Analysis of the impact of vegetation cover variation on pollution levels in the La Caimanera marsh (Colombia) using remote sensing and sensor imagen

Study area

The La Caimanera marsh is situated in the Gulf of Morrosquillo subregion, in the northern part of the Department of Sucre (Colombia). The coordinates of the La Caimanera marsh correspond to 9°25’57.42” latitude and 75°37’53.64” west W.

Source of data

The data were obtained from various satellite sources. Their high value in the analysis of mangrove ecosystems using remote sensing is highlighted. Images from the Planet Application Program Interface (https://www.planet.com/) were used, with a spatial resolution of 4.7 meters and blue (B), green (G), red (R), and near-infrared (NIR) spectral bands. In addition, images from Landsat 7, 8, and 9 satellites (https://landsat.gsfc.nasa.gov/) were included, which provide resolutions of up to 30 m, which is ideal for medium-scale environmental monitoring. For time series and advanced analyses, MODIS collection data available from Earth Engine were used, such as MOD44B.006 Terra Vegetation Continuous Fields Yearly Global, with a resolution of 250 m per pixel, and MCD12Q1.061 MODIS Land Cover Type Yearly Global, with a resolution of 500 m. The MOD11A2 V6.1 product was used for land surface temperature (LST) estimation.

For the analysis of land use and land cover, Sentinel images from the Dynamic World platform, which manages a global scale with a resolution of 10 m (https://dynamicworld.app/), were accessed. To evaluate mangrove dynamics and their evolution, data from the Global Mangrove Watch (GMW) (https://www.eorc.jaxa.jp/ALOS/en/dataset/gm_e.htm), developed in collaboration with the Japan Aerospace Exploration Agency (JAXA) and Aberystwyth University, were used, which provided accurate estimates of the extent and changes in mangrove forests at 11 annual epochs between 1996 and 2020 (https://www.globalmangrovewatch.org/). In addition, NASA Landsat SEDAC Global Mangrove Forest Distribution data at 30-meter resolution were used to provide a detailed analysis of spatial and temporal patterns of mangrove cover. Data were obtained from the European Union’s Global Surface Water Explorer platform (https://global-surface-water.appspot.com/). Figure 1 illustrates the methodology used for the acquisition and processing of the satellite images used in this study.

Figure 1. Methodological framework used for image acquisition and processing.

However, the Random Forest algorithm was used, which is often preferred for its ability to generate robust and accurate results. This machine learning model uses the combined output of several decision trees trained using the CART (Classification and Regression Tree) algorithm, allowing the creation of subsets of data that optimize land cover classification. The Random Forest mathematical model is detailed in the following equation:

Where, $T$ is the total number of trees in the forest, $h_i\left(x\right)$ is the prediction of tree i for input $x$ , $H\left(x\right)$ is the final prediction, $I$ is the indicator function that becomes 1 if the prediction of tree $i$ is class $c$ and $0$ otherwise. The following equation was used as the regression model for the Random Forest.

Considering that correlations between land surface temperature (LST) and the Normalized Difference Vegetation Index (NDVI) were employed, it was necessary to ensure that both products had the same spatial resolution. To this end, NDVI, as a time series, was also extracted from the MODIS data. Specifically, the Terra MODIS MOD13Q1 product, extracted using the Google Earth Engine, was used ( Figure 1). The NDVI and LST values for each year were adjusted to the 90th percentile to minimize the influence of pixels affected by atmospheric variations. NDVI is a widely recognized index that allows the quantification of vegetation health and density.¹⁹

The integration of MODIS data is essential for analyzing the temporal dynamics of vegetation health and thermal fluctuations, compensating for the temporal and spectral limitations of Landsat imagery. This approach allows filling gaps in the Landsat time series, providing a more complete long-term analysis. LST and NDVI results extracted from the Google Earth Engine were exported in raster format and processed in QGIS 3.38.2’Grenoble’ to generate detailed maps. For the analysis, 450 points were generated within the interior areas of the bog, avoiding aquatic areas to mitigate the effect of water on the correlation between NDVI and LST. The extracted values were used in a correlation analysis performed in Python (version 3.12) to study the relationship between land surface temperature and vegetation health in the study area.

