Multitemporal Analysis Using Remote Sensing and GIS to Monitor Wetlands Changes and Degradation in the Central Andes of Ecuador (Period 1986–2022)

Abstract

Wetlands are transitional lands between terrestrial and aquatic systems that provide various ecosystem services. The objective of this study was to evaluate the change in wetlands in the Chimborazo Wildlife Reserve (CR) in the period 1986–2022 using geographic information systems (GISs), multitemporal satellite data, and field data from the 16 wetlands of the reserve. Images from Landsat satellite collections (five from Thematic Mapper, seven from Enhanced Thematic Mapper, and eight from Operational Land Imager and Thermal Infrared Sensor) were used. Image analysis and processing was performed, and the resulting maps were evaluated in a GIS environment to determine the land cover change and growth rate of hydrophilic opportunistic vegetation (HOV) according to hillside orientation. The results show that there are negative annual anomalies in the water-covered areas, which coincide with the increase in HOV. This shows that the constancy or increase in the rate of increase in HOV, which varies between 0.0018 and 0.0028, causes the disappearance of these ecosystems. The importance of the study lies in its potential contribution to the decision-making process in the management of the CR.

1. Introduction

Wetlands are transitional areas between terrestrial and aquatic regions, which are permanently or temporarily covered by shallow water [1]. They are considered a desirable habitat for a variety of animal and plant species by providing food and shelter [2,3]. They are part of the key sites that showcase the Earth’s biodiversity [4] and are currently recognized as key ecosystems for the provision of ecosystem services [5,6,7], accounting for 40.6% of the total value of environmental services (ES) [8,9].

However, due to human activities, more than half of the wetland ecosystems existing at the beginning of the 20th century have been lost in the Americas, Europe, Australia, and China [5] due to processes such as global warming [10,11], changes in land use [12,13,14,15], including industrial runoff, road construction, and the collection and introduction of plants or animals [1].

The Americas is the continent with the largest area covered by wetlands (>3 × 106 km²), representing more than 20% of its surface area [16]. Additionally, the environmental heterogeneity of the continent generates a wide variety of wetland types [3,17,18]. Possibly, the main drivers of this diversity are the longitudinal and latitudinal extent of the continent, the altitudinal range from sea level to 6962 m.a.s.l., and its geological and climatic diversity, with the influence of the Pacific and Atlantic Oceans [3,19].

In the Americas, and particularly in Ecuador, wetlands are among the most important ecosystems, both in terms of their size and the diversity of the ecosystems they host. Ecuador has identified 13 Ramsar wetlands [20] and 59 wetlands, covering a total area of 286,659 hectares [21], in 11 highland provinces that make up the high Andean zone of Ecuador [22]. Eighty-six percent of these ecosystems are within the National System of Protected Areas, to which the RC has belonged since 1987, created with the objective of protecting and promoting, under appropriate ecological parameters, the habitat of the native camelids of the Andes [23]. In addition, it seeks to preserve the páramo and puna ecosystems [24,25] and thus promote the conservation of the 20,518.03 ha [21] of wetlands found in the jurisdiction of the highest snow-capped mountain in the world, measured from the center of the Earth.

Understanding temporal changes in wetlands is crucial for their conservation and management. Wetland identification, mapping, and analysis are often performed using remote sensing [26], a tool that offers well-documented advantages such as a synoptic view, multispectral data collection, multitemporal coverage, and is cost-effective [27,28,29]. Recent advances in remote sensing technologies have become the main source of spatial information on the Earth’s surface cover [26].

Over the past five decades, remote sensing technology has been used in many areas of wetland research such as (1) land use/land cover changes or mapping in wetland regions [11,26,30,31]; (2) carbon cycling and climate warming in wetland environments [32,33]; (3) carbon release from peatland fires [34,35,36]; and (4) hydrological processes in wetlands [37,38]. New multitemporal analysis techniques have been incorporated such as hyperspectral and synthetic aperture radar (SAR) imagery [39] and multitemporal classification using SAR data such as TerraSAR-X [40] techniques that have allowed for the mapping and monitoring of changes in the wetlands of the Yellow River and the Chesapeake and Delaware Bays [41] as well as the monitoring of hydrological regimes in several wetlands such as the Zhalong Wetland [42]. This demonstrates that the integration of time series data such as Landsat has improved the monitoring of hydrological regimes, as the LANDSAT archive offers a rich dataset with more than 40 years of freely available multispectral decametric observations with clear utility for wetland characterization [43].

