PHASE: A geostatistical model for the Kriging-based spatial prediction of crop phenology using public phenological and climatological observations

Abstract Detailed information on plant developmental stages, referred as phenological phases, can assist research, applications and synergies e.g., in land use, climate science and remote sensing. Usually, detailed ground information about phenological phases is only available as point observations. However, in most application scenarios of spatially interpolated phenological information is required. In this article, we present an approach for modeling and interpolation of crop phenological phases in temperate climates on the example of the total area of Germany using statistical analysis and a Kriging prediction process. The presented model consists of two major parts. First, daily temperature observations are spatially interpolated to retrieve a countrywide temperature data set. Second, this temperature information is linked to the day of year on which a phenological event was observed by a governmental observation network. The accumulated temperature sum between sowing and observed phenological events is calculated. The day on which the temperature sum on any location exceeds a phase-specific critical temperature sum, which indicates the day of entry of the modeled phase, is finally interpolated to retrieve a countrywide data set of a specific phenological phase. The model was applied on the example of eight agricultural species including cereals, maize and root crops and 37 corresponding phases in 2011. The results for most of the tested crops and phases show significantly lower root mean squared errors (RMSE) values and higher goodness of fit ( R 2 ) values compared to results computed using Ordinary Kriging (OK) and Inverse Distance Weighting (IDW). The modeling accuracy varies between 2.14 days and 11.45 days for heading and emergence of winter wheat, respectively. The uncertainty of the majority of the modeled phases is less than a week. The model is universally applicable due to automatic parametrization, but model accuracies depend on the crop type and increase during a growing season. The possibility to enhance the model by additional explaining variables is demonstrated by consideration of soil moisture within an extended model setting.