Arctic sea ice thickness (SIT) have been mostly retrieved from microwaved and visible altimeters since the 2000s.
However, the repeatability of altimeters and their spatial coverage limit SIT estimates spatially and temporally.
On the other hand, the passive microwave (PMW) radiometer have daily basin-scale coverage of the Arctic.
In this study, we proposed a SIT retrieval from PMW observations, based on a statistical inversion technique.
It is based on the evidence of hig correlations between PMW observations and existing altimetric satellite-derived SIT, especially at 36 GHz.
Lidar ICESat-2 SIT products were used to train a neural network with multiple combinations of brightness temperatures between 1.4 and 36 GHz as inputs over the 2018-2019 time period. The PMW retrieved SIT can mimic the lidar SIT product over the full winter over the Arctic, with a correlation of 0.85, and a RMSE of 0.54 cm.
Results were also compared with the altimeter CS2SMOS and the Nucleus for European Modelling of the Ocean (NEMO) SIT products and with the Operation IceBridge QuickLook SIT measurements.
The Neural Network (NN) SIT retrieval with all frequencies from 1.4 to 36 GHz has good performance, a correlation of 0.72 and a RMSE of 57 cm when compared to OIB-QL measurements, for large sea ice thickness (mostly above 3 m), under multi-year ice environments.
The NN SIT retrieval using only 18 and 36 GHz has also shown satisfactory performances, paving the way for the creation of long time series, these two microwave channels being available since the 1980s.