The Accuracy of Self-Reported Crop Yield Estimates and Their Ability to Train Remote Sensing Algorithms

Despite recent advances in remote sensing, one of the major constraints that still remains is collecting the ground data needed to calibrate and validate remote sensing algorithms at large spatial and temporal scales. This is particularly challenging when mapping continuous variables such as yield, where calibration data often do not exist at the field-scale and are difficult to obtain through visual interpretation of high-resolution imagery. While crop cut estimates of crop yield are widely used to calibrate satellite yield estimation models, these data are time and cost intensive to collect. In this study, we examine the ability of self-reported yield estimates, which are much faster and easier to collect at large scales, to train satellite yield estimation models. We assess the accuracy of self-reported yield data and identify whether it is possible to increase self-reported accuracy by providing more information to farmers about the study design and potential benefits. Our results showed that farmers' self-reported crop yields were not accurate, and that self-reported crop yields led to inaccurate satellite yield estimation models when used for calibration. We also found that providing more information to farmers about the study design and benefits of satellite yield estimation did not improve self-reported accuracy. These results suggest that even though self-reported yield estimates may be a faster and lower cost way to collect field-level yield estimates, they likely are not an adequate data source to train satellite yield prediction models and should be used with caution.