Climate change in the Andes poses new challenges that require the monitoring and control of vegetation cover for current and future land-use management [44,45]. In this research, the multitemporal analysis of bofedales was performed based on satellite remote sensing data, the same data that provide important information to detect changes in landscapes over long periods of time in contrast to conventional approaches that grant short-term data [46].

Taking into consideration that the water systems of this region are experiencing rapid recession rates [47,48,49,50], the importance of this study lies in the fact that the water systems of the reserve constitute the tributaries of several downstream watersheds and that the water supply of the populations that benefit from it would be strongly compromised, especially during drought years. The objective of this study was to analyze Landsat satellite images (5 TM, 7 ETM, and 8 OLI/TIRS) between 1986 and 2022 of the CR wetlands to identify, monitor, and detect changes in the spatial and temporal variability of these ecosystems and promote their conservation.

2. Materials and Methods

2.1. Study Area

The wetlands of the CR (Figure 1) are a hydrologically dynamic and ecologically significant landscape for this protected area. The 16 wetlands are located in the interior of the Andes at altitudes ranging from 3840 to 4314 m.a.s.l. in the provinces of Bolívar (six wetlands), Chimborazo (four wetlands), and Tungurahua (six wetlands). The temperature ranges between −3 and 14 °C, there is an average annual precipitation of 1000 mm, and humidity of 70 to 85% [23].

2.2. Data Collection

The data used in this research were divided into satellite data and field data. Field data included primary field data for the land cover/use classes and area of each wetland. Field points were collected using a global positioning system (GPS) from February to September 2022, which were used for image classification and overall assessment of the accuracy of the classification results. On the other hand, the satellite data for the 1986–2022 study period consisted of multispectral data acquired by the Landsat satellite. The mosaics had less than 15% cloudiness and pixels affected by clouds and shadows were eliminated through the Fmask algorithm [51]. Additionally, quality bands of Landsat images were used to discard pixels with high uncertainty [52]. Factors such as soil type, altitude, and precipitation were considered indirectly through the construction of multitemporal mosaics of the satellite images. These elements influence the spectral reflectance of the land surface.

The sampling points were determined in the field in the jurisdiction of the CR. However, due to the pixel sizes of the satellite images, each point corresponded to training areas of 50 × 50 m [14].

2.3. Satellite Data

All synthetic surface reflectance images from Landsat satellite missions (Landsat 5 TM, Landsat 7 ETM, and Landsat 8 OLI/TIRS) in the study period 1986–2022 were obtained. All data were acquired as Level 2 surface reflectance from Collection 2, available on the freely accessible Google Earth Engine (GEE) platform (https://earthengine.google.com). (accessed on 30 May 2024)

The images were cropped in the study area and visually examined for cloud cover. Consequently, due to the limited availability of satellite images due to cloud cover in the study area, the mosaic was generated from the collection of satellite images generated in years with an annual cumulative precipitation less than the average recorded in the study period 1980–2022 (854 mm/year). The number of images used is shown in

Table 1. Number of satellite images used in the study area (CR).

Satellite	1986	1991	1996	2001	2009	2013	2016	2022 August and December
Landsat-5	15	13	7
Landsat-7				17	16	14
Landsat-8							17	9

.

This criterion was considered because high Andean wetlands are sensitive to changes in precipitation and temperature, since there is a direct relationship between the development of the vegetation that conforms them and the temporal variability of these meteorological variables [53,54]. The mosaics generated corresponded to the years 1986, 1991, 1996, 2001, 2009, 2013, and 2016; in the case of the year 2022, because the annual accumulated precipitation was higher than average, the mosaics were formed from images of periods with low monthly accumulated precipitation (less than 50 mm).

