Climate affects how crops grow. To eliminate that effect and only study agricultural inputs, we divided each crop into climate zones or bins according to growing degree days (GDD) and precipitation. We then analyzed and modelled each bin individually.
Check out the climate bins of all crops below.
The effects of agricultural input shocks can also be examined through production, or total volume of cultivated crop.
Production is yield * harvested area. In this barplot you can examine shock effects on the global production of different crops individually.
In the tabs All crops you can see the effects on all 12 crops summed together. In the tab All countries production decreases are summed for whole countries.
Model performance can be measured by comparing results generated by the model to results we know previously. In this case, random forest modelled yields that we compared to known yields. NSE or Nash-Sutcliffe Efficiency is one measure of the models predicting power. A model with NSE score 1 is a perfect model, and models between 0-1 are acceptable. Models with NSE > 0.65 are good and NSE > 0.75 are very good.
RMSE or Root Mean Square Error is another measure of model performance. Here we see RMSE-values calculated from when training the model (train) and then again when testing with external data (test). If the RMSE-scores are not significantly different between train and test, the model does not overfit.