Using soil moisture remote sensing to assess systematic biases impacting long-term streamflow forecasts

Recent research has provided new insight regarding the strength of the link between antecedent soil moisture conditions and subsequent runoff generation. This insight has obvious implications for hydrologic forecasting - but can be properly leveraged only in cases where forecasting models accurately characterize the relationship between antecedent soil moisture and lagged runoff.

Specifically, there are two key processes that must be represented correctly by a hydrologic model. The first is the temporal autocorrelation (i.e., memory) of root-zone soil moisture (RZSM) anomalies relative to an expected seasonal cycle. The level of water storage capacity in the soil column root zone is potentially sufficient for such autocorrelation to persist over multiple months - and therefore contribute to the skill of long-term streamflow forecasts. The second is the link between future RZSM levels and the concurrent infiltration capacity of the soil column. As a result of this link, the capacity of the land surface to absorb rainfall or snow melt is progressively degraded as RZSM increases.

In this presentation, we will utilize new satellite-derived soil moisture products acquired from the NASA Soil Moisture Active/Passive (SMAP) mission to evaluate the ability of the National Water Model (NWM) to accurately capture both of these processes - and thus properly leverage the forecasting skill of its internal soil moisture predictions. Results demonstrate that relative to SMAP soil moisture and observed streamflow, the NWM tends to under-represent the overall lagged correlation between antecedent RZSM conditions and subsequent 90-days streamflow sums. Observed RZSM versus streamflow under coupling in the NWM is strongest for the trans-winter case of RZSM obtained during the late fall and streamflow sums obtained during the following springtime. This suggests that the NWM is missing, or perhaps mis-parameterizing, processes that could otherwise be exploited to leverage late-fall/early-winter RZSM values for seasonal streamflow forecasts extending into the following spring.

By applying a simple causal model, and a correction factor for attenuation bias in observed correlations, we will demonstrate that negative NWM soil moisture versus streamflow bias is due to the combined impact of both: i) a general underestimation of concurrent (i.e., non-lagged) RZSM versus streamflow coupling and ii) a seasonally varying bias in NWM RZSM memory (i.e., a tendency for the NWM to underestimate trans-winter RZSM memory). These systematic biases degrade the ability of the NWM to serve as the modelling core of a long-term, hydrologic forecasting system. Viewed more broadly, these results provide an important example of using satellite-based soil moisture retrievals as a diagnostic tool to detect, and correct for, systematic errors present in land-surface and hydrologic models.