Soil moisture is a key variable in regional water cycle and has important applications in water resources and agricultural drought management. Most global soil moisture products are retrieved from microwave remote sensing data. However, currently at the national scale, there is a lack of spatially clear, temporally continuous, and high-resolution soil moisture information. This dataset uses the Random Forest (RF) algorithm to generate a 1-kilometer soil moisture dataset (ChinaCropSM1 km) for dryland wheat and corn in China from 1993 to 2018 based on a large number of daily in-situ observations of soil moisture.
This data was independently trained using on-site observation data from the National Agricultural Meteorological Stations (AMSs) (181327 samples) (164202 samples), and tested using other data (17125 samples). Firstly, develop irrigation modules based on crop types (i.e. wheat, corn), soil depth (0-10 cm, 10-20 cm), and phenological characteristics. Four daily datasets were created based on crop type and soil depth, and their accuracy was satisfactory (wheat r 0.93, ubRMSE 0.033 m3 m − 3; Corn r 0.93, ubRMSE 0.035 m3 m − 3). The global soil moisture products, including the Surface Soil Moisture Dataset based on Global Remote Sensing (RSSSM) and the European Space Agency (ESA) Climate Change Initiative (CCI) SM, show significantly better spatiotemporal resolution and accuracy of ChinaCropSM1 km compared to the former (with a 116% increase in r and a 64% decrease in ubRMSE). This method can be applied to other regions and crops around the world, and the improved dataset has important value for research and field management such as agricultural drought monitoring and crop yield prediction.
On site observation data from National Agricultural Meteorological Stations (AMSs).
A 1-kilometer soil moisture dataset (ChinaCrop SM1 km) for dryland wheat and corn in China from 1993 to 2018 was generated based on daily in-situ observations of a large amount of soil moisture using the Random Forest (RF) algorithm.
The ChinaCropSM1 km predicted by the RF model is in good agreement with the in-situ SM observation results (ubRMSE range is 0.028-0.037, deviation range is -0.0011-0.0009, r range is 0.925-0.944, R2 range is 0.860-0.895).
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
CSTR:11738.11.NCDC.ZENODO.DB6453.2024
10.5281/zenodo.7582755