Conference Agenda

All over the world, and more precisely in China, freshwater is an increasingly scarce resource suffering from rapidly changing environments due to human activities Study location Poyang and Dongting Lake are China’s first and second largest freshwater lake in the middle reaches of Yangtze River catchment. Its lakes and associated wetlands deliver important ecosystem functions such as freshwater supply, water purification, flood and climate regulation, and biodiversity’s sanctuaries. The development of comprehensive water resource management and nature conservation strategies requires detailed mapping and monitoring of inland waters. The generation of such information requires either large human resources for conventional ground surveying or expensive data. In addition to costly methods, the monitoring of large wetlands such as the Poyang and Dongting lakes study site with a water surface of up to 3.500 km² is difficult due to its inaccessibility during annual flood period.

Remote sensing offers a mature and comprehensive tool to solve this task with large area coverage at very low costs. Until today, in addition to satellite data with lower temporal resolution such as Envisat ASAR, for SAR sensors, or HJ1A and Landsat for the optical High resolution satellite, as well as daily middle resolution MODIS satellite data and the MERIS sensors onboard of Envisat, were frequently selected source for capturing lake dynamics. Satellite data from the new European Sentinel and Sentinel2 fleet has been available since 2014 and provides high-resolution information. In this paper we present the application of Sentinel-1and Sentinel 2 time series data for spatio-temporal high-resolution wetland and lakes mapping. New is the level of detail that can be achieved with Sentinel data. Potential and limitations are analyzed. Water surface obtained from the Sentinel were also compared with water bodies database such the GRW of Pekel as well as the ones generated by the UCLA lakes group.

Wetlands have two distinctive features, which are spatiotemporal dynamics and spatial heterogeneity. The dynamic characteristic of the water makes its cover fraction show seasonal changes. At the same time, the spectrum of wetland vegetation has a high similarity with agricultural land vegetation, causing accurately distinguish wetlands difficultly by remote sensing technology.

The amount of cloud in the mid-latitude region is relatively abundant, and a single sensor cannot provide the time density required for accurate identification of wetlands. Multi-sensor technology has been applied well in monitoring urban landscapes and invasive species mapping. Compared with previous remote sensing data, Sentinel-1/2 and GF-1 data have an advantage of high temporal and spatial resolution, but their application in wetland classification is worthy of the further exploration.

The landscape types of wetland ecosystems not only have large heterogeneity in space, but also have strong dynamic characteristics with changes in water bodies. This brings many challenges to accurately map and monitor wetlands. The normalized difference vegetation index (NDVI) is effective in monitoring multi-temporal vegetation phenological changes. There are many studies using time series remote sensing data which has a significant advantage of high observed density to monitor highly dynamic systems.

Dongting Lake is located in the northern part of Hubei Province, the middle reaches of the Yangtze River Basin. It is the second largest freshwater lake in China. As one of the two remaining natural rivers in the middle and lower reaches of the Yangtze River, Dongting Lake plays an extremely important role in regulating runoff and protecting biodiversity.

In this paper, the Dongting Lake wetlands, which contains three important international wetlands is considered as study area. The data source is the Sentinel-1/2 and GF-1 time series data acquired in 2017, the Sentinel-2 and GF-1 data construct the NDVI time series, and the Sentinel-1 data constructs the backscatter coefficient time series. we fuse the Sentinel-2 and GF-1 data to improve temporal and spatial resolution. In order to avoid the errors caused by a single classifier, the study area is classified using support vector machine and random forest classifier. Results showed the following: (1) The accuracy of the three types of data using the RF classifier is not much different from that of the SVM classifier. Prove that in the aspect of time series wetland classification, the impact of the data source is more significant than which classifier to choose. (2) The overall classification accuracy of Sentinel-2, GF-1 and Sentinel-1 time series are 93.93%, 84.30% and 48.04%, the accuracy of seasonal marshes that symbolizes the dynamic characteristics of wetlands is 92.23%, 79.43% and 48.32%. The classification method based on optical remote sensing time series data can fill the needs of wetland dynamic mapping, while the wetland mapping making use of SAR time series data is not available. (3) The classification accuracy of Sentinel-2+GF-1 fusion time series is 94.09%, and the commission rate of each class is significantly lower than that of the three sensors. The time series mapping based on multi-source data fusion has a wonderful robustness in the aspect of wetland classification.

In inland and coastal waters, organic carbon undergoes seasonal and spatial variations in relation to river run-off and biological processes. Particulate organic carbon, while being a smaller fraction of the total organic carbon with respect to dissolved organic carbon, plays an important role in the local and regional carbon dynamics. In fact, it is the variation of POC that can be used to examine carbon sequestration and well as carbon sink. Particulate matter strong influences the optical, chemical and biological conditions of most inland and coastal aquatic ecosystems.

There are numerous challenges to identifying particulate carbon, and most importantly specifying particulate carbon classes. These latter are dominated largely by productive particulate carbon from phytoplankton and detrital particulate organic carbon. These two pools play very different roles in the aquatic carbon dynamics, with high temporal and spatial variability within a single waterbody in relation to local sources and seasonal changes. However, many studies examine particulate carbon dynamics with a single algorithm, which leads to poor estimation where multiple numerous sources and sinks are present, typical for inland and coastal water environments.