2.4. Image Processing and Classification

The processing of satellite images prior to the characterization of changes is immensely necessary and has the main and only objective of establishing a more direct affiliation between the acquired data and biophysical phenomena [55]. To determine the spatiotemporal changes of the wetlands in the study area, we took as a starting point (year 2022) the sampling points obtained in the field and later, supervised classification in the GEE platform.

Based on the spectral characteristics of the current sampling points, reference areas were defined for the classification of the mosaics corresponding to the years 1986, 1991, 1996, 2001, 2009, 2013, and 2016.

To improve the classification, the variable altitude was added through a digital elevation model (DEM) SRTM with a spatial resolution of 30 m, which improved it considerably as recommended [56,57].

The most popular and adopted supervised classification methods for land cover change studies are Smile CART (classification and regression tree), Smile random forest (RF), and support vector machine (SVM) [57]. In this study, the parameters used for each classifier were as follows: for Smile Cart, smileCart: maxNodes default (null)—minLeafPopulation default (1), for RF: numberOfTrees 90, and for SVM: kernelType: ‘LINEAR’, gamma: null, cost: 10. To determine the best classification method, several tests were performed with all classifiers, and the accuracy of the results was evaluated from the calculation of the overall accuracy and Kappa index [58,59].

The RF decision model was selected over alternative classifiers such as support vector machines (SVMs) and the Smile CART decision tree model within GEE. The RF model is based on running multiple decision trees randomly, providing the most likely classification for the assigned categories according to the categories of the sampling points [57,59]. Thus, the RF model offers several advantages over the Smile CART and SVM models, particularly in data management and in the management of feature interactions and the provision of tools for the feature reduction in the data obtained on the CR wetlands. We obtained as a result that the highest percentages of overall accuracy and Kappa index in five out of the eight years of study were in 1986, 1991, 1996, 2009, 2013, and 2016, which were attained in the application of the RF model (

Table 2. Performance of the Smile CART, RF, and SVM classifiers.

Year	Classifier	Overall Accuracy	Kappa Index	Year	Classifier	Overall Accuracy	Kappa Index
1986	SVM	64%	0.55	2009	SVM	69%	0.59
	CART	84%	0.8		CART	81%	0.76
	RF	89%	0.87		RF	86%	0.81
1991	SVM	64%	0.55	2013	SVM	59%	0.48
	CART	86%	0.83		CART	84%	0.8
	RF	86%	0.83		RF	85%	0.81
1996	SVM	70%	0.62	2016	SVM	61%	0.49
	CART	82%	0.78		CART	80%	0.79
	RF	85%	0.81		RF	83%	0.88
2001	SVM	62%	0.52	2022	SVM	58%	0.47
	CART	87%	0.85		CART	85%	0.81
	RF	85%	0.81		RF	85%	0.81

).

All satellite data were studied by assigning signatures per pixel and differentiating the study area into six classes based on the specific digital number (DN) value of different landscape elements. To minimize this effect, training polygons were selected (i.e., homogeneous coverage with a minimum size of 3 × 3 pixels (90 × 90 m)), ensuring representativeness and avoiding the impact of spectral variability [56,57]. The classes delineated were bare soil/rocks, HOV, agriculture, snow, water, and grassland (

Table 3. Classes delineated on the basis of supervised classification and sampling points.

Class No.	Class Name	Description	Number of Sampling Points
1	Bare soil	Exposed soil land surfaces and arid areas influenced by human influence.	12
2	Hydrophilic opportunistic vegetation (HOV)	Consisting of plant communities closely related to the aquatic environment or to soils permanently saturated with water.	16
3	Agriculture	Cropland and fallow land	28
4	Snow	Snow-covered terrain	14
5	Water	Wetlands, lagoons and streams	7
6	Pajonal	Low and flooded land, covered with thatch and other associated species, typical of humid places.	25

). For each of the predetermined land use/land cover types, training samples were selected by delineating polygons around representative sites [14].

