Mapping Phenological Functional Types (PhFT) in the Indian Eastern Himalayas using machine learning algorithm in Google Earth Engine

Abstract

Phenological studies involve capturing information on dates of recurrent and seasonal biological events in plants and animals. In plants, phenological events of leaf flushing, full bloom, autumn discolouration, leaf fall etc. can be used to distinguish vegetation into separate classes. We refer here such phenological distinction as Phenological Functional Types (PhFT). The PhFT can be considered as the precursor of Plant Functional Types (PFTs) where PFT uses additional traits (e.g. leaf area, tree height, leaf structure, rate of evapotranspiration and photosynthesis, etc.) to classify vegetation. The PFT classification is essentially needed for developing and running dynamic global vegetation models (DGVMs) used in studying impacts of climate change on vegetation. We used archived long time series satellite remote sensing data, Landsat 5,7 and 8 (1985–2019), for classifying a landscape into appropriate PhFT. Normalized Difference Vegetation Index (NDVI) was calculated for the given period in the Google Earth Engine (GEE). The GEE is a cloud-based platform developed for the retrieval and processing of remotely sensed images and other data. The monthly median values of NDVI for the mentioned period was used to label each pixel into appropriate PhFT classes using Random Forest (RF) algorithm in GEE to obtain four distinct classes of evergreen forest, deciduous forest, agriculture and non-forest. The comparison of PhFT map was done with reference maps of global MCD12Q1 and the forest type map of India; with an overall moderate agreement of 68.55% and 66.22%, respectively. MCD12Q1 has an accuracy of 73% for the land cover map and the forest type map of India has 75% accuracy, whereas, we achieved an overall accuracy of 78%. The PhFT classification accuracy can further be improved using additional indices and topographic variables. The methodology demonstrated in this study can be adopted for classifying a landscape into distinct PhFT/PFT classes.