Coastal Wetland Mapping Using Ensemble Learning Algorithms: A Comparative Study of Bagging, Boosting and Stacking Techniques

Coastal wetlands are a critical component of the coastal landscape that are increasingly threatened by sea level rise and other human disturbance. Periodically mapping wetland distribution is crucial to coastal ecosystem management. Ensemble algorithms (EL), such as random forest (RF) and gradient boosting machine (GBM) algorithms, are now commonly applied in the field of remote sensing. However, the performance and potential of other EL methods, such as extreme gradient boosting (XGBoost) and bagged trees, are rarely compared and tested for coastal wetland mapping. In this study, we applied the three most widely used EL techniques (i.e., bagging, boosting and stacking) to map wetland distribution in a highly modified coastal catchment, the Manning River Estuary, Australia. Our results demonstrated the advantages of using ensemble classifiers to accurately map wetland types in a coastal landscape. Enhanced bagging decision trees, i.e., classifiers with additional methods to increasing ensemble diversity such as RF and weighted subspace random forest, had comparably high predictive power. For the stacking method evaluated in this study, our results are inconclusive, and further comprehensive quantitative study is encouraged. Our findings also suggested that the ensemble methods were less effective at discriminating minority classes in comparison with more common classes. Finally, the variable importance results indicated that hydro-geomorphic factors, such as tidal depth and distance to water edge, were among the most influential variables across the top classifiers. However, vegetation indices derived from longer time series of remote sensing data that arrest the full features of land phenology are likely to improve wetland type separation in coastal areas.

Modeling Intersecting Processes of Wetland Shrinkage and Urban Expansion by a Time-Varying Methodology

Continuous urban expansion worldwide has resulted in significant wetland degradation and loss. A limited number of studies have addressed the coupling of wetland and urban dynamics, but this relationship remains unclear. In this study, a time-varying methodology of predicting wetland distribution was developed to support decision-making. The novelty of the methodology is its ability to dynamically simulate wetland shrinkage together with urban expansion and reveal conflicts and potential tradeoffs under different scenarios. The developed methodology consists of three modules: a historical change detection of wetland and urban areas module, a spatial urban sprawl simulation and forecasting module that can accommodate different development priorities, and a wetland distribution module with time-varying logistic regression. The methodology was applied and tested in the Tonghu Wetland as a case study. The wetland and urban extents presented a spatially intersecting shift, where wetlands lost more than 40% of their area from 1977 to 2017, while urban areas expanded by 10-fold, threatening wetlands. The increase in the relative importance metric of the time-varying regression model indicated an enhanced influence of urban expansion on the wetland. An accuracy assessment validated a robust statistical result and a good visual fit between spatially distributed wetland occurrence probabilities and the actual distribution of wetland. Incorporating the new variable of urban expansion improved modeling performance and, particularly, realized a greater ability to predict potential wetland loss than provided by the traditional method. Future wetland loss probabilities were visualized under different scenarios. The historical trend scenario predicted continuously expanding urban growth and wetland shrinkage to 2030. However, a specific urban development strategy scenario was designed interactively to control the potential wetland loss. Consideration of such scenarios can facilitate identifying tradeoffs to support wetland conservation.

Monthly gridded data product of northern wetland methane emissions based on upscaling eddy covariance observations

