Identifying driving hydrogeomorphic factors of coastal wetland downgrading using random forest classification models.

Coastal wetlands provide critical ecosystem services but are experiencing disruptions caused by inundation and saltwater intrusion under intensified climate change, sea-level rise, and anthropogenic activities. Recent studies have shown that these disturbances downgraded coastal wetlands mainly through affecting their hydrological processes. However, research on what is the most critical driver for wetland downgrading and how it affects coastal wetlands is still in its infancy. This study examined drivers of three types of wetland downgrading, including woody wetland loss, emergent herbaceous wetland loss, and woody wetlands converting to emergent herbaceous wetlands. By using random forest classification models for the wetland ecosystems in the Alligator River National Wildlife Refuge, North Carolina, USA, during 1995-2019, we determined the relative importance of different hydrogeomorphic processes and the dominant variables in driving the wetland downgrading. Results showed that random forest classification models were accurate (> 97 % overall accuracy) in classifying wetland downgrading. Multiple hydrogeomorphic variables collectively contributed to the coastal wetland downgrading. However, the dominant control factors varied across different types of wetland downgrading. Woody wetlands were most susceptible to saltwater intrusion and were likely to downgrade if the saltwater table was shallower than 0.2 m below the land surface. In contrast, emergent herbaceous wetlands were most vulnerable to inundation and drought. The favorable groundwater table for emergent herbaceous wetlands was between 0.34 m above the land surface and 0.32 m below the land surface, beyond which the emergent herbaceous wetland tended to disappear. For downgraded woody wetlands, their distance to canals/ditches played a crucial role in determining their fates after downgrading. The machine learning approach employed in this study provided critical knowledge about the thresholds of hydrogeomorphic variables for the downgrading of different types of coastal wetlands. Such information can help guide effective and targeted coastal wetland conservation, management, and restoration measures.

Assessing responses of hydrological processes to climate change over the southeastern Tibetan Plateau based on resampling of future climate scenarios.

With global warming, hydrological regimes in the headwater basins of the Tibetan Plateau (TP) have significantly changed. Investigating the responses of hydrological processes to climate change in TP has become more and more important to make robust strategies for water resources management. However, using just a few GCMs may constrain the uncertainty in assessment of climate impacts. Therefore, a framework is proposed in this study to generate ensemble climate change scenarios and then investigate changes of hydrological processes under climate change in the upper reaches of Yarlung Zangbo River basin (UYZR) and Lancang River basin (ULR). Firstly, the Latin Hypercube Simulation (LHS) is used to generate an ensemble of future climate change scenarios by resampling change factors of meteorological variables from 18 GCMs under emission scenarios RCP2.6 and RCP8.5. The inherent dependence structures of change factors, i.e. the correlations of change factors among 12 months for different meteorological variables, are also considered in ensembles. Secondly, the HBV hydrological model coupled with a degree-day snowmelt model is applied to explore the potential change of runoff in the future period 2041-2070. Results show that: 1) the resampling method is effective and can provide a wide ensemble of climate change scenarios. 2) Precipitation, temperature and potential evapotranspiration in the UYZR and ULR basins are expected to increase under the two scenarios, particularly under RCP8.5. 3) The total runoff also shows a moderately upward trend in two basins, both mainly due to increased precipitation. In the UYZR basin, fast runoff accounts for a larger proportion in total runoff than slow runoff, while in ULR, both almost play the same role in total runoff. Furthermore, snowmelt-induced runoff in both basins would be less and rainfall-induced runoff will probably become more important in the future.