Forest height mapping using inventory and multi-source satellite data over Hunan Province in southern China

Background:Accurate mapping of forest canopy heights at a fine spatial resolution over large geographical areas is challenging.It is essential for the estimation of forest aboveground biomass and the evaluation of forest ecosystems.Yet current regional to national scale forest height maps were mainly produced at coarse-scale.Such maps lack spatial details for decision-making at local scales.Recent advances in remote sensing provide great opportunities to fill this gap.Method:In this study,we evaluated the utility of multi-source satellite data for mapping forest heights over Hunan Province in China.A total of 523 plot data collected from 2017 to 2018 were utilized for calibration and validation of forest height models.Specifically,the relationships between three types of in-situ measured tree heights(maximum-,averaged-,and basal area-weighted-tree heights)and plot-level remote sensing metrics(multispectral,radar,and topo variables from Landsat,Sentinel-1/PALSAR-2,and SRTM)were analyzed.Three types of models(multilinear regression,random forest,and support vector regression)were evaluated.Feature variables were selected by two types of variable selection approaches(stepwise regression and random forest).Model parameters and model performances for different models were tuned and evaluated via a 10-fold cross-validation approach.Then,tuned models were applied to generate wall-to-wall forest height maps for Hunan Province.Results:The best estimation of plot-level tree heights(R2 ranged from 0.47 to 0.52,RMSE ranged from 3.8 to 5.3 m,and rRMSE ranged from 28%to 31%)was achieved using the random forest model.A comparison with existing forest height maps showed similar estimates of mean height,however,the ranges varied under different definitions of forest and types of tree height.Conclusions:Primary results indicate that there are small biases in estimated heights at the province scale.This study provides a framework toward establishing regional to national scale maps of vertical forest structure.

Mechanism-learning coupling paradigms for parameter inversion and simulation in earth surface systems

Building the physics-driven mechanism model has always been the core scientific paradigm for parameter estimation in Earth surface systems,and developing the data-driven machine learning model is a crucial way for paradigm transformation in geoscience research.The coupling of mechanism and learning models can realize the combination of“rationalism”and“empiricism”,which is one of the most concerned research hotspots.In this paper,for remote sensing inversion and dynamic simulation,we deeply analyze the internal bottleneck and complementarity of mechanism and learning models and build a coupling paradigm framework with mechanism-learning cascading model,learning-embedded mechanism model,and mechanism-infused learning model.We systematically summarize ten specific coupling methods,including preprocessing and initialization,intermediate variable transfer,post-refinement processing,model substitution,model adjustment,model solution,input variable constraints,objective function constraints,model structure constraints,hybrid,etc.,and analyze the main existing problems and future challenges.The research aims to provide a new perspective for in-depth understanding and application of the mechanism-learning coupling model and provide theoretical and technical support for improving the inversion and simulation capabilities of parameters in Earth surface systems and serving the development of Earth system science.

A fully automatic and high-accuracy surface water mapping framework on Google Earth Engine using Landsat time-series

Efficient and continuous monitoring of surface water is essential for water resource management.Much effort has been devoted to the task of water mapping based on remote sensing images.However,few studies have fully considered the diverse spectral properties of water for the collection of reference samples in an automatic manner.Moreover,water area statistics are sensitive to the satellite image observation quality.This study aims to develop a fully automatic surface water mapping framework based on Google Earth Engine(GEE)with a supervised random forest classifier.A robust scheme was built to automatically construct training samples by merging the information from multi-source water occurrence products.The samples for permanent and seasonal water were mapped and collected separately,so that the supplement of seasonal samples can increase the spectral diversity of the sample space.To reduce the uncertainty of the derived water occurrences,temporal correction was applied to repair the classification maps with invalid observations.Extensive experiments showed that the proposed method can generate reliable samples and produce good-quality water mapping results.Comparative tests indicated that the proposed method produced water maps with a higher quality than the index-based detection methods,as well as the GSWD and GLAD datasets.

Land-surface temperature retrieval at high spatial and temporal resolutions based on multi-sensor fusion

Land-surface temperature(LST)is of great significance for the estimation of radiation and energy budgets associated with land-surface processes.However,the available satellite LST products have either low spatial resolution or low temporal resolution,which constrains their potential applications.This paper proposes a spatiotemporal fusion method for retrieving LST at high spatial and temporal resolutions.One important characteristic of the proposed method is the consideration of the sensor observation differences between different land-cover types.The other main contribution is that the spatial correlations between different pixels are effectively considered by the use of a variation-based model.The method was tested and assessed quantitatively using the different sensors of Landsat TM/ETM,moderate resolution imaging spectroradiometer and the geostationary operational environmental satellite imager.The validation results indicate that the proposed multisensor fusion method is accurate to about 2.5 K.