How well do gridded products represent soil moisture signatures in natural ecosystems during precipitation events?

Soil moisture (SM) is a key factor influencing the interactions between the atmosphere and processes at the Earth’s surface. Recent advances in remote sensing and land surface modelling have improved the estimation of soil moisture in ungauged areas. This study evaluates the performance of four state-of-the-art gridded SM products -SPL4SMAU, GLDAS, ERA5 and ERA5-Land- compared to in situ measurements at ten sites located in near-natural shrublanbd and native forest ecosystems of the semi-arid and humid regions of central and southern Chile (five in the semi-arid north and five in the humid south). The unbiased root mean square error (ubRMSE), Pearson’s product-moment correlation coefficient (r) and modified Kling-Gupta efficiency (KGE') were used as performance metrics to evaluate the representation of surface soil moisture (SSM) and root zone soil moisture (RZSM). In addition, event rising time (RT) and amplitude (A) were used as SM signatures to assess the dynamic aspects of the soil moisture time series and to enable process-based model evaluations. Our results show that SPL4SMAU achieves the lowest ubRMSE for both SSM and RZSM, especially in the northern region. However, ERA5 and ERA5-Land outperformed SPL4SMAU in terms of linear correlation and KGE', with particularly good results in the humid south. In terms of SM responses to the first precipitation event of the year, SSM amplitude was generally higher in the humid south, with SPL4SMAU and ERA5-Land very close to in situ values, while GLDAS showed a lower sensitivity to precipitation. As expected, all datasets showed a slower response for RZSM compared to SSM, with GLDAS showing the longest rising times in both regions. On the other hand, SPL4SMAU and GLDAS showed a stronger increase in SSM amplitude in the south for the most intense precipitation event of the year, while ERA5-Land showed more moderate rising times, which is consistent with the in-situ data. Overall, ERA5-Land and ERA5 proved to be reliable datasets for representing the spatio-temporal variability of SM in central and southern Chile, especially in the southern ecosystems, while SPL4SMAU performed well in terms of uRMSE but showed large variability in the other metrics analysed. We gratefully acknowledge the financial support of ANID-Fondecyt Regular 1212071, 1210932, ANID-PCI NSFC 190018, and ANID/FONDAP 1523A0002.