THE RELEVANCE OF GEOGRAPHIC LOCATION FOR ACCURATELY ESTIMATING CROP SIGNALS FROM MODIS GRIDDED IMAGERY Campagnolo, Manuel (1), Carmona, Vitor (2) (1) Instituto Superior de Agronomia, Universidade de Lisboa, Portugal (mlc@isa.ulisboa.pt) (2) Instituto de Financiamento da Agricultura e Pescas, Portugal CHARACTERIZATION OF SPATIAL RESOLUTION OF MODIS GRIDDED REFLECTANCE PRODUCTS Most MODIS gridded products are the result of a resampling of swath data over the sinusoidal grid • The swath data depend on the orbit and on the sensor point spread function (sensor PSF); • The sinusoidal grid (SIN) geometry depends on the latitude and longitude. As a result, spatial resolution of MODIS gridded products depends on geographic location. It also depends on the sensor view zenith angle and it is anisotropic. The gridded reflectance point spread function can be estimated for any direction using standard image processing techniques. Conclusions: 1. For a given range of view zenith angles, spatial resolution of MODIS gridded products depends on latitude and longitude, being optimal over the Equator and the Prime Meridian [1]. 2. The gridded reflectance point spread function (PSF) is aligned with the sinusoidal grid. RELEVANCE OF SPATIAL RESOLUTION HETEROGENEITY AND ANISOTROPY FOR CROP MONITORING Estimation of the spatial resolution: (1 st ) in one direction, over a perfect edge; (2 nd ) rotate the edge in all directions. SIN grid over swath observations Grid cell values after resampling Edge spread function PSF (all directions) Example: Kansas (lat=39.08N, long=96.56W), sensor Aqua, view zenith angle = 22.6 o MODIS SIN grid and orbit geometry Example: mean response for 250m nominal MODIS Aqua gridded imagery of the same crop pattern at different geographic locations. The radial plots represent PSF estimates for vza ranging from 0 o to 40 o . Ellipses represent the areas on the ground that contribute 50%, 75% and 90% to the response at the grid cell under a Gaussian model. The gray areas in the background represent one grid cell footprint at the same scale. Research has shown that dense temporal series of well-calibrated remotely sensed images can be used to forecast crop production [2]. This supposes that the crop signal and the image response are highly correlated. However, the signal (e.g. NDVI) of the surface is spatially averaged according to the point spread function which varies in magnitude and direction with geographic location. This lead to location dependent degradation in the image response. Below, we illustate this effect on a given crop pattern. Conclusions and recomendations: 1. The gridded image can be deconvoluted using the estimated location dependent point spread function. 2. If the crop boundaries are known, a robust metric for crop monitoring – e.g. the weighted average of the NDVI of the 5% purest grid cells [2] – should be used to mitigate the boundary effects. 3. Swath data should be used preferentially; if gridding is necessary, the grid should suit the location. References: [1] M.L. Campagnolo and E.L. Montaño (2014) Estimation of effective resolution for daily MODIS gridded surface reflectance products, IEEE Transactions on Geoscience and Remote Sensing. [2] I. Becker-Reshef et al. (2010) A generalized regression-based model for forecasting winter wheat yields in Kansas and Ukraine using MODIS data, Remote Sensing of Environment 114, 1312-1323 Location dependent response variability