2.5. Accuracy Assessment

An assessment of the classification accuracy of the 1986 and 2022 images was conducted to determine the quality of information derived from the data in the CR. For classification data to be useful in change analysis detection, an assessment of individual classification accuracy is essential [60]. To evaluate the accuracy of land cover maps extracted from satellite images, a stratified random method was used to represent different land cover classes of the area. An accuracy assessment was carried out using 100 points, based on real data and photointerpretation. The comparison of the baseline data and the classification results was performed statistically using error matrices. In addition, a non-parametric Kappa test was also performed to measure the degree of accuracy of the classification, as it takes into account not only the diagonal elements, but all elements of the confusion matrix [61]. Kappa is a measure of agreement between the producer’s predefined ratings and the ratings assigned by the user. This is calculated by the following formula (Equation (1)):

$$K=P\left(A\right)-P\left(E\right)/1-P\left(E\right)$$

(1)

where P (A) is the number of times the K raters agree, and P(E) is the number of times the raters are expected to agree only by chance [62,63].

2.6. Detection of Land Use/Land Cover Change in the CR

The study employed the post-classification change detection technique performed in ArcGIS 10. This technique has been used successfully due to its efficiency in detecting the location, nature, and rates of change [64]. Additionally, to obtain the changes in land cover/land use of the CR during the period 1986–2022, the overlay procedure was used. The application of this procedure was used to describe the main types of change in the study area. To determine the number of conversions of a particular land cover and its corresponding area during the evaluated period, a pixel-by-pixel cross-tabulation analysis was performed. This resulted in new thematic layers reflected in eight maps of six classes containing different combinations of change class [14].

2.7. HOV Growth Rate in CR by Aspect

Once the HOV surfaces were obtained for the mosaic of each study year, they were intersected with the hillside orientation variable, which was generated with the geo-tool “Aspect” from the SRTM DEM, with a spatial resolution of 30 m.

For the data analysis, generalized linear models (GLMs) were used and applied to the time series. The dependent variable was the surface area with HOV for each of the hillside orientations (north, northeast, east, southeast, south, southwest, west, and northwest), while time was included as a covariate and also considered as an independent variable. Initially, a generalized linear mixed model (GLMM) was tested, incorporating time as a random effect; however, the model did not converge, suggesting that the interannual variability was not significant enough to justify its inclusion as a random effect. Therefore, the final analysis was performed using a GLM with time as a fixed effect, ensuring a stable and interpretable model [65].

To guarantee the validity of the results obtained with the GLM, statistical tests were performed to evaluate compliance with its fundamental assumptions. The Shapiro–Wilk test was applied to assess the normality of the residuals, homoscedasticity was verified with the Breusch–Pagan test, and finally, the independence of the residuals was evaluated using the Durbin–Watson statistic [66].

Finally, Infostat software version 2020 was used to estimate a Poisson-type distribution and a log-family fit. In this way, the equation that describes the phenomenon of HOV increase in the wetland ecosystem as a function of aspect [67] is (http://www.infostat.com.ar): (accessed on 6 June 2024)

$$\mathrm{Log}(\mathrm{Area}\_\mathrm{HOV}\_\mathrm{t})=\beta 0+\beta 1\times \mathrm{t}+\beta 2\times \mathrm{Orientation}$$

(2)

where:

-

Log (HOV_Area_t): is the natural logarithm of the area covered by HOV at time t.

-

β0: is the intercept of the model.

-

β1: represents the temporal rate of change (growth rate).

-

β2: captures differences specific to hillside orientation.

Calculation of the HOV Surface Area at Each Orientation

• The HOV polygons were delimited by the supervised classification of satellite images (random forest) and crossed with the hillside orientation map, generated from the DEM.
• The “Aspect” tool was used to categorize the hillsides into eight main orientations (N, NE, E, SE, S, SW, W, NW).
• For each year of analysis, the classified HOV polygons were intersected with the hillside orientations, and the area (in hectares) of HOV in each direction was calculated.

