Estimation of Cultivated Area in Small Plot Agriculture in Africa for Food Security Purposes
Holecz, Francesco; Collivignarelli, Francesco; Barbieri, Massimo

Food Security Information Systems in Africa are currently based on Earth Observation to follow the status of the vegetation and its progress throughout the growing season. The most common indicators - which are inferred from low spatial resolution (25km to 500m) sensors - are Rainfall Estimates, Cold Cloud Duration, Normalized Difference Vegetation Index, Evapo-transpiration, Soil Moisture and Water Requirement Satisfaction Index. These indicators, together with other information such as weather indexes, market information, management practices and socio-economic data, among others, are used to monitor food security and produce regular bulletins on the food situation.

Area estimation is one of the most logic Earth Observation applications, because remote sensed images offer geospatial information. Moreover, if a multi-temporal data set is available, land cover changes can be identified and mapped. Looking at agricultural applications, where the spatial-temporal dynamic is significant, the use of frequently and regularly acquired high resolution Earth Observation data can doubtless contribute in i) the crop area estimation and ii) the crop growth development, hence strongly reducing (time consuming) field surveys. Besides the fact that Synthetic Aperture Radar (SAR) can penetrate into the clouds - thereby the capability to provide high resolution data on a frequent basis, conditio sine qua non for this application - these sensors have the characteristic to be sensitive to roughness and moisture content. As far as agricultural surfaces are concerned, soil/plant roughness and moisture content, as well as their evolution over time, are not casual. In fact, knowledge of land practices, it has been shown that multi-temporal high (10-20m) resolution SAR data offer valuable information to identify at the earliest stage of the crop season, when and where (ensemble of) fields are prepared, and later on, the phenological crop’s development. As example in Malawi, an accuracy of 80% has been obtained using 15 m resolution SAR data. Indeed, in that country, it was recognized that the limiting the limiting factor was the coarse spatial resolution, if compared to the field dimension.

Two are the possible ways to overcome to this limitation: either to use sensors with higher spatial resolution (i.e. better than 5m) or to combine in situ point data with high resolution remote sensing data. In synthesis:

1. The use of very high resolution SAR data cannot be applied country-wide on a multi-temporal basis for obvious reasons. Therefore, these data will be exclusively utilized to accurately identify the fields or ensemble of fields. Multi-temporal high resolution SAR data are acquired prior and during the crop season in order to estimate the potential crop extent and the effective cultivated extent. In this approach, a key issue is to appropriately plan and acquire the very high resolution data in a time-dependant way, in order to identify the prepared fields prior to the plant growth.

2. Regression estimators, or alternatively, procedures based on confusion matrices, are traditional ways to combine accurate (possibly unbiased) information observed on a sample with inaccurate (biased) information known for the whole population or for a larger sample set. Remote sensing data are used for stratification and to improve, a posteriori, the estimates accuracy. With respect to the ground samples, it should be noticed, that the amount of collected samples - whose collection is based on a dedicated statistical sampling design - is considerable: for instance, for a country like Malawi (approximately 100,000 sqkm), several tens of thousands samples must be collected. Moreover, in order to limit the field visits, the field survey is usually performed in the mid of the crop season or prior to the harvesting time, hence too late for food security purposes.

The implemented and demonstrated methodology consists on the synergetic use of multi-resolution, multi-temporal remote sensing data. In particular, the service is based on the utilization of i) multi-annual ALOS PALSAR-1 Fine Beam Dual (15m) data acquired during the dry season, ii) very high resolution 1-day interferometric Cosmo-SkyMed StripMap data (3m) acquired at start of the crop season, and finally, iii) multi-temporal ENVISAT ASAR Alternating Polarization (15m) data acquired along the whole crop season. Products derived from these three independent and complementary systems are fused at semantic level, enabling the generation of the cultivated area. Note that each intermediate product has a well defined purpose in agriculture and food security: i) the potential crop extent provides information relevant to the field status prior to the crop season; ii) the potential area at start of the crop season gives a crucial information of the prepared cropped area; iii) the crop growth extent during the rainfed crop season monitors the phenological status and development. The proposed methodology has been operationally implemented and demonstrated in various agro-ecological zones and landscapes of Malawi. The resulting cultivated area has been validated with an overall accuracy better than 90%.