models explaining variation in p_i with time, suggest that slope may play a more important role than elevation in land cover change. We found ample evidence for a combined effect of slope and elevation on the two land cover metrics, which would be consistent with strategies directed to mitigate site-specific landslide-associated risks. Overall, our work shows strong feedbacks between lowland and upland areas, raising questions about the sustainable production of WES.

1. Introduction

An increasing frequency of extreme atmospheric events [1,2] is challenging our basic knowledge about the resilience mechanisms [3,4,5] that mediate macrosystem, i.e., multicomponent and multiscale systems, responses to such events [6]. Theory suggests that the connectivity and spatiotemporal heterogeneity of a macrosystem strongly influences the propagation of external perturbations [3,7]. Specifically, perturbations propagate effectively when connectivity is high and heterogeneity is low [8,9]. A corollary is that urban, exurban, and rural regions increasingly connected through the supply and demand of ecosystem services are likely to be affected by perturbations affecting one or more components of these linked systems (Figure 1). Mountains prone to landslides and the flow of water-derived ecosystem services are a case in point. In these environments, rural and exurban regions are becoming more connected with urban regions, and also less heterogeneous due to land use change, opening the possibility for complex interactions among geophysical, biological, and socio-cultural components, and ultimately the resilience of these systems to extreme events (Figure 1) [10]. One consequence of these complex interactions may entail a reduction of ecosystem flows due to an increase of landslide-associated risks, which has led to the implementation of diverse mitigation strategies [11,12,13]. Chief among the latter have been guided efforts to change land cover and land use [14,15]. Yet, in the tropics, unlike temperate regions [16], there have been limited efforts to evaluate these mitigation strategies targeted at reducing landslide risks. The integration of land cover data from historic and remote sensing sources [17] may provide unique opportunities to examine the results of interventions aimed at influencing landscape characteristics in tropical small mountainous watersheds (SMW), while providing baseline information for years to come.

Increased connectivity or hyper-connectivity and loss of heterogeneity in rural and exurban areas prone to landslides has been enhanced by human activities with multiple consequences on the production and consumption of ecosystem services along urban–rural gradients (Figure 1, [13,18,19]). First, water infrastructure projects built to move water from production to consumption sites, have linked regions that typically were not previously connected [20,21,22]. Landslides triggered by extreme atmospheric events may contribute large volumes of debris and sediments that may test the ability of human-built systems to produce and deliver clean water for human consumption and hydropower generation, far beyond production sites. Second, the development of road and telecommunication networks has facilitated access to land and services, ultimately supporting people livelihoods [20,21,22,23]. Yet, in areas prone to landslides with impoverished economies, limited planning, and poor social networks, this built infrastructure network may become an added vulnerability to people and livelihoods [24,25]. Third, incentives that promote a limited set of agricultural activities may lead to landscape homogenization over large areas [26,27,28,29]. This has the potential to create ecosystem disservices, including the reduction of biodiversity, water yields, and air quality, as well as the increase of driftwood from landslides [26,29]. These changes not only impact the livelihoods of rural communities, but also the supply of diverse agricultural goods along urban–rural gradients [30].

In mountainous areas, increasing tree cover through the management of land-cover and land-use has been an important strategy to mitigate landslide risk while enhancing the flow of ecosystem services. Two important assumptions underly this strategy. The first is that landslides have negative impacts on ecosystems without considering their possible role in the maintenance of diversity or ecosystem function [31]. The second assumption is that trees increase slope stability [14,15,28], although this may not always be the case [16,32]. In addition, an increase of tree cover contributes to the regulation of the hydrological and carbon cycles [26]. The latter function has become particularly important at a time when carbon sequestration has become a major strategy to mitigate climate change [33]. Global and regional studies at biome and/or government unit scales have documented changes in forest cover in mountainous regions [26,34,35,36], yet their ability to reveal drivers or mechanisms of change have been limited. Regional to local studies, on the other hand, have the potential to reveal complex institutional arrangements, diverse mechanisms to increase tree cover, and variation of the scales at which these mechanisms operate, e.g., [28,37,38]. Among local studies, those focusing on watersheds can be particularly informative about drivers of land-cover and land-use change due to the diverse ways in which these systems are connected, and the variability of geophysical, biological, and socio-cultural factors therein [39]. This connectivity and variability may be important in discussions that center on the sustainability, and more broadly speaking the resilience of SMW.

1.1. Overall Hypothesis and Objectives

We propose to integrate various sources of data such as historic accounts, land cover and land use maps, and remotely sensed open, analysis-ready data (ARD; [40]), to monitor long-term interventions aimed at mitigating landslide-associated risks and ultimately, enhance water-derived ecosystem services (WES) in tropical SMW. Integrating site-specific knowledge will contextualize the observations in an effort to understand drivers of change. We hypothesized that the demand for WES increased the connectivity between lowland and upland regions, and decreased the heterogeneity of our target SMW. Focusing on four tropical SMW located in the Central Andes of southwestern Colombia we addressed three questions. First, to what extent changes in forest cover reflect variation in the flow of WES at regional and watershed-scales? Second, do changes in forest cover vary with elevation or slope, in a manner consistent with the enhancement of WES or the mitigation of landslide-associated risks, respectively? Lastly, do changes in forest cover vary with slope but depending on elevation in a manner consistent with strategies directed to mitigate site-specific landslide-associated risks? Addressing these questions is important to anticipate unforeseen consequences associated with an increasing demand for ecosystem services.

1.2. Case Study

In this application of remote sensing, we examine temporal changes in land cover and land use in four SMW located in the Middle Cauca river region of the Valle del Cauca department, the third largest contributor to Colombia’s population and GDP (Figure 2A,B [41,42]). Here, the Cauca river runs in a south–north direction through one of the most fertile regions of Colombia—the Cauca valley—and receives the waters from numerous tributaries draining the east- and west-facing slopes of the Western (mean elevation 1464 m a.s.l.; range 884–4014) and Central (2097 m a.s.l.; range 890–4214) Cordillera of the Andes, respectively. The interplay of geologic, geomorphologic, climatologic, biogeographic, and anthropogenic processes not only underlies the enormous diversity of landforms, soils, life-zones, landscapes, ecosystems, and communities within this region [43], but also disturbance regimes that include earthquakes, volcanic eruptions, droughts, flooding, and landsliding [44]. Altogether, these processes have played a key role in the history of occupation of the region [45].

