Detection of irrigation dates and amounts on maize plots from the integration of Sentinel-2 derived Leaf Area Index values in the Optirrig crop model

The increase in food production due to the expansion of agricultural lands has led to the intensive use of (mainly) fresh water for irrigation. A key challenge for irrigated agriculture has thus become to optimize the use of available water resources to fit environmental constraints while satisfying the increasing food demand, achieving efficient uses of irrigation water, i.e. well-thought sequences of irrigation dates and amounts. In coherence, we developed a methodology based on the automated acquisition of Leaf Area Index (LAI) values, derived from remote sensing data, confronted with predictions drawn from the Optirrig crop growth and irrigation model, to solve the inverse problem of detecting irrigation dates and amounts, at the plot scale and for maize crops grown in the Occitanie region, France. The method consisted of seeking possible irrigation events (dates, amounts) between two cloud-free Sentinel-2 (S2) optical images and detecting the most probable of these events, responsible for the least difference between the predicted and observed, S2-derived LAI values (LAIS2). The approach was first tested with synthetic noisy values to encompass the effects of errors on the observed and modeled LAI values, and these of increased duration between available observations (cases ∆S2=5, 10,and 15 days), promoting the possibility to use daily-interpolated LAI values as a starting point for the inverse problem (∆S2 is fixed to 10 and 15 days then values are interpolated and recorded on a 5 days basis, cases ∆S2=5mod10 and 5mod15 days, respectively). From the synthetic dataset, irrigation dates detection results showed that the best performance is obtained for ∆S2=5 days or when using daily interpolated LAI values when ∆S2=5mod10 or 5mod15 days with an F−score near 85%. Most irrigation dates were detected with errors between 0 and 3 days, while irrigation amounts (20, 30 or 40 mm) were correctly identified in over 80% of cases, when simulating dry climatic conditions typical of the Mediterranean ring. For the documented real cases, the irrigation dates were detected with an overall recall value of 81.6% when evaluated using daily-interpolated LAIS2. The irrigation amounts are correctly identified for only 28.5% of the detected irrigation dates for the plots located in Montpellier. In contrast, the detection of the irrigation amounts was not possible over the plots in Tarbes. This weakness in the detection of irrigation amounts seems related to the fact that Optirrig simulates the exact crop irrigation requirements, based on a soil water balance equation and accurate soil moisture calibration, while farmers’ decisions are taken on different grounds in the field. Overall, the obtained results prove the relevance of the combined used of Optirrig and optical remote sensing data for the detection of irrigation dates, and possibly amounts, at field scale.