Introduction

Peat-accumulating wetlands are ecosystems characterized by thick organic soils that accumulate over centuries to millennia1,2. When these ecosystems are undisturbed, they provide a multitude of ecosystem services (ES) for nature and society, including carbon storage, water regulation, biodiversity provision, as well as food, fiber, and cultural resources for local communities3,4. It has often been noted that the safeguard of pristine peatlands from drainage and reclamation, in combination with the restoration of drained peatlands, will help mitigate climate change, improve water resilience, and maintain biodiversity5,6.

Across the tropical belt (~ 20°N to 20°S), peatlands may occupy up to 1.02 Mkm27, but there remain large discrepancies in peatland estimates across existing maps1,8,9,10,11, in part due to many countries lacking official peatland inventories (including Costa Rica). As the tropics are subject to some of the highest rates of land-use change and degradation globally12, improving peatland maps and deepening knowledge of tropical peatlands constitute critical steps to help improve the management of these ecosystems and ensure their wise use, as endorsed by the Convention on Wetlands (Ramsar).

The past decade has witnessed an increase in continental- to global-scale peatland maps that are supported by two main approaches: inventory-based and probability-based. While both types of products offer a much-needed, high-level estimate of the general location and extent of peatlands globally, they tend to lack specificity and they are often heavily biased due to missing or inappropriate input data. In the case of inventory-based maps, the reliance is on regional and national peatland inventories1, histosol maps8, a combination of inventories and histosol data10, or other proxy data such as flooded vegetation or species-specific vegetation maps7. While these products offer the distinctive advantage of being based on ground-truthed and established relationships between peat extent and specific vegetation or soil types at regional and national scales, a caveat is that they only include ecosystems that have been previously described and mapped. This means that they probably underestimate peatland extent in unmapped regions, but overestimate peat deposits at the local scale at the same time. For example, the documented relationship between a specific wetland plant community and the presence of peat may not apply everywhere this plant community is found13. As for probability-based maps, they combine environmental and topographic data to predict the location of peatlands9,11. Predictive mapping has the power to explicitly identify areas where the conditions hypothesized to form and accumulate peat are found. It is, however, limited by the fact that the algorithms tend to generalize the conditions of peat formation and accumulation and its vegetation - which are incredibly diverse across the globe. Lastly, for both types of maps, the spatial resolution tends to be too coarse for local assessment. Overall, these different mapping techniques all come with their advantages and limitations and have yielded largely different peatland extent across the world, but also across the Caribbean region (Table 1).

Table 1 Estimated peatland extents for the Caribbean region.
Full size table

In Central and South America, direct peatland mapping started in the 1980s and 1990s14,15,16. In these studies, peatlands have been mapped in soil surveys as organic soils, histosols, or swamps, primarily to identify large peat deposits that could be exploited as energy resources through mining. The resulting maps tend to be local in scope and thus only provide snapshots of potential peatland locations. In the 2000s, the trend of identifying and mapping wetlands and peatlands on a local scale continued, however the focus shifted to using the peat deposits as archives for paleoecological reconstructions17. In general, however, the lack of country-level peatland inventories – as seen across much of Central America – has resulted in the use of the aforementioned, and poorly ground-referenced, global-scale predictive datasets1,8,9,10,11. The work done by18 to create an inventory of Caribbean peatlands is the first in recent years to focus on this region.

The objective of this study is to generate a probabilistic map of peat presence for Costa Rica to further the accuracy of peatland maps for the country. Our map was created using a probabilistic approach based on several high-resolution proxy datasets that are indicative of peat presence, including vegetation communities, ecosystem types, soil organic carbon content, areas of seasonally and permanently flooded vegetation, and the probability of fibric, hemic, and sapric histosol presence. This objective responds to the specific commitments of Costa Rica within the international regulatory framework related to the Ramsar Convention, including protecting all of the country’s coastal wetlands, restoring priority wetland areas, and developing wetland management and monitoring plans.

