Godinho

Satellites for small farms crop mapping: exploring the use of Sentinel-2A in a small farm Mediterranean landscape

Sérgio Godinho *±1 – Nuno Guiomar 1 – Teresa Pinto Correia 1

1 ICAAM – Instituto de Ciências Agrárias e Ambientais Mediterrânicas, Universidade de Évora, Portugal
Speaker
± Corresponding author: (sgodinho@uevora.pt)


Introduction

The Food and Agriculture Organization of the United Nations (FAO) states that there are more than 570 million farms in the world, and that the vast majority of these are small or very small. About 94% of the world’s farms are less than 5 hectares in size (FAO, 2014). In many developing countries, farm sizes are becoming even smaller, where small parcels with typically ≤ 2 ha represents approximately 50% of rural populations (Morton, 2007). Despite the widely recognized role of the small farms for food production and security, the extent and distribution of crop types, crop production and cropping intensity in a small-scale farming context still remain uncertain or unknown (e.g. crop production and cropping intensity). As a result, small-scale crop production, even in more developed countries with sophisticated agricultural monitoring systems, is under-reported, although those small farms could produce a substantial percentage of food for local/regional consumption. Accurate information on the extent of crop area, type, and yield estimates is needed to better quantify the crop production capabilities of small farms, which is crucial information for the political authorities in the agricultural sector. Therefore, one important first step in doing this is to map accurately the spatial distribution of crop types in small farms (crop mapping). Such data may be obtained cost-efficiently from Earth Observation Satellites (EOS) due to their multispectral, synoptic, and systematic characteristics that allow the identification of different objects on the landscape (e.g. Jain et al., 2013). One of the most recent and ambitious Earth observation project is the Copernicus Program from the European Commission (EC) in partnership with the European Space Agency (ESA). In general, this program aims providing accurate, up-to-date and easily accessible satellite-based information in order to support scientific research and policy-making in the fields of agriculture, forest and environmental management, climate change and civil protection (ESA, 2015). To achieve the main goals of Copernicus Program, ESA is developing a constellation of satellites called Sentinel, which will be launched over time until 2021. Regarding this Sentinel satellites constellation, Sentinel 2A (acquiring and providing high-resolution multispectral data) is already operational and sending information to the Earth. Sentinel 2B (S2B) will be launched in 2017 and together with S2A they will both provide a global coverage of the Earth’s land surface every 5 days. Concerning the crop mapping in the small-scale farming systems, it is expected that S2A and S2B will bring together new research opportunities (e.g. detailed analysis of crop areas, types and yield estimates in small farms context) due to their high spatial (up to 10 m), spectral (13 bands) and temporal resolutions (each 5 days).

Particularly in the Mediterranean basin, the number of small farms remains significant, while at the same time no much is known on the distribution, typology and production capacity of these farms, as well as on their integration in the local and regional food systems. In the ongoing Horizon 2020-funded research project called ‘Smaller farms, Small Food Businesses and Sustainable Food and Nutrition Security (SALSA)’ the aim is to produce integrated knowledge on the small farms and their role in sustainable food and nutrition security, but also to progress further beyond the state of the art in the use of new methodological approaches and tools. Thus, the purpose of the present study, developed under the scope of the SALSA project, is to perform a crop classification by using S2A-derived data and to examine their suitability in mapping crop types at small-scale farms. Specifically, there were three main research objectives in this study: 1) exploit the usefulness of multi-temporal S2A imageries in crop classification accuracy; 2) compare the object-based and the pixel-based classification approaches; and 3) evaluate the performance of each spectral band to the classification accuracy.

Materials and Methods

To meet these objectives, a 10x10km2 square around the Montemor-o-Novo city in southern Portugal, with a typical small farm Mediterranean landscape, was selected as a case study to perform all the tests. As a classification statistical method the non-parametric Stochastic Gradient Boosting (SGB) machine learning algorithm was used for the classifications tests. Twelve cloud-free S2A images of the same 10×10 km2 area were acquired (between 30th April and 27th September 2016) in order to develop a multi-temporal classification scheme. The images were downloaded and atmospherically corrected (DOS1 method) using the Semi-Automatic Classification (SCP) plugin implemented in the QGIS software. Field work and Google Earth image-interpretation was performed to collect reference data for nine crop types existing in the study site.

Results and Discussion

Based on the first tests it was found that the integration of several S2A images from different dates into the classification scheme resulted in a significant improvement in the accuracy of crop classification (increase of 34.26% and 0.28 in overall accuracy and Kappa index, respectively). The comparison between the two tested classifications approaches showed that pixel-based (overall accuracy=91.91% and Kappa index=0.91) can produce better results than object-based (overall accuracy=84.30% and Kappa index=0.81), presenting an overall accuracy improvement of 7.61% and a Kappa increase of 0.10. By using the variable importance function from the SGB algorithm, it was found that shortwave infrared (B11 and B12) and red-edge bands (B5) were the most useful for crop mapping. Overall, it has been demonstrated that S2A has the capability for producing accurate crop map in a complex Mediterranean small farm context.

References

Food and Agriculture Organization of the United Nations (FAO).: 2014. The State of Food and Agriculture 2014: Innovation in Family Farming

Jain, M., Mondal, P., DeFries, R.S., Small, C., Galford, G.L.: 2013. Mappingcropping intensity of smallholder farms: A comparision of methods using multiple sensors.

Morton, J.F.: 2007. The impact of climate change on smallholder and subsistence agriculture. Proceedings of the National Academy of Sciences, 104(50), 19680-19685.

European Space Agency (ESA).: 2015. Copernicus observing the Earth. Available at http://www.esa.int/Our_Activities/Observing_the_Earth/Copernicus.