Spatiotemporal prediction of grass curing in Victoria, Australia using Multi-Source remote sensing and deep learning models

Degree of Curing (DoC) is an important ecological indicator for assessing grassland dryness and vegetation flammability, and accurate monitoring and prediction are critical for wildfire risk assessment in fire-prone regions. Focusing on Victoria, Australia, this study developed and compared several deep learning models for monthly spatiotemporal DoC prediction using multi-source remote sensing datasets derived from MODIS and ERA5-Land. Among the tested models, the Temporally Enhanced Convolutional Long Short-Term Memory model (TE-ConvLSTM) achieved the best overall performance, with an RMSE of 9.18 and an R 2 of 0.834. By integrating explicit temporal encoding and multi-scale spatial feature extraction, the model effectively captured seasonal DoC dynamics, more accurately reproduced elevated curing conditions and seasonal peaks during the fire-prone period and better represented spatial gradients and regional heterogeneity across the study area. The results demonstrate that the proposed temporally enhanced convolutional framework provides an effective approach for modeling spatiotemporal DoC dynamics and establishes a promising foundation for future investigations of predictive grass curing modelling for regional wildfire risk assessment and monitoring.