Inspiration
For a mix of reasons, people are looking to move away from almonds, the biggest crop in the entirety of California. Something has to be grown in its place, so we tried to see if we could find the best fits.
What it does
The first model predicts whether for the coming years, more precisely precipitation in this proof of concept. Our second model is trained to estimate yield (pounds per acre) based on the predicted weather forecast. This would be a powerful tool for people with economical and agricultural knowledge to make decisions on acreage.
How we built it
We used a recurrent neural network long short term memory model (or RNN LSTM) for the weather prediction, we fed up to 70 years’ worth of monthly precipitation data into the model, gathered from stations around the San Joaquin Valley. For yield prediction, we used a Random Forest model to predict yield based on historical yield data and the corresponding historical weather data. Then we had it estimate over the prediction of the first model. Our model takes the unweighted average of all decision trees within the Random Forest.
Challenges we ran into
Data collection takes time and old records are tricky to manipulate and interpret. Machine Learning is of the same quality of the data it takes in, and feeding it quality data was the central struggle for our project. Making the models work with the data we have was also a challenge, since the timestamps for weather collection and crop reports are different.
Accomplishments that we're proud of
This is a good proof of concept if we may say so ourselves. Weather prediction is always tricky. The Random Forest was really strong and we believe we avoided overfitting for the most part.
What we learned
We became more familiar with AI models that some of us had not worked with before. We learned that our idea is definitely something that can become a reality, and the most crucial part of it is data and people who can work with it.
What's next for San Joaquin Valley Weather and Crop Prospects
San Joaquin Valley is going to make a decision, AI or not, we can only try to make our part to ensure we walk to the future with confidence and knowledge and we believe this can be an important part of that.
Log in or sign up for Devpost to join the conversation.