Methods

Study region

The country of Costa Rica is located between 8–11°N and 82–86°W in Central America. It is bordered by the Caribbean Sea on the east and the Pacific Ocean on the west. A mountain range (Cordillera de Talamanca) runs through the south-central portion of the country, largely separating the climatic regions on the eastern and western sides of the country19. Generally, the country’s climate is controlled by the seasonal movements of the Inter-Tropical Convergence Zone20.

Noteworthy is that the ‘peatlands of Costa Rica’ are very seldom described as such. Instead, peat-accumulating ecosystems are integrated into more general ‘wetland’ classifications. Indeed, in Costa Rica and many other countries of Latin America, the term ‘peat’ has not been integrated into ecosystem classifications, making peatland mapping particularly challenging. Instead, wetland classification is primarily based on plant communities21. The wettest peat-accumulating systems include mangroves and grass-dominated marshes; they tend to be found along the Caribbean coast. Mangrove ecosystems are dominated by Rhizophora mangle (red mangrove) and Avicennia germinans (black mangrove)22. Marshes are dominated by an array of grasses, herbs, and sedges21. Slightly drier (because of a slight elevation difference) are the peat swamps, which are composed of palms (Raphia taedigera), trees (Campnosperma panamense), or herbaceous (Myrica sp.) plant communities22. Those ecosystems can be found along coastlines (in back-barrier areas) as well as within river floodplains. Lastly, Sphagnum-dominated peat bogs have been described in the high mountains of Costa Rica, usually above 2500 m21,23.

Peatland mapping methods

Given the large differences in topographic settings, vegetation communities, and peatland sizes between high-elevation and lowland peatlands, two approaches were used to map those peatland types.

High-elevation peatland mapping

High-elevation peatlands in Costa Rica have been found in small depressions in the mountainous region of the Cordillera de Talamanca21,23. While these ecosystems only account for a small fraction of the country’s peatlands, they may be important environments for local biodiversity and water regulation. Their flora is also unique within Central America21. Because these peatland areas are so small, they cannot be automatically mapped using freely available satellite imagery sources, which have a spatial resolution too coarse to differentiate them from the surrounding environment. We therefore used the World Imagery basemap in ArcGIS Pro, with up to 0.5-m spatial resolution, to manually delineate these small peatland areas at a 1:4,000-meter scale (first author: E. Rabel). The procedure was based on a set of confirmed high-elevation peatland occurrences above 700 m.a.s.l. collated from previously published articles, during our team’s fieldwork in 2023, and supported by expert knowledge regarding mountain peatlands (co-authors). These input data provided the context to develop visual (color and texture) indicators and to identify the geomorphological setting of confirmed peatlands. Both have been extrapolated across the mountain ranges in Costa Rica to identify additional peatlands on the satellite imagery. Once a new potential peatland area was identified visually, the 3-D terrain feature in Google Earth Pro was used to assess whether or not the terrain could support the accumulation of peat (‘geomorphological setting’). Two classes of polygons were then drawn: (1) areas of high confidence of being peatlands, vs. (2) areas of lower confidence of being peatlands. The latter have some but not all the characteristics of confirmed peatlands in terms of image color and texture, and/or terrain (inferred from viewing the 3-D landscape in Google Earth Pro). Areas with a slope greater than 15 degrees were excluded from both categories, as peat is unlikely to accumulate under such slope angles. Specific slope values were not used as a differentiating factor between the two confidence classes.

Lowland peatland mapping

Spatial data describing the main vegetation communities, ecosystem classes, and soil information for the entire country of Costa Rica were gathered for the probabilistic analysis. Our general methodology follows that described by24] and [25, where the different data sources are given different weights in a probability analysis, describing different levels of accuracy and pertinence. The following datasets were used:

(1) Vegetation communities and species: were derived from the Registro Nacional de Humedales vector layer22. While this is an ‘ecosystems’ dataset (see below), it also includes the distribution of some indicative wetland communities and species as polygons (Raphia taedigera, Campnisperma panamense, Prioria copaifera, Mauritia flexuosa; Table 3), which were included in the ‘vegetation raster’ and got high weights in this probabilistic analysis (Table 2).

