基于稀疏样点的南方丘陵地区耕地土壤有效磷制图

Mapping soil available phosphorus of cultivated land in hilly region of southern China based on sparse samples

绘制耕地表层土壤有效磷空间分布图对精准农业管理和土壤环境评估具有重要意义。目前土壤磷数字制图研究大多面向充足土壤样点的平坦地区,基于稀疏样点的南方丘陵地区耕地土壤有效磷制图效果尚不清楚。本文以典型南方丘陵地区福建省建瓯市为研究对象,基于96个稀疏土壤实测样点,利用空间分辨率为10 m的Sentinel-2遥感影像获取的遥感变量,联合气象变量和地形变量建立随机森林(Random Forest, RF)模型预测建瓯市耕地表层土壤(0~20 cm)有效磷含量,并对比5种不同环境变量组合下的RF模型精度。结果表明,加入遥感变量后,地形、气象和pH组合的RF模型预测有效磷含量的精度显著提升[决定系数(R;)从0.36提升至0.59],联合全部变量(遥感、地形、气象和土壤pH)的RF模型预测精度最佳。遥感变量、气象变量、地形变量和土壤pH分别可以解释土壤有效磷含量的22.87%、30.64%、30.38%和16.11%,其中年均温、pH、地形湿度指数和高程是影响南方丘陵地区耕地土壤有效磷空间分布的主导因素。因此,利用遥感、气象、地形和土壤pH组合的RF模型是样点数量有限情况下预测南方丘陵地区县市域耕地土壤有效磷含量的有效方法。

Spatial distribution mapping of topsoil available phosphorus content of cultivated land is essential for precise agricultural management and soil environmental assessment. Most research has focused on sufficient soil samples to map topsoil phosphorus content of cultivated land in flat areas. However, there are few studies on soil available phosphorus mapping based on sparse samples in the hilly areas of southern China. Jian’ou City was selected as the study area, which is a hilly area in southern China and has the largest cultivated land area among all county-level cities in Fujian Province. A total of 96 soil measurements, Sentinel-2 remote sensing data with a spatial resolution of 10 m, and climate and topographical variables were used to predict topsoil(0–20 cm) available phosphorus content. Random forest models with five combinations of environmental variables were constructed and their performance was compared for model prediction. Three assessment criteria, namely the coefficient of determination(R;), mean absolute error(MAE), and root mean square error(RMSE) were used to evaluate the performance of random forest models for five combinations of environmental variables. The results showed that the prediction accuracy of the random forest model using climate variables, topographic variables, and soil pH values significantly improved after adding remote sensing variables, with an R;increase from 0.36 to 0.59 and an RMSE decrease of 20.34%. In addition, the random forest model using all variables(remote sensing, topography, climate,and soil pH) obtained the optimal performance(R;= 0.59, MAE = 19.04 mg·kg;, RMSE = 25.26 mg·kg;) among five combinations of environmental variables. Therefore, remote sensing variables are of great value for the mapping of soil available phosphorus based on sparse samples, and we suggested that the use of remote sensing variables should be increased in future studies to improve prediction accuracy. Remote sensing variables, climate variables, topographic variables, and soil pH could explain 22.87%, 30.64%,30.38%, and 16.11% of the topsoil available phosphorus content, respectively. Furthermore, the spatial distribution of soil available phosphorus content in the study area was found to be mainly affected by the mean annual temperature, soil pH, soil moisture index,and elevation. The spatial distribution maps of soil available phosphorus content by the five random forest models were similar. High values of topsoil available phosphorus content were distributed in the central and western regions, whereas low values were distributed in the eastern and southern regions. The spatial variation of the soil available phosphorus in the distribution map produced by the optimal random forest model with total environmental variables was the most precise. Therefore, a random forest model that uses all variables(i.e., soil pH, topographic, remote sensing, and climate) can be used as a robust method to resolve soil available phosphorus content mapping with sparse soil samples in the hilly regions of southern China. Thus, this research can provide some guidance for other researchers interested in mapping the soil available phosphorus content in the hilly regions of China.