Remote Sensing-based Monitoring Of Insect Stress On Yunan Pine Forests Using Radiative Transfer Models

In recent years,pine shoot beetles(PSB)broken out in Yunnan province,has cuased extensive destruction of Yunnan pine plantation,which poses a great challenge to the healthy and sustainable management of forests.With the development and application of remote sensing technology,unprecedented richness of satellite remote sensing data can be used for free,and remote sensing has gradually became the main way of large-scale monitoring and forecasting of forest insects.However,most of the current monitoring models are based on empirical methods,which are difficult to popularize.Based on the radiation transfer theory,the mathematical relationship between remote sensing signals and vegetation parameters can be used for parameter inversion,which has great potential for monitoring forest insect damage.In this study,we aim to explore the detection of PSB using the reflection model at leaf and canopy scales.We firstly developed a spectral linear mixing method aided by leaf radiative transfer(RT)model(e.g.,LIBERTY and PROSPECT-D)and a new Yellow Index(YI)to simulate the needle reflectance of different damage degree at leaf scale.Then,based on leaf RT model simulation,the sensitivity of biochemical parameters(e.g.,leaf chlorophyll content)of different damage level were analyzed using the extended Fourier amplitude sensitivity test(EFAST)global sensitivity analysis method.Finally,leaf chlorophyll content(Cab)of the damaged needles was retrieved using inversion of LIBERTY and PROSPECT-D with spectral linear mixing method.We systematically compared one-dimensional(1D)and three-dimensional(3D)RT models in the Cab,leaf area index(LAI),and shoot dieback retrieval performance using Sentinel-2 and unmanned aerial vehicle(UAV)remote sensing data for the damaged pine forests at canopy scale.For 1D RT model,INFORM canopy model was coupled with introduced spectral linear mixing method of PROSPECT-D leaf model to invert Cab,LAI,and shoot dieback parameters using look-up table(LUT)-based invertion.For 3D RT models,a shoot dieback indicator and tree allometric equations were used to parameterize RAPID model to estimate beetles damage.Specifically,the vertical heterogeneities of both the biophysical parameters and optical properties within the canopy were carefully considered in RAPID.Firstly,we developed an approach for reconstructing damaged forests based on individual tree segmentation from the lidar data and in situ measurements.Secondly,the nadir reflectance images were simulated from the reconstructed heterogeneity of the forest scenes,which were compared with the actual hyperspectral and Sentinel-2(S2)reflectance data.After the comparisons,the parameterized RAPID model was used to create a LUT.Furthermore,tree allometric equations were used to regularize the LUT inversion approach to improve the parameter estimation accuracy.The main results are as follow:1.for needle scale:(1)The spectral linear mixing method can effectively improve the reflectance simulation accuracy of Yunnan pine needles with different damage degree,especially for the serious degree of needles.The needle reflectance simulation accuracy with mean relative error(MRE)is 1.4%-18% for LIBERTY model,and 3%-15% for PROSPECT-D model.These results indicate that PROSPECT-D can be used to describe the optical properties of needles.In addition,the simulated Yellow index can quantitatively assess the damage degree of needles.(2)The results of sensitivity of biochemical parameters showed that Cab and Cw were the main parameters caused the change of reflectance in visible and short wave infrared region for healthy needles.The contribution of Cab and Cw to reflectance decreases for the deepening damage degree of needle.Especially for the serious damage degree of needle,the contribution of cab and Cb to the reflectance is equal.(3)The accuracy of Cab estimation was improved by LIBERTY and PROSPECT-D models with spectral linear mixing method for the damaged needles.Furthermore,the Cab estimation accuracy of PROSPECT-D was better than that using LIBERTY.This result proves that PROSPECT-D is also suitable for the extraction of physical and chemical parameters of needles.2.for canopy scale:(1)The 1D INFORM model was suitable to retrieve LAI.However,the estimation of Cab and shoot dieback has larger error when comparing with the field-measured values.The results showed that the 1D model has great limitations in describing the characteristics of heterogeneous tree crowns.(2)RAPID-simulated forest reflectance was highly consistent with Sentinel-2 and UAV hyperspectral images observations.The results revealed that 3D RT models(e.g.,RAPID)were suitable to describe the spectral characteristics of heterogeneous forest.(3)The accuracy of Cab and shoot dieback retrieval using 3D RAPID model was better than using1 D INFORM model for the damaged forests.The results show that the 3D RT model has more advantages in the inversion of the biophysical and biochemical parameters of the heterogeneous canopy.(4)The spatial-temporal distribution of different damage level of forests from 2015 to 2019 was mapped using RAPID model with the optimal LUT inversion.The results showed that the area and severity of forest damage decreased year by year,which was consistent with the field measurements.This study proved that the advantage of 3D RAPID model in inversion of biophysical and biochemical parameters for the damaged forests.A RT model-based method was proposed to detect the PSB damage of Yunnan pine forests.