The growing demands on agriculture, coupled with the climate crisis, are pressurising Earth’s valuable freshwater resources and making assessment of water practices ever more important. A study combining data from ESA’s SMOS mission with those from the Copernicus Sentinel-2 and Sentinel-3 missions, enables accurate tracking of irrigation on a field level to help improve agricultural water management.

Modern agricultural practises require efficient water irrigation systems, ideally with a move away from water intensive flood irrigation towards more targeted approaches, such as drip irrigation methods. A key part of modernising irrigation systems, however, is gaining accurate information on current water management methods.

Researchers at isardSAT and the Spanish Institute of Agrifood Research and Technology (IRTA), have used remote sensing data from multiple ESA Earth observation missions to classify irrigation systems on a field level irrespective of crop type [1]. This new method availed of a machine learning model (ResNet) using remote sensing data over the intensively farmed Catalan region of Lleida in Spain, to produce accurate maps of irrigation systems. The maps show a shift towards drip and sprinkle irrigation approaches in the region, previously unseen in local water management databases. 

Maria José Escorihuela, Principal Investigator at isardSAT, says, “Our model presents a useful tool at the administrative level to monitor changes in irrigation systems and help modernise water management methods. The precision of our irrigation maps was only possible by combining high resolution soil moisture data from SMOS with evapotranspiration data from Sentinel-2 and Sentinel-3.”

There are three main types of irrigation systems: flood, sprinkle and drip methods. Flood irrigation is a traditional and water intensive method, which relies on flooding the field every few weeks and can be easily detected by Earth observation soil moisture data. 

Sprinkling systems - adding water to the field daily – are more efficient but also use a large excess of water and result in detectable soil moisture changes. On the other hand, the water efficient drip method is very localised and currently difficult to detect with soil moisture data, which is why the new model combined SMOS soil moisture data with evapotranspiration data from the Sentinel-2 and Sentinel-3 satellites of the European Union’s Copernicus Programme. 

“Soil moisture data tell us what water the farmer puts on their land, while evapotranspiration data tell us about the transpiration of water from the plant. Evapotranspiration is high when the plant is in good health and well-watered, even when the soil is not moist. By using a synergy of these data our model could help us distinguish between all three types of irrigation, as well as non-irrigated fields,” explains Escorihuela.

The SMOS satellite is one of ESA’s Earth Explorer missions, a family of missions designed for research purposes to monitor parts of Earth’s system. SMOS hosts the microwave imaging radiometer with aperture synthesis (MIRAS) instrument, which tracks variations in soil moisture by monitoring the L-band microwave radiation emitted over land. 

In this research, a study area was selected in the northeast of the Iberian Peninsula in the agricultural region of Lleida, where irrigation usually starts in mid-March and lasts until the end of November. The study area was divided into eight irrigation districts, covering a total surface area of around 3000 km2. The ResNet machine learning models were trained with ground truth data from more than 300 fields within this area, collected during a field campaign in 2020. Results showed good precision in predicting irrigation types compared to field data. 

Experiments were also run to examine if differences in crop types – varying from winter cereals, alfalfa, olives to fruit and nut trees – had any bearing on the prediction of irrigation systems. Results showed that the model performed equally well in classifying irrigation systems for all crop types. 

References

[1] G. Paolini, et al, “Classification of Different Irrigation Systems at Field Scale Using Time-Series of Remote Sensing Data,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 10055-10072, 2022, DOI: https://doi.org/10.1109/JSTARS.2022.3222884