2.8. Protection of Change and Increase in HOV in the CR Wetlands

To determine the rate of increase in the HOV for each wetland, a field trip was made to delimit the polygons of each of these ecosystems (16) using geolocation equipment and georeferenced points. Obtaining the polygons that represented the wetlands, a weighted average was made with the total surfaces for each of the hillside orientations (aspect) and the rates of increase in HOV for each of the orientations, thus having an individualized rate for each wetland of the increase in HOV.

Finally, the change and increase in HOV was modeled using an exponential growth formula, which allowed us to project its future expansion as a function of time. The formula used is as follows:

$$A_{t }=A_0 · e^{r.t}$$

(3)

where:

-

A_t: is the projected area of HOV at time t.

-

A₀: is the initial area of HOV.

-

r: is the exponential growth rate specific to each hillside orientation or wetland.

-

t: is the time elapsed since the base year.

This formula, widely used in ecological studies, was adjusted for each hillside orientation and wetland using data observed during the study period. Values of r were calculated using nonlinear regressions within a GLM, which accounted for variability between the orientations and wetlands. Initially, a GLMM was tested, but due to convergence issues, the analysis was finalized using a GLM with time as a fixed effect, ensuring a stable and interpretable model.

The model allowed for predictions of the estimated time for each wetland to be completely covered by HOV, considering the trends observed in recent decades. This approach provides a key tool for evaluating future scenarios and planning management and conservation strategies based on these projections.

3. Results

3.1. Changes in Land Cover/Land Use of the CR

Figure 2 shows the land cover classification map of the CR for the period 1986–2022. For this classification, the RF algorithm worked with a 70% and 30% partition for training—evaluation and the execution of 90 decision trees [68]. The overall classification accuracies achieved and the kappa statistics are shown in

Table 4. General classification accuracies and kappa statistics by year.

Satellite	1986	1991	1996	2001	2009	2013	2016	2022 August and December
Overall accuracy	89%	86%	85%	89%	85%	84%	83%	82%
Kappa index	0.87	0.83	0.81	0.87	0.81	0.80	0.88	0.87

. According to [69], accuracy assessment reports require an overall classification accuracy greater than 80% and kappa statistics greater than 0.8, which were successfully achieved in the present investigation.

The results of the CR land cover classification for the period 1986–2022 are summarized in

Table 5. Land cover/land use classes and areas in hectares.

Category	1986	1991	1996	2001	2009	2013	2016	2022
Bare soil	8910.61	11,197.21	5874.87	13,114.39	8385.21	9468.57	6705.01	9536.88
(HOV)	3050.32	3835.42	4583.12	4797.70	5970.94	6315.64	6317.79	6469.38
Agriculture	7568.86	10,033.21	9258.54	8355.15	10,698.78	8219.97	8772.18	7755.22
Snow	1733.49	1420.70	1176.14	1896.58	2361.38	1614.42	1756.86	1493.40
Water	1994.65	901.15	921.40	1091.11	866.44	343.32	4498.15	912.24
Pajonal	28,965.98	25,467.84	30,453.53	23,661.16	24,341.64	26,650.08	24,870.11	26,346.95

. The number of hectares of the classes delineated in this study shows the practices and variability in land cover and land use in the reserve during 1986–2022.

The comparison of each class from 1986 and 2022 showed that there had been a marked change in land use and land cover during the study period (36 years). The results showed a significant increase with respect to the HOV in the CR, highlighting that it was the only supervised classification class that maintained constant growth, which is reflected in the increase of 3419.06 ha in 2022 with respect to the initial year of study (1986) while the classes corresponding to bare soil, crops, snow, water, and grassland showed significant levels of variability in each year of study.

3.2. HOV Growth in the CR by Hillside Orientation

Figure 3 shows the different hillside orientations in the CR. Once the intersection of this variable with the areas with HOV was carried out, the surface areas in hectares were obtained for each year of study according to the hillside orientation (

Table 6. Surface area in hectares of HOV according to hillside orientation.

