Predictive power of remote sensing versus temperature‐derived variables in modelling phenology of herbivorous insects

Abstract

AbstractApplication of remote sensing datasets in modelling phenology of heterotrophic animals has received little attention. In this work, we compare the predictive power of remote sensing versus temperature‐derived variables in modelling peak flight periods of herbivorous insects, as exemplified by nocturnal moths. Moth phenology observations consisted of weekly observations of five focal moth species (Orthosia gothica, Ectropis crepuscularia, Cabera exanthemata, Dysstroma citrata and Operophtera brumata) gathered in a national moth monitoring scheme in Finland. These species were common and widespread and had peak flight periods in different seasons. Temperature‐derived data were represented by weekly accumulating growing degree days (GDD) calculated from gridded temperature observations. Remote sensing data were obtained from three sources: (1) snow melt‐off date from the MODIS daily snow maps, (2) greening date using the NDWI from MODIS data and (3) dates of start, maximum and end of growing season based on the JRC FAPAR products. Peak phenology observations of moths were related to different explanatory variables by using linear mixed effect models (LMM), with 70% of the data randomly selected for model calibration. Predictive power of models was tested using the remaining 30% of the data. Remote sensing data (snow melt‐off and vegetation greening date) showed the highest predictive power in two moth species flying in the early and late spring, whereas in the three other species none of the variables showed reasonable predictive power. Flight period of the spring species coincides with natural events such as snow melt or vegetation greening that can easily be observed using remote sensing techniques. We demonstrate the applicability of our methodology by predictive spatial maps of peak flight phenology covering the entire Finland for two of the focal species. The methods are applicable in situations that require spatial predictions of animal activity, such as the management of populations of insect pest species.