A while ago, I wrote a short article centered around this question “do plant-based irrigation scheduling methods work?” I would like to mention a few more words about this and also discuss what role AI (artificial intelligence) might play here.
My team and I have carried out extensive research on developing new plant-based water stress detection methods and evaluating irrigation scheduling algorithms for about eight years now (mostly at Washington State University). We followed in the footsteps of renowned scientists (reference in the field) from USDS-ARS at Bushland, TX who already had over 15 years of experience in the area of thermal sensing-based irrigation water management.
Interestingly enough, like many other researchers our goal was to replace soil moisture sensors with plant-based measurements. However, the more we tried the more we realized that at the end of the day everything goes back to what is happening in the soil. Our conclusion was that we always need soil water content as a variable in any irrigation scheduling algorithm!
What I have personally learned, after over 20 years of trying different plant-, soil and microclimate-based irrigation scheduling approaches, is that:
“A reliable irrigation scheduling algorithm looks at everything,”
meaning that soil moisture, plant parameters (e.g. surface temperature, water potential), and microclimate are all taken into consideration.
There are commercial systems that might measure all these variables, but they cannot do much because there is no algorithm running in the background to connect the dots. As I always say, raw data is useless! More raw data with no algorithm just means more headaches.
An algorithm here, if you’re wondering, is simply a set of rules and mathematical equations / models that help make sense of raw data. The output of an algorithm can be number(s), which can be used in decision-making.
There are people who claim to have developed artificial intelligence (AI) algorithms that solve all agricultural problems. AI ( > machine learning > deep learning) sounds very fancy, but are these claims true?
As fancy as it may sound, AI is not going to provide solutions to our problems especially if we don't know what our problems are (which is sadly true in most cases in agriculture)! I’m going to mention only one thing about AI (related to this writing):
"We are far from using artificial intelligence in many areas of agriculture, especially irrigation scheduling."
An AI needs training. We need to feed it with tons and tons of reliable data (especially in the case of deep learning), in order for it to work. An AI algorithm is nothing w/o good data. It is like a human with no education, training or experience, and like a human, if it’s fed by incorrect information it is going to be corrupt.
Another issue is that, we are talking about agriculture here. We do not have access to resources that, say, Google has. We are not dealing with Internet of Things here. We don’t have hundreds of millions of connected farms or plants (Internet of Plants - IoP).
In most cases, we might not have “enough data” to feed and train the algorithms. Before talking AI, we need to install as many sensors as we can in as many farms as we can. We need to collect as much crop data (e.g. yield) as we can.
Knowing that less than 11% of the U.S. agriculture has been exposed to soil and plant sensors, we need to be realistic here.
Talk AI, when you could Google plants on the Internet, when you could have “Internet of Plants”!
It may take decades to get to that point, but until then don’t take any Ag-AI claim seriously.