Abstract
The Global Ecosystem Dynamics Investigation (GEDI) is a spaceborne lidar instrument that collects near-global measurements of forest structure. While expansive in scope, GEDI samples are spatially sparse and cover a small fraction of the land surface. Converting the sparse samples into spatially complete predictive maps is of practical importance for a number of ecological studies. A complicating factor is that GEDI collects measurements over forested and non-forested land alike, with no automatic labeling of the land type. Such classification is important, as it categorically influences the probability distribution of the spatial process and the ecological interpretation of the observations/predictions. We propose and implement a spatial mixture model, separating the observations and the greater spatial domain into two latent classes. The latent classes are governed by a Bernoulli spatial process, with spatial effects driven by a Gaussian process. Within each class, the process is governed by a separate spatial model, describing the unique probabilistic attributes. Model predictions take the form of scalar predictions of the GEDI observables as well as discrete labeling of the class membership. Inference is conducted through a Bayesian paradigm, yielding rich quantification of prediction and uncertainty through posterior predictive distributions. We demonstrate the method using GEDI data over Wollemi National Park, Australia, using optical data from Landsat 8 as model covariates. When compared to a single spatial model, the mixture model achieves much higher posterior predictive densities on the true value. When compared to a random forest model, a common algorithmic approach in the remote sensing community, the random forest achieves better absolute prediction accuracy for prediction locations far from observed training data locations, but at the expense of location-specific assessments of uncertainty. The unsupervised binary classifications of the mixture model appear broadly ecologically interpretable as forest and non-forest when compared to optical imagery, but further comparison to ground-truth data is required.
Similar content being viewed by others
References
Akay AE, Oğuz H, Karas IR, Aruga K (2009) Using LiDAR technology in forestry activities. Environ Monit Assess 151:117–125
Banerjee S, Carlin BP, Gelfand AE (2003) Hierarchical modeling and analysis for spatial data. Chapman and Hall/CRC, Boca Raton
Barry S, Elith J (2006) Error and uncertainty in habitat models. J Appl Ecol 43(3):413–423
Bates D, Maechler M (2021) Matrix: sparse and dense matrix classes and methods. R package version 1.3-4
Belgiu M, Drăguţ L (2016) Random forest in remote sensing: a review of applications and future directions. ISPRS J Photogramm Remote Sens 114:24–31
Bolin D, Lindgren F (2011) Spatial models generated by nested stochastic partial differential equations, with an application to global ozone mapping. Annal Appl Stat 1:523–550
Breiman L (2001) Random forests. Mach Learn 45:5–32
Burns P, Clark M, Salas L, Hancock S, Leland D, Jantz P, Dubayah R, Goetz SJ (2020) Incorporating canopy structure from simulated GEDI lidar into bird species distribution models. Environ Res Lett 15(9):095002
Calder C, Lavine M, Müller P, Clark JS (2003) Incorporating multiple sources of stochasticity into dynamic population models. Ecology 84(6):1395–1402
Dubayah R, Armston J, Healey SP, Bruening JM, Patterson PL, Kellner JR, Duncanson L, Saarela S, Ståhl G, Yang Z et al (2022) GEDI launches a new era of biomass inference from space. Environ Res Lett 17(9):095001
Dubayah R, Blair JB, Goetz S, Fatoyinbo L, Hansen M, Healey S, Hofton M, Hurtt G, Kellner J, Luthcke S et al (2020) The global ecosystem dynamics investigation: high-resolution laser ranging of the Earth’s forests and topography. Sci Remote Sens 1:100002
Dubayah RO, Drake JB (2000) Lidar remote sensing for forestry. J Forest 98(6):44–46
Duncanson L, Kellner JR, Armston J, Dubayah R, Minor DM, Hancock S, Healey SP, Patterson PL, Saarela S, Marselis S et al (2022) Aboveground biomass density models for NASA’s Global ecosystem dynamics investigation (GEDI) lidar mission. Remote Sens Environ 270:112845
Finley AO, Banerjee S, MacFarlane DW (2011) A hierarchical model for quantifying forest variables over large heterogeneous landscapes with uncertain forest areas. J Am Stat Assoc 106(493):31–48
