Improving NASA SMAP Soil Moisture Using Upper Missouri River Basin Data
In response to severe drought and flood events in the Upper Missouri River Basin (UMRB) between 2011 and 2019, the Infrastructure Investment and Jobs Act provided funding to improve water monitoring in the region. The UMRB Data Value Study, led by NOAA’s National Integrated Drought Information System (NIDIS), is the assessment element of this multi-component, multi-agency project.
This project, Improving NASA SMAP Soil Moisture Using Upper Missouri River Basin Data, was led by NASA’s Goddard Space Flight Center as a component of the UMRB Data Value Study. This project explored how data from new, higher-density soil moisture stations in the UMRB could enhance soil moisture products derived from NASA’s Soil Moisture Active Passive (SMAP) satellite. The researchers also described potential uses for these more accurate gridded soil moisture products to enhance land surface models, which can lead to improved drought and flood monitoring.
Project Approach and Findings
Calibration and validation of satellite-derived products is more effective when multiple in situ stations are present at the scale of product’s spatial resolution (grid cell size). However, in situ soil moisture stations are often sparse relative to land area. Beginning in 2022, the U.S. Army Corps of Engineers is leading a build-out of hundreds of new mesonet stations across the UMRB. Even mid-way through the installation process, the UMRB station build-out provides a unique opportunity to assess the accuracy of NASA SMAP’s soil moisture estimations and improve land surface models that simulate soil moisture and the water cycle in the region.
This project found assimilating in situ UMRB soil moisture data to correct SMAP satellite soil moisture estimates improved the accuracy of these products, which are commonly used in land surface models for tracking drought or forecasting floods. Land surface model products that used the in situ-adjusted SMAP soil moisture inputs performed better at representing real-world conditions than model products that used un-adjusted SMAP soil moisture estimates as an input.
Two tests were run to determine impacts of using in situ-adjusted SMAP data in land surface models for drought and flood monitoring. First, a test run simulating regional flooding in 2019 in South Dakota using in situ-adjusted SMAP soil moisture products in land surface models improved the skill of model-generated streamflow forecasts. Using adjusted soil moisture data and the Normalized Difference Vegetation Index (NDVI) to track drought using the relationship between soil moisture and vegetation health had mixed results. In many regions of the UMRB, the land surface model’s performance in representing drought improved, but in other parts of the basin, skill decreased when the in situ-adjusted SMAP products were used. This is potentially a consequence of inadequate periods of record for soil moisture data for regions where skill decreased.
As a separate test, the research team used in situ soil moisture data to train a Random Forest model—a machine learning method that builds many decision trees, collectively called a “forest,” to make a prediction. The trained model was then used to generate proxy data for missing soil moisture depths and for days where there were gaps in in situ records, leveraging soil moisture data from existing UMRB stations to create more complete records at desired depths.
For more information, please contact Elise Osenga (elise.osenga@noaa.gov).
Research Snapshot
John Bolton, NASA Goddard Space Flight Center
Manh-Hung Le, Science Applications International Corp, NASA Goddard Space Flight Center; Kristen Whitney, Earth System Science Interdisciplinary Center (ESSIC), NASA Goddard Space Flight Center
Key Findings from This Research
- Because of its station density, scientists can use the recently expanded UMRB mesonet network for satellite calibration and validation research.
- Using in situ soil moisture data to improve inputs to land surface models can reduce model uncertainty for products and forecasts related to drought and flood events.
- In this analysis, actual in situ data (where available) improved model accuracy more than the Random Forest proxy soil moisture data. However, both improved land surface model accuracy, and the Random Forest modeled data was helpful if no in situ data were available at a given depth.
- For areas where multiple stations were located within a SMAP grid cell, links between soil moisture and vegetation health were stronger when using the in situ-adjusted SMAP soil moisture products versus the unadjusted SMAP products. Tracking and understanding links between soil moisture and plant health could better support informed decision-making and communications around fire risk, forage production, and ecological drought.
- The team also developed a map indicating relative wetness or dryness of each grid cell where a station is located. This reveals how distinct soil moisture conditions at an individual station are relative to stations on adjacent land areas. This comparison can provide context for existing networks to interpret regional soil moisture conditions. Similar maps could also help identify key regions of interest for future soil moisture monitoring in the UMRB or elsewhere.