(2) Ecosystem classes: a total of three datasets were combined to create the ‘ecosystems raster’. First, the Registro Nacional de Humedales22 vector layer was selected (same dataset as above but now all polygons). Second, a total of 15 wetland types provided in 450 m resolution have been included from the Global Lakes and Wetlands Database v2 (GLWDv2)26. Thirdly, we used the flooded vegetation areas from the Sentinel-2 10-m Land Use/Land Cover (LULC) raster dataset produced by Esri and Impact Observatory27. All the selected specific categories from those ecosystem datasets generate high confidence in their ability to indicate peat (Table 2).

(3) Global Soil Organic Carbon Map (GSOC; FAO): this dataset was accessed using the Global Soil Information System (GLOSIS) portal28. This raster dataset has a spatial resolution of ~ 930 m in Costa Rica; this gridded product provides interpolated SOC stocks on the basis of local soils samples that were collected to a target depth of 30 cm28. A total of 1518 samples from Costa Rica were used to generate this mapping product28. We used the GSOC data for Costa Rica to identify areas characterized by high SOC (Table 3). The GSOC layer received the lowest weight because, while peat soils are composed of high amounts of SOC, not all organic carbon-rich soils are peat (Table 2). Additionally, this layer has the tendency to increase SOC with elevation while underestimating lowland areas with high SOC.

(4) SoilGrids 250 m resolution: Fibric, hemic, and sapric histosol layers were extracted from the “SoilGrids250m - Predicted probability WRB 2006 subgroup classes” database30. This histosol dataset was given a high weight in our analysis, since histosols very often refer to peat soils (Table 2). However, we acknowledge that this product is in fact a probabilistic map, which is guided by regional inventories that are scarce in Costa Rica and thus, it may be inaccurate.

All selected datasets were uploaded into ArcGIS Pro 3.2.2. and then converted into individual polygons (when applicable) to enable independent scoring of the included information. For example, the fibric, hemic, and sapric histosols layers from the SoilGrids250m database were scored separately and analyzed as individual layers. Each dataset received a score based on spatial resolution, with lowest resolution datasets receiving a score of 1 and highest resolution datasets receiving a score of 5. Each polygon within each dataset received a data score for the type/suitability of the peat-indicator data provided. For polygon scoring, we also used a scale from 1 to 5, with 1 being the least likely to indicate the presence of peat and 5 being the most likely to indicate the presence of peat. For example, a vegetation polygon representing a species that is strongly associated with peatlands was given the highest score of 5. The histosol and GSOC raster layers were given initial scores of 1 to 5 using natural breaks to divide their values into five classes. The lowest raster value class was given a score of 1, and the highest raster value class was given a score of 5 (Table 3). Each data polygon was then given an individual ‘source score’ that was the average of the spatial resolution score and the data score (Table 3).

Once every data polygon was scored from 1 to 5, six scored raster datasets were created representing the four categories described above (1- Vegetation, 2- Ecosystems, 3- GSOC, and 4- Histosols; Supplementary Figure S1). Those rasters were then put into the Suitability Modeler in ArcGIS Pro and given weights (Table 2). Determining the weight of each collated dataset is an integral part of this analysis and needs to consider several important factors, such as the original intended use of the dataset, whether it is ground referenced, and how accurate it is in indicating peat presence. For the initial probabilistic map, we gave an equal weight to the Vegetation and Ecosystem rasters (30% each), 10% to each of the three Histosols datasets (total of 30%), and 10% to the GSOC (Table 2). The effect of different initial scoring on the mapping results will be discussed below.

