Researchers in Andalusia Develop AI Tool to Improve Irrigation Efficiency

Summary

Researchers at the University of Córdoba have devel­oped an AI tool, LSTMHybrid, that can pre­dict water needs for irri­ga­tion a week in advance with 80% accu­racy, aim­ing to help farm­ers save water and energy. The tool sim­pli­fies data input for farm­ers and irri­ga­tion man­agers, allow­ing for more pre­cise bud­get­ing of water needs and opti­miza­tion of water and energy costs.

Researchers from the University of Córdoba’s agron­omy depart­ment have devel­oped an arti­fi­cial intel­li­gence tool that will help farm­ers pre­dict how much water they need for irri­ga­tion a week in advance.

The researchers added that this lat­est tool, LSTMHybrid, is part of a broader effort to dig­i­tize irri­ga­tion, which they said will help farm­ers lower pro­duc­tion costs by sav­ing water and energy.

The lat­est tool is based on the Cangenfis model, devel­oped in 2021 and trained using four years of cli­matic data from Zújar in the Andalusian province of Granada. When deployed, it could pre­dict long-term water needs for irri­ga­tion with 80 per­cent accu­racy.

However, the first iter­a­tion of the AI-pow­ered tool could only pre­dict over­all water needs for var­i­ous crops, includ­ing rice, corn and toma­toes.

“The big dif­fer­ence with respect to pre­vi­ous mod­els is that it is the first time that it has been done on a seven-day scale,” said Rafael González, one of the three lead researchers involved in both projects.

LSTMHybrid allows farm­ers to bud­get their water needs more pre­cisely and over­lay expected irri­ga­tion require­ments with the dif­fer­ent tar­iff peri­ods. The hope from researchers is this more pre­cise data will help farm­ers make the most eco­nom­i­cally and agro­nom­i­cally informed deci­sions to opti­mize water and energy.

The need to mod­ern­ize Spain’s irri­ga­tion sys­tem, which researchers said has tra­di­tion­ally been guided by his­tor­i­cal expe­ri­ence and not pre­dic­tive data, has been made all the more nec­es­sary by the endur­ing drought and per­ilously low reser­voir lev­els.

While CANGENFIS used hun­dreds of neural net­works that take half a mil­lion dif­fer­ent fac­tors into con­sid­er­a­tion, LSTMHybrid makes its pre­dic­tions based on aver­age tem­per­a­ture, ref­er­ence evap­o­tran­spi­ra­tion, humid­ity and pre­vi­ous irri­ga­tion records.

The new model can also save pre­vi­ously-entered data to help improve its abil­ity to pre­dict year-over-year.

This sim­pli­fi­ca­tion allows farm­ers and irri­ga­tion man­agers to man­u­ally input weekly data into the sys­tem via an ordi­nary com­puter, pre­dict­ing how much water is needed for irri­ga­tion the fol­low­ing week.

“Knowing the demand for water sev­eral days in advance will facil­i­tate the man­age­ment of the sys­tem and will help to opti­mize the use of water and energy costs,” said Juan Antonio Rodríguez, another researcher involved in both projects.

Along with improv­ing water man­age­ment, Antonio Rodríguez added that the new pre­dic­tive capa­bil­ity would help the region’s tran­si­tion to renew­able energy by pro­vid­ing more accu­rate fore­casts for agri­cul­tural energy demand.

“The knowl­edge is there, and the tech­nol­ogy has been tested and works,” said the third lead researcher Emilio Camacho. ​“Now we have to develop the tool that allows the com­mu­ni­ties to use this tech­nol­ogy in a sim­ple way so that the com­pa­nies that are going to pro­vide the tech­no­log­i­cal solu­tion to the irri­ga­tion com­mu­nity intro­duce these advances.”

• University of Córdoba
• University of Córdoba