Improving remote sensing of extreme events with machine learning: land surface temperature retrievals from IASI observations

Abstract Retrieving weather extremes from observations is critical for forecasting and climate impact studies. Statistical machine learning methods are increasingly popular in the remote sensing community. However, these models act as regression tools when dealing with problems such, they not always well-suited estimation of extreme states. This study firstly introduces two error types that arise such statistical methods: (a) ‘dampening’ refers to reduction range variability retrieved values, a natural behavior models; (b) ‘inflating’ opposite effect (i.e. larger ranges) due data pooling. We then introduce concept localization intends better take into account local conditions model. Localization largely improves retrievals states, can be used both retrieval at pixel level or image processing techniques. approach tested on land surface temperature using infrared atmospheric sounding interferometer observations: dampening reduced 1.9 K 1.6 K, inflating 1.1 0.5 respectively.