1.The New Generation Vegetation Map of China (1:500,000)

Vegetation maps are important sources of information for biodiversity conservation, ecological studies, vegetation management and restoration, and national strategic decision making. The current Vegetation Map of China (1:1000000) was generated by a team of more than 250 scientists with an effort that lasted over 20 years starting from the 1980s, which won The National Natural Science Award of China (Second Prize) in 2011. However, 3.3 million km2 of vegetated areas of China changed their vegetation type group during the rapid development of China in the past three. The current vegetation map needs to be updated urgently to better represent the distribution of current vegetation distribution. Under this background, our lab takes the project of “Big Earth Data Science Engineering Project (CASEarth)”, which aims to map the new generation vegetation map of China at a finer spatial scale, i.e., 1:500,000. In this project, we will develop deep learning-based strategic to build a robust model for vegetation classification by fusing climate surface data, terrain data, time-series satellite data (e.g., sentinel 1, sentinel 2, Landsat, and MODIS), and filed data. A mobile phone APP (LiVegetation) and an online mapping system have been developed to help to increase the efficiency of collecting crowdsourcing field data. Moreover, we will also build a permanent platform that shares both data and tools for researchers in the field of plant science, ecology, and remote sensing.

2.Software and hardware developments of LiDAR and its application in agroforest and urban ecosystems

To reduce the errors from traditional algorithms, such as “local maximum filter” and “watershed”based on the Canopy Height Model(CHM), our group proposed three kinds of segmentation algorithm directly from point cloud data. We utilized spatial structure and intensity information of point cloud data to develop “top-down” and “bottom-up” segmentation algorithm for the airborne LiDAR data, and  applied graph theory and ecological metabolic theory for the terrestrial LiDAR data segmentation, and extraction accuracy are more than 90%.  Through these algorithms, we can extract the single treeand its structural parameters in forest and urban ecosystem.

We designed and developed a near-surface remote sensing platform using UAV as the carrier. It’s equiped with LiDAR, CCD camera and hyperspectral sensor. This platform can work stably and flexibly with high precision, which allowed us to extract dynamic three-dimensional structure information of vegetation from the individual, community, local landscape. The accessibility of canopy height, volume, biomass, especially the profile of the vertical distribution of biomass will strongly promote the research of vegetation and biodiversity, and provide accurate and reliable decision-making information to the local government.

3.Crop3D :Integrated system of hardware and software for high-throughput crop phenotyping

Molecular designer breeding is an  effective solution to improve crop yield and quality and has been considered as the most advanced technology that makes crop genetic improvement worldwide. However, the difficulties in measuring phenotypes non-destructively during crops’ whole growth period, especially in the field conditions impede the process.

Recent progress in plant phenomics show that a suite of new technologies, including Light Detection and Ranging (LiDAR) technology, intelligent robot technology, unmanned aerial vehicle (UAV) remote sensing technology, could accelerate progress in understanding genetic function and environmental responses. With the support of “the Strategic Program of Molecular Module-Based Designer Breeding Systems” and “the Instrument Developing Project of the Chinese Academy of Sciences”, we developed a series of phenotyping platforms to characterize agriculturally relevant traits used in greenhouse and field conditions, which were named as Crop 3D. By integrating LiDAR, high-resolution camera, thermal camera and hyperspectral imager, Crop 3D platforms can acquire multi-sources data during the whole growing period automatically and synchronously. The platform in greenhouse applied “sensor-to-plant” working style to avoid disturbance on plant growth. For single plant in greenhouse, phenotype parameters such as plant height, crown, leaf length and width, leaf angle, etc. were extracted. Comparing the height derived by Crop 3D with the data measured by Electronic Distance Measuring Device, the R2 was more than 0.94; field-based platform precisely measured expression of physiological traits at canopy level under heterogeneous and fluctuating environment. Canopy height, canopy cover, plant area density and other relevant traits were measured to enhance our understanding of relationship between genotype and environment; moreover, in large region, UAV platform was applied to estimate crop progress, condition and yield.

4.Species distribution model algorithm (geographic one-class data) and its application in simulating biodiversity pattern

The species distribution model(SDM)mainly utilizes the species distribution data and environmental data to estimate the niche of the species according to the specific algorithm, and projects it into the realistic landscape, then reflects the probability of species occurrence, the suitable habitat for species, or species richness.

The traditional species distribution model requires not only the location information (positive samples) of the species but also the location information (negative samples) where the species does not exist. The lack of negative sample data or inaccurate negative sample will cause the uncertainty and unavailability in intraditional SDM. Our group proposed a new species distribution model algorithm, Presence and Background Learning (PBL). PBL only needs “positive” samples and background samples of species to unbiasedly estimate the probability of existence of the species, and the predicted results are significantly more accurate than traditional SDM (eg, MaxEnt). Moreover, we also proposed a new set of accuracy assessment methods for species distribution models to compensate the deficiencies of “negative” samples in the traditional precision evaluation system. Software ModEco was developed by our group, through combing PBL model with other SDMs, which has been downloaded by more than 500 research institutions in different regions (North America, South America, Europe and Asia).

5.Simulating and predicting the effect of climate change and land use change on structure and function of terrestrial ecosystem at reginal and global scale, through combination of multi-source remote sensing data and ecological model.

The global change with the main characteristics of increasing temperature has profoundly affected the structure and function of terrestrial ecosystem. Based on the ground observation data, satellite remote sensing data and lidar data, our group researched on the carbon cycle process and the terrestrial ecosystem carbon pool system based on various mathematical statistics model and terrestrial ecological model. The main research results include:

(1) The establishment of a new high-resolution global climate product database (Alvarez et al., 2013). Climate data are the basic data for ecological studies and model simulations. The research group collected more than 20,000 global meteorological station data, analyzed the interpolation accuracy of various types of remote sensing vegetation data and terrain data as interpolation covariates for meteorological products , and established a global 1km high resolution meteorological database.

(2) Estimate the global and global forest biomass (Hu et al., 2016; Su et al., 2016). Forest ecosystems are the largest terrestrial ecosystem carbon sinks. Based on the advantages of the technology, the research group tried to establish a method to estimate the forest biomass on the large scale range by combining the data of spaceborne lidar data, optical image data and forest inventory data. Based on this method, the global average forest aboveground biomass were 120 ± 60 mg / ha and 210.09 ± 109.31 mg / ha, respectively. The method proposed in this study to estimate the aboveground biomass of forest is of great significance to monitor the dynamic change of large-scale forest carbon storage.

(3) Improve the ecological model and estimate global water use efficiency (Xue et al., 2015). Based on the ground station data and remote sensing data, the research group improved the phenology module of the Integrated Biosphere Model (IBIS), and improved the simulation accuracy of global carbon sinks. Using IBIS model and MODIS data, the spatial and temporal changes of global water use efficiency were calculated independently. The results of the study group’s global average water use efficiency is about 1.7gC / kg H2O and increases with climate change.