SpTimer: Spatio-Temporal Bayesian Modeling Using R

Despite a broad pattern of warming in minimum temperatures over the past 50 years, regions of southeastern Australia have experienced increases in frost frequency in recent decades, and more broadly across southern Australia, an extension of the frost window due to an earlier onset and later cessation. Consistent across southern Australia is a later cessation of frosts, with some areas of southeastern Australia experiencing the last frost an average 4 weeks later than in the 1960s (i.e. mean date of last frost for the period 1960–1970 was 19 Septemberversus22 October for the period 2000–2009). We seek to model the spatial changes in frosts for a region exhibiting the strongest individual station trends, i.e. northern Victoria and southern New South Wales. We identify statistically significant trends at low‐lying stations for the month of August and construct and validate a Bayesian space–time model of minimum temperatures, using rates of greenhouse gas (GHG) emissions, as well as other well‐understood causal factors including solar radiation, the El Niño Southern Oscillation (ENSO3.4) and times series data relating to the position (STRP) and intensity (STRI) of subtropical highs and blocking high pressure systems. We assess the performance of this modelling approach against observational records as well as against additive and linear regression modelling approaches using root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) as well as false alarm and hit rate metrics. The spatiotemporal modelling approach demonstrated considerably better predictive skill than the others, with enhanced performance across all the metrics analysed. This enhanced performance was consistent across each decade and for temperature extremes below 2 °C.

Skilful and localised daily weather forecasts for upcoming seasons are desired by climate-sensitive sectors. Various General circulation models routinely provide such long lead time ensemble forecasts, also known as seasonal climate forecasts (SCF), but require downscaling techniques to enhance their skills from historical observations. Traditional downscaling techniques, like quantile mapping (QM), learn empirical relationships from pre-engineered predictors. Deep-learning-based downscaling techniques automatically generate and select predictors but almost all of them focus on simplified situations where low-resolution images match well with high-resolution ones, which is not the case in ensemble forecasts. To downscale ensemble rainfall forecasts, we take a two-step procedure. We first choose a suitable deep learning model, very deep super-resolution (VDSR), from several outstanding candidates, based on an ensemble forecast skill metric, continuous ranked probability score (CRPS). Secondly, via incorporating other climate variables as extra input, we develop and finalise a very deep statistical downscaling (VDSD) model based on CRPS. Both VDSR and VDSD are tested on downscaling 60 km rainfall forecasts from the Australian Community Climate and Earth-System Simulator Seasonal model version 1 (ACCESS-S1) to 12 km with lead times up to 217 days. Leave-one-year-out testing results illustrate that VDSD has normally higher forecast accuracy and skill, measured by mean absolute error and CRPS respectively, than VDSR and QM. VDSD substantially improves ACCESS-S1 raw forecasts but does not always outperform climatology, a benchmark for SCFs. Many more research efforts are required on downscaling and climate modelling for skilful SCFs.