基于CSSA-RF模型的小流域土壤pH空间分布预测

Spatial Prediction of Soil pH Distribution in Small Watershed Based on CSSA-RF Modelling

    摘要
  • 摘要:
    目的 为提升小流域土壤pH的空间预测精度，建立高效、准确的土壤pH空间预测模型。
    方法 在重庆市涪陵区渠溪小流域采集137个0 ~ 20 cm土壤表层样本，测定其pH值。结合16个关键环境因子，通过随机抽样，将70%的样本作为训练集，采用鸡群算法（CSO）与麻雀搜索算法（SSA）相融合的策略对随机森林超参数进行优化，分别构建了多元线性回归（MLR）、普通克里格（OK）、支持向量机（SVM）、随机森林（RF）、麻雀搜索算法优化的随机森林（SSA-RF）、改进麻雀搜索算法优化的随机森林（CSSA-RF）土壤pH空间预测模型，并使用30%的样本作为测试集检验模型精度，以平均绝对误差（MAE）、均方根误差（RMSE）、决定系数（R2）和Lin协调相关系数（CCC）作为模型精度的评价指标。
    结果 6种土壤pH空间预测方法在测试集上的RMSE排序为CSSA-RF < SSA-RF < RF < SVM < OK < MLR，MAE排序为SVM < RF < CSSA-RF < SSA-RF <OK < MLR，CCC排序为CSSA-RF > SSA-RF > SVM > RF > OK > MLR，R2排序为CSSA-RF > SSA-RF > RF > SVM > OK > MLR。其中CSSA-RF模型展现出最佳的预测精度和泛化能力，其测试集的RMSE、MAE、CCC和R2分别为0.3563、0.2923和0.8828和0.8142，模型的预测精度最高。
    结论 在土壤样点密度稀疏的小流域尺度下，辅以环境协变量的CSSA-RF模型其土壤pH的空间预测精度最高，较于其他未优化的模型预测性能更佳。

     

    Abstract:
    Objective The aim was to improve the spatial prediction accuracy of soil pH in small watersheds and to establish an efficient and accurate spatial prediction model for soil pH.
    Method Total 137 soil surface samples of 0 ~ 20 cm were collected in the Quxi small watershed of Fuling district, Chongqing, and their pH values were measured. Combining 16 key environmental factors, 70% of the samples were randomly selected as the training set. The hyperparameters of the random forest (RF) were optimized by the strategy of integrating the chicken swarm optimization (CSO) and the sparrow search algorithm (SSA). Six soil pH spatial prediction models, namely Multiple Linear Regression (MLR), Ordinary Kriging (OK), Support Vector Machine (SVM), Random Forest (RF), Random Forest for optimization of Sparrow search algorithm (SSA-RF), and Random Forest for optimization of Chicken Sparrow Search Algorithm (CSSA-RF), were constructed. And 30% of the samples were used as the test set to evaluate the model accuracy, with mean absolute error (MAE), root mean square error (RMSE), coefficient of determination (R2), and Lin’s coordinated correlation coefficient (CCC) as the evaluation indicators of the model accuracy.
    Result The RMSE ranking of the six soil pH space prediction methods on the test set was CSSA-RF < SSA-RF < RF < SVM <OK < MLR, and the ranking of MAE was SVM < RF < CSSA-RF < SSA-RF <OK < MLR. The sorting of CCC was CSSA-RF > SSA-RF > SVM > RF > OK > MLR, and the sorting of R2 was CSSA-RF > SSA-RF > RF > SVM > OK > MLR. Among them, the CSSA-RF model demonstrated the best prediction accuracy and generalization ability. The RMSE, MAE, CCC and R2 of its test set were 0.3563, 0.2923 and 0.8828 and 0.8142 respectively, and the prediction accuracy of the model was the highest.
    Conclusion At the small watershed scale with sparse soil sample point density, the CSSA-RF model supplemented with environmental covariates has the highest spatial prediction accuracy of soil pH and better prediction performance than other unoptimized models.

     

    • HTML全文
    • 参考文献(37)
    • 相关文章
    • 施引文献
  • 资源附件(0)

/

返回文章
返回