Data analysis

The data analysis in this study was carried out using different tools, depending on the type of information and its specific purpose. QGIS 3.38.2’Grenoble’ and Google Earth Engine (GEE) were used to process images from Planet, Landsat, MODIS and Sentinel satellites ( Figure 1). In particular, Landsat images were processed using QGIS, while GEE was used to handle Dynamic World data, vegetation indices such as NDVI and NDWI, land surface temperature (LST) obtained from MODIS, and water quality indices such as OC3 and QAA quality. The region of interest (ROI) was delimited by creating a mosaic of marsh images, with a focus on land cover variation, specifically mangrove areas. Major water bodies within the study boundaries were excluded from a more focused vegetation analysis. For land cover classification, a supervised classification process was conducted using JavaScript in the GEE. One hundred training samples were collected for each land-cover class based on visual interpretation and the researchers’ prior knowledge of the region. These samples allowed the surface to be categorized into five classes: dense mangroves, sparse mangroves, bare land, built-up areas, and waterbodies. For the classification algorithm, Random Forest (RF) with a set of 150 trees was used because of its ability to increase the classification accuracy by combining the results of multiple decision trees.^20,21

Characteristics of the study area

Figure 2 illustrates the study area used to develop the corresponding analysis.

Figure 2. Geographic location of the La Caimanera marsh (Colombia).

Figure 2 illustrates the zoning of the La Caimanera marsh in the two visually contrasting areas. The first corresponds to the area covered by mangroves (in red), predominantly constituted by at least four different species of mangroves that form a dense tree canopy at medium to high altitudes, creating a closed forest landscape characteristic of these ecosystems.²² In the center of the swamp, there is a section of open water that lacks surface vegetation, which makes it a favorable area for tourism, recreation, and artisanal fishing. The La Caimanera marsh is bordered to the north by the Atlantic Ocean, with which it is connected by a natural channel that plays a fundamental role in the hydrological and ecological dynamics of the system. This channel facilitates the exchange of water between the marsh and sea, directly influencing critical parameters such as salinity, water quality, and the stability of the mangrove ecosystem, promoting the health and resilience of these key habitats for the region.

The La Caimanera marsh is distinguished by its highly diverse habitats, including beaches, mangroves, mudflats, and wetlands, making it a crucial ecosystem for the development of a wide range of animal species at different trophic levels.²³ These habitats harbor a wide variety of life forms, from autotrophic bacteria that decompose organic matter to aquatic mammals, including diverse invertebrates such as annelids, mollusks, and arthropods, as well as vertebrates such as fish, amphibians, reptiles, birds, and terrestrial mammals, many of which depend on mangroves to complete the fundamental stages of their life cycle.²⁴ The mangrove structure in this area is characterized by a fringe-type mangrove swamp approximately 500 m wide, starting from the main channel in the Boca de la Ciénaga sector, coinciding with that reported by the Corporación Autónoma Regional de Sucre.¹⁸ In the first strip, the mangrove is predominantly monospecific and dominated by the red mangrove (Rhizophora mangle). As one moves inland, the ecosystem becomes more heterogeneous, with co-dominance of black mangroves (Avicennia germinans) and white mangroves (Laguncularia racemosa). After this zone, the forest adopts a basin-type structure, with flooded areas during the rainy season fed mainly by the Petalaca and San Antonio streams. Black mangroves predominate here, accompanied by white mangroves, and to a lesser extent, red mangroves.¹⁷

