Sugarcane abandonment mapping in Rio de Janeiro state Brazil

The mapping of sugarcane plantations and their changes is relevant to the economy and the environment, notably due to sugarcane's interface in the biofuel industry through ethanol. The necessary mapping of sugarcane crop changes is especially challenging when plantation occurs under smallholders' land ownership structures used for heterogeneous crop management. We evaluated two approaches to address this challenge with the example of sugarcane abandonment in the Norte Fluminense Region, Northeastern Rio de Janeiro state, Brazil. The region is characterized by a large share of smallholders. We trained a random forest classifier for sugarcane for 2018 based on all available Landsat imagery. Based on the concept of temporal generalization, we applied the classifier trained in 2018 for the period from 1986 until 2020. The resulting annual sugarcane probability maps were used as input for two abandonment mapping methods: LandtrendR and muti-temporal cropland abandonment mapping. The performance of both approaches was evaluated based on a stratified sampling approach. We detected three distinct trajectories for sugarcane farmland: i) permanently abandoned sugarcane, ii) fallow sugarcane, and iii) and stable sugarcane. The multi-temporal cropland abandonment mapping performed better for the sugarcane abandonment class (F1 = 0.84) than the LandtrendR approach (F1 = 0.21). The LandtrendR results revealed a higher omission (PA = 0.12) in mapping the sugarcane abandonment class. For the multi-temporal cropland abandonment mapping, we found that 66% (67,353 ha) of the stable sugarcane areas were abandoned between 1990 and 2016. The highest abandonment rates occurred between 1990 and 1994 and between 2010 and 2016. The spatial distribution of abandonment was heterogeneous. The earliest abandonment was concentrated in the northern part of the study area. The most recent abandonment was more extensive in the southern part of the study site. Our results highlight the advantages and challenges of using Landsat time series to map sugarcane abandonment in a heterogeneous management system. Our results also highlight the spatially and temporal heterogeneous pattern of sugarcane abandonment in the region and provide the necessary database for subsequent studies to identify underlying and proximate causes for the abandonment. • Training in different sugarcane management patterns enables high accuracy mapping. • Sugarcane mapping model can be transferred across years to avoid sample collection. • The Multi-temporal mapping outperformed the LandtrendR method approach. • By 2018, more than half of the stable sugarcane were abandoned in Rio de Janeiro. • Sugarcane abandonment in Rio de Janeiro was heterogeneous in time and space.