Spatial and Temporal Validation of In-Situ and Satellite Weather Data for the South West Agricultural Region of Australia

Seasonal variations in weather have significant impacts on crop yields. The accuracy of weather data is therefore an important consideration for crop yield models. This study uses an independent in-situ weather station network to validate the accuracy of monthly temperature and precipitation data across the South West Agricultural Region of Australia from the in-situ weather station network operated by the Bureau of Meteorology (BOM), interpolated grids from this network and satellite weather data. This region covers five classes of the Koppen-Geiger climate classification system and is responsible for $10 billion AUD of agricultural produce annually. A strong bias was found for the maximum temperatures in the Copernicus Land Surface Temperature (LST) satellite product. This bias was linearly correlated to the in-situ temperature and exceeded 20C in warmer months. Due to the bias’s linear nature, a linear correction was able to reduce the RMSE of the Copernicus LST product by 82%. This process was tested for other regions of Australia. The initial bias correction reduced RMSE by 80% and region-specific bias correction reduced RMSE by 84%. The validation process demonstrated that BOM’s interpolated grid dataset reliably has the lowest RMSE. Nearest neighbour in-situ weather stations generally had the next lowest RMSE, followed by weather-station corrected satellite products and lastly the non-weather station corrected satellite products. While the gridded product generally had the lowest RMSE, there were spatial and seasonal variations. Monthly maximum temperatures were more accurately measured by the bias-corrected Copernicus LST product in the northern and eastern extents (where there is a lower density of in-situ stations). Copernicus LST monthly minimum temperatures had similar to slightly better RMSE than the gridded product for the southern half of the study area and the rain-gauge corrected Global Satellite Mapping of Precipitation product performed similarly to the gridded product in drier months.