Can a sudden drop in production be predicted?

Something tremendously important in a crop as technified as #cotton.

⚠️Entre the 2020-21 and 2021-22 harvests, our client’s production fell from almost 6 T/ha to about 4, and in the case of zafrinha there was a decrease of 1.5 T/ha. Yields of zafras and zafrinhas range from 2 to 6 tons per hectare, with the former generally being the most productive.

Our team was assigned the task

👨💻 to generate a machine learning model (#ml ) that took into account elements such as #climate, soil and plant material. But first we had to diagnose the data and understand the specific dynamics of our client’s crop.

The effect of plant genetics carried little weight in our analysis, as 90% of the production came from only two varieties, with the one that accounted for almost 70% of the area being the most productive.

Even more remarkable was that the third most used variety (3.45% of the cultivated area) had very variable yields depending on the type of soil where it was located.

In addition, we took advantage of the data analysis to assess whether the crop coefficient we used for the irrigation recommendation, based on the studies developed by #fao The customer’s crop, it was in line with how the customer’s crop had performed historically, and we found that in order to be more efficient we had to develop a specific coefficient.

Identified

From the initial investigations, we identified that precipitation and temperature, as well as their temporal distribution, had been one of the main factors contributing to the decrease in production, and we saw how these circumstances were reflected in the spectral indices that we also introduced into the model.

Result

⭐️Con all this information we were able to train our model and identify the information that was really useful.
The ultimate goal of our work is that all this agronomic information can be transformed into very accurate #crop #predictions, which help to make decisions that protect the quantity and quality of production, with enough time to implement them.

At HEMAV we build custom models for each client, providing our machine learning with the best data quality, driving more responsible and #efficient productions for the future of the planet.