Our focal watersheds, Nima, Bolo, Fraile, and Desbaratado (Figure 2C,D), have been long-time, key players in the economy of the Cauca valley, Valle del Cauca department, and Colombia (Figure 3; Supplementary Materials File S1). Starting in the late 19th and early 20th century, these watersheds began to supply water for domestic, agricultural, and industrial uses, ultimately contributing to the transformation and homogenization of the Cauca valley [46,47,48]. The four SMW differ in their geophysical, biological, and socio-cultural settings, yet they also share similarities, including their high susceptibility to landslides [38,49]. Not surprisingly, landslide-associated risks and the intensive management of water have been intertwined throughout the recent history of this region due to its dependency on WES (Figure 3; Supplementary Materials File S1).

Figure 2. Location of studied watersheds within (A) Colombia (green), (B) the Department of Valle del Cauca (green dots; from left to right the cities of Palmira, Cali, and Buenaventura; shaded grey region the focal watersheds; polygon in fuchsia the Amaime watershed), (C) digital elevation model with focal watersheds (Nima, Bolo, Fraile, and Desbaratado), (D) characteristics of the watersheds, and (E) Global Precipitation Measurements (GPM_3IMERGM V06; data bounding box −76.05, 3.45, −76.05, 3.45) [50].

Figure 2. Location of studied watersheds within (A) Colombia (green), (B) the Department of Valle del Cauca (green dots; from left to right the cities of Palmira, Cali, and Buenaventura; shaded grey region the focal watersheds; polygon in fuchsia the Amaime watershed), (C) digital elevation model with focal watersheds (Nima, Bolo, Fraile, and Desbaratado), (D) characteristics of the watersheds, and (E) Global Precipitation Measurements (GPM_3IMERGM V06; data bounding box −76.05, 3.45, −76.05, 3.45) [50].

In 1954, the upper and middle Cauca river, including tributaries, began to be administered by the newly created Corporacion Autónoma Regional del Cauca-CVC (Figure 3; Supplementary Materials File S1 [46]). In 1970, the CVC formulated a watershed plan for Nima, the first for any watershed within CVC’s jurisdiction and one of the first for Colombia [38,51]. This management plan and others that followed, are extremely valuable for at least two reasons. First, they provide base line data to investigate the “developmental trajectories” of tropical SMW driven by an increasing demand of WES. Second, these management plans describe natural hazards, including landslide-associated risks affecting WES. In 1994 the CVC was re-structured and renamed as the Corporación Autónoma Regional del Valle del Cauca to reflect jurisdictional and functional changes aligned with new national legislation (Colombia’s environmental Law 99 of 1993 [46]); from now on, we use CVC_or and CVC to refer to the original and restructured institutions, respectively.

2. Methods

2.1. Watershed Delineation

Watershed delineation either by CVC_or or CVC has been based on management criteria. For example, Nima is a sub-watershed of Amaime, but due to historical reasons, it was managed independently until 2013 (Figure 3; Supplementary Materials File S1). This explains why current water balances are estimated for Amaime and not Nima. Additionally, in the watershed management plans developed by CVC_or, the watersheds were delineated based on outlet points located at the mountain foothills/Cauca valley boundary (~900 m a.s.l.); it is at these points where the water from the rivers begins to be distributed for agricultural uses. Based on these considerations, we used ArcHydro Tools 10.7 [52] in ArcMap 10.7.1 to generate a stream network and delineate the watersheds of interest based on biophysical features. The input data included a 30-m DEM (ASTGTM v003 [53]) and an official stream network (Rios [54]).

2.2. Water Balances

The water balance estimates [difference between supply (precipitation and surface waters) and demand (domestic, industrial, agricultural)] of the target watersheds were obtained from CVC [55]. Based on these figures, the watersheds were ranked from the highest to the lowest deficit as follows: Bolo > Amaime (Nima) > Fraile > Desbaratado. This ranking was based on scenario 3, which expects that 90% of the precipitation will be observed.

2.3. Historic Land Cover Datasets

We identified four historic forest cover maps, two for the Nima (late 1960s–early 1970s and late 1980s [49,51]) and two for the Bolo-Fraile-Desbaratado (mid-1970s and mid-1980s [56,57]) watersheds (

Table 1. Historical datasets, including, elements digitized, original formats, scale, and sources.

Period	Watershed	Year	Map	Elements Digitized	Format	Scale	Source
1970s	Nima	1969	Current land use map	Forest, natural and artificial grasslands, permanent crops, annual crops, fallow, areas undergoing erosion, water bodies, rock surfaces in paramo.	PDF; extracted from watershed management report; converted to TIFF	1:100,000	Proyecto para el Manejo de la Cuenca Superior del Rio Nima. Mapa No. 8. Uso Actual de la Tierra	Ecopedia (https://ecopedia.cvc.gov.co/ accessed on 1 November 2020)	[51]
	Bolo-Fraile-Desbaratado	1977	Forest map	Forest type	PDF; extracted from watershed management report; converted to TIFF	1:100,000	Dibujo No. 722-09-6-1975 Cuenca Hidrográfica de los íos Bolo, Fraile y Desbaratado. Mapa de Bosques	CVC-Biblioteca	[57]
	Bolo-Fraile-Desbaratado	1977	Land use map	Forest, natural and artificial grasslands, permanent crops, annual crops, fallow, areas undergoing erosion, water bodies, rock surfaces in paramo.	PDF; extracted from watershed management report; converted to TIFF	1:100,000	Dibujo No. 722-10-11-1975 Cuenca Hidrográfica de los íos Bolo, Fraile y Desbaratad.Mapa de Uso Actual del Terreno	CVC-Biblioteca	[57]
		1969		Forest	Paper scanned to PDF; converted to TIFF.	1:25,000	Plancha No. 280-IV-C. 1:25,000	IGAC	[58]
		1969		Forest	Paper scanned to PDF; converted to TIFF.	1:25,000	Plancha No. 280-IV-D. 1:25,000	IGAC	[59]
		1969		Forest	Paper scanned to PDF; converted to TIFF.	1:25,000	IGAC. Carta General. Plancha No. 300-II-A. 1:25,000	CVCor	[60]
1980s	Nima	1989	Land use map	Natural forest; [planted forest]	Paper scanned to PDF (resolution 2160 x 2696); converted to TIFF.	1:25,000	Dibujo No. 713-60-21 Proyecto para el Manejo de la Cuenca Superior del Rio Nima. Mapa de Uso Actual	CVCor	[49]
	Bolo-Fraile-Desbaratado	1989	Land use map	Natural forest, planted forest	Paper scanned to PDF; converted to TIFF.	1:50,000	Dibujo No. 722-09-19 UMC Rios Bolo-Fraile Desbaratado. Mapa de Uso Actual	CVCor	[56]
		1984		Forest	Paper scanned to PDF; converted to TIFF.	1:25,000	Plancha No. 300-II-B. 1:25,000	IGAC	[61]

). In addition, we used the first edition (1960s) of Colombia’s topographic maps covering parts of our study region [58,59,60,61]. According to the data sources, the maps were derived from the photointerpretation of aerial photographs and/or field verification.

