Volume XLII-3/W11
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W11, 35–42, 2020
https://doi.org/10.5194/isprs-archives-XLII-3-W11-35-2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-3/W11, 35–42, 2020
https://doi.org/10.5194/isprs-archives-XLII-3-W11-35-2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

  14 Feb 2020

14 Feb 2020

AGRICULTURALLY CONSISTENT MAPPING OF SMALLHOLDER FARMING SYSTEMS USING REMOTE SENSING AND SPATIAL MODELLING

A. Crespin-Boucaud1,2, V. Lebourgeois1,2, D. Lo Seen1,2, M. Castets1,2, and A. Bégué1,2 A. Crespin-Boucaud et al.
  • 1CIRAD, UMR TETIS, F-34398 Montpellier, France
  • 2TETIS, Univ Montpellier, AgroParisTech, CIRAD, CNRS, IRSTEA, Montpellier, France

Keywords: Landscape, Crop type mapping, Cropping practices, Sentinel-2, Spatio-temporal modelling, Smallholder agriculture

Abstract. Smallholder agriculture provides 90 % of primary food production in developing countries. Its mapping is thus a key element for national food security. Remote sensing is widely used for crop mapping, but it is underperforming for smallholder agriculture due to several constraints like small field size, fragmented landscape, highly variable cropping practices or cloudy conditions. In this study, we developed an original approach combining remote sensing and spatial modelling to improve crop type mapping in complex agricultural landscapes. The spatial dynamics are modelled using Ocelet, a domain-specific language based on interaction graphs. The method combines high spatial resolution satellite imagery (Spot 6/7, to characterize the landscape structure through image segmentation), high revisit frequency time series (Sentinel 2, Landsat 8, to monitor the land dynamic processes), and spatiotemporal rules (STrules, to express the strategies and practices of local farmers). The method includes three steps. First, each crop type is defined by a set of general STrules from which a model-based map of crop distribution probability is obtained. Second, a preliminary crop type map is produced using satellite image processing based on a combined Random Forest (RF) and Object Based Image Analysis (OBIA) classification scheme, after which each geographical object is labelled with the class membership probabilities. Finally, the STrules are applied in the model to identify objects with classes locally incompatible with known farming constraints and strategies. The result is a map of the spatial distribution of crop type mapping errors (omission or commission) that are subsequently corrected through the joint use of spatiotemporal rules and RF class membership probabilities. Combining remote sensing and spatial modelling thus provides a viable way to better characterize and monitor complex agricultural systems.