Combining Remote Sensing and Low-Cost Sensors for LULC and Irrigation Characterization in the South of France

In the face of climate change, Mediterranean regions, such as the South of France, are increasingly struggling with drought, water scarcity, and low groundwater levels. For agricultural regions relying on irrigation systems to guarantee summertime crop productivity, this is a central issue. Consequently, optimizing agricultural water uses and understanding the impact of irrigation systems on local and regional hydrological processes is indispensable. At larger scales, another challenge is to identify crop types as well as cropping and irrigation patterns for irrigation water management, reservoir operation, and real-time resource allocation. In this context, remote sensing provides a promising approach.  

This study focuses on combining land use - land cover (LULC) analyses based on Sentinel-1 and -2 data and in-situ measurements realized using innovative low-cost sensors, to characterize irrigation water use in two Southern French case study areas. The first of these, the Crau area in Provence, is specialized in using gravity irrigation to make the production of high-quality hay possible even during the arid summer months. The second area is a viticultural one, centred around the Canal de Gignac approximately 100 km further West, in which the majority of vines are sustained using drip irrigation, provided consistent water access is possible. In both cases, the study aimed first to identify irrigated plots, and then to further characterize the irrigation practices with regard to agricultural water use efficiency. 

The LULC analysis was carried out in Google Earth Engine, using a Gradient Tree Boosting (GTB) algorithm on combined Sentinel-1 and -2 imagery from which several spectral indices as well as Haralick texture features were calculated. The detection of irrigated grassland plots further relied on a temporal characterization of phenological stages. Subsequently, a comparative implementation of different irrigation monitoring approaches was carried out, using soil moisture estimates derived from Sentinel-1 and different optical spectral indices. Data from low-cost sensors and local water user associations was used for calibration and validation. 

Preliminary results indicate that combining these diverse approaches make an operational detection and monitoring of irrigation practices possible. For the detection of irrigated vineyard and grassland plots during the 2023 growing season, overall accuracies of 92% and 95% respectively were achieved. The comparison of different irrigation monitoring approaches showed that the Normalized Difference Moisture Index (NDMI, p=0.002), the Shortwave Infrared Water Stress Index (SIWSI, p=0.001) and the Specific Leaf Area Vegetation Index (SLAVI, p=0.001) showed the highest potential for accurate irrigation detection.