Some maps were obtained as hard copies from CVC_or or the Instituto Geográfico Agustín Codazzi, in which case we scanned them into PDF files; other maps were extracted from digital reports (mid-1970s and mid-1980s [51,56,57]) in PDF format (Table 1). The PDF files were converted into TIFF images, and geo-referenced in ArcGIS Pro version 2.4 using the Bogotá Datum as shown on the original maps. Subsequently, the maps were projected to the MAGNA Colombia Oeste system based on the Molodensky–Badekas model [62].

For each of the two periods, we digitized the elements of interest, i.e., natural and planted forests, and homogenized the Nima and Bolo-Fraile-Desbaratado maps in QGIS version 3.13. Data homogenization was important to (1) align the limits of the mapped watersheds between themselves and with those that we generated, and (2) align the elements of interest, in our case forests, in border areas.

2.4. Current Land Cover Datasets

The current forest cover (1999–2020) was derived from the classification of Landsat Analysis Ready Data supplied by the Global Land Analysis and Discovery laboratory (GLAD) [63] of the Department of Geographical Sciences at the University of Maryland. The GLAD Landsat ARD products represent a 16-day time-series of tiled Landsat normalized surface reflectance composites with minimal atmospheric contamination [40,64,65].

Our study area fell within tile 076W_03N in the GLAD reference system. We identified a total of 506 multispectral images (January 1999 (image No. 438) to December 2020 (image No. 943)) suitable for our work and proceeded to fill pixels of low quality, including cloud contamination, using the GLAD Landsat ARD Tools v1.1 [40,66]. Subsequently, we generated 89 phenological metrics that together with two topographic variables (elevation and slope) derived from the SRTMGL1 v003 DEM [67] were used as input for land cover classification purposes [65,68,69].

Obtaining forest cover maps involved two major steps. In a first step, we generated training sets through the digitization of small polygons that were classified based on a six-level hierarchical classification system. The coarsest level (A) represented forest and non-forest or background areas. The other five levels contained information on dominant life-form (B), use (C), density of dominant life form (D), degree of deciduousness (E), and leaf type (F). For the purposes of this work, we focused on level A. Thus, forest include natural and planted forests, as well as permanent crops dominated by trees. The polygons were digitized in ArcGIS Pro 2.6 using base-map imagery (Maxar Vivid: 2/22/2015, 1/28/2016, 2/27/2016, 10/10/2017; 2/12/2018, 8/16/2018; Maxar Digital Globe: 9/20/2016) and two GLAD Landsat ARD images (2018 (image No. 878 and 883)). The shapefile was subsequently split in two, one containing the forest and the other the background polygons. In a second step, we used the 89 metrics, topographic variables, and training sets to run a classification tree for the year 2018 with GLAD tool v1.1 [40]. The resulting binary map of forest versus non-forest areas was visually inspected, and to improve the results we revised the training sets and repeated the entire process for a second time. The producer’s and user’s accuracy were 98% and 97%, respectively, and the overall accuracy 97%. The adjusted classification tree was used to classify the remaining 19 images (2002–2020). In a Linux server, we created a virtual Windows machine with Vagrant (https://www.vagrantup.com/ accessed on 1 November 2020) to run GLAD tool v1.1.

2.5. Data Analysis

Our work focuses on four SMW that exhibit some similarities, but also differences that may explain variation in the trajectories of these watersheds (Figure 2 and Figure 3). Based on this, we chose three scales of analyses to investigate changes in land cover as a function of elevation and slope. Our regional scale encompasses all four watersheds (All or R4), our watershed scale focuses on each individual watershed (WS_i), and the sub-regional scales encompasses three watersheds excluding the target ones (R3; All-WS_i). The latter represent to some extent a mirror image of any given watershed, allowing us to understand the relative contribution of each individual watershed (WS_i).

We used two landscape metrics, namely forest fractional or proportional cover and forest aggregation, to characterize the spatial and temporal variation of land cover. Fractional or proportional forest cover (p_i; 0–1) is a compositional metric that quantifies the abundance of cover class i. He’s [70] aggregation index (AI_i; 0–100) is a spatial configuration metric that quantifies aggregation levels within a given class i. This index is based on e_i,I the pixel edges shared with class i and max e_i,I, the largest possible number of shared edges for class i. We reasoned that AI_i, would characterize forest connectivity. For simplicity, we will refer to p_i and AI_i as forest cover and forest aggregation, respectively.

The aforementioned metrics were examined as a function of elevation and slope. We reclassified the SRTMGL1 v003 DEM into four categories (Low or tropical (EL, 0–1000), Medium or premontane (EM, 1000–2000), High or lower montane (EH, 2000–3000), Very high or montane plus paramo (EVH, 3000–4500) elevations; units m a.s.l.) following Colombia’s life zones sensu Holdridge [71]. The very high elevations include paramo ecosystems that are key for the water economy of the Valle del Cauca and Colombia [72,73]. The same DEM was used to generate a slope map that was subsequently reclassified into three classes (Low or gentle (SG, 0–15), Medium or steep (SS, 15–35), and High or very steep (SVS, 35–75) slopes; units in degrees). These classes are based on a slope stability study in southeast Asia documenting variation in landslide frequency as a function of slope [74]. In Colombia, historic information indicates that the protection of forest in slopes >22 degrees was of concern as early as 1953 [75].