Evaluation of indexes in the marsh area

Figure 3 shows a climogram of the maximum, minimum, and average temperatures in the study area over the last 5 years. These data were obtained from the MODIS Land Surface Temperature and Emissity (MOD11) satellite and data taken directly by researchers in the study area using low-cost sensors, such as temperature (DS18B20), solar radiation (TSL2591), and air quality (MQ-135) sensors connected to an ESP32 S3 controller to validate the satellite information obtained. This information is important for identifying the climatic patterns of the study area and the times of the year when the dry and rainy seasons occur. The Google Earth Engine (GEE) platform and satellite images from the European Space Agency (ESA) were used to generate a time series of NDVI and NDWI indices in four critical areas of the marsh. The first area, located in the southern sector, was characterized by a high density of vegetation and abundant mangroves, represented by the NDVI_south index. In contrast, the northern sector, which connects the swamp with the sea and acts as a tourist channel for residents and visitors, was monitored using the NDVI_North index. Additionally, a point was identified in the transition zone between the beach and the swamp, captured in the NDVI_beach index, while the last area of interest was located in the interior of the swamp, characterized by open waters and scarce presence of mangroves, reflected in the NDVI_swamp. The NDWI index was used in similar areas of the marsh ( Figure 3b).

Figure 3. a. Climogram of the La Caimanera marsh (mean value for the period 2018-2023); b. Location of sample points for the indices.

The values shown in Figure 4 with the yellow line correspond to NDVI_In, which represents the inner part of the marsh, characterized by open water and the scarce presence of nearby mangroves. Finally, NDVI_beach refers to the area of the marsh bordering the beach and surrounding communities.

Figure 4. NDVI and NDWI index comparisons in four points of the La Caimanera marsh.

Figure 4 presents the NDVI variation in the four zones of the La Caimanera marsh from January 2019 to January 2024, highlighting the patterns in vegetation cover. The NDVI_south (blue) shows high and stable values, reflecting dense and consistent vegetation. In contrast, NDVI_Beach (red) shows low and fluctuating values, indicating sparse vegetation, because it is a construction zone, roads, little vegetation, and high salinity. NDVI_North (green) exhibits moderate vegetation, with significant mangrove density, lower than that of the southern zone, with seasonal variations or human influences. NDVI_In (orange) shows high variability but with a tendency toward negative values because the zone is mostly unvegetated. In parallel, NDWI in Figure 4 (dotted lines) shows the humidity in each zone. The southern region (blue line) presents high negative values because the high density of mangroves prevents the detection of water by the satellite. The beach (red line) fluctuates near zero, indicating the low humidity levels characteristic of this zone. In the north (green line), the predominantly negative values throughout the period reflect moderate humidity. In contrast, the interior zone of the marsh (orange line) shows high peaks associated with the presence of open water during most of the year.

Figure 5 presents various indices applied to the study area, including OC3 and QAA, which are fundamental for estimating the chlorophyll concentration and, therefore, to assess the health of phytoplankton in the ROI. In particular, the Normalized Difference Vegetation Index (NDVI) in the La Caimanera marsh, Colombia, was analyzed for July 2010, 2015, 2020, and 2023, revealing significant dynamics in both vegetation and water bodies in the area.

Figure 5. Satellite evaluation of the area of interest using NDVI, NDWI, MNDWI and OC3 and QAA quality parameters.

The NDVI is a key tool in remote sensing and is used to measure the density and health of vegetation, providing an accurate representation of the spatial distribution of ecosystems. This temporal analysis allowed the identification of crucial variations in vegetation patterns and changes in the extent of the main water body, providing essential information for environmental management and ecosystem conservation. In 2010, most of the terrestrial area showed moderately dense vegetation (light green), with some areas of mangroves surrounding the inner water body of the marsh, and very dense vegetation (dark green). The main water body, represented in blue, is clearly visible, whereas a small interior water body, corresponding to the marsh, is distinguished by lighter tones. By 2015, a significant increase in dense vegetation was observed, particularly in the mangrove zone in the center-south of the study area. The marsh appears to have slightly increased in size, which could be related to changes in water levels or an increase in the surrounding vegetation. Despite this growth, moderately dense vegetation continued to predominate in areas farther from the coast. In 2020, there was a general decrease in dense vegetation, mostly replaced by moderately dense or sparse vegetation (light green to greenish-yellow). The inland water body maintained its size, but the change in vegetation density suggests possible variations in environmental conditions or land management during this period. Finally, in 2023, a partial recovery of dense vegetation was detected, although it did not reach the levels observed in 2015. The landscape appears more heterogeneous, with a mixture of areas of dense and moderate vegetation, reflecting greater variability in the ecosystem conditions. The marsh, on the other hand, appears to have been slightly reduced in extent, which could be due to variations in water levels or changes in riparian vegetation.

