Time window and influencing factors analysis of tillage soil texture remote sensing in the typical black soil region

Abstract: Spatial distribution of cultivated soil texture is generally crucial to the precision management of agriculture and farmland protection in black soil areas. Remote sensing technology can be an effective way to rapidly acquire the spatial distribution of soil texture. Taking the Youyi Farm in Heilongjiang Province of China as the study area, this study aims to evaluate the best time window for the remote sensing inversion of soil texture in the cultivated land. 25 Sentinel-2 images were selected in the study area from 2019 to 2021. The band and spectral index of each image were input into the random forest model, in order to establish a remote sensing inversion model of soil texture. A comparison was made on the model accuracy of soil texture inversion from the images in different periods. The most suitable image was determined for the remote sensing inversion of soil texture. The spatial distribution of soil texture was obtained to evaluate the accuracy of soil texture inversion from the Sentinel-2 multi-spectral images on different dates. The results showed that: 1) From the visible light to short wave infrared 1 (1 565-1 655nm) spectral reflectance increased with the increase of wavelength, from the short wave infrared 1 to short wave infrared 2 (2 100-2 280nm) spectral reflectance decreased significantly. The spectral reflectance decreased with the increase of silt and clay content, whereas there was an increase with the increase of sand content. 2) There was the maximum accuracy of inversion model in the silt and sand on May 7, 2020 (the coefficient of determination (R2) of silt was 0.785 in 25 Sentinel-2 images, and Root Mean Square Error (RMSE) was 6.697%; the R2 of sand was 0.776, and RMSE was 8.296 %). The maximum accuracy was also achieved in the clay inversion model on May 3, 2019 (R2=0.776, RMSE=1.600%). 3) The appropriate time of satellite images was selected as an important impact on soil texture inversion. The best time window was from late April to mid-May in the study area. The time window and good-quality spectral data were obtained to develop a stable spectral model for the spatial distribution of soil texture. 4) Different inversion accuracy of soil texture were obtained using the 25 Sentinel-2 images from 2019 to 2021. This data was attributed to the soil water content and straw mulching. 5) The silt and clay particles were distributed more in the northeast, north, and south, and less in the middle and southwest of the study area. There was the opposite trend of the sand, especially the generally high sand content in the middle of the study area. Therefore, the high-precision remote sensing inversion model was achieved in the soil texture. The finding can provide the key technologies for regional soil texture mapping and farmland protection in the black soil areas.