We combined analysis of covariance (Ancova) and regression analyses to examine temporal trends in the two metrics function of scale, and elevation and slope within scale. In most Ancova models, a significant interaction between the covariate (year) and independent factors (elevation and slope) indicated that the assumption of slope homogeneity was not met (Supplementary Material Table S1 [76]). Thus, we used regression analysis. The raster analyses were conducted with the raster (3.4–13), rgdal (1.5–27) [77,78], and landscapemetrics [79] R packages in RStudio server 1.4.1717 running R 4.1.1. [80].

2.6. Data and Analysis Limitations

Our historic reconstruction of forest cover has some limitations that became evident while developing the time line of events that shaped water and land-use relationships (Figure 3; Supplementary Materials File S1). First, our sources of information indicate that prior to the creation of CVC_or, both lowland and upland forests were experiencing large-scale transformations. Second, CVC_or used the limits of the Cauca valley (see 2.1 above) to identify the outlets of the SMW. Thus, the management plans prepared by CVC_or did not include lowland regions. This could have been important to examine connectivity within and across regions. Future studies may undertake an in-depth review of cadastral documents, property titles, and permit water and forest use requests submitted to various agencies. Third, comparisons among the historic datasets, and between these and the current ones, may not be straight forward. This is due to differences in the original data sources, mapping methodologies, mapping scales, and georeferencing ([17] Table 1). Nevertheless, these datasets provide a unique record of watershed management paradigms, as well as processes and patterns shaping our target watersheds. Lastly, the index of aggregation, AI_i, seemed to be sensitive to the mapping source.

3. Results

3.1. Basin-Scale Land Cover and WES Relationships

Our first question focused on basin or whole watershed patterns. At the regional scale (R4; bottom, right Figure 4A and Figure 5A), mean forest cover (p_i) was 0.13 and 0.44 in the historic and current datasets, respectively, whereas mean forest aggregation (AI_i) was 96.7 and 93.2. In the current dataset, AI_i but not p_i increased significantly over time (Table S1), both at watershed and regional scales (WS_i and R4; Figure 4A and Figure 5A, bottom row).

We focused on the current data set and included time and watershed (WS_i) in two Ancovas. In the first, forest cover varied significantly with WS_i, but not time nor their interaction (Table S2A). A post hoc Tukey test on WS_i revealed that the mean forest cover of Nima (0.64) differed significantly from that of Bolo (0.37), Fraile (0.45), and Desbaratado (0.46); in turn, the mean forest cover of Bolo differed from that of Desbaratado and Fraile, whereas the mean forest cover of the latter two did not differ. In the second Ancova, forest aggregation varied significantly with WS_i and time, but not their interaction (Table S2A). These results together with regression analyses (Table S1) show that forest aggregation increased significantly over time, and that the rate of increase was indistinguishable among WS_i.

We partitioned the data (Nima, Fraile, Desbaratado, and R4 vs. Bolo, Fraile, Desbaratado, and R4) to examine relationships between forest cover and the water balance (wb) estimates (the lowest value was subtracted from each data point to eliminate negative numbers and fit logarithmic models). The first model described a negative (p_i = −0.05ln(wb) + 0.69, R² = 0.92), whereas the second described a positive (p_i = 0.017ln(wb) + 0.37 R² = 0.99) relationship between the two variables.

3.2. Land Cover Change and the Independent Effects of Elevation and Slope

The second question examined the independent effects of slope and elevation on forest cover (p_i) and forest aggregation (AI_i). In the current dataset and irrespective of scale (R4, WS_i), most of the Ancovas did not meet the assumption of homogeneity of slopes, that is the effect of elevation or slope on p_i and AI_i was not consistent across time (Table S2B). This prompted us to (1) examine broad patterns of variation in forest cover and forest aggregation within the historic and current datasets, and (2) examine temporal trends using linear models.

Historic mean forest cover varied with elevation in a manner suggestive of an inverted U (Figure S2A), and increased with slope in a non-linear fashion, both at regional (R4) and watershed (WS_i) scales (Figure S2B). These patterns were stronger in the current dataset (Figure S2A,B). Forest aggregation varied in slightly different ways. First, mean AI_i was larger in the historic than in the current dataset (Figure S2C,D). Second, the historic and current mean values of AI_i increased with elevation in a non-linear fashion or reached an asymptote. This trend, observed both at regional (R4) and watershed (WS_i) scales, is suggestive of an increase in connectivity among forest patches with elevation (Figure S2C). In the historic dataset, the lowest AI_i was for Nima at mid elevations. In contrast, AI_i varied with slope in a manner suggestive of an inverted U with the largest mean AI_i observed at steep slopes (Figure S2D).

Differences in the number of significant, linear models explaining variation in forest cover (p_i) with time, suggest that slope may play a more important role than elevation in land cover change. Among the elevation models, p_i significantly increased with time in five out of nineteen models (high elevations at R4 and the four watersheds; Table S3A; Figure 4A); in contrast, p_i significantly decreased in one out of nineteen (low elevation Fraile; Table S3A; Figure 4A). Among the slope models, p_i significantly increased with time in eight out of fifteen (steep and very steep slopes in R4, Nima and Bolo, and very steep slopes in Fraile and Desbaratado; Table S3A; Figure 4B) models. In contrast, p_i decreased with time in three out of fifteen (gentle slopes in Nima, Fraile, and Desbaratdo; Table S3A; Figure 4B) models.

Not only were more models of forest aggregation (AI_i) versus time significant, but they also involved steeper slopes compared to p_i. In regard to elevation, AI_i significantly increased with time in 11 out of 19 (mostly very high and high elevations at R4, Bolo, Fraile, and Desbaratado, high elevations at Nima, and medium elevation at Nima and Desbaratado; Table S3B; Figure 5A) models. In regard to slope, AI_i significantly increased in 11 out of 15 (very steep slopes in R4 and the four watersheds as well as steep slopes in Nima and Bolo; Table S4B; Figure 5B) models.

Figure 4. Temporal variation of fractional forest cover (p_i) as a function of (A) elevation and (B) slope at regional (black; R4), watershed (green; WS_i), and sub-regional (grey: R3) scales. The vertical dotted line separates the historic and current datasets. Numbers within rectangles represent significant slopes in linear models.