Regarding the NDWI, for the years 2010 to 2015, there were slight increases in the intensity of blue in the marsh, which could indicate an increase in water content or an improvement in water quality. Similarly, for the time window from 2015 to 2020, the coloration remains similar, suggesting a certain stability in the water conditions; for the period from 2020 to 2023, a slight decrease in the intensity of blue in the marsh is clearly observed, which would be related to a reduction in the volume of water or changes in water quality. However, the changes in the NDWI and MNDWI index colors for the area around the bog show lighter shades of blue, which change subtly over the years. This indicates variations in the soil moisture or surrounding vegetation. Finally, the evaluation of both indices showed some darker spots scattered in the area, which represent small bodies of water (jabueyes) or areas of higher humidity. These values appear to have remained relatively constant over the years. Regarding water quality in the marsh, the OC3 and Quasi-Analytical Algorithm (QAA) parameters were used for 2020 and 2023, generating the following information: the OC3 algorithm is commonly used to estimate chlorophyll-a concentration in oceanic and coastal waters, and the images show orange and purple tones as predominant colors with a slight intensification of purple color in 2023 compared to 2020, especially in the marsh area and some terrestrial areas. This indicates a small increase in the chlorophyll-a concentration or changes in the optical properties of the surrounding water and land.

The QAA algorithm was used to derive the inherent optical properties of water such as light absorption and scattering. The predominant colors were green, blue, and yellow tones, with the bog having a lighter blue-green tone. Figure 5 shows a slight reduction in the intensity of the blue-green color in the marsh in 2023 compared to 2020, indicating a lower amount of sediment and less turbidity in the water. Finally, the images were downloaded using the Copernicus Browser platform service of the European Space Agency, using a cloudiness filter of 10% allowed, with high-resolution visualization, WSG 84 (EPSG 4326) coordinate system using the Senttinel-2 L2A satellite sensor.

Variations in vegetation cover

Figure 6 illustrates LULC analysis using the dynamic world platform for the corresponding study area.

Figure 6. LULC analysis using the Dynamics World platform for the study area.

The analysis of LULC images of the La Caimanera marsh, performed from images captured by the Dynamics World platform during the summers of 2015, 2017, 2021, and 2024, revealed remarkable ecosystem resilience. Mangrove cover, predominant in the ROI, showed a slight expansion between 2015 and 2017 and remained stable in the following years, suggesting effective conservation management. In parallel, a subtle initial reduction in the marsh water body was observed, followed by stabilization, possibly attributable to climatic variations or adjustments in local water management. Urban dynamics in the coastal zone exhibited a pattern of controlled growth, as evidenced by a modest expansion of built-up areas between 2015 and 2017, which subsequently stabilized. This phenomenon, together with the absence of predominant agricultural areas, indicates a balance between human development and preservation of the natural ecosystem. Minor fluctuations in grass and shrub cover suggest a dynamic interaction between the natural processes of ecological succession and anthropogenic changes in land-use. Together, these patterns point to integrated and sustainable management of the coastal landscape, where mangrove conservation and urban growth regulation coexist harmoniously, promoting ecosystem resilience in the face of environmental and anthropogenic pressures. The La Caimanera marsh ecosystem shows remarkable stability over the period 2015-2024 with subtle but significant changes in the distribution of its main landscape components, as can be seen in the distribution of the territory in the La Caimanera marsh for 2015, 2017, 2021, and 2024, as illustrated in Figure 7.

Figure 7. Territory distribution in the La Caimanera marsh for 2015, 2017, 2021 and 2024.

