Clustering of Remotely Sensed Time Series using Functional Principal Component Analysis to Monitor Crops

The monitoring of cropland areas and in particular the capability to evaluate the performance of a field over space and time is becoming a crucial activity to schedule agronomic operations (e.g., fertilization) properly. In particular, the use of remotely sensed data opened new ways for this kind of analysis. In this work, we present a methodology based on Functional Data Analysis that starting from remotely sensed time-series data generates cluster maps of a cropland area. Starting from vegetation index time-series data, Functional Principal Component Analysis (FPCA) was applied to derive FPCA scores and components. FPCA scores are then clusterized to obtain maps that embed the dynamics of crops over space and time. The derived maps can be used to optimize agronomic tasks such as fertilization also acting as base layers to create management zones and then prescription maps.