Figure 4. Temporal variation of fractional forest cover (p_i) as a function of (A) elevation and (B) slope at regional (black; R4), watershed (green; WS_i), and sub-regional (grey: R3) scales. The vertical dotted line separates the historic and current datasets. Numbers within rectangles represent significant slopes in linear models.

Figure 5. Temporal variation of index of aggregation (AI_i) as a function of (A) elevation and (B) slope at regional (black; R4), watershed (green; WS_i), and sub-regional (grey: R3) scales. The vertical dotted line separates the historic and current datasets. Numbers within rectangles represent significant slopes in linear models.

Figure 5. Temporal variation of index of aggregation (AI_i) as a function of (A) elevation and (B) slope at regional (black; R4), watershed (green; WS_i), and sub-regional (grey: R3) scales. The vertical dotted line separates the historic and current datasets. Numbers within rectangles represent significant slopes in linear models.

3.3. Land Cover Change and the Combined Effect of Elevation and Slope

The third question examined the combined effect of slope and elevation on forest cover (p_i) and forest aggregation (AI_i). The historic data, shows that mean p_i varied with elevation in a manner suggestive of an inverted U, yet the effect was not consistent among slopes; this was observed both at the region (R4) and watershed (WS_i) scales. Specifically, mean p_i was smaller on gentle than steep and very steep slopes at high to very high elevations (Nima, Bolo, R4; Figure S3A). The historic data also showed that p_i was highest in Nima on very steep slopes at high elevations (Figure S3A).

In the current dataset the aforementioned patterns were very strong, that is the relationship between mean forest cover (p_i) and elevation clearly described an inverted U. In addition, the curves for the gentle slopes departed considerably from those of the steep and very steep slopes (Figure S3A). Interestingly, mean p_i is similar among slopes at high elevations. In contrast, mean forest aggregation (AI_i) increased or reached an asymptote with elevation on steep and very steep, but not gentle slopes. Furthermore, mean AI_i was lowest on gentle than steep and very steep slopes particularly at high and very high elevations (Figure S3B). Interestingly, mean AI_i on very steep slopes was intermediate between steep and gentle slopes. Patterns observed in the current data mirrors those from the historic data with the exception of Desbaratado (Figure S3B).

The linear models describing temporal changes in the two land cover metrics indicate that most change is occurring on steep and very steep slopes at high elevations, and gentle slopes at different elevations. In 18 out of 49 (R4 and WS_i;

Table 2. Results of linear models between two forest metrics and time as a function of the combined effects of Elevation and Slope. . <0.1, * <0.05, ** <0.01, *** <0.001.

			Proportion of Forest (pi)				Forest Aggregation Index (AIi)
Watershed	Elevation	Slope	Intercept	Slope	R2	p	Intercept	Slope	R2
All (R4)	Very High	Very steep	0.790	0.000	0.002	8.4 × 10−1	82.171	0.028	0.289	1.7 × 10−2 *
		Steep	0.661	0.000	0.003	8.2 × 10−1	85.189	0.049	0.788	3.9 × 10−7 ***
		Gentle	0.433	−0.005	0.152	9.9 × 10−2 .	70.911	−0.075	0.159	9.0 × 10−2 .
	High	Very steep	0.796	0.008	0.637	4.2 × 10−5 ***	80.445	0.150	0.681	1.3 × 10−5 ***
		Steep	0.763	0.006	0.472	1.1 × 10−3 **	85.431	0.134	0.486	9.0 × 10−4 ***
		Gentle	0.728	0.003	0.139	1.1 × 10−1	63.706	0.047	0.110	1.6 × 10−1
	Medium	Very steep	0.710	0.006	0.419	2.7 × 10−3 **	69.758	0.168	0.498	7.3 × 10−4 ***
		Steep	0.686	0.003	0.106	1.7 × 10−1	85.955	0.052	0.078	2.4 × 10−1
		Gentle	0.246	−0.003	0.147	1.0 × 10−1 .	76.515	−0.145	0.172	7.7 × 10−2 .
	Low	Very steep
		Steep
		Gentle	0.061	−0.002	0.074	2.5 × 10−1	64.336	−0.008	0.000	9.7 × 10−1
Nima	Very High	Very steep	0.778	0.001	0.032	4.6 × 10−1	82.491	0.019	0.245	3.1 × 10−2 *
		Steep	0.719	−0.001	0.037	4.3 × 10−1	86.256	−0.009	0.034	4.5 × 10−1
		Gentle	0.455	−0.007	0.273	2.1 × 10−2 *	72.895	−0.229	0.523	4.6 × 10−4 ***
	High	Very steep	0.858	0.007	0.626	5.5 × 10−5 ***	82.338	0.093	0.516	5.3 × 10−4 ***
		Steep	0.823	0.006	0.488	8.7 × 10−4 ***	87.380	0.146	0.543	3.2 × 10−4 ***
		Gentle	0.811	0.002	0.031	4.7 × 10−1	67.500	−0.046	0.052	3.4 × 10−1
	Medium	Very steep	0.769	0.007	0.396	3.9 × 10−3 **	68.674	0.183	0.484	9.4 × 10−4 ***
		Steep	0.768	0.003	0.140	1.1 × 10−1	88.036	0.072	0.178	7.1 × 10−2 .
		Gentle	0.318	−0.005	0.324	1.0 × 10−2 **	79.158	−0.193	0.271	2.2 × 10−2 *
	Low	Very steep
		Steep
		Gentle
Bolo	Very High	Very steep	0.790	0.004	0.411	3.1 × 10−3 **	82.319	0.052	0.685	1.2 × 10−5 ***
		Steep	0.701	0.003	0.203	5.3 × 10−2 .	86.654	0.049	0.620	6.2 × 10−5 ***
		Gentle	0.445	0.001	0.022	5.4 × 10−1	69.653	0.012	0.007	7.4 × 10−1
	High	Very steep	0.742	0.008	0.535	3.7 × 10−4 ***	79.764	0.168	0.556	2.4 × 10−4 ***
		Steep	0.720	0.007	0.434	2.1 × 10−3 **	85.704	0.145	0.369	5.8 × 10−3 **
		Gentle	0.634	0.005	0.327	1.0 × 10−2 *	65.795	0.090	0.232	3.6 × 10−2 *
	Medium	Very steep	0.568	0.009	0.361	6.5 × 10−3 **	67.501	0.253	0.357	6.8 × 10−3 **
		Steep	0.596	0.005	0.218	4.3 × 10−2 *	85.252	0.067	0.073	2.6 × 10−1
		Gentle	0.207	0.000	0.001	9.1 × 10−1	75.590	−0.030	0.007	7.3 × 10−1
	Low	Very steep
		Steep
		Gentle	0.063	−0.001	0.016	6.0 × 10−1	64.381	0.031	0.001	9.0 × 10−1
Fraile	Very High	Very steep	0.782	−0.002	0.048	3.6 × 10−1	82.358	0.005	0.005	7.8 × 10−1
		Steep	0.614	−0.002	0.043	3.9 × 10−1	83.450	0.066	0.720	4.4 × 10−6 ***
		Gentle	0.438	−0.006	0.214	4.6 × 10−2 *	71.857	−0.043	0.041	4.0 × 10−1
	High	Very steep	0.783	0.009	0.708	6.3 × 10−6 ***	80.843	0.198	0.770	8.0 × 10−7 ***
		Steep	0.739	0.005	0.446	1.7 × 10−3 **	83.909	0.110	0.459	1.4 × 10−3 **
		Gentle	0.745	0.002	0.077	2.4 × 10−1	59.272	0.062	0.288	1.7 × 10−2 *
	Medium	Very steep	0.770	0.005	0.299	1.5 × 10−2 *	71.012	0.158	0.549	2.8 × 10−4 ***
		Steep	0.722	0.000	0.000	9.5 × 10−1	85.988	0.022	0.016	6.0 × 10−1
		Gentle	0.258	−0.006	0.314	1.2 × 10−2 *	75.817	−0.288	0.317	1.2 × 10−2 *
	Low	Very steep
		Steep
		Gentle	0.051	−0.002	0.238	3.4 × 10−2 *	64.380	−0.209	0.053	3.4 × 10−1
Desbaratado	Very High	Very steep	0.815	−0.001	0.004	7.9 × 10−1	82.921	0.024	0.207	5.0 × 10−2 **
		Steep	0.645	−0.002	0.046	3.7 × 10−1	85.527	0.048	0.782	5.1 × 10−7 ***
		Gentle	0.357	−0.005	0.152	9.9 × 10−2 .	66.144	−0.073	0.154	9.6 × 10−2 *
	High	Very steep	0.869	0.006	0.518	5.1 × 10−4 ***	80.654	0.091	0.522	4.7 × 10−4 ***
		Steep	0.831	0.006	0.420	2.7 × 10−3 **	85.504	0.128	0.519	5.0 × 10−4 ***
		Gentle	0.860	0.000	0.001	8.9 × 10−1	55.747	0.010	0.002	8.5 × 10−1
	Medium	Very steep	0.793	0.005	0.346	8.0 × 10−3 **	73.643	0.106	0.282	1.9 × 10−2 *
		Steep	0.764	0.002	0.106	1.7 × 10−1	85.312	0.069	0.165	8.4 × 10−2 .
		Gentle	0.303	−0.007	0.349	7.7 × 10−3 **	76.437	−0.280	0.317	1.2 × 10−2 *
	Low	Very steep
		Steep
		Gentle	0.071	−0.003	0.186	6.5 × 10−2 .	66.008	−0.086	0.015	6.2 × 10−1

