Estimation and uncertainty analyses of grassland biomass in Northern China: Comparison of multiple remote sensing data sources and modeling approaches

Abstract

a b s t r a c tAccurate estimation of grassland biomass and its dynamics are crucial not only for the biogeochemical dynamics of terrestrial ecosystems, but also for the sustainable use of grassland resources. However, estimations of grassland biomass on large spatial scale usually suffer from large variability and mostly lack quantitative uncertainty analyses. In this study, the spatial grassland biomass estimation and its uncertainty were assessed based on 265 field measurements and remote sensing data across Northern China during [2001][2002][2003][2004][2005]. Potential sources of uncertainty, including remote sensing data sources (DATsrc), model forms (MODfrm) and model parameters (biomass allocation, BMallo, e.g. root:shoot ratio), were determined and their relative contribution was quantified. The results showed that the annual grassland biomass in Northern China was 1268.37 ± 180.84 Tg (i.e., 532.02 ± 99.71 g/m 2 ) during 2001-2005, increasing from western to eastern area, with a mean relative uncertainty of 19.8%. There were distinguishable differences among the uncertainty contributions of three sources (BMallo > DATsrc > MODfrm), which contributed 52%, 27% and 13%, respectively. This study highlighted the need to concern the uncertainty in grassland biomass estimation, especially for the uncertainty related to BMallo.

Renewed Estimates of Grassland Aboveground Biomass Showing Drought Impacts

“…The MODIS‐derived α value with an AGB model of [ f AGB = α × NDVI + β] for the grasslands was 839.9 g/m 2 (Xue et al, ) or 864.36 g/m 2 (Li et al, ) for the central IMAR, 638.19 g/m 2 (Xu et al, ) or 680.15 g/m 2 (Gao et al, ) for the western IMAR, and 1,156.72 (Li et al, ) or 1,161.13 g/m 2 (Xu et al, ) for the eastern IMAR. However, all of these models carried low R 2 values (e.g., 0.39 in Xu et al, ; 0.60 in Jia et al, for northern China), suggesting that large uncertainties existed in predicting AGB from these models. When the above model for northern China was used to estimate the grassland AGB in IMAR with α = 220.98 g/m 2 (Jia et al, ), the regional average would be significantly underestimated, as we found that the M scalar for the same region was 253.62 g/m 2 .…”

“…However, all of these models carried low R 2 values (e.g., 0.39 in Xu et al, ; 0.60 in Jia et al, for northern China), suggesting that large uncertainties existed in predicting AGB from these models. When the above model for northern China was used to estimate the grassland AGB in IMAR with α = 220.98 g/m 2 (Jia et al, ), the regional average would be significantly underestimated, as we found that the M scalar for the same region was 253.62 g/m 2 . More importantly, the M scalar across the IMAR grasslands varied substantially, ranging from 2 g/m 2 to 1,470 g/m 2 (Figure a).…”

“…However, these models estimating VI have two potential flaws embedded in remote sensing data: (1) they are satellite dependent because of the differences in optical sensors applied among the satellites, and (2) the model coefficients are likely not constant among the study regions. For example, Jia et al () used three types of satellite data (MODIS, AVHRR, and Landsat TM) to model AGB for grasslands in northern China and found large differences among the models' parameters and lower R 2 values (0.54–0.60). Li et al () compared the results from MODIS and Landsat TM in central IMAR and reached a similar conclusion ( R 2 = 0.71–0.85).…”

“…Direct measurements of biomass provide us with the most accurate estimates for the time being, but scaling up from in situ measurements to broader spatial and temporal scales has to be conducted in conjunction with satellite remote sensing technology and ecosystem models (Chen et al, ; Field et al, , ; Moreau et al, ; Rouse et al, ; Roy & Ravan, ; Running et al, ). With the rapid development and availability of diverse satellite products and improvements in ecosystem models, a large amount of effort has been placed toward upscaling endeavors to the landscape scale (e.g., Moreau et al, ; Zheng et al, ), regions (Jia et al, ; Paruelo et al, ; Xie et al, ), and even global scales (Hese et al, ). These upscaling efforts are often derived through statistical models based on the vegetation index (VI).…”

“…For example, the normalized difference vegetation index (NDVI) (Rouse et al, 1974) and the enhanced vegetation index (EVI) (Huete et al, 1997) have been widely applied in estimating aboveground biomass (AGB). Yet a similar level of NDVI or EVI can be found among forests, grasslands, or even desert ecosystems although AGB are very different (e.g., Brown et al, 1989;Jia et al, 2016;Xu et al, 2008;Zheng et al, 2004). In modeling AGB for grasslands across Inner Mongolia using a similar approach (i.e., f AGB = α × VI + β), we found that AGB with NDVI of 0.4 can vary from 21 g/m 2 to 275 g/m 2 , with a predicted AGB of 113 g/m 2 .…”