• Remote Sensing Technology and Application
  • Vol. 39, Issue 5, 1261 (2024)
Junchen FENG, Hao DONG, Peng HAN, Yuanbin LI..., Jingyu LIU and Yunhong DING|Show fewer author(s)
Author Affiliations
  • School of Computer Science and Information Engineering,Harbin Normal University,Harbin150025,China
  • show less
    DOI: 10.11873/j.issn.1004-0323.2024.5.1261 Cite this Article
    Junchen FENG, Hao DONG, Peng HAN, Yuanbin LI, Jingyu LIU, Yunhong DING. Prediction of Forest Burned Area based on MODIS-EVI2 and Ensemble Learning[J]. Remote Sensing Technology and Application, 2024, 39(5): 1261 Copy Citation Text show less
    Forest fires in Australia in 2016, the red dot is the location of the fire
    Fig. 1. Forest fires in Australia in 2016, the red dot is the location of the fire
    Processing the satellite image flow of MOD13A1 products using ENVI5.3
    Fig. 2. Processing the satellite image flow of MOD13A1 products using ENVI5.3
    Hierarchical Fusion structure Diagram of Stacking Model
    Fig. 3. Hierarchical Fusion structure Diagram of Stacking Model
    The prediction results of different models using EVI2 and NDVI as explanatory variables
    Fig. 4. The prediction results of different models using EVI2 and NDVI as explanatory variables
    The accuracy of different models using EVI2 and NDVI as explanatory variables
    Fig. 5. The accuracy of different models using EVI2 and NDVI as explanatory variables
    Experimental procedure of predicting Fire Burned Area by Stacking-XRSK Model
    Fig. 6. Experimental procedure of predicting Fire Burned Area by Stacking-XRSK Model
    The REC curves for the Stacking-XRSK model and base models
    Fig. 7. The REC curves for the Stacking-XRSK model and base models
    数据数据描述来源
    火灾发生位置经度LONGFA1
    纬度LAT
    火灾发生时间时间MONGFA1
    气象因素平均温度TEMPNOAA2
    平均风速WDSP
    最大持续风速MXSPD
    最高温度MAX
    最低温度MIN
    降雨量PRCP
    地形因素海拔ALTGE3
    植被因素

    两波段增强植被指数

    归一化差值植被指数

    EVI2

    NDVI

    ESDS4
    过火面积燃烧面积SIZEGFA1
    Table 1. Detailed description of each variable
    模型参数
    XGBoost

    max_depth=3;learning_rate=0.003

    n_estimators=1000;booster=’gbtree’;min_child_weight=1;early_stopping_rounds=10;eval_metric=’rmse’

    RFmax_depth=5;n_estimators=500; min_samples_split=2;min_samples_leaf=1
    SVMC=3;kernel=’rbf’;degree=3;tol=0.001
    KNNn_neighbors=7
    Table 2. The parameter setting of four kinds of regression models
    MAEMSER2
    XGBoostEVI20.068 20.012 90.811 1
    NDVI0.076 10.016 80.753 5
    RFEVI20.066 30.017 70.740 6
    NDVI0.075 10.021 80.680 1
    SVMEVI20.090 10.0190.721 6
    NDVI0.098 50.021 80.680 9
    KNNEVI20.079 50.022 30.672 9
    NDVI0.084 10.0240.648 3
    Table 3. Evaluation results of the eight models with EVI2 as explanatory variable and NDVI as explanatory variable
    MAEMSER2
    Stacking-XRSK0.063 10.012 20.815 5
    XGBoost0.084 70.015 20.769 6
    RF0.067 70.014 10.787 2
    SVR0.092 70.0170.743 2
    KNN0.0790.019 50.704 9
    Table 4. Evaluation results of Ensemble Learning Model and single Base Model
    Junchen FENG, Hao DONG, Peng HAN, Yuanbin LI, Jingyu LIU, Yunhong DING. Prediction of Forest Burned Area based on MODIS-EVI2 and Ensemble Learning[J]. Remote Sensing Technology and Application, 2024, 39(5): 1261
    Download Citation