) linear models, there was a significant and positive relationship between p_i and time (Figure S4A–D). These models mostly described relationships on steep and very steep slopes at high elevations, and very steep slopes at medium elevations (Figure S4A–D). The relationship was negative in 6 out of 49 (R4 and WS_i; Table 1) linear models; of these, all but one model corresponded to gentle slopes at medium and very high elevations (Figure S4A–D). Twenty-nine out of Forty-nine linear models relating AI_i with time (R4 and WS_i; Table 2) were significant (Figure 5A–D), and most described a positive relationship; those exhibiting a negative relationship corresponded to gentle slopes at very high and low elevations (Figure S5A–D).

4. Discussion

We combined historic maps, open, analysis-ready remotely sensed data (GLAD Landsat ARD), and site-specific knowledge to investigate changes in land cover at four tropical SMW where landslide-associated risks and the intensive management of water have been intertwined throughout their recent history (Figure 3; Supplementary Materials File S1). We hypothesized that the demand for WES increased the connectivity between lowland and upland regions, and decreased the heterogeneity of our target watersheds.

The first question sought to establish relationships between the flow of WES and forest cover. Using water balance estimates as a surrogate for WES, we distinguished two possible relationships with landscape metrics. The second question examined variation in landscape metrics either with elevation or slope. Overall, the relationship among these variables were non-linear, with landscape metrics showing an increase from low to very high elevations or from gentle to very steep slopes. The temporal trends in our metrics suggest that slope, i.e., a surrogate for landslide-associated risks, has been an important factor driving change in our study area. The third question focused on the combined influence of elevation and slope on land cover. Landscape metrics varied non-linearly with elevation, but the extent of the observed variation was strongly influenced by slope and scale. In particular, land cover on gentle slopes exhibited the largest variation across elevations. Temporal trends suggest that landscape metrics are increasing on steep and very steep slopes at medium and high elevations, and decreasing on gentle slopes irrespective of elevation.

In the discussion that follows we use the site-specific knowledge to contextualize our findings and think about ways in which landslide-associated risks in upland areas and the intensive management of water in lowland areas may have altered the connectivity and spatiotemporal heterogeneity of these SMW.

4.1. Land Cover and Water-Derived ES Relationships Colombia

The site-specific history shows that the consolidation and expansion of an export, industrial based-agricultural system and urban centers starting in the mid-19th century, was accompanied by an increased demand of WES and homogenization of the lowlands (Figure 3; Supplementary Materials File S1). A diverse landscape comprised of tropical dry forests, pastures, tree gardens, orchards, and wetlands was transformed into high-water demanding sugarcane fields. The nearby upland areas, began to experience high rates of deforestation with the conversion of forests to upland subsistence agricultural systems, including pastures [81,82,83,84]. This raises the question about whether an increasing demand for fuelwood, charcoal, and food in the lowlands, triggered changes in upland regions to satisfy those needs. In these upland regions, torrential flows and landsliding together with deforestation began to impact water supplies, triggering a second wave of land cover change. This time, upland agricultural systems began to be converted to planted and natural forests, and forests were set aside for protection (Figure 3; Supplementary Materials File S1). In 2011, policies that supported the production of biofuels provided a new impetus for the expansion of sugarcane in lowland regions [85].