Table 2 Initial weights used for each dataset in the probabilistic analysis.
Full size table
Table 3 Procedure for evaluating and scoring peatland indicator data. GSOC values are given in units of tonnes of C per hectare (or MgC/ha− 1), and all histosol values are given in units of estimated probability of occurrence28,30.
Full size table

Agreement analysis

To visually compare our results with those from previously published studies7,9,10, an overlay analysis was performed. Raster cells in each dataset (n = 3) were given a value of 1 if peat was not indicated, and a value of 2 if peat was indicated. These rasters were then summed to create a ‘master raster’, with values ranging from 3 (no mapped peatlands in any of the datasets) to 6 (peatlands mapped in all 3 datasets). Under this scheme, a value of 4 indicates an area that was mapped as peatland by 1 of the 3 studies, while a score of 5 indicates that 2 of the 3 studies have mapped that area as peatlands. The PEAT-ML map11 was excluded from this analysis because the grid cells of this map have values that reflect the percentage of peatland coverage per grid cell, therefore the map is not directly comparable to the others.

Results and discussion

High-elevation peatland distribution and extent in Costa Rica

Our map of high-elevation peatlands (> 700 m.a.s.l.) returned a total extent of 23.08 km2, with individual peatlands often smaller than 0.05 km2 in size (Fig. 1). The largest peatland that was identified was only 0.75 km2 in size. The ‘high-confidence’ mapped peatlands accounted for 20% of the total high-elevation area (4.5 km2), whereas the ‘lower-confidence’ areas accounted for 18.5 km2. To yield higher confidence, the collation of additional ground references is inevitable. We note that this mapping was performed visually and manually because the small peatland size precluded the use of the probabilistic method using coarser resolution datasets. The Costa Rican high-elevation peatland types include Sphagnum bogs23, and bogs dominated by Puya dasylirioides and Blechnum occidentale19.

Fig. 1
figure 1

High-elevation peatland distribution and extent in Costa Rica. The dark gray mask indicates areas below 700 m elevation, which were excluded from our classification.

Full size image

Low-elevation peatland extent and distribution in Costa Rica

Our probabilistic map of low-elevation peatlands indicates a total peatland extent of 1433 km2, when using the > 50% probability classes (Fig. 2). Spatial resolution varies from 230 to 450 m across this map, based on the resolution of the layers present in any given area. The majority of these peatlands (~ 85%) are found along the Caribbean coast, accounting for 1189 km2 (Fig. 3B). They mostly consist of wooded and palm swamps dominated by Prioria copaifera and Raphia taedigera, respectively, and occasional occurrences of herbaceous swamps. These ecosystems tend to be found in back-barrier settings, behind mangroves and/or sandy beaches18. In the northern, continental portion of the country (including Caño Negro in the Alajuela Province), large riverine and palustrine wetland complexes comprise peat-accumulating systems, which might account for ~ 10% of Costa Rica’s lowland peatlands (109 km2; Fig. 3A). These peat-accumulating systems may include herbaceous marshes and wooded swamps, accompanied by shallow lagoons and riparian forests22. These peatlands are found in the floodplains of larger rivers such as the Frío and the Medio Queso, such that wetland hydrology is heavily impacted by seasonal flooding. For instance, some of the peatlands found in the Caño Negro Wildlife Refuge (RAMSAR site 541), all found in floodplains, experience over two meters of flooding above the surface level during the rainy season (April-November). This flooding is thought to suppress soil organic decay during the wet months of the year, which allows for a net positive carbon balance in these soils and thus the formation and preservation of peat. The third main area captured by our probability map is found near the base of the Osa Peninsula, on the southern Pacific coast. This area accounts for less than 5% of the lowland peatland extent (82 km2) (Fig. 3C). The peatlands of this area include herbaceous swamps and wooded swamps22.

Fig. 2
figure 2

Low-elevation (0–700 m.a.s.l.) peatland extent and distribution in Costa Rica, based on a probabilistic mapping approach using vegetation, ecosystem, and soil inventory data; shades of greens indicate probability classes.

