Rainfall validates MODIS-derived NDVI as an index of spatio-temporal variation in green biomass across non-montane semi-arid and arid Central Asia

- NDVI increases with rainfall spatially and temporally in Central Asia from 2000 to 15.
- The slope of the NDVI-rainfall relation is steeper in more vegetated regions.
- Ecoregion does not affect the NDVI-rainfall slope and vegetation relation strength.
- The explanatory power of the NDVI-rainfall relation increases with livestock.
- NDVI can soundly measure spatial and temporal variation in Central Asian vegetation.

As satellite-derived normalized difference vegetation index (NDVI) is related to vegetation biomass, it may provide a proxy for habitat quality across extensive species ranges where ground-truth data are scarce. However, NDVI may have limited accuracy in sparsely-vegetated arid and semi-arid environments due to signal contamination by substrate reflectance. To validate NDVI as a vegetation proxy in the low-altitude deserts of Central Asia, we examine its response to precipitation across the migratory corridor of Asian Houbara Chlamydotis macqueenii, a threatened gamebird occupying deserts from the Middle East to China. Restricting NDVI data by altitude (masking higher elevations unoccupied by n = 61 satellite-tracked houbara) and 2009 Globcover land cover (excluding cropland and built-up area), we relate moderate-resolution imaging spectroradiometer (MODIS) NDVI data to Global Precipitation Climatology Project precipitation data across five World Wildlife Fund semi-arid ecoregions (totaling 4.06 million km2). We examine this both spatially (per 1°cell, mean annual NDVI and mean precipitation over 16 years, 2000-2015); and temporally (annual NDVI and annual precipitation) using separate temporal General Linear Models per cell and an overall Generalized Linear Mixed Model (GLMM) (including cell ID as a random effect). We sought to explain spatial variation in the NDVI-precipitation relation among temporal per degree-cell models, in terms of the slope (strength) and adjusted (adj.) R2 (explanatory power), using inter-annual mean NDVI (2000-2015) and Gridded Livestock of the World livestock density. NDVI increases with precipitation, both spatially (adj. R2 = 0.58, p < 0.001) and temporally (mean adj. R2 across n = 244, 1° cells = 0.44; GLMM across cells p < 0.001). More vegetated regions show a stronger temporal response of vegetation biomass for a given precipitation increment (slope of NDVI to precipitation in per cell temporal models increases with inter-annual mean NDVI; adj. R2 = 0.38, p < 0.001), reinforcing the conclusion that NDVI provides a proxy for vegetation abundance. The slope of this relation did not differ among ecoregions. Although livestock density is generally assumed to degrade vegetation and weaken the NDVI-precipitation relationship, explanatory power (adj. R2 of per cell NDVI-precipitation models) is weakly, but positively, related to livestock density (adj. R2 = 0.02, p = 0.011). This may be because we assess livestock at a coarse grain, at scales where overall stocking density is positively associated with vegetation abundance, but may also indicate that livestock are not degrading vegetation at regional landscape-scales despite potential localized effects. The strong signature of rainfall shows MODIS NDVI offers a potentially powerful proxy for spatial and temporal variation in arid and semi-arid vegetation at a resolution of 1° and 1 year over the houbara's breeding and wintering range, and probably also at finer spatial resolutions. NDVI can therefore be used in analyses relating (a) staging and wintering site selection to variation in habitat among potential wintering locations, and (b) variation within and between localities to demographic carry-over effects.