Developing a satellite-based frost risk model for the Southern African commercial forestry landscape

Abstract

Frost is a sporadic meteorological event affecting the productivity of commercial forests in South Africa. Severe frost occurrences may cause irreversible damage to forest stands, slowing down tree growth or leading to tree mortality. Using the Moderate Imaging Spectrometer (spatial resolution: 1 km by 1 km, swath: 1 200 km by 1 200 km) night-time land surface temperatures between 2002 and 2011, this study mapped frost risk classes using six satellite-derived variables at the landscape level. These variables included calculated thresholds of minimum temperature, probability of frost occurrence, mean temperature, total number of frost days, frost duration and the
frost severity index. Results show that, using an unsupervised random forest approach with partitioning around medoids, clustering was successful in mapping frost risk using eight optimal clusters. The methodology developed in this study contributes to building a robust frost-risk model to manage and mitigate forest frost damage.

Keywords: MODIS LST, frost risk, unsupervised random forests, PAM clustering