Year	East	Northeast	Northwest	North	West	South	Southeast	Southwest
1986	495.0	343.9	268.2	264.9	287.0	493.1	527.2	357.8
1991	908.8	545.0	245.8	316.4	280.1	481.3	747.3	306.5
1996	887.7	694.2	382.8	470.1	372.1	590.0	797.4	381.7
2001	759.6	580.3	504.3	479.5	562.2	656.4	693.8	551.8
2009	1064.6	844.9	550.1	596.2	555.8	828.8	897.6	596.3
2013	1127.0	918.5	529.8	626.5	602.8	875.4	990.5	638.1
2016	1189.0	856.5	513.0	571.4	578.7	936.4	1002.4	653.3
2022	1092.6	890.4	622.0	654.0	676.4	908.5	931.9	664.9

). This is because the effect of environmental conditions, such as topography, climatic season, soils, hillside orientation, etc., can modify the growth and structure of plant communities [70]. In the case of hillside orientation, this can lead to different behaviors in plant communities via incident radiation and different values of potential atmospheric demand [71,72,73].

After obtaining the data on the surface areas in hectares of the HOV by hillside orientation, a data analysis was performed using a GLM applied to the time series to describe the increase in HOV as a function of aspect. Initially, a GLMM was tested, but due to convergence issues, a GLM with time as a fixed effect was selected to ensure a more stable and interpretable model. To validate the results obtained with the GLM, statistical tests were performed to assess compliance with its fundamental assumptions including the Shapiro–Wilk test for normality, the Breusch–Pagan test for homoscedasticity, and the Durbin–Watson statistic for the independence of residuals.

The normality of the residuals was analyzed using the Shapiro–Wilk test (p = 0.1016), confirming that they followed an approximately normal distribution. Homoscedasticity was verified with the Breusch–Pagan test (p = 0.1735), indicating that the variance of the residuals was constant. In addition, the independence of the residuals was evaluated through the Durbin–Watson statistic (2.528), ruling out problems of autocorrelation.

Table 7. Increased hydrophilic opportunistic vegetation as a function of aspect.

Aspect	a	b	r2	p-Value
East	0.00200	−3.94	0.72	0.0078
Northeast	0.00260	−5.09	0.84	0.0013
Northwest	0.00280	−5.4	0.86	0.0009
North	0.00270	−5.26	0.90	0.0003
West	0.00290	−5.66	0.89	0.0005
South	0.00250	−4.86	0.95	<0.0001
Southeast	0.00170	−3.32	0.77	0.0039
Southwest	0.00260	−5.11	0.89	0.0004

shows that the phenomenon of the highest increase occurred in the south orientation with 0.95, while in the east orientation, the phenomenon of the lowest increase was evidenced with a value of 0.72. Additionally, the analysis included the values of p, r2, a, and b for each orientation, providing a more exhaustive understanding of the statistical significance and dynamics of the changes observed.

3.3. Change and Increase in HOV in the CR Wetlands

Table 8. Change and increase in HOV by wetland.

Wetland	Area (ha)	Rate of Increase in HOV per Wetland	Area of HOV (Year 2022)	Time Frame Wetland Extirpation	Projected Area % Coverage	% Cover	Year in Which the Wetland Will Disappear
W1	29.3	0.0027	18.58	168	29.31	100%	2190
W2	154.4	0.0023	5.83	1440	154.81	100%	3462
W3	35.5	0.0027	27.00	100	35.34	100%	2122
W4	57.8	0.0024	49.71	65	58.02	100%	2087
W5	18.8	0.0022	7.74	400	18.75	100%	2422
W6	13.3	0.0027	11.52	52	13.27	100%	2074
W7	9.4	0.0021	0.28	1660	9.43	100%	3682
W8	1.8	0.0026	0.40	604	1.85	100%	2626
W9	8.8	0.0025	4.47	270	8.79	100%	2292
W10	0.3	0.0028	0.12	325	0.29	100%	2347
W11	12.8	0.0028	9.97	90	12.88	100%	2112
W12	12.2	0.0024	8.05	175	12.16	100%	2197
W13	21.4	0.0023	2.02	1015	21.31	100%	3037
W14	15.2	0.0028	2.16	707	15.25	100%	2729
W15	7.6	0.0023	4.74	200	7.57	100%	2222
W16	25.7	0.0018	11.53	435	25.67	100%	2457