The BioGeoLakes team has been working on an absorption-based approach was used to determine surface particulate organic carbon based on the specification of local POC absorption characteristics of dominant POC sources; phytoplankton or detritus based. This specification was made using a new particulate organic matter index, which was tested across a range of modelled and real lake/coastal conditions. Based on remote sensing reflectance in four wavebands, the model provided a good separation of organic particulate types and a good estimate of organic particulate concentrations in shallow lakes in the Yangtze River valley and estuary. These study lakes include Taihu, Chaohu and Poyang while work in the coming year will include a large number of smaller lakes in the Valley using the OLCI/Sentinel-3 satellite data. The approach shows a good potential to quantify particulate carbon dynamics in ecosystems where multiple organic carbon sources are present

As the second largest freshwater lake of Yunnan Province,Erhai Lake is an important drinking water source in Dali. In recent decades. With the rapid development of economy, the impact of human activities on Erhai is becoming increasingly prominent. Water quality deterioration and eutrophication led to the occurrence of cyanobacterial blooms and affected the normal ecological function of lakes.

Based on multi-source remote sensing data, this paper studies the spatiotemporal distribution of cyanobacteria in Erhai lake from 1999 to 2016 and analyzes the effects of both meteorological and human activites on the cyanobacteria.

By combining the results of random forest classification with the results of DMSP(Defense Meteorological Satellite Program-the Operational Linescan System) nightlight remote sensing images, the method decision tree was used to extract the new building land in Erhai Basin. Remote sensing monitoring of cyanobacteria in Erhai sea, extracted by Landsat TM, ETM + and Sentinel 2A, shows that from 1999 to 2016, there was an increasing trend in the scale and duration of water blooms.

Based on the analysis of the correlation between the scale, duration, frequency and first occurrence time of water and the new building rate, it is found that since 2003, the algal bloom frequency of Erhai has been positively correlated with the growth rate of the building, especially in the double porch and Taoyuan Town.

For short-term effect factor researches, we obtained the data of cyanobacterial blooms in Erhai Lake in 2013 and analyzed the effects of temperature, sunshine, precipitation, air pressure and wind, we summarized the meteorological characteristics when blooms breaking-out, that is the blooms often occurred when the sun shines again after rain, the strong sunshine and higher daily temperature variation were the main factor of blooms, low wind speed and lower air pressure were in favour of the formation of blooms.

Biodiversity stakes within Yangtze watershed are very important at national level but also international ones. These very rich ecosystems, being key wintering areas for many waterfowl of SE Asia, are suffering from rapidly changing environments due to human activities. It’s crucial to understand what the key factors and their effects are. As data within large spatial and temporal scales are difficult to get, remote sensing and spatial analysis technology turns out to be a useful tool to access information. Coupling environmental information obtained from remote sensing and biological and ecological data could help understand what factors are driving observed phenomena. Studies are on progress within the Yangtze River basin to understand the link between the lakes water surfaces and the date of arrival and departure of some bird species during migration.

Being the eldest of the Copernicus program, the Sentinel -1 mission has been imaging the Earth since April 2014 using a C-band Synthetic Aperture Radar. The high temporal and spatial resolutions (12 days at the equator, 10x10m pixel spacing in full resolution GRD) allow to create dense and accurate time series. Standing water has a distinguishable signature using radar wavelengths, which are very rarely disturbed by cloud coverage. This makes the use of radar imagery and thus Sentinel-1 more than relevant for the study of water body dynamics. However data quality comes with the price of high data volume. As these data flows are exponentially increasing, the use of “cloud computing” becomes more and more appealing. It allows to process information without downloading teraoctets of data. Google EarthEngine is an interface proposing such solution. The extensive catalog of available data allows advanced processing over large amount of data without being limited by local computational capacities.

The general aim is to use the backscatter GRD Sentinel-1 images to extract water surfaces of lakes of interests in the Yangtze River basin for years 2015, 2016, 2017 and 2018, and then put them in relation with ecological data. The first intermediate objective was to export monthly aggregated water surfaces for above years and for the months of January, February, July, November and December. A total of 513 images, distributed among 227 dates were processed. The water class was created using a simple k-mean algorithm and corrected with a DEM, considering standing water presence possible only on slopes inferior to 5°. In a second time, each Sentinel-1 GRD image is processed independently and the water extents are extracted. This captures faster dynamics, but classifications are more difficult to conduct due to noisier signal compared to a multi date composition. The final time series is composed of 340 images in 114 dates, from October to December for the years 2016, 2017 and 2018. The use of a k-mean algorithm makes the pre-processing step almost inexistent and keeps the processing time within minutes; the most time consuming step is the download of the final binary rasters. These time series of water extraction, put in relation with bird species presence, will allow to study if water levels in studied lakes have an impact on the arrival and departure dates of the flocks.

Radar signal is by itself a very powerful tool for water classification and the Sentinel-1 mission makes access to such products very easy. Even though confusion exist between water and some smooth surfaces such as roads, airports or desert areas during classification, the amount of data available is enormous and the use of cloud infrastructures like Google EarthEngine makes computing significantly faster and easier, allowing to study large areas for a long period of time. This allows to generate the water dynamics for more than 80 lakes of the Yangtze watershed, from the very large Poyang Lake to smaller ones such as Wuchang, Shengjin or Baitu Lakes. This study provides the first analysis of the water dynamics of these keys lakes in terms of biodiversity with such a time frequency.