Full size image
Fig. 3
figure 3

The three lowland areas most likely to comprise peatland complexes in Costa Rica, based on our probabilistic approach. (a) Caño Negro, (b) the northern Caribbean coast -Caribe Noreste, and c the Osa Peninsula.

Full size image

Comparison with other Costa Rican peatland maps

Figure 4A shows the high-elevation peatland map combined with the low-elevation peatland map (> 50% probability) for Costa Rica as derived in our study. This final map yields a total peatland extent of 1456 km2, of which the vast majority (1433/1456 km2, or ~ 98%) is found in the lowlands. As described above, the lowland vs. high-elevation sites are characterized by different peatland types, with distinct plant communities and hydrological requirements. The cool temperatures and moist conditions needed to initiate, build, and maintain the high-elevation Sphagnum bogs found in the Cordillera de Talamanca (RAMSAR site 1286) are rarely found in the tropical belt, thereby explaining the small percentage of mountain peatlands when compared to the more widespread lowland peat swamps and other lowland peat-accumulating riverine wetlands. A comparison with other maps of Costa Rican peatlands (Fig. 4B-E)7,9,10,11 indicates that all studies agree on the presence of peat along the Caribbean coast, particularly in the northern portion of the country (Fig. 4). Three of the four other studies also agree on peat presence in the Alajuela Province7,9,10,11. However, total peatland area estimates vary widely across those studies, from ~ 577 to 2670 km2 (Table 4). Other differences across the mapping products are found along the Pacific coast; the SWAMP map9 and the PEATMAP10 suggest peatlands in these regions, whereas our approach yielded low probabilities. The Global Peatland Map 2.07 has the smallest proposed extent of peatland for Costa Rica (Table 4), with this area being exclusively found along the Caribbean coast (Fig. 4D). As for the high-elevation areas, the map by PEATMAP10 does not capture the small peatland areas identified in our study, but rather suggests the potential for peatlands along water bodies (e.g., streams and associated floodplains). High-elevation areas are neither covered in the SWAMP map9, the Global Peatland Map 2.07 nor the PEAT-ML map11.

Table 4 Extent estimates for Costa Rican peatlands, including this study.
Full size table
Fig. 4
figure 4

Synthesis of peatland distribution and extent for Costa Rica. (a) This study. (b) SWAMP map9. (c) PEATMAP10. (d) Global Peatland Map 2.07. (e) PEAT-ML map12 taken from grid cells of > 20% peatland coverage. (f) Agreement analysis of the SWAMP map, PEATMAP, and Global Peatland Map 2.0.

Full size image

We also compared our new peatland map with local information (Fig. 5)14,15,16, which was either obtained from national soil subgroup maps or from our own field work in 2023. Some areas proposed in these local products14,15 extend farther inland than suggested by our probabilistic model. In the Cordillera de Talamanca, peatland extent estimates overlap with peatland areas from our study14,15, though theirs only point to a general area containing small peatlands, rather than delineating each peatland as presented in our study. As for our own expedition in the Alajuela Province (in the north-central portion of the country, bordering Nicaragua), our model did not predict peatlands in this area, although there is a wetland complex (Medio Queso) with documented peat in this region29; Fig. 5. Areas such as Medio Queso, which do not align with the predictive map, warrant further evaluation of the model and highlight the need for additional ground referencing. In this case, land-use might have biased the results, as Medio Queso is heavily managed for cattle grazing and was therefore excluded on the basis of its land-use type.

Fig. 5
figure 5

This study’s peatland extent estimate for Costa Rica, overlaid by national and local peatland/histosol extent estimates from Mora et al.14, Obando & Malavassi15, and Jong16. Peatland sites confirmed during our 2023 field expedition are also presented.

