There are several techniques for mapping wetlands. In this study, we examined four statistical models to assess the potential distribution of wetlands in the South-Kivu province by combining optical and SAR images. The approach involved integrating topographic, hydrological, and vegetation indices into the four most used classifiers, namely Artificial Neural Network (ANN), Random Forest (RF), Boosted Regression Tree (BRT), and Maximum Entropy (MaxEnt). A wetland distribution map was generated and classified into ‘wetland’ and ‘non-wetland.’ The results showed variations in predictions among the different models. RF exhibited the most accurate predictions, achieving an overall classification accuracy of 95.67% and AUC and TSS values of 82.4%. Integrating SAR data improved accuracy and precision, particularly for mapping small inland wetlands. Our estimations indicate that wetlands cover approximately 13.5% (898,690 ha) of the entire province. BRT estimated wetland areas to be ~ 16% (1,106,080 ha), while ANN estimated ~ 14% (967,820 ha), MaxEnt ~ 15% (1,036,950 ha), and RF approximately ~ 10% (691,300 ha). The distribution of these areas varied across different territories, with higher values observed in Mwenga, Shabunda, and Fizi. Many of these areas are permanently flooded, while others experience seasonal inundation. Through digitization, the delineation process revealed variations in wetland areas, ranging from tens to thousands of hectares. The geographical distribution of wetlands generated in this study will serve as an essential reference for future investigations and pave the way for further research on characterizing and categorizing these areas.
Wetlands are highly abundant habitats within the Congo basin, particularly in the Democratic Republic of the Congo (DRC) and its eastern provinces. These wetlands play a crucial role in the South-Kivu province, providing goods and services to the local community while supporting biodiversity. With their diverse range of uses and significant agricultural potential, it is imperative to pay special attention to the conservation and management of these ecosystems. Although the importance of wetland ecosystem services is widely recognized, more detailed inventories are needed to ensure the effective implementation of conservation strategies. Many wetlands in the region still need to be identified and are not represented on publicly available maps. This knowledge gap poses challenges to their conservation and sustainable use. More definition are available for wetland, the one proposed by Amler et al. was adopted: wetland in the East African landscape refers to “a diverse and dynamic ecosystem characterized by the presence of water, both permanent and seasonal, along with distinct ecological features, including habitats such as marshes, swamps, floodplains, inland valleys, coastal, mangroves, and shallow lakes, with varying water depths (most lower than 1 m) and vegetation types”.
The vast geographic expanse and complex distribution of wetlands in eastern DRC present significant challenges for conducting comprehensive inventories. However, recent advancements in technology, such as the availability of high-resolution georeferenced field data archives and open access to high-spatial-resolution remote sensing data, coupled with the application of artificial intelligence (AI) techniques, have opened up new possibilities for accurate and detailed wetland mapping. Despite the importance of wetlands, there still needs to be more knowledge regarding their distribution and status. Closing this knowledge gap requires assessing the potential distribution and characterizing wetlands at national and provincial levels. Effective management and monitoring methods are essential for conserving and protecting wetlands, as these ecosystems face multiple pressures from human activities, invasive species, and climate change. The loss or degradation of wetlands significantly impacts their ability to sustain biodiversity, maintain water quality, mitigate floods, and sequester carbon. Accurate mapping of wetlands with high spatial and thematic precision plays a crucial role in their effective management and monitoring. These maps help identify potential risks and pressures on wetlands and assess the effectiveness of wetland conservation programs.
The initial studies on wetland mapping in the Democratic Republic of the Congo (DRC) date back to 2010, with researchers such as Bwangoy et al.; and Lee et al. exploring various aspects of wetland classification and monitoring. To gather data, these studies used optical sensors, specifically Landsat 5 MSS and 7 TM. However, the integration of Synthetic Aperture Radar (SAR) imagery was also incorporated due to its ability to penetrate through vegetation canopies and its sensitivity to moisture conditions. For SAR data, the PALSAR radar and SRTM datasets were utilized. Bwangoy et al. demonstrated that the integration of optical and SAR data resulted in high accuracy levels, surpassing existing maps such as Africover (77%) and The JRC/GRFM Regional Flooded Forest Map of Central Africa (73.0%), with a Kappa coefficient exceeding 0.70. Utilizing this approach, Bwangoy et al. estimated the coverage of wetlands in the DRC to be approximately 440,000 km, accounting for 19.2% of the country’s total area.
The eastern region of the Democratic Republic of the Congo (DRC) and eastern Africa as a whole showcase a diverse range of landscapes. However, small inland wetlands, characterized by their size (< 500 ha), often go unnoticed and receive limited attention in conservation and restoration efforts, mainly due to the challenges associated with identifying them within large regions. Many of these wetlands exhibit seasonal variations in water levels and vegetation, making remote sensing a valuable tool for their detection. Despite their potential for agricultural production and various uses in South-Kivu province and eastern DRC, many of these wetlands remain undocumented on official maps. The lack of official recognition has resulted in their unsustainable exploitation.
While other regions in Africa employed classic mapping methods like supervised or unsupervised classification, including Maximum Likelihood, ISODATA, PCA, or K-means, the DRC utilized a "decision tree" model for wetland mapping and identifying emerging wetland forests. However, the limitations of the satellite images used, characterized by low spatial and spectral resolutions, hindered the production of maps suitable for provincial or territorial decision-making. Consequently, these studies have yet to achieve results at such scales. Nevertheless, these studies served as a foundation for subsequent mapping efforts at the national level.
