Estimating Ecosystem Respiration in the Grasslands of Northern China Using Machine Learning: Model Evaluation and Comparison

Abstract

While a number of machine learning (ML) models have been used to estimate RE, systematic evaluation and comparison of these models are still limited. In this study, we developed three traditional ML models and a deep learning (DL) model, stacked autoencoders (SAE), to estimate RE in northern China’s grasslands. The four models were trained with two strategies: training for all of northern China’s grasslands and separate training for the alpine and temperate grasslands. Our results showed that all four ML models estimated RE in northern China’s grasslands fairly well, while the SAE model performed best (R2 = 0.858, RMSE = 0.472 gC m−2 d−1, MAE = 0.304 gC m−2 d−1). Models trained with the two strategies had almost identical performances. The enhanced vegetation index and soil organic carbon density (SOCD) were the two most important environmental variables for estimating RE in the grasslands of northern China. Air temperature (Ta) was more important than the growing season land surface water index (LSWI) in the alpine grasslands, while the LSWI was more important than Ta in the temperate grasslands. These findings may promote the application of DL models and the inclusion of SOCD for RE estimates with increased accuracy.

Model Selection for Ecosystem Respiration Needs to Be Site Specific: Lessons from Grasslands on the Mongolian Plateau

“…Second, we constructed our models with half-hourly data that were processed through a series of data processing (e.g., coordinate rotation, air density fluctuation, u* filtering). As a result, the measured values are not continuous, which may not fully reflect the extreme weather events influencing ecosystem respiration that occurred during this period [52]. Third, the temporal and spatial variations of ecosystem respiration were affected by productivity and site structure [53].…”

Quantitative Analysis of the Research Trends and Areas in Grassland Remote Sensing: A Scientometrics Analysis of Web of Science from 1980 to 2020

“…Classification: From 1980 to 2020, the selection of data sources has undergone drastic changes. Due to the development of machine learning technology [20,56], the applications of GRS research have greatly increased. GRS research data sources have developed from a single type of multispectral data (such as MODIS and Landsat TM) to the currently widely used hyperspectral data [12,87], multi-angle data [8,17,74], laser radar data [18], high-resolution data [88], unmanned aerial vehicle (UAV) data [89], and multi-source data.…”

“…[93], and used the overall coverage, NPP, biomass, etc., for aims such as determining grassland degradation [16] and estimating grass yield and rough protein content. Due to the wide application of high-resolution and hyperspectral remote sensing data, quantitative analysis of the spectral characteristics of different grassland species, and in-depth research on multitemporal monitoring of different vegetation indices [107][108][109][110], model parameters have been continuously optimized [11,12,14,20]. These parameters can be used to effectively identify different grassland types in the community, and determine the height, coverage, and area ratio of the grassland, thus providing complete monitoring of the succession processes of grassland communities.…”

“…Through comprehensive analysis of high spatiotemporal resolution remote sensing data and ground survey data, the macro surveying speed of grassland resources has been significantly accelerated [14], and the accuracy and timeliness of grassland dynamic change monitoring have been improved [15]. The advantages of remote sensing technology are not only reflected in the active monitoring of grassland resources at a sizeable regional scale [6,8,9,[16][17][18][19], but also play an essential role in the adequate description of ecosystem structure [9], function [20], process [21], and dynamic cascade relationships [12,22]. Therefore, they facilitate revealing the mutual feedback relationship between ecosystem functions and services, achieving the optimization of ecosystem services and promoting sustainable development management of grassland ecosystem [3,16,23].…”

Quantifying the Spatial Representativeness of Carbon Flux Footprints of a Grassland Ecosystem in the Semi‐Arid Region

“…The eddy covariance technique is a micrometeorological detection method used to measure the turbulent transport of matter and energy between vegetation and the atmosphere (Baldocchi, 2003; Yu et al., 2014). High‐frequency turbulence data are used as training data for machine learning (ML) to be upscaled to regional scales and can be used to validate gridded data products (Sun et al., 2021; Xiao et al., 2014; Zhu et al., 2020). However, the heterogeneity of in‐situ flux data is often neglected in upscaling and validation, which ultimately leads to uncertainties in data products due to the lack of “footprint awareness,” namely the absence of a function describing the relationship between the spatial distribution of surface sources or sinks in the near‐surface layer and observed fluxes (Metzger, 2018).…”