In spite of similarities among four SMW, their trajectories have been to some extent different. Shared characteristics include an intensive use of water for irrigation, a high susceptibility to landsliding [38,86] and the presence of paramo ecosystems at very high elevations [73]. Yet, variation in their proximity to agroindustry and urban centers, and history of human occupation may explain differences in their trajectories. The creation of Palmira in the early 19th century in the vicinity of Nima and its subsequent development as an important urban and sugarcane agribusiness center starting in the early 20th century most likely explains why this watershed has the most diverse and intense water uses among the four watersheds, had the largest proportion of forest in the early 1970s, and currently has the largest amount of protected land (Figure 3; Figure S1; Supplementary Materials File S1). Although the sugarcane agribusiness, including demand for WES, expanded quickly in the lowlands of the neighboring Bolo, Fraile, and Desbaratado watersheds, this did not translate into the use of water for the generation of electricity, the creation of public protected areas (Figure 3; Supplementary Materials File S1, but see [82]), nor the expansion of the small urban centers of Pradera, Florida, and Miranda. It is very likely that the elevated susceptibility to landsliding and torrential flows, and presence of indigenous communities influenced the trajectories of the Bolo, Fraile, and Desbaratado watersheds (Figure 3; Supplementary Materials File S1) [57].

Our expectation for the first question was that an increased demand for WES would translate into an increase of forest cover at watershed-scales, but this was not exactly the case. First, estimates of the water balances—a surrogate for WES—in our four watersheds indicated that Bolo followed by Amaime (Nima) exhibited deficits, whereas Fraile and Desbaratado did not [55]. Second, Nima exhibited the largest proportion of forest cover and Bolo de lowest; Fraile and Desbaratado were in between. This led us to explore two non-mutually exclusive models. In the first model that included Amaime (Nima) but excluded Bolo, we observed a negative relationship between the water balance and forest cover (p_i). This may support the idea that an increasing demand for WES flows increases investments in watersheds, and this may be particularly true in areas prone to landsliding as observed in Nima (Figure 3; Supplementary Materials File S1). In the second model that excluded Amaime (Nima), but included Bolo, we observed a positive relationship between the water balance and forest cover. This may suggest lack of investment in watersheds for which there is a high-demand for WES flows. A third possibility based on all the data is a lack of relationship between water balance and forest cover due to chance events driving the trajectories of our study watershed and/or use of water subsidies, including ground water.

4.2. Land Cover Change and the Independent Effects of Elevation and Slope

Our second question was aimed at examining variation in land cover either with elevation or slope that could be indicative of long-term interventions in our target watersheds. Forest cover consistently increased at high elevations in the four watersheds and tended to decrease at very high and medium elevations. We also observed that forest cover increased on very steep (high) slopes across the four watersheds.

The site-specific history of our watersheds offers valuable information to contextualize these results. Concerns about the quantity and quality of the water in the Nima watershed are evident since the late 1930s (Figure 3; Supplementary Materials File S1). These concerns became explicit in the first management plans of our watersheds (Nima in 1970 and Bolo, Fraile, and Desbaratado in 1977 [38,57]). These plans aimed at “providing sufficient water for domestic, agricultural, and industrial use, stimulating agricultural and forest productivity without affecting natural resources, and improving the quality of life of people living in the watersheds.” The plan called for changes in land use based on the biophysical conditions of the watersheds, as well as changes in land tenure. Elevation followed by slope and/or susceptibility to landsliding and erosion were used as guidelines to outline changes in land use, that in the Nima watershed suggested a diverse mix of uses at watershed- and farm-scales.

In the Nima watershed, changes in land tenure would follow farm-scale plans which required a long-term commitment of the colonos, sharecroppers, and farmers if they wanted to gain access to resources, credits, and land rights. In addition, the farm-scale plans would either involve the acquisition of land, intensification of production trough appropriate techniques, or selling the parcels if they could not meet production goals. Exclusive purchasing rights were given to the municipality of Palmira to fulfill general interests or the maintenance of the watershed [38]. In the Bolo, Fraile Desbaratado the plan established management zones aligned to a large extent with elevation. A first zone in the upper reaches of the watersheds (>2200 m a.s.l.) that included the largest settlements would be devoted to natural regeneration (21,138 ha) and commercial reforestation (13,525 ha). A second zone in the middle reaches (1400–2200 m a.s.l.; 15,312 ha) would be subjected to land-use re-structuring to promote adequate agricultural practices. Finally, a third zone in the in low reaches (1000–1400 m a.s.l.; 12,600 ha) would be devoted to agricultural and livestock activities [57].

Over roughly 60 years forest cover has increased at some elevations or slopes. In Nima, a significant proportion of the land has been protected, in Nima and Bolo, an unknown number of people have been displaced, and across the region, landslides continue to occur. In addition, the water balances estimated by CVC show the increasing importance of water wells to supply the needs of the agroindustry in the lowlands [55].

4.3. Land Cover Change and the Combined Effect of Elevation and Slope

Our third question focused on the variation of land cover with slope and elevation as interacting variables. Understanding the interactions between these two variables may suggest interventions consistent with the mitigation of site-specific, landslide-associated risks. Indeed, our results showed that forest cover increased on steep and/or very steep slopes at high elevations. Our results also showed that forest cover decreased on gentle slopes at very high elevations in some watersheds, and definitively on gentle slopes at low elevations across the region.

The historical data shows that improving water quantity was not the only concern among water users. Reducing the amount of sediments that entered the streams as a result of landslide, gully, and laminar erosion was very important (Figure 3; Supplementary Materials File S1). The Nima plan reads “it is therefore necessary to artificially recover areas affected by landslides-to limit sediment transport that has such an impact on the hydroelectric plants, treatment of the waters feeding the Aqueduct, and in the economy of the farms” [38]. The Bolo, Fraile, and Desbaratado plan proposes that agricultural lands on steep slopes should be reconverted into protective–productive forests. The historical data for Nima is particularly rich given its strategic importance; this may explain the establishment of several protected areas targeting steep to very steep slopes at high elevations in the Nima watershed (Figure 3; Supplementary Materials File S1).

