Satellites read the moisture of the soil almost everywhere on land, but not under dense vegetation: over a forest, most of what they receive comes from the trees. This project developed a way of reading soil moisture there too. The method is published, and its daily maps of the globe are open.
At a glance
- Theme: Soil moisture under dense vegetation
- Observations: SMAP microwave brightness at two polarisations; ground stations for checking
- Approach: The Hybrid Dual-Channel Algorithm (HDCA): soil moisture and two properties of the canopy found together
- Product: Daily, 36 km, all land, April 2015 to December 2025
- Status: Published, data open

The basics
Everything on the Earth's surface gives off faint microwave radiation, and satellites such as NASA's SMAP measure it from orbit. Wet soil gives off less of it than dry soil, and that is how soil moisture is read from space. Over bare ground the reading is direct. Plants complicate it in two ways: they weaken the radiation that comes up from the soil, and they add radiation of their own, part of it scattered on the way. The satellite receives the sum of the two.

Over grassland and crops the plants' share is small and can be allowed for. A forest holds so much water in its trunks, branches and leaves that little of the soil's signal gets through, and the standard soil moisture maps mark their values there as unreliable. Yet forests cover almost a third of the land.
The challenge
To tell the soil from the trees, one has to know how strongly the canopy weakens the signal and how much of it the canopy scatters. The scattering is usually treated as a fixed property of each kind of vegetation. But a forest changes through the year: leaves grow and fall, and the trees hold more water in a wet season than in a dry one. Changes of the canopy that are not allowed for are mistaken for changes in the soil.
The idea
For each place and day the satellite delivers two numbers: the brightness of the ground at vertical and at horizontal polarisation. Three things are wanted from them: the soil moisture, how strongly the canopy weakens the signal, called the vegetation optical depth, and how much of it the canopy scatters, called the scattering albedo. Two measurements cannot settle three unknowns. The usual algorithms therefore set the scattering in advance, one value for each kind of vegetation, and hold it constant.
The Hybrid Dual-Channel Algorithm, HDCA for short, drops that assumption. It starts from the physics of how microwaves pass through a canopy and includes the radiation that has been scattered once on the way. From this follows a link between the two unknowns of the canopy: the ratio of scattering to weakening goes with how much the two polarisations differ, which the satellite measures anyway. With that link as a third equation, all three unknowns can be found for every day, and the scattering is free to change through the year.

The whole record is then read twice. The first pass gives what is usual for each place: ten-day means and the normal of each month. In the second pass each day is read again with those values as a guide, so that a single noisy measurement cannot pull the result far from the days around it. The second pass is the published product.
What changes on the map

The large pattern is the same in both: deserts dry, the humid tropics and the far north wet. The difference lies in the forests, where the new retrieval is drier. Part of what an algorithm with fixed scattering reads as soil that is always wet is in fact the canopy.
Checked against the ground

Soil moisture measured in the ground under forest is rare. Three sites allow a test. At Millbrook, where the forest is lighter, all products follow the seasons, and the new retrieval has the smallest bias, 0.008 m³/m³. At Harvard Forest the existing products are too wet by 0.18 to 0.30 and follow the ground only loosely, with correlations of 0.37 to 0.54; the new retrieval is too wet by 0.025 and reaches a correlation of 0.72. At Tambopata, in the rainforest, the existing products are too wet by 0.26 to 0.44, with correlations of 0.05 to 0.37; the new retrieval is too wet by 0.018 and reaches 0.51.
Away from forests little changes. At the core validation sites of the SMAP mission, most of them farmland and grassland, the new retrieval follows the ground as closely as the existing products and has the smallest random error, 0.031 m³/m³ on average.
Which part does what

Leaving parts of the algorithm out, one at a time, shows what each is for. Reading the record twice makes the retrieval follow the ground more closely from day to day and lowers its random error. Counting the radiation that was scattered once removes a bias of about 0.14 m³/m³, towards dry in this set-up. And letting the scattering change with time matters most where the canopy is densest, at Harvard Forest and Tambopata.
The data and their limits
The product gives soil moisture, vegetation optical depth and scattering albedo for every day on the satellite's 36 km grid, for all land, from April 2015 to December 2025. It is open under a CC BY 4.0 licence and can be explored on the lab's data portal.
Three cautions go with it. A cell of 36 km averages many kinds of land. The quality has been checked at the sites named above and is not verified elsewhere. And the scattering albedo, which makes the soil moisture better, is not yet a tested vegetation product in its own right: there are no measurements on the ground to compare it with, so the paper could check it only indirectly.
Why it matters
Forests store much of the carbon on land and return large amounts of water to the air. Knowing how wet their soils are helps to follow drought and fire risk, and to see how forests respond to a changing climate. It also fills the largest blank on the soil moisture maps that weather and water forecasts rely on.
The paper
Lee, J., Yueh, S., Entekhabi, D., Colliander, A., Im, J., Park, C.-H., and Kim, H. (2027). Soil moisture retrieval for dense vegetation considering time-varying first-order scattering from SMAP L-band dual-polarized brightness temperatures. Remote Sensing of Environment, 349, 115787. The work was done with colleagues at NASA's Jet Propulsion Laboratory, the Massachusetts Institute of Technology, the Finnish Meteorological Institute, UNIST and Ajou University.












