This study uses a random forest model to capture the highly non-linear relationship between the surface soil moisture and land surface features (and precipitation). In the end, to produce the long-term surface soil moisture at a global scale of 0.25 degrees.
This repo contains the sciprt to produce the insitu constrained raindom forest surface soil moisture, including:
-
Obtaining the in-situ surface soil moisture from International Soil Moisture Network (ISMN), data is available on the official website: https://ismn.geo.tuwien.ac.at/en/, the core package in this part of work: https://pypi.org/project/ismn/.
-
Downloading the land surface features from Google Earth Engine (GEE), including:
Land surface temperature MOD11A1
NDVI and EVI MOD13A1
Precipitation ECMWF/ERA5
And synchronizing the land surface (/atmosphere) features with the in-situ SSM in spatial- and temporal- resolution (daily, 1km).
-
Training and testing the Random Forest Model with 70% of the data, and validating and evaluating with the rest 30%.
-
Applying the Trained RF model on the gridded land surface features to get the long-term in-situ contained global surface soil moisture.
The author thanks R.Zhuang, Y.Zeng, B.szabo, S.Manfreda, Q.Han and Z.Su for their help with the result discussion.
LZhang ([email protected])