Assessing how spatiotemporal fusion enables an agricultural drought index to represent ecosystem evaporative stress dynamics at continent-to-field scales

Remote sensing (RS)-based drought indices have been well explored and documented 1–4 , with the majority specifically designed for agricultural drought monitoring 5,6 . However, due to inherent trade-offs between spatial resolution and temporal frequency of RS data, it remains underexplored how an RS-based drought index could represent the variability of ecosystem evaporative stress (ES), which represents the observable but highly uncertain land surface fluxes as ecohydrological consequences of drought 7–9 . Here, we present a continental-to-field drought assessment for Australia, emphasising the use of spatiotemporal fusion 10,11 to downscale a modified Soil Moisture Agricultural Drought Index (SMADI) 12,13 to 100 m resolution at weekly frequency, thereby linking broad-scale drought monitoring with field-scale ES dynamics. Firstly, we integrated climatology-removed soil moisture (SM) 14 , land surface temperature (LST) 15 , and the Normalised Difference Vegetation Index (NDVI) 16 , to generate SMADI at three spatial resolutions (5 km, 100 m, and flux tower footprint) across Australia during 2016–2023. A baseline analysis at 5 km resolution showed moderate correlations between SMADI and a Standardised Precipitation Evapotranspiration Index (SPEI) 17 across Australia at multiple accumulation timescales, with the highest agreement observed at the 12-month scale (median absolute R = 0.57 for agricultural regions). Secondly, the downscaled SMADI (dSMADI), generated through unbiased spatiotemporal fusion 18 , captured weekly 100 m water stress variability during the 2019 “Tinderbox” drought in southeast Australia 19 , demonstrating spatial heterogeneity that was obscured in 5 km resolution SMADI. Thirdly, stratified analysis between dSMADI and the Ecosystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) ES index (ESI) 20 demonstrated spatial agreements and comparable spatial variability especially for irrigated croplands, while a temporal validation against flux tower observations 21,22 further confirmed the temporal reliability of dSMADI. Fourthly, and finally, a Taylor series-based variance decomposition 23 of dSMADI was implemented to quantify the relative importance of each component (condition indices of SM, LST and NDVI) in driving SMADI variability across spatial resolutions. Lagged vegetation response explained most SMADI variability in drylands 24 , where sparse vegetation has low values that, as denominators in SMADI, amplified the index sensitivity continentally. At field scales, the distribution of component contributions became more nuanced, where SM and LST generally shared comparably stronger contributions in ecosystems with constant access to water (e.g., evergreen vegetated systems). Together, these results demonstrated the broader value of spatiotemporal fusion for translating a coarse-resolution drought index into field-relevant information in heterogeneous, including irrigated, agricultural landscapes 25 , and positioned dSMADI as a scalable approach for continental-to-field agricultural drought monitoring.