Vectorized dataset of silted land formed by check dams on the Chinese Loess Plateau.

Check dams on the Chinese Loess Plateau (CLP) have captured billions of tons of eroded sediment, substantially reducing sediment load in the Yellow River. However, uncertainties persist regarding the precise sediment capture and the role of these dams in Yellow River flow and sediment dynamics due to the lack of available spatial distribution datasets. We produced the first vectorized dataset of silted land formed by check dams on the CLP, combining high-resolution and easily accessible Google Earth images with object-based classification methods. The accuracy of the dataset was verified by 1947 collected test samples, and the producer's accuracy and user's accuracy of the dam lands were 88.9% and 99.5%, respectively. Our dataset not only provides fundamental information for accurately assessing the ecosystem service functions of check dams, but also helps to interpret current changes in sediment delivery of the Yellow River and plan future soil and water conservation projects.

BRefine: Achieving High-Quality Instance Segmentation.

Instance segmentation has been developing rapidly in recent years. Mask R-CNN, a two-stage instance segmentation approach, has demonstrated exceptional performance. However, the masks are still very coarse. The downsampling operation of the backbone network and the ROIAlign layer loses much detailed information, especially from large targets. The sawtooth effect of the edge mask is caused by the lower resolution. A lesser percentage of boundary pixels leads to not-fine segmentation. In this paper, we propose a new method called Boundary Refine (BRefine) that achieves high-quality segmentation. This approach uses FCN as the foundation segmentation architecture, and forms a multistage fusion mask head with multistage fusion detail features to improve mask resolution. However, the FCN architecture causes inconsistencies in multiscale segmentation. BRank and sort loss (BR and S loss) is proposed to solve the problems of segmentation inconsistency and the difficulty of boundary segmentation. It is combined with rank and sort loss, and boundary region loss. BRefine can handle hard-to-partition boundaries and output high-quality masks. On the COCO, LVIS, and Cityscapes datasets, BRefine outperformed Mask R-CNN by 3.0, 4.2, and 3.5 AP, respectively. Furthermore, on the COCO dataset, the large objects improved by 5.0 AP.

Substantial decrease in CO2 emissions from Chinese inland waters due to global change.

Carbon dioxide (CO2) evasion from inland waters is an important component of the global carbon cycle. However, it remains unknown how global change affects CO2 emissions over longer time scales. Here, we present seasonal and annual fluxes of CO2 emissions from streams, rivers, lakes, and reservoirs throughout China and quantify their changes over the past three decades. We found that the CO2 emissions declined from 138 ± 31 Tg C yr-1 in the 1980s to 98 ± 19 Tg C yr-1 in the 2010s. Our results suggest that this unexpected decrease was driven by a combination of environmental alterations, including massive conversion of free-flowing rivers to reservoirs and widespread implementation of reforestation programs. Meanwhile, we found increasing CO2 emissions from the Tibetan Plateau inland waters, likely attributable to increased terrestrial deliveries of organic carbon and expanded surface area due to climate change. We suggest that the CO2 emissions from Chinese inland waters have greatly offset the terrestrial carbon sink and are therefore a key component of China's carbon budget.

Inferring floodplain bathymetry using inundation frequency.

This study proposes a new method to retrieve the bathymetry of turbid-water floodplains from the inundation frequency (IF) data derived from over 32 years of composite optical remote sensing data. The new method was tested and validated over the Curuai floodplain in the lower Amazon River, where the entire bathymetry was surveyed in 2004, and water level gauge data has been available since 1960. The depth was estimated based on the relationship derived from IF and surveyed depth data, and the results were compared to those retrieved from bare-Earth DEM. We further assessed the sensitivity of the approach by analyzing the deepest part of the lake (i.e., permanent water body ~ 8m) with high IF, as well as the effect of gradual sedimentation in the lake over time. The results showed that the model is highly accurate and sensitive to IF changes even in the permanent water body areas, suggesting that this model can be used in other seasonal lakes worldwide with turbid-waters, where large-scale bathymetry surveys are not feasible due to high operation costs.

Dramatic decrease of flood frequency in the Mekong Delta due to river-bed mining and dyke construction.

