Estimation of crop gross primary production (GPP): II. Do scaled MODIS vegetation indices improve performance?

Abstract

a b s t r a c tSatellite remote sensing estimates of gross primary production (GPP) have routinely been made using spectral vegetation indices (VIs) over the past two decades. The Normalized Difference Vegetation Index (NDVI), the Enhanced Vegetation Index (EVI), the green band Wide Dynamic Range Vegetation Index (WDRVI green ), and the green band Chlorophyll Index (CI green ) have been employed to estimate GPP under the assumption that GPP is proportional to the product of VI and photosynthetically active radiation (PAR) (where VI is one of four VIs: NDVI, EVI, WDRVI green , or CI green ). However, the empirical regressions between VI*PAR and GPP measured locally at flux towers do not pass through the origin (i.e., the zero X-Y value for regressions). Therefore they are somewhat difficult to interpret and apply. This study investigates (1) what are the scaling factors and offsets (i.e., regression slopes and intercepts) between the fraction of PAR absorbed by chlorophyll of a canopy (fAPAR chl ) and the VIs and (2) whether the scaled VIs developed in (1) can eliminate the deficiency and improve the accuracy of GPP estimates. Three AmeriFlux maize and soybean fields were selected for this study, two of which are irrigated and one is rainfed. The four VIs and fAPAR chl of the fields were computed with the MODerate resolution Imaging Spectroradiometer (MODIS) satellite images. The GPP estimation performance for the scaled VIs was compared to results obtained with the original VIs and evaluated with standard statistics: the coefficient of determination (R 2 ), the root mean square error (RMSE), and the coefficient of variation (CV). Overall, the scaled EVI obtained the best performance. The performance of the scaled NDVI, EVI and WDRVI green was improved across sites, crop types and soil/background wetness conditions. The scaled CI green did not improve results, compared to the original CI green . The scaled green band indices (WDRVI green , CI green ) did not exhibit superior performance to either the scaled EVI or NDVI in estimating crop daily GPP at these agricultural fields. The scaled VIs are more physiologically meaningful than original un-scaled VIs, but scaling factors and offsets may vary across crop types and surface conditions.

Integrating chlorophyll fAPAR and nadir photochemical reflectance index from EO-1/Hyperion to predict cornfield daily gross primary production

“…We found that NDVI calibrated with fAPARchl, also referred to as scaled NDVI, can improve performance of GPP estimation over original un-calibrated NDVI (Zhang, Cheng et al 2015). Our current study and other studies support the use of fAPARchl in place of fAPARcanopy in plant GPP modeling activities Croft et al 2015;Zhang et al 2014;Zhang et al 2009;Flanagan et al 2015).…”

Vegetation Productivity in Natural vs. Cultivated Systems along Water Availability Gradients in the Dry Subtropics

“…Regarding annual GPP, uncertainties are related to the apparent need of site-specific empirical scaled EVI-fAPAR functions [67], and the variability of annual LUE (affected by vegetation structural and functional traits–e.g. photosynthetic syndrome–, and soil and climatic conditions) [67–72], sometimes solved by means of look-up tables (based on biome type and climatology) [37,73] or more recently by means of the carotenoid-sensitive “Photochemical Reflectance Index” (PRI) [38,74,75]. However, the accuracy of annual EVI-GPP relationships seems to improve in land covers with high annual EVI ranges and summer rainfalls [28], as in our case.…”

Monitoring Grassland Seasonal Carbon Dynamics, by Integrating MODIS NDVI, Proximal Optical Sampling, and Eddy Covariance Measurements

“…Several versions of the LUE model driven by satellite remote sensing of NDVI have been in wide use for many years [24,31,65,66]. While these models can depict broad global patterns, their agreement with local field measurements varies considerably across ecosystems or with different model formulations [67,68].…”