Tree cover, predominantly composed of mangroves, has experienced a slight decrease from 60.17% in 2015 to 59.82% in 2024, reflecting a marginal but persistent loss of vegetation. In parallel, the area covered by water showed a similar trend, decreasing from 29% in 2015 to 28.33% in 2024, corroborating the visual observation of a subtle contraction of the marsh water body. In contrast, built-up areas exhibit a gradual increase, rising from 10.12% in 2015 to 11.26% in 2024. This 1.143% increase in built-up areas, although moderate, indicates increasing pressure from human development in the natural ecosystem. The relative stabilization of these percentages between 2021 and 2024 suggests that land management policies may be effective in balancing development demands with conservation imperatives. These quantitative data reinforce the initial interpretation of a resilient ecosystem under careful management, where gradual changes in land cover reflect a delicate balance between mangrove conservation, marsh-water dynamics, and controlled urban growth. The persistence of a high proportion of tree and aquatic cover, despite developmental pressures, underscores the importance and effectiveness of conservation strategies implemented in the region.

Mangrove cover in the La Caimanera marsh

According to Bunting² and data obtained through the Global Mangrove Watch platform developed by the Japan Aerospace Exploration Agency (JAXA) and Aberystwyth University, the total area of mangrove habitat in the La Caimanera marsh in 2020 was 12.24 km², with a coastal linear cover of 7.55 km. Figure 8a shows the changes in mangrove cover, highlighting an increase of approximately 0.17 km² between 1996 and 2020, equivalent to a 1.3% increase in the total area. It is important to note that, as of 2016, a significant acceleration in this gain of mangrove area was observed, which evidences the joint efforts of the community and environmental management entities in the conservation and restoration of this fundamental ecosystem in the Gulf of Morrosquillo region.

Figure 8. a. Variations in mangrove area in the La Caimanera marsh between 1996 and 2020; b. Distribution of mangrove covers losses and gains.

This growth in mangrove cover underscores the success of the protection, restoration, and sustainable management initiatives that have been implemented in the area. Mangroves are not only vital for local biodiversity but also play a key role in climate change mitigation, coastal protection, and the livelihoods of local communities. The acceleration in the recovery of these areas since 2016 suggests greater effectiveness of conservation strategies, which could be replicable in other mangrove ecosystems globally, requiring intervention and recovery mechanisms in density.

Regarding the distribution of mangrove biomass areas in the swamp, that is, the amount of living plant material (leaves, branches, trunks) present per unit area, expressed in tons per hectare (t/h), in 2020 the average amount of aerial biomass was 61.66 (t/h). Detailed percentage distribution of aboveground biomass within different ranges: 31.78% of the mangrove area has a biomass between 0 t/h and 50 t/h, which indicates that a third of the mangrove is young or is being affected by disturbances or is in recovery. Approximately 58.41% of the area has a biomass between 50 and 100 t/h, which is the predominant category, suggesting that most mangroves are composed of healthy adult trees. Likewise, 9.81% had between 100 t/h and 150 t/h, that is, a mature forest with healthy adult trees. Figure 8b shows the recovery of mangrove areas in the La Caimanera marsh over the last 20 years (represented by blue dots) as well as the mangrove areas that have been lost over the same period (represented by red dots). This information is consistent with the minimal changes in land use observed in the data in Figure 6 over the last 10 years. On the other hand, the classification of mangroves according to their height is important for determining the structure and state of the mangrove ecosystem in that area. Using JAXA satellite images, it was found that the maximum canopy height was 12.4 meters, and that the distribution of the entire mangrove area, in terms of height, was established, as shown in Figure 9a. Only 3.4% of the mangrove vegetation in this region consists of trees with heights between 0 and 5 m. This indicates that a very small portion of the mangroves are young or of low stature, which may reflect recent regeneration or areas where mangrove growth is limited, possibly because of salinity and water depth. The highest concentration of mangroves in the area is 10–25 m in height, which is more than 83% of all vegetation. Therefore, the mangrove ecosystem in this area is dominated by mature, well-established trees. This is a sign of a healthy and mature ecosystem in which trees have reached an optimal growth stage. However, it is not without concern that almost one-fifth of the mangroves (19.7 %) are trees that exceed 20 m in height, representing very tall and possibly older trees, which may be key to the overall stability and structure of the ecosystem but also represent the end of the life cycle and should be considered for early renewal.