Here we present a proof of concept evaluation of the impacts of riverbed-mining on river-wetland connectivity by analyzing the temporal trends of the flood frequencies in the Vietnamese Mekong Delta (VMD), while accounting for the effect of dyke constructions. We focus on the Long Xuyen Quadrangle (LXQ), which is significant in terms of biodiversity and economic contribution to the VMD as it is one of the most important food baskets of Southeast Asia that depends on seasonal flooding. Our results indicate that the flood frequency in LXQ has decreased significantly over the past 20 years (1995-2015). Time-series analyses of water level data at Chau Doc, Tan Chau, and Can Tho stations confirmed that the overall descending trend is statistically significant (p-value < 0.001 and tau ~ 0.1). However, the river discharge at Kratie showed no significant trend (p-value = 0.98) over the same period. This indicates that the flood frequency is associated with the lowering of the riverbed (incision) other than climatic factors. The connectivity analysis also revealed a remarkable drop in the inundation duration after early 2000, which corresponds to the previous observations of the shifting shoreline of the VMD from construction to shrinking. Finally, regression and principal component analyses underpinned the strong causality between the riverbed-mining and the decreased seasonal flooding patterns in LXQ, while accounting for the effect of the dyke system over the last decades (R2 = 0.75). This study offers compelling evidences on the relationship between sand-mining in the river and the disrupted flood regimes in VMD. The reduction in water and sediments that is necessary for sustaining current rates of agricultural production in the long term would endanger the livelihoods of millions of VMD inhabitants.

Effective soil erosion control represents a significant net carbon sequestration.

The debate over whether soil erosion is a carbon (C) sink or atmospheric CO2 source remains highly controversial. For the first time, we report the magnitude of C stabilization associated with soil erosion control for an entire large river basin. The soil erosion of the Yellow River basin in northern China is among the most severe worldwide. Progressive soil conservation has been implemented by the Chinese government since the 1970s, including the largest ever revegetation programme, the Grain-for-Green Project, which began in 1999. Based on compiled hydrological records and organic carbon (OC) data, together with primary production estimates, we evaluated the sequestered OC resulting from soil conservation. Compared with that at baseline in 1950-1970, in which significant soil conservation did not occur, the fate of erosion-induced OC was substantially altered in the period from 2000-2015. Approximately 20.6 Tg of OC were effectively controlled per year by soil conservation efforts. Simultaneously, the decomposition of erosion-induced soil organic carbon (SOC) declined from 8 Tg C yr-1 to current 5.3 Tg C yr-1. The reduced C emissions (2.7 Tg C yr-1) within the Yellow River basin alone account for 12.7% of the mean C accumulation acquired via forest expansion throughout all of China previously assessed. If the accumulated C in restored plants and soils was included, then 9.7 Tg C yr-1 was reduced from the atmospheric C pool during this period, which represents a tremendous C-capturing benefit. Thus, the increased C storage obtained via soil conservation should be considered in future C inventories.

Drastic change in China's lakes and reservoirs over the past decades.

Using remote sensing images, we provided the first complete picture of freshwater bodies in mainland China. We mapped 89,700 reservoirs, covering about 26,870 km(2) and approximately 185,000 lakes with a surface area of about 82,232 km(2). Despite relatively small surface area, the total estimated storage capacity of reservoirs (794 km(3)) is triple that of lakes (268 km(3)). Further analysis indicates that reservoir construction has made the river systems strongly regulated: only 6% of the assessed river basins are free-flowing; 20% of assessed river basins have enough cumulative reservoir capacity to store more than the entire annual river flow. Despite the existence of 2,721 lakes greater than 1 km(2), we found that about 50 lakes greater than km(2) have formed on the Tibetan Plateau resulting from climate change. More than 350 lakes of ≥1 km(2) vanished in four other major lake regions. Although the disappearance of lakes happened in the context of global climate change, it principally reflects the severe anthropogenic impacts on natural lakes, such as, the excessive plundering of water resources on the Inner Mongolia-Xinjiang Plateau and serious destruction (land reclamation and urbanization) on the eastern plains.