This paper attempts to predict average daily bicycle volumes on a nationwide level using crowdsourced GPS data from the CITYCYCLING campaign in Germany. The data source was 514 permanent counting sites across the country as well as the campaign-generated GPS bicycle volumes of about 300,000 participants and 7.5 million tracks from a smartphone app. For model building, Gradient Boosting Regression and Support Vector Regression were selected. The results show a medium to high model fit for the prediction of bicycle volumes at sites with permanent counters. To illustrate this, the models are applied to the road network of a district of the city of Dresden, Germany.