InternationalJournalofBiometeorologyhttps://doi.org/10.1007/s00484-018-1534-2ORIGINAL PAPERMachinelearningmodelingofplantphenologybasedoncouplingsatelliteandgriddedmeteorologicaldatasetBartosz Czernecki 1 ·Jakub Nowosad 2 ·Katarzyna Jab?o´ nska 3Received:12October2017/Revised:23February2018/Accepted:22March2018©TheAuthor(s)2018AbstractChanges in the timing of plant phenological phases are important proxies in contemporary climate research. However,most of the commonly used traditional phenological observations do not give any coherent spatial information. Whileconsistentspatialdatacanbeobtainedfromairbornesensorsandpreprocessedgriddedmeteorologicaldata,notmanystudiesrobustly benefit from these data sources. Therefore, the main aim of this study is to create and evaluate different statisticalmodels for reconstructing, predicting, and improving quality of phenological phases monitoring with the use of satellite andmeteorological products. A quality-controlled dataset of the 13 BBCH plant phenophases in Poland was collected for theperiod 2007–2014. For each phenophase, statistical models were built using the most commonly applied regression-basedmachine learning techniques, such as multiple linear regression, lasso, principal component regression, generalized boostedmodels, and random forest. The quality of the models was estimated using a k-fold cross-validation. The obtained resultsshowed varying potential for coupling meteorological derived indices with remote sensing products in terms of phenologicalmodeling; however, application of both data sources improves models’ accuracy from 0.6 to 4.6 day in terms of obtainedRMSE. It is shown that a robust prediction of early phenological phases is mostly related to meteorological indices, whereasfor autumn phenophases, there is a stronger information signal provided by satellite-derived vegetation metrics. Choosing aspecific set of predictors and applying a robust preprocessing procedures is more important for final results than the selectionof a particular statistical model. The average RMSE for the best models of all phenophases is 6.3, while the individualRMSE vary seasonally from 3.5 to 10 days. Models give reliable proxy for ground observations with RMSE below 5 daysfor early spring and late spring phenophases. For other phenophases, RMSE are higher and rise up to 9–10 days in the caseof the earliest spring phenophases.Keywords Phenophase · Phenology modeling · BBCH scale · Machine learning · MODIS · E-OBSIntroductionPhenologyoftheplantsismainlyinfluencedbyphotoperiodand temperature (Swanton et al. 2000). Previous studies? Bartosz Czerneckinwp@amu.edu.pl1Department of Climatology, Faculty of Geographicaland Geological Sciences, Adam Mickiewicz University,Krygowskiego 10, 61 680 Pozna´ n, Poland2Space Informatics Lab, Department of Geography and GIS,University of Cincinnati, 219 Braunstein Hall, Cincinnati,OH, 45221, USA3Institute of Meteorology and Water Management - NationalResearch Institute, Podle´ sna 61, 01 673 Warsaw, Polandhave shown that global warming determines the advanceof phenological events (Bradley et al. 1999; Root et al.2003; Menzel et al. 2006; Parmesan 2006; Cleland et al.2007), and some of the plants currently approach theirphysiological limits (Iler et al. 2013). The consequencesof changes in plant phenology due to climate change cancreate more feedbacks that alter biogeochemical cyclingand species interactions (Melillo 2014), and may affectall Earth’s spheres (Elmendorf et al. 2016). Monitoringof phenological processes and plant reaction to currentlyobserved climate change is therefore of high importance.Changesintimingofphenologicalphasesarealsoimportantproxies in contemporary climate research, such thatphenological data are commonly used in the reconstructionof long-time temperature time-series due to its longercoverage compared to instrumental observations (Schleip
