Detection of irrigated and rainfed crop fields in temperate areas using Sentinel-1, Sentinel-2 and rainfall time series

Human activities have an impact on the different components of the hydrosphere and 80 % of the world's population is now facing water shortages that will worsen with global warming. Faced with this emergency situation, it is necessary to develop adaptation strategies to monitor and manage water resources for the entire population and to maintain agricultural activity. One of the adaptation strategies that has been favoured is the management of crop irrigation to optimize the use of scarce water ressources.

To meet this objective, it is necessary to have explicit information on irrigated areas. However, up to now, this information is missing or imprecise at the field scale (it is only produced as aggregated statistics or maps at the regional or national scales). In this work, we propose a method for detecting irrigated and rainfed plots in a temperate areas (Adour-Amont watershed of 1500 km² located in south-western France) jointly using optical (Sentinel-2), radar (Sentinel-1) and rainfall (SAFRAN) time series, through the random forest classification algorithm. This spectral information was synthesized in the form of cumulative monthly indices corresponding to the sum of the spectral information for each element (optical, radar, rainfall). This cumulative approach makes it possible to reduce the redundancy of the spectral information and the calculation time of the classification process.

The summer crops studied were maize, soybean and sunflower, representing respectively 82%, 9% and 8% of the crops cultivated of the studied area, but only part of these crops were irrigated. In order to make the distinction for the same crop, we assume that the speed and amplitude of canopy development differs between irrigated and rainfed crop. Five scenarios were used to evaluate the performance of classification models. They have been built according to the different spatialized data, i.e (Optic; Radar; Optic & Radar; Optic, Radar & Rainfall and 10-day images, which is reference scenario without the cumulative monthly indices). Finally, generated classification maps were evaluated using ground truth data collected during 2 years with contrasted meteorological conditions.

The use of separate radar and optical data gives low results (Overall Accuracy (OA) < 0.5) compared to the combined classifications of the cumulated data set (optical & radar), which gives good results (OA ± 0.7). The use of the monthly cumulated rainfall allows a significant improvement of the Fscore of the irrigated and rainfed crop classes. Our study also reveals that the use of cumulative monthly indices leads to performances similar to those of the use of 10-day images while considerably reducing computational resources.