shows the rate of increase in HOV for each wetland. Overall, in all wetlands, an increase in HOV was observed, inside and outside the perennial ponds and seasonally flooded areas. In these ecosystems, vegetation is currently dominated by different species, such as Bartramia potosica, Thuidium peruvianum, Agrostis breviculmis, and Leptodontium ulocalyx, among others [24], all of which share the common characteristic of retaining moisture and inhabiting flooded depressions, banks, and ditch margins [24,74,75,76]. These species have had to adapt to tolerate alternating periods of flooding and drought, factors that have contributed to the dominance of the species with high growth rates and aggressive reproductive strategies [77,78,79].

In this research, each sample site presented different changes and growth rates; however, the wetlands La Lazabanza BNI, Puente Ayora AI, and Puente Ayora BNI were the wetlands with the highest HOV growth rates (0.0028). In contrast, the wetlands Los Hieleros ANI and Casa Cóndor BI showed the lowest rates of HOV increase with 0.0018 and 0.0021, respectively.

Additionally, Table 8 shows the impact of the increase in HOV on these ecosystems. Thus, the expansion of fractions related to this vegetation makes it clear that the more this flora predominates over the coverage of the CR, the more evident the loss of perennial wetlands and temporarily flooded areas will be. According to this study, the Culebrillas Wetland will become the first ecosystem in which temporary or permanent water will disappear within the reserve’s jurisdiction, since the projection shows that in 52 years, the advance of the HOV will have completely covered this particular ecosystem. On the other hand, the Casa Condor Wetland, with the lowest growth rate, is the ecosystem in which the water resource will last the longest, with a prediction period of 1660 years.

The results show that from the year 2074, which shows the loss of the first wetland (Culebrillas), to the year 3682, which shows the extinction of the last wetland (Casa Condor) in the Chimborazo Reserve, over a period of 1608 years, all of the aquatic ecosystems of the highest snow-capped mountain in the world, measured from the center of the Earth, will have disappeared (Figure 4).

4. Discussion

This study clarifies the importance of incorporating remote sensing and GIS for the study of land cover and land use change detection in an area, as it provides crucial information on the spatial distribution and nature of land cover change [14].

The overall 90% accuracy of land use and land cover maps indicates that the integration of the supervised classification of satellite imagery with visual interpretation is an effective method for documenting land use and land cover changes in an area [80]. Given that evidence from field work and spatially resolved satellite images suggests that the reserve’s wetlands have undergone complex spatiotemporal variability in ecohydrology and habitat degradation, it is possible to identify the cover classes that have predominated and those that have been affected by these changes over the period 1986–2022. Specifically, at the beginning of the study period (1986), the HOV only covered 3050.31 ha. of the total study area; however, over the course of 36 years, this class has come to cover a total of 6469.37 ha. within the protected area. This is possibly due to the alternating periods of flooding and drought and the presence of low nutrient availability conditions [81,82,83], which has allowed species to develop strategies for reproduction and interaction among them, the latter becoming one of the most important factors for the regulation of species dominance, since the presence of associations helps to develop self-regulating dynamics that facilitate their establishment and reproduction [77,78,79].

During the period of 1986–2022, the hectares comprising the bare soil class underwent considerable changes of increase and decrease; thus, for example, in 2001, this class covered the largest amount of hectares of its record (13,114.38 ha); while 2016 was the year in which the soils of the reserve were covered by any type of vegetation cover, which decreased the amount of bare soil within the reserve (6705 ha). This was demonstrated by a study conducted by Paula et al. [84], in which they identified considerable changes in this class in the period 1962–2010, thus, the authors determined that in 1987, a value of 26.61% of bare soil was recorded, a value that increased to 27.50% by the year 2000. This increase in the area of bare soil may possibly have been due to the rapid deforestation in the area that eliminated the vegetative cover of the soil and left it arid and exposed. Similar trends were observed in the Swat district study by Qasim et al. [85]. In addition to deforestation, extensive livestock grazing has deformed the plants present in the area into wastelands [86,87].