The combination of optical spectral and SAR indices has proven to be suitable for wetland mapping, as indicated by studies conducted by Kulawardhana et al., García and Lleellish; Farda et al.; Alves et al.; Sun et al.; López-Tapia et al.; Islam et al.; Saha et al.; and Pham et al. to mention just a few. For decades, Synthetic Aperture Radar (SAR) are currently integrated for flood process, wetlands mapping, and vegetation monitoring. SAR data is particularly beneficial for wetland mapping because it can penetrate vegetation canopies (depending on the wavelength) to identify inundation and is sensitive to moisture conditions. While initial wetland mapping research predominantly relied on optical satellite images, SAR sensors offer the advantage of acquiring data even in the presence of clouds, haze, and other atmospheric disturbances, as they emit their own incident radiation. However, weather conditions like wind, rain, and cold temperatures can impact SAR data quality. Additionally, the integration of multi-sensor images allows for the consideration of both water and vegetation factors, which can influence wetland-mapping accuracy.
Among the various methods used for mapping and delineating wetlands, those incorporating topographic features, hydrological processes, and vegetation aspects tend to offer high accuracy. These approaches consider all three essential factors in wetland definition. Similar methodologies have been tested in different regions, including Canada, Nigeria, South Africa, and other areas, with varying levels of accuracy depending on the image and model types used. Classifiers such as Decision Tree (DT), Support Vector Machine (SVM), Artificial Neural Network (ANN), Logistic Regression Model (LRM), and Maximum Entropy (Maxent) have been utilized, with Random Forest (RF) and SVM showing promising results in terms of accuracy. These models have also been tested for wetland distribution and other applications, proving effective. Overall, both DT, SVM, ANN, RF, BRT, KNN classifiers are the most famous worldwide classification algorithms used for wetland mapping.
Despite abundant data and tools, there still needs to be more knowledge regarding which models can accurately assess the practical distribution map of small inland wetlands and delineate them. Debates persist regarding the choice of models and the types of images or indices to employ for this purpose. However, advancements in technology have significantly contributed to the field of Geographic Information Systems (GIS) and Remote Sensing (RS), enabling the modeling and prediction of wetland ecosystems at both small and macro scales, as well as the assessment of distribution factors. Modern geo-statistics and techniques integrated into GIS tools facilitate efficient modeling of wetland distribution, while RS provides valuable imagery for detecting, digitizing, and estimating wetland distribution. These advancements have enhanced our understanding and capability to assess wetland distribution accurately, contributing to effective wetland management and conservation efforts.
To address the need for identifying and delineating these low-lying wetlands in the eastern region of the DRC, we propose an approach that combines optical images from the Sentinel satellites with Synthetic Aperture Radar (SAR) images. The methodology draws upon previous research by Mwita et al., and Garba et al., with modifications such as replacing Landsat images with Sentinel (1 and 2) and ALOSPALSAR. Additionally, we evaluate the performance of four widely used statistical classifiers in wetland mapping.
These statistical models have been extensively studied and proven to enhance the accuracy of wetland distribution predictions. Over the past decade, these techniques have garnered significant attention in ecosystem modeling and forecasting thanks to their ability to improve predictive capabilities. One key advantage of these mathematical models is their utilization of different types of independent and dependent variables, including categorical and quantitative variables. This versatility extends their applicability beyond wetland mapping and serves as a valuable reference for researchers in various fields of science, such as biology, sociology, and agronomy.
The overall objective of this study is to contribute to the identification and study of wetlands in the Democratic Republic of the Congo (DRC) by developing improved methods for mapping small inland wetlands. Specifically, this study aims to achieve the following objectives: (1) Identify the critical explanatory variables derived from remote sensing data, including Sentinel-1 and Sentinel-2, as well as ALSOPALSAR, and field data, that are relevant for modeling the distribution of small wetlands in eastern DRC; (2) Evaluate the capabilities of single-date Sentinel optical data and their combination with Synthetic Aperture Radar (SAR) data for mapping small wetlands in the South-Kivu province.
This evaluation will include assessing the accuracy of these mapping methods and identifying potential errors; (3) Digitize and characterize the identified wetlands, including analyzing their morphological characteristics such as area and perimeter; (4) Discuss the strengths and limitations of the mapping methods employed in this study, providing an overview of the advantages and challenges associated with integrating optical, topographic, and SAR indices, as well as using novel classifiers, to achieve accurate mapping results.
To address these objectives, it is hypothesized that integrating optical, topographic, and SAR indices with new classifiers will result in a more accurate method for mapping small wetlands. Additionally, it is suggested that fully polarimetric SAR imagery can provide valuable information about surface scattering mechanisms, allowing for a more precise distinction of small wetlands. Furthermore, including SAR data and novel vegetation indices is expected to improve the mapping process. Finally, it is anticipated that the delineation of wetlands after digitization will reveal various areas with distinct morphological characteristics. By addressing these research hypotheses, this study will enhance our understanding of wetland mapping methodologies in the context of the DRC, with a specific focus on small wetlands.