A drought in the late 1980s and early 1990s that severely impacted the supply of WES to the lowland regions may further explain localized changes in land cover. For example, the largest increases in forest cover are observed on very steep slopes at high elevations, whereas decreases are observed on gentle slopes at different elevations in all but the Bolo watershed. The drought prompted major water-users in the low-lying areas of our study region to form private associations that in addition to reducing water disputes, would work towards the conservation of water in the upper reaches of the watersheds (Figure 3; Supplementary Materials File S1) [46,87]. The associations worked together with regional and local authorities to promote a diverse array of strategies that included land purchasing, enclosure of small headwater streams, establishment of vegetation buffer zones along streams, and environmental education programs among others [88]. Later on, these associations with the support of international and local non-profit conservation organization, and the sugarcane industry, became key players in the establishment of water funds, including payment for ecosystem services (PES) mechanisms [37,87,89]. Local actions aimed at increasing forest cover most likely catalyzed natural revegetation in these watersheds. At the same time, the establishment of protected areas could have contributed to the displacement of people at some combination of elevation and slopes, and deforestation on gentle slopes.

Over the years, the mission, activities, and projects of these associations, including CVC, would change to meet new challenges and realities, including the achievement of biodiversity, climate change, and sustainability targets feeding into national ones [37,90]. In addition to the water markets, carbon and renewable energy markets have become an added driver of land cover change not only in the four watersheds but also in other SMW draining the Central Cordillera in the Valle del Cauca Department [91,92]. This raises questions regarding the types of interaction established between the regions supplying diverse ecosystem services and those consuming those services, types of investments, land tenure, and livelihoods among others (Figure 1). The water associations and agribusinesses largely depend on the flow of these ecosystem services. Peasants, including indigenous communities living in the Bolo, Fraile, and Desbaratado, rely on the land for their livelihoods (Figure 3; Supplementary Materials File S1).

The sustainable production of multiple ecosystem services in areas influenced by torrential rains, landsliding, and droughts will require a substantial change in the ways we manage these coupled systems. Specifically, the production of WES cannot depend solely on the management of upland areas. It requires a reformulation of the agribusiness models in the lowlands to accommodate the uncertainties that regional to local climate changes is bringing. The resilience [3,4,5] of these coupled systems will require thinking about the connectivity and heterogeneity of lowland and upland regions.

4.4. Land-Cover Change, Mountains, and Water-Derived ES

Focusing on four watersheds in southwestern Colombia we aimed to understand how an increasing demand for WES may have influenced broad patterns of land cover change. Even though the watersheds share important similarities, their trajectories slightly differed most likely due to the timing, type, location, and pace of the interventions. For example, forest cover (p_i), but not rates of change varied among watersheds; the latter were indistinguishable from zero (range ~0.1 to 0.2 percent yr⁻¹; Table S1). Within watersheds, not only forest cover, but also rates of change varied as a function of slope, but depending upon elevation (Nima −0.7 to 0.7 percent yr⁻¹, Bolo 0.0 to 0.9 percent yr⁻¹, Fraile −0.6 to 0.9 percent yr⁻¹, and Desbaratdo −0.7 to 0.6 percent yr⁻¹; Table 1).

This variability becomes even larger upon examination of other regions that have been “designed” for the supply of WES. At one extreme is China’s “Grain for Green” and Panama’s Canal Watershed initiatives. Both aimed at increasing forest cover to protect and enhance WES in the area of influence of national strategic projects with global impacts [13,27,28,29]. In China, fractional forest cover change has been estimated at 0.15% year⁻¹ [26], whereas in Panama at 2.0% yr⁻¹ [93]. In these large areas, government-led strategies that included the creation of protected areas, tree planting based on a small subset of species, expropriation of land, subsidies, compensations, and payment for ecosystem services, among others, brought significant ecological, economic, and social change [94,95]. The outmigration of hundreds to millions of people is an example of the social impacts of these strategies.

At the other extreme, are municipal and water fund initiatives focusing on smaller areas such as those of this study; surprisingly, these initiatives may have potentially large impacts. For example, in our study region the number of water funds have increased in number from roughly four in the late 1990s to fifteen in 2015 [87,96]. This increase in the number of water funds coincides with the diversification of the sugarcane industry now an important player in the production of renewable energy [97,98,99]. In addition to the production of ethanol from sugarcane and the combustion of bagasse, the sugar mills of our study region produce biodiesel from oil palms grown hundreds of kilometers away in the Llanos Orientales de Colombia contributing in significant ways to land cover change [97,98,99,100]. The same agribusinesses are promoting the construction of a main national road linking the Llanos Orientales, the Cauca valley, and the port of Buenaventura in the Valle del Cauca (Figure 2A). This road passes through the Fraile watershed and the small urban center of Florida [101]. Understanding these connections within and among regions becomes important to address the sustainability of WES, and more broadly speaking ecosystem services. These connections, in turn, may help understand the emergence and growth of a macrosystem that originally connected lowland and upland regions, and now different biomes.

5. Conclusions

In the Andes, high and very high elevation forests, including paramo, play important roles in the maintenance of base flows during the dry season and regulation of torrential flows during the rainy season [73,102,103]. Whereas the first function is critical for the maintenance of water quantity to support “thirsty” crops, industries, and cities in the low-lying regions, the second is critical for the maintenance of water quality and natural hazard risk reduction. Upland and lowland regions are increasingly connected by numerous biophysical and biological processes, yet an increasing demand of WES in the lowlands has strengthen this connectivity largely mediated through the establishment of infrastructure to deliver those services (Figure 1). In our study region, the supply of WES has translated in important changes in land cover and land use both in the upper (this work) and lower reaches [104] of our target watersheds. Furthermore, the lowlands have undergone an increasing homogenization, that is also becoming visible in upland regions as “mountain livelihoods” become less viable and/or more areas are set aside for conservation. Over roughly 120 years, our four SMW catalyzed similar transformations in neighboring watersheds, which has the potential to increase the vulnerability of this region to diverse hazards, including climate change. For example, the homogenization of lowland areas could feedback into local and regional climates amplifying climate change effects (e.g., [105]). Additionally, the alteration of river corridors may have consequences on sediment and water fluxes, including channel geomorphology [106,107]. The homogenization of upland and lowland regions may also limit the production of high-quality food for years to come, which is unfortunate given the important role that mountains play in food systems [30].