Isabelle Piccard

Sub-pixel classification of SPOT-VEGETATION time series for the assessment of regional crop areas in Belgium

Abstract Global time series of low resolution images are available with high repeat frequency and at low cost, but their analysis is hampered by the presence of mixed pixels and the difficulty in locating detailed spatial features. This study examined the potential of sub-pixel classification for regional crop area estimation using time series of monthly NDVI-composites of the 1xa0km resolution sensor SPOT-VEGETATION. Belgium was selected as test zone, because of the availability of ample reference data in the form of a vectorial GIS with the boundaries and cover type of the large majority of agricultural fields. Two different methods were investigated: the linear mixture model and neural networks. Both result in area fraction images (AFIs), which contain for each 1xa0km pixel the estimated area proportions occupied by the different cover types (crops or other land use). Both algorithms were trained with part of the reference data and validated with the remainder. Validation was repeated at three different levels: the 1xa0km pixel, the municipality and the agro-statistical district. In general, the neural network outperformed the linear mixture model. For the major classes (winter wheat, maize, forest) the obtained acreage estimates showed good agreement with the true values, especially when aggregated to the level of the municipality ( R 2 xa0≈xa085%) or district ( R 2 xa0≈xa095%). The method seems attractive for wide-scale, regional area estimation in data-poor countries.

Crop mapping in countries with small-scale farming: a case study for West Shewa, Ethiopia

Remote sensing is nowadays considered to be a valuable input for the annual collection of crop statistics. Derived crop maps can serve as a baseline for yield or area estimation or to target next years census. For subsistence farming, where small parcels are mixed with other land use, crop mapping remains very challenging. This article evaluates the potential of discriminating crops in West Shewa, an area with small-scale farming in central Ethiopia. A hard classification of high-resolution (30 m) images, yielding good results for commercial farming, could not deal with mixed pixels due to the small parcels. Very high resolution (4 m) images have a more appropriate pixel size, although they only cover subsets of the region. The very high resolution classification was used to calibrate a neural network for sub-pixel classification of the high resolution images. The accuracies were not satisfactory, but did at least demonstrate the potential of this approach.

Ten-daily global composites of METOP-AVHRR

Systematic scanning of the earth surface could be achieved for the first time in 1978, with the launch of the earth observation system NOAA-AVHRR. Some twenty years later, the SPOT-VEGETATION instrument introduced significant improvements at the levels of image quality, timeliness and availability. Since the start in April 1998, VITO is responsible for the central processing, archiving and distribution of the VEGETATION data. This paper briefly announces how a similar service is being established at VITO to provide the same kind of image data from the recently launched METOP-AVHRR.

Potato monitoring in Belgium with “WatchITGrow”

WatchITGrow is a web-based application developed for potato monitoring in Belgium. The different components encompass a back-end with biophysical parameters derived from high resolution satellite imagery, agrometeorological algorithms, phenological development and crop models; and a front-end with dashboards to visualize spatio-temporal information and insert potato field information.

Assessment of AquaCrop for winter wheat using satellite derived fCover data

The integration of crop growth models with remote sensing has presented great potential in (regional) crop yield forecasting; although so far few field-level applications exist. Based on crowd/farm-sourced observations (phenological stages and yield measurements) and a basic assimilation procedure using satellite (DMC) and digital hemispherical pictures (DHP) derived green fractional cover data (fCover), the AquaCrop plug-in model was assessed for winter wheat fields in Belgium. A semi-automated R-environment was developed to simultaneously run, assess and evaluate the ensemble of field-level simulations. The root mean square error (RMSE) was 0.8 ton/ha. It was concluded that the presented approach might be promising for large scale field-level yield forecasting.

An Integrated Long Time Series of 1km Resolution NDVI for Europe from the NOAA-AVHRR and SPOT-VEGETATION Sensors

This paper describes the newly processed 1 km resolution NDVI AVHRR archive over Europe and the integration of the data with the SPOT-VEGETATION NDVI archive. Investigation of the agreement between the two NDVI datasets pointed out a high linear correlation. The RMSE between the datasets is presented and varies for most of the area between 0.05 and 0.10. When applying the relative adjustment functions to account for the difference in spectral response function between the sensors, the RMSE was considerably reduced. The results presented here will be used to integrate the NDVI datasets from the various AVHRR sensors and from SPOT-VEGETATION.