A spatiotemporal synthetic NDVI generation model for agricultural fields

Although Normalized Difference Vegetation Index (NDVI) has ever been one of the widely used indices for remote sensing based agricultural analysis, its non-periodic and asynchronous availability in terms of phenological phase of the vegetation restricts applicability in precision farming. In this study we proposed a new model that generates spatiotemporal synthetic NDVI data that can be used on parcel level analysis. Continuous time fractional vegetation cover (FVC) measurement from spatially distributed agricultural observation network and asynchronous multi-temporal NDVI data from high resolution remote sensing satellite images are used in an adaptive manner. High resolution satellite images are needed to create NDVI time series and to detect changes at parcel resolution. The disadvantages of having too few images for a given season could be overcome by carrying out additional ground measurements. Spectral measurements are laborious and prone to errors and thus automatic measurements have to be performed. In studies like Kastens et all or Calera et all, it has been demonstrated the linear relation between NDVI and Leaf Area Index (LAI) and also between NDVI and plant cover. In this paper we generated parcel based continuous synthetic NDVI time series using instant NDVI values computed from satellite images together with continuously recorded digital images of parcels. The model capturing the linear relationship between NDVI and FVC was initiated using the oldest dated satellite image. As new satellite images are obtained, the model and estimated values are updated to increase accuracy of generated synthetic NDVIs. The mean absolute error of the predicted synthetic NDVI values with respect to actual parcel NDVI obtained from Spot5 images is at most 0.05 which represent only 7% of total NDVI.