2.5. Model error analysis

To further explore the sources of model errors, we conducted an error analysis on all the models worldwide. We mainly assessed climate and vegetation factors related to our study, including the following six factors: mean annual temperature (MAT), mean annual precipitation (MAP), temperature change (ΔTMP), interannual variability of temperature (IAT), interannual variability of precipitation (IAP), and interannual variability of vegetation (IAV). The errors of the models are expressed by the CV. Among these factors, the MAT and MAP of each grid point are calculated from CRU climate data averaged over 1982–2015, the ΔTMP is calculated as the average temperature of 2011–2015 minus the average temperature of 1982–1986, the IAT and IAP are calculated from the CV for CRU climate data from 1982 to 2015, and the IAV is calculated from the CV in the NDVI data from 1982 to 2015.

We adopted linear regression method and calculate correlations between the model error CV with the related variables to analyze the relationships between different variables.

In addition, the vegetation variability is related to the land cover type, and we also analyzed the influence of different land cover types on the model performance. The global vegetation classification data are derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) land cover product MOD12C1 (http://glcf.umd.edu/data/lc/). The MOD12C1 data product has an annual resolution and a spatial resolution of 0.05°. We used land cover data from 2012 as the base map with which to divide the global vegetation cover areas. The MOD12C1 data product adopts the International Geosphere‐Biosphere Program scheme, which divides the globe into 17 land use types. In this paper, after removing areas of water and snow, 15 land cover types are used for the analysis. The land cover data were interpolated to a 0.5° resolution to match the NDVI and climate data. The dominant land cover type (the land cover type with the highest proportion within the 0.5° pixel) was used as the land cover type of a given grid point.