Full size image

Key uncertainties and next steps

The accuracy of the peatland map of Costa Rica is ultimately limited by data availability and quality, as well as ground-referencing information. The probabilistic nature of our approach also adds a level of complexity to the interpretation of the map. First, the SoilGrids250m and GSOC layers were probabilistic models themselves, which may inherently add uncertainty to our model28,31. They are global in extent, but the regional records from Central America that are feeding them are scarce28,31. Moreover, for the SoilGrids250m presents large gaps in training data for the tropics and wetland areas, as well as unstandardized laboratory data and soil descriptive variables, and cautioned against the use of legacy soil profile data in the making of their map30. In the case of our use of histosol data, we worked under the assumption that all histosols are peatlands, which is a simplification that may add inaccuracy. Uncertainties in the GSOC layer were similar to that of the SoilGrids250m layer (i.e., the histosol layers), resulting from discrepancies in soil sampling, measurements of soil properties, and mapping techniques among countries28. As for the vegetation and ecosystems datasets that were used, the layer produced by SINAC is useful as it was produced in-country with regional appropriate classes and ground-referenced data22. The largest source of error in the GLWDv2 dataset26 is the ever-evolving nature of the wetlands themselves and changes due to anthropogenic causes (diversion and clearing for agriculture, dam construction, etc.). These changes lead to temporal misalignments between source datasets, which are especially pronounced in areas where multiple datasets overlap26. The Sentinel-2 10-m LULC map contains uncertainty in classifying spectrally similar pixels, as these are difficult for satellite-based machine learning models and human annotators to distinguish31. The same uncertainty also lies within the GLWDv2 dataset, as land cover types are ever-changing due to natural and anthropogenic causes.

The weight given to each raster has an influence on our probabilistic map. We developed a series of sensitivity tests to assess the importance of weights on the final product. First, we gave the histosol layers the highest weight (60% rather than our initial 30%), which yielded a map with high-probability peatlands only found along the Caribbean coast (Supplementary Figure S2). In contrast, giving the GSOC layer the highest weight (60% rather than our initial 10%) showed a significant rise in probability along the Cordillera de Talamanca, as this dataset generally shows a strong increase of SOC towards higher altitudes. However, the highest probabilities were still found along the northern Caribbean coast (RAMSAR site 811).

Lastly and as mentioned above, ground referencing is the ultimate tool needed to improve this map and others. We have completed a field expedition in the summer of 2023, where we confirmed the presence of peat in 7 areas across Costa Rica (Fig. 5)13, but this effort remains too small to effectively validate all peatland areas we mapped in this study. Much more exploration is needed to confirm the maps and improve our models. In return, maps such as the one presented in this paper are valuable to guide targeted field work. For example, high- vs. low-probability areas for peatlands can be delineated and used to inform future expeditions.

Climate space for Costa Rican peatlands

Precipitation- and moisture-based indices are becoming popular to help understand the distribution of ecosystems, including peatlands32,33. Here we combined our predictive map of Costa Rican peatlands along with high-resolution gridded climate data for 1970–200034 to uncover the climate space of Costa Rican peatlands (Fig. 6). Both mean annual precipitation (ranging from ~ 2000 to ~ 5000 mm) and precipitation seasonality (from 25 to 85%) span wide ranges. Precipitation seasonality is a measure of the variability of monthly precipitation amounts when compared to the annual mean; low values refer to small differences across the months, and vice versa. This index is calculated using the ratio of the standard deviation of the monthly total precipitation versus the mean monthly total precipitation (i.e., the coefficient of variation). Note that the four main peatland types/regions are also clearly delineated within the climate space (Fig. 6), with the Caribbean peatlands being the least seasonal as well as the wettest overall.

Fig. 6
figure 6

Climate envelope analysis of the Costa Rican peatland distribution, as estimated by this study. Climate data extracted from WorldClim version 2.134.

Full size image

In conclusion, this study presents the first published effort aimed at developing a country-based, probabilistic peatland map for Costa Rica. While this map should be used with caution, it can provide guidance for land management, policymaking, and future science endeavors.