Remote Sensing Monitoring Of Insect Pests In Larch Forest Based On Physical Model

In the recent ten years,gradual research progresses have been made on the remote sensing application in the forest pest monitoring area,especially on the monitoring model.However, most monitoring models were developed based on statistical analysis,which is highly depending on the specific field data.Therefore,these models are limited for only small regions and not suitable for other study areas.Hence,it is necessary and urgent to develop more general models. Meanwhile,most research used two physiological characteristics,leaf litter and the change of the leaf colour,to monitor forest status after the trees being attacked.Although they are suitable for the loss assessment,these researches are not so good to detect the insect pests in the early stage.That is because it's usually in the middle and latter stage of the insects attacking when the two physiological characteristics changes can be detectable by remote sensing data.Moreover, most VIS and NIR data were applied in the pest monitoring research,while,the application of TIR data are rare.In this paper,the key scientific problems of the forest insect pests monitoring by remote sensing were studied based on the theories of the ecology,forest protection and quantitative remote sensing.Through the analysis on the physiological characteristics changes after the trees being attacked,three indicative factors of the early forest insect pests monitoring were brought forward,which are CWC(represents the total water content of forest canopy),TVDI(represents the relative water content of soil) and the LAI(leaf area index).In order to evaluate the feasibility of the CWC inversion from the Landsat data,the sensitivity of the reflectance characteristics of coniferous needles on the needle water content was studied based on the filed data and the LIBERTY(Leaf Incorporating Biochemistry Exhibiting Reflectance and Transmittance Yields) model.The results show that three spectrum indices,which are MSI (Moisture Stress Index),NDWI(Normal Difference Water Index) and GVMI(Global Vegetation Moisture Index) were highly correlated with the water content.Especially,GVMI is not sensitive to whether the needle is clustered or not.Moreover,GVMI uses two broad bands (NIR and SWIR),which can be calculated from Landsat TM/ETM+ data.Therefore,it is confirmed that Landsat TM/ETM+ data is feasible to inverse the forest canopy water content, which lays a solid theoretical foundation for the following research.Then,the inversion research on the three indicative factors was performed using VIR,NIR and TIR data,combined with physical-based RT(Radiative Transfer) model and ANN(Artificial Neural Network).Firstly,the inversion method(RT-ANN) for LAI and CWC by combining the 5-scale RT model and BP ANN was put forward based on the historical research.And the method was customized on Landsat data to inverse the two indicative factors.Field data and MODIS data were both validate the reasonable inversion precision.Finally,twelve images of LAI and CWC were obtained based on twelve Landsat images during the ten years.The images show that LAI and CWC decrease gradually.Then,the inversion of TVDI(Temperature/Vegetation dryness Index) which can repsent the soil relative water content was studied.TVDI is determined by using the Ts/NDVI featrue space.Pre-analysis results show that the Landsat TM₆ data must be corrected to remove atmopsheric effects and emissivity effects before retriving TVDI.By combining the field rainfall data,the inversed TVDI image series show that this index is good to indicate the soil dryness and usefull for monitoring soil relative water content.Eventually,the decision rules for the larch forest insect pests monitor were performed based on the three inversed indices.The rules are then applied in the YIERSHI forest farm.The insect pest forest compartments which extract from the images is highly consistent with the ground field data.Considering practicability,a general and physically-based forest insect pest monitoring model is developed by speeding the inversion algorithm.