Integration of Multisensor Remote Sensing Data for the Retrieval of Consistent Times Series of High-Resolution NDVI Images for Crop Monitoring in Landscapes Dominated By Small-Scale Farming Agricultural

Abstract:

There is a need for timely and accurate information of food supply and early warnings of production shortfalls. Crop growth models commonly rely on information on vegetation dynamics from low and moderate spatial resolution remote sensing imagery. While the short revisit period of these sensors captures the temporal dynamics of crops, they are not able to monitor small-scale farming areas where environmental factors, crop type and management practices often vary at subpixel level. Although better suited to retrieve fine spatial structure, time series of higher resolution imagery (circa 30 m) are often incomplete due to larger revisit periods and persistent cloud coverage. However, as the Landsat archive expands and more fine resolution Earth observation sensors become available, the possibilities of multisensor integration to monitor crop dynamics with higher level of spatial detail are expanding.

We have integrated remote sensing imagery from two moderate resolution sensors (MODIS and PROBA-V) and three medium resolution platforms (Landsat 7- 8; and DMC) to improve the characterization of vegetation dynamics in agricultural landscapes dominated by small-scale farms. We applied a data assimilation method to produce complete temporal sequences of synthetic medium-resolution NDVI images. The method implements a Kalman filter recursive algorithm that incorporates models, observations and their respective uncertainties to generate medium-resolution images at time steps for which only moderate-resolution imagery is available. The results for the study sites show that the time series of synthetic NDVI images captured seasonal vegetation dynamics and maintained the spatial structure of the landscape at higher spatial resolution.

A more detailed characterization of spatiotemporal dynamics of vegetation in agricultural systems has the potential to improve the estimates of crop growth models and allow a more precise monitoring and forecasting of crop productivity.