Land data assimilation in China: A review (Invited Presentation)

Land data assimilation (LDA) plays a key tool in the earth system science (ESS) by bridging the gap between land models and land observations. Progresses on methods and applications of LDA have been made in China during the past decade. This paper provides a review of LDA in China. Rather than attempt to cover all topics, we have particularly chosen a few significant issues to be introduced.

       Many state-of-the art LDA methods have been explored, applied or modified. Those LDA methods include (1) continuous DA methods based on simulated annealing and other heuristic optimization algorithm, (2) dual-pass DA methods based on ensemble Kalman filter (EnKF) and variational DA, (3) the proper orthogonal decomposition-based ensemble four-dimensional variational DA method (PODEn4DVar). Meanwhile, the performances of a variety of major LDA methods were compared using numerical models and real land surface models (e.g., EnKF vs SIR-PF, ES vs EnKF, EnKF vs EnKS), and the applications of some nonlinear non-Gaussian filtering algorithms in DA, such as unscented particle filter (UKF), particle filter (SIR-PF), unscented filter (UPF), gauss and particle filter (GSPF), local ensemble transform kalman filter (LETKF), also have been explored. Additionally, an evolutionary algorithm-based error parameterization method for DA had been proposed, this LDA method searches for the most ideal error adjustment factors in order to obtain better assimilation results in LDA.

       Some multivariate, multi-source and multi-purpose land data assimilation systems (LDAS) of China have been developed by different groups. At present, four of which are well known in China. The first one is the Chinese Land Data Assimilation System (CLADS), which is the first LDAS in China. The CLADS has multiple model operators (CoLM, SiB2, SHAW, and snow model) and multiple data types being assimilated. The second one is the Global Microwave Land Data Assimilation System (GMLDAS-ITP), which is a basic LADS and mainly focus on the land surface state at the Tibetan plateau. GMLDAS-ITP has a function to automatic calibrate model parameters with assimilated satellite data. Such a calibration can reduce the biases due to specifying default parameter values in a model. The third one is the Joint Data Assimilation System (Tan-Tracker), which is based on the PODEn4Dvar assimilation method, in assimilating Greenhouse gases Observing SATellite (GOSAT) carbon dioxide (CO2) data. The last one is the Python Multivariate Land Data Assimilation System (DasPy), it is an open source multivariate data assimilation system. The objective of DasPy development is for the multi-sources and multivariate land data assimilation applications, such as soil moisture, soil temperature and joint state and parameter estimation. DasPy also is a very useful tool for the beginner to study how to carry out DA experiments.

       Corresponding observation operators have been explored and constructed in order that multi-sources observations can be assimilated, including the forward model for snow radiance, and the forward model for soil moisture radiance. Therefore, various kinds of multi-sources observations and data products have been operationally assimilated in practical LDA experiments. Assimilated data in LDAS include a series of remote sensing data, such as MODIS LST products, AMSR-E data, SSM/I, ASCAT soil wetness index; as well as a lot of in-situ observations generated by some allied observation experiments (e.g., WATER, HiWATER). Additionally, some novel kind of observation, such as cosmic-ray, also had been tentatively used to estimate soil moisture.

       Various ESS datasets have been produced using LDAS,for example, soil properties, soil moisture and temperature, surface flux, evapotranspiration (ET), snow depth, snow water equivalent, precipitation and crop growth. Moreover, LDAS also was used to implement parameter estimations and structure error identification of land model, ecological model, hydrological model using multiple data sets with different DA algorithms, in order to improve and perfect the parameterization schemes of those models in term of physical mechanisms, and therefore improve the DA skills of those models.

       Generally, model parameters and forcing data are play basic roles in LDAS, researchers in China have developed a lot of basic dataset (e.g., 1:100,000 land cover map, vegetation map, wetland map, glacier map, soil map and the MODIS land cover product) that enable researchers to conduct various LDA experiments. In addition, some high-quality forcing datasets also have been produced, the frequently-used dataset of which is China meteorological forcing dataset for resolution of 0.1 deg and temporal resolution of 3-hr, this dataset contain temperature, press, humidity, wind speed, longwave/shortwave radiation and precipitation and so on.

       In addition to the above progresses, many other achievements also have been made in the following areas, including (1) constructed some labs with the capability of high performance computing, (2) enhanced observation network provide more abundant data (e.g., multi-layer soil moisture and temperature, surface flux, snow depth) for validating DA results, (3) cultivated a large number of young scientists of LDA.

       In summary, LDA was introduced in China at the beginning early 2000, and experience a vigorous development in China. At present, however, LDA is still not perfect and some urgent issues need to be addressed. In the future, researchers of LDA in China should set their sights on the following aspects: (1) System refinement and validation; (2) Operational application in eco-hydrological forecasting and water resource management; (3) Development of methods to quantify observation errors, especially the representativeness error of both in situ and remote sensing observations.