Natural wetlands constitute the largest and most uncertain source of methane (CH4) to the atmosphere and a large fraction of them are found in the northern latitudes. These emissions are typically estimated using process (“bottom-up”) or inversion (“top-down”) models. However, estimates from these two types of models are not independent of each other since the top-down estimates usually rely on the a priori estimation of these emissions obtained with process models. Hence, independent spatially explicit validation data are needed. Here we utilize a random forest (RF) machine-learning technique to upscale CH4 eddy covariance flux measurements from 25 sites to estimate CH4 wetland emissions from the northern latitudes (north of 45∘ N). Eddy covariance data from 2005 to 2016 are used for model development. The model is then used to predict emissions during 2013 and 2014. The predictive performance of the RF model is evaluated using a leave-one-site-out cross-validation scheme. The performance (Nash–Sutcliffe model efficiency =0.47) is comparable to previous studies upscaling net ecosystem exchange of carbon dioxide and studies comparing process model output against site-level CH4 emission data. The global distribution of wetlands is one major source of uncertainty for upscaling CH4. Thus, three wetland distribution maps are utilized in the upscaling. Depending on the wetland distribution map, the annual emissions for the northern wetlands yield 32 (22.3–41.2, 95 % confidence interval calculated from a RF model ensemble), 31 (21.4–39.9) or 38 (25.9–49.5) Tg(CH4) yr−1. To further evaluate the uncertainties of the upscaled CH4 flux data products we also compared them against output from two process models (LPX-Bern and WetCHARTs), and methodological issues related to CH4 flux upscaling are discussed. The monthly upscaled CH4 flux data products are available at https://doi.org/10.5281/zenodo.2560163 (Peltola et al., 2019).

Geomorphological analysis of wetland distribution on various spatial scales

<p><strong>Abstract.</strong> This study introduces some case analyses of wetland distribution on various spatial scales, from nationwide to the area of a wetland group, with a focus on geomorphological feature. Then described the usefulness of GIS analysis in wetland research. The nationwide wetland distribution in Japan showed that wetland density was high at less than 200&amp;thinsp;m and around 1600&amp;ndash;2000&amp;thinsp;m. Wetlands in mountainous regions were concentrated in snowy Quaternary volcanic regions from the center to the northern part of Japan. This implied snow accumulation and topography of volcanic mountains are important for wetland formation. Secondly, we clarified that wetlands were mainly distributed on the gentle slope of original volcanic surfaces and in landslides in the Hachimantai volcanic groups, in the northern Japan, using 10-m grid DEM and aerial photo interpretation. With the higher-resolution data, it was clear that wetlands were arranged depending on the microtopography of landslides and volcanic surfaces and groundwater. Using data with resolution suitable for the target topographical size and combining the results of multiple spatial scales/resolutions, we can understand the origin of wetlands in more detail.</p>

Predicting Wetland Distribution Changes under Climate Change and Human Activities in a Mid- and High-Latitude Region

Wetlands in the mid- and high-latitudes are particularly vulnerable to environmental changes and have declined dramatically in recent decades. Climate change and human activities are arguably the most important factors driving wetland distribution changes which will have important implications for wetland ecological functions and services. We analyzed the importance of driving variables for wetland distribution and investigated the relative importance of climatic factors and human activity factors in driving historical wetland distribution changes. We predicted wetland distribution changes under climate change and human activities over the 21st century using the Random Forest model in a mid- and high-latitude region of Northeast China. Climate change scenarios included three Representative Concentration Pathways (RCPs) based on five general circulation models (GCMs) downloaded from the Coupled Model Intercomparison Project, Phase 5 (CMIP5). The three scenarios (RCP 2.6, RCP 4.5, and RCP 8.5) predicted radiative forcing to peak at 2.6, 4.5, and 8.5 W/m2 by the 2100s, respectively. Our results showed that the variables with high importance scores were agricultural population proportion, warmness index, distance to water body, coldness index, and annual mean precipitation; climatic variables were given higher importance scores than human activity variables on average. Average predicted wetland area among three emission scenarios were 340,000 ha, 123,000 ha, and 113,000 ha for the 2040s, 2070s, and 2100s, respectively. Average change percent in predicted wetland area among three periods was greatest under the RCP 8.5 emission scenario followed by RCP 4.5 and RCP 2.6 emission scenarios, which were 78%, 64%, and 55%, respectively. Losses in predicted wetland distribution were generally around agricultural lands and expanded continually from the north to the whole region over time, while the gains were mostly associated with grasslands and water in the most southern region. In conclusion, climatic factors had larger effects than human activity factors on historical wetland distribution changes and wetland distributions were predicted to decline remarkably over time under climate change scenarios. Our findings have important implications for wetland resource management and restoration because predictions of future wetland changes are needed for wetlands management planning.