Figure 9. a. Mangrove height classification in La Caimanera Marsh; b. RGB image Mangrove area in 1997; c. Mangrove classifications of 1997; d. Mangrove area 2021; e. Mangrove gain between 1997 and 2021.

The multitemporal analysis of the mangrove ecosystem in the La Caimanera marsh was based on a comparison of processed satellite images. The base image in RGB composition ( Figure 9b) clearly distinguishes the two predominant coverages: water bodies and mangrove forests. The application of the binary overlay analysis method between the mangrove classifications of 1997 ( Figure 9c) and 2021 ( Figure 9d) revealed a significant transformation of the ecosystem during this period. The spatial distribution of mangroves in 1997 exhibited a remarkably fragmented pattern, characterized by a discontinuous canopy arrangement, particularly evident in the peripheral areas. In contrast, the 2021 image shows a significantly denser and more homogeneous mangrove cover with a closed and continuous canopy, indicative of a more robust and mature forest structure.

The positive evolution observed during these 24 years suggests effective regeneration of the ecosystem, attributable to multiple converging factors: the successful implementation of conservation measures, the remarkable natural resilience of the system, and an apparent reduction in anthropogenic pressures. This transformation was particularly significant in the central and southern sectors of the study area, where previously fragmented spaces have evolved into continuous and vigorous forest cover. The results obtained not only show a substantial improvement in ecosystem health but also position the La Caimanera marsh as a relevant case study for mangrove restoration and conservation initiatives at regional and global levels, especially considering the increasing pressures that these coastal ecosystems are currently facing. Figure 9e corresponds to the mangrove losses that have occurred in a different time window from 2021 to 2023. It is noticeable that the amount of vegetation cover lost is small, but it has come to interrupt the good growth process that, for the previous time window, 1997 to 2021, showed the mangrove area of the La Caimanera marsh. Similarly, an analysis of changes in mangrove cover in recent years, depicted in Figure 9e, reveals a contrasting trend for the period 2021-2023. Although cover losses during this interval are quantitatively smaller, their occurrence interrupts the pattern of recovery and significant expansion that characterized the previous period (1997-2021). This inflection in the dynamics of vegetation cover, although moderate in extent, is particularly relevant because it marks a break in the positive trajectory experienced by the La Caimanera marsh ecosystem. This change in trend suggests the possible emergence of new pressures on the ecosystem or the intensification of existing stress factors, which merits special attention from the perspective of management and conservation of the area.

Figure 10a presents an analysis of the changes in surface water bodies in the La Caimanera marsh based on annually updated data provided by the application developed by Global Surface Water (https://global-surface-water.appspot.com/).²⁵ This tool uses over 32 million Landsat and Copernicus satellite images, with a spatial resolution of 30 m, to quantify the variation in global surface water cover. In the analysis, increases in the presence of water are represented in green, whereas decreases are shown in red. A significant change was highlighted in the water masses of the upper sector of the swamp, specifically in the open-water zone located in the center. These changes have been observed between 1984 and 2021. The black areas indicate zones where no significant change in the presence of water was recorded during the same period, while the intensity of the color indicates the degree of change in percentage terms.

Figure 10. a. Mapping of surface water history in La Caimanera Marsh; b. Monthly water recurrence in marsh; c. Water history in the study area.

Using this application to specifically examine the temporal evolution of surface water presence in the loss area ( Figure 10b, red circle), it was possible to visualize temporal profiles at the pixel level. These profiles showed monthly recurrence over 37 years ( Figure 10b), as well as a history of water presence sorted by seasonality and year over the same period ( Figure 10c). Overall, the plots indicate that at this location (latitude 9.422947°, longitude -75.6332587°), there were no valid observations for 1985, 1988, 1990, 1992-1995, 2003, and 2010 (represented as gaps in the lower plot). Between 1996 and 1999, a high water level was observed, which decreased markedly after this time window. Throughout the 37 years of observation, no water was detected between January and June or September; the months with recurrent water presence were July, August, October, and November ( Figure 10b), in agreement with the rainy season in the region.