The third class that faced land cover change during the study period was crops. This increasing trend of land cover and land use change in the CR area reinforces the fact that economic forces are commonly an important driver of anthropogenic land change [88] and is the main reason why the soil in this protected area has undergone different changes. The water bodies and temporarily flooded areas in the reserve have changed from other land cover to agricultural cover. This is in spite of MAATE’s efforts to regulate agricultural activities.

According to the information revealed by the classification results, the area covered by the water class also experienced changes between 1986 and 2022, but mostly negative changes in coverage. This is evidenced by the loss of 1082.42 ha. since 1986. Changes in land cover and land use observed in all other classes affected the water class over three decades. The easy accessibility to the water resource will have repercussions on its depletion and conversion to dry tributaries and its inevitable replacement by compacted surfaces or arid lands. Another possible reason for the contraction could include an increase in the rate of surface runoff due to the lack of plant roots to retain water. As runoff would exceed the groundwater recharge capacity, this in turn would cause the water table in the reserve’s water bodies to drop. As previously mentioned, the increased rate of deforestation also contributes to increased surface runoff and is responsible for the downward flow of nutrients and sediment [89,90,91].

Finally, the area covered by the snow class also experienced significant changes of increase and decrease in the period of 1986–2022. This is evidenced by studies conducted on the glaciers of the Chimborazo volcano, which showed that this class has experienced a drastic retreat in recent decades, with a 72% loss of area between 1962 and 2016 [92]. All this, as a result of the global warming process and the fall of ash from the Tungurahua Volcano, has influenced the accelerated melting process so that the landscape of Chimborazo now looks different. The rock-solid ice—known as “perpetual snow”—is now only found in the highest areas of the Andean colossus [93].

In addition, the study made predictions that included all the changes observed in the coverage of the wetlands of the CR, mainly in the advance and predominance of HOV within the wetlands of the reserve. These changes constitute real environmental problems, not only for the species that inhabit these ecosystems, but also for the local populations that use this resource for irrigation purposes. Thus, the expansion of this type of cover demonstrates that the more this flora predominates over the cover of the CR, the more evident the loss of perennial wetlands and temporarily flooded areas will be. According to the results of this research, the Culebrillas Wetland will become the first ecosystem in which temporary or permanent water will disappear, since the projection made in this study shows that over a period of 52 years, the advance of the HOV will have completely covered this particular ecosystem, followed by the Cruz del Arenal ANI Wetland, which will disappear in 65 years, and so on, until all the wetlands in the reserve have disappeared as a result of the aforementioned changes.

The information presented in this article focuses on the dynamics of change that have occurred in the wetlands of the Chimborazo Reserve and puts into perspective how these patterns may continue to be replicated in the coming years if immediate conservation measures are not taken. In this context, it is important to mention that in different regions of the world, different projects have already been implemented for the protection and integral and systematic management of mountains, rivers, forests, fields, lakes, pastures, and sand as well as the protection of forests and natural areas [94,95]. This shows that if actions are taken to recover the wetlands of the RC, positive results could still be achieved in the region.

5. Conclusions

Based on the results obtained through the use of GIS and remote sensing applications to achieve the specific research objectives, it can be concluded that the land cover and land use practices in the study area have altered significantly in 36 years and will alter even more in the projection years of the study. The change in land cover of the CR wetlands was evidenced by the continuous changes occurring in the study area (increase and decrease) in the bare soil, crops, snow, water, and scrubland classes and the increase in the area covered by HOV (3419.06 ha) since 1986. The major impact of this expansion is subject to the disappearance of these ecosystems and the loss of perennial pools and seasonally flooded areas over time. These hydrological changes will result in the expansion of widespread but not precisely uniform vegetation throughout the study area, as measured by the fraction of HOV. Therefore, proper management of the CR’s water resources is required, because without proper management, this valuable resource will soon be lost or will no longer be able to play the important ecological role it plays within the protected area.