We are now in an era where crop yields can be predicted from climate data.
With the development of AI and machine learning, it has become possible to predict future conditions from various data. Additionally, the evolution of measurement and recording devices has enabled us to obtain a vast amount of data. This is what is known as big data. By analyzing and combining these data, we can now make reasonable predictions about future conditions that were previously unthinkable. One of the most familiar examples is the “weather forecast.” Other fields, such as the advertising industry, are also making effective use of big data.
Although agriculture accounts for a small percentage of the world economy (4%), it is essential for human survival, and in developing countries, it makes up a large portion of the GDP. Predicting regional production is crucial for supporting stable crop production, and accurate predictions are significant for humanity from the perspective of appropriate distribution. The most suitable data for yield prediction is “weather data,” and several types of data frames aggregating past weather data are publicly available and can be used by anyone.
U.S. weather data is abundant and has high predictive accuracy
As a developed country, the United States has abundant weather data. Yield predictions using this data can predict yields with high accuracy. In the U.S., detailed yield data is also provided, so there is a wealth of both “material” and “results” for predictions, which leads to high predictive accuracy. Not only in the U.S., but in other developed countries as well, accurate statistical information is publicly available, so similar analyses and research are likely being conducted. Each region has its own parameters, and localized prediction tools are being developed worldwide.
Using U.S. models to predict yields in data-poor Africa
On the other hand, areas like Africa, which are concentrated with developing countries, lack devices and facilities to sample weather data, leading to lower quality weather data. Additionally, detailed yield data, which are the results of predictions, are not available, making machine learning for predictions difficult. In such cases, prediction models created in other areas are applied. At first glance, it may seem that the prediction accuracy would decline, but prediction models that capture the key points can provide a certain level of accuracy as long as the core data exists, even if the region changes. Even if the accuracy is somewhat lower, having predictions versus not having them makes a significant difference in responses. It seems that agriculture around the world is becoming more interconnected through data.
Increasing the resolution of the world
Some crops have higher yields when temperatures rise, while others have lower yields. Some crops’ yields change with increased humidity. In the field of crop production, there are various correlations. It is difficult to predict final results like yields by looking at individual data alone. By combining multiple pieces of information, discrepancies with the final results become smaller, and predictive accuracy increases. It feels as though the “resolution” of the world is increasing.
Let’s make use of big data!
コメント