Abstract: There is provided a computer implemented method of training a machine learning model, comprising: computing at least one in-colony feature indicative of an internal state of a sample honeybee colony positioned for pollination of at least one crop in a geographical area, from output of at least one internal sensor monitoring the honeybee colony, computing at least one out-colony feature indicative of an external environment of the sample honeybee colony, from output of at least one external sensor monitoring the environment of the honeybee colony, creating a multi-record training dataset, wherein a record comprises: a combination of at least one in-colony feature and at least one out-colony feature, and a ground truth label indicating pollination effectiveness of the sample honeybee colony, and training a machine learning model on the multi-record training dataset.