Fuglstad G-A, Simpson D, Lindgren F, Rue H (2019) Constructing priors that penalize the complexity of Gaussian random fields. J Am Stat Assoc 114(525):445–452
Gelfand AE, Kim H-J, Sirmans C, Banerjee S (2003) Spatial modeling with spatially varying coefficient processes. J Am Stat Assoc 98(462):387–396
Gelfand AE, Schliep EM (2016) Spatial statistics and Gaussian processes: a beautiful marriage. Spatial Stat 18:86–104
Gneiting T, Raftery AE (2007) Strictly proper scoring rules, prediction, and estimation. J Am Stat Assoc 102(477):359–378
Heaton MJ, Datta A, Finley AO, Furrer R, Guinness J, Guhaniyogi R, Gerber F, Gramacy RB, Hammerling D, Katzfuss M et al (2019) A case study competition among methods for analyzing large spatial data. J Agric Biol Environ Stat 24:398–425
Held L, Schrödle B, Rue H (2010) Posterior and cross-validatory predictive checks: a comparison of MCMC and INLA. Festschrift in honour of ludwig fahrmeir, Statistical modelling and regression structures, pp 91–110
Hoffrén R, Lamelas MT, de la Riva J, Domingo D, Montealegre AL, García-Martín A, Revilla S (2023) Assessing GEDI-NASA system for forest fuels classification using machine learning techniques. Int J Appl Earth Obs Geoinf 116:103175
Liaw A, Wiener M (2002) Classification and regression by randomForest. R News 2(3):18–22
Lindgren F, Rue H (2015) Bayesian spatial modelling with R-INLA. J Stat Softw 63:1–25
Lindgren F, Rue H, Lindström J (2011) An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach. J R Stat Soc: Series B (Statis Methodol) 73(4):423–498
Marselis SM, Keil P, Chase JM, Dubayah R (2022) The use of GEDI canopy structure for explaining variation in tree species richness in natural forests. Environ Res Lett 17(4):045003
Marselis SM, Tang H, Armston J, Abernethy K, Alonso A, Barbier N, Bissiengou P, Jeffery K, Kenfack D, Labrière N et al (2019) Exploring the relation between remotely sensed vertical canopy structure and tree species diversity in Gabon. Environ Res Lett 14(9):094013
McRoberts RE (2011) Satellite image-based maps: Scientific inference or pretty pictures? Remote Sens Environ 115(2):715–724
Molto Q, Rossi V, Blanc L (2013) Error propagation in biomass estimation in tropical forests. Methods Ecol Evol 4(2):175–183
Neelon B, Gelfand AE, Miranda ML (2014) A multivariate spatial mixture model for areal data: examining regional differences in standardized test scores. J R Stat Soc: Series B Appl Stat 63(5):737
Paciorek CJ, Schervish MJ (2006) Spatial modelling using a new class of nonstationary covariance functions. Environ Off J Int Environ Soc 17(5):483–506
Potapov P, Li X, Hernandez-Serna A, Tyukavina A, Hansen MC, Kommareddy A, Pickens A, Turubanova S, Tang H, Silva CE et al (2021) Mapping global forest canopy height through integration of GEDI and Landsat data. Remote Sens Environ 253:112165
Qi W, Dubayah RO (2016) Combining Tandem-X InSAR and simulated GEDI lidar observations for forest structure mapping. Remote Sens Environ 187:253–266
Qi W, Saarela S, Armston J, Ståhl G, Dubayah R (2019) Forest biomass estimation over three distinct forest types using TanDEM-X InSAR data and simulated GEDI lidar data. Remote Sens Environ 232:111283
R Core Team (2020) R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria
Sampson PD, Guttorp P (1992) Nonparametric estimation of nonstationary spatial covariance structure. J Am Stat Assoc 87(417):108–119
Smith I, Velasquez E, Pickering C (2021) Quantifying potential effect of 2019 fires on national parks and vegetation in South-East Queensland. Ecol Manag Restor 22(2):160–170
Sothe C, Gonsamo A, Lourenço RB, Kurz WA, Snider J (2022) Spatially continuous mapping of forest canopy height in Canada by combining GEDI and ICESat-2 with PALSAR and sentinel. Remote Sens 14(20):5158
Stein ML (1999) Interpolation of spatial data: some theory for kriging. Springer Science & Business Media, USA
Vanhatalo J, Foster SD, Hosack GR (2021) Spatiotemporal clustering using Gaussian processes embedded in a mixture model. Environmetrics 32(7):e2681
Verhelst K, Gou Y, Herold M, Reiche J (2021) Improving forest baseline maps in tropical wetlands using GEDI-based forest height information and Sentinel-1. Forests 12(10):1374
Wall MM, Liu X (2009) Spatial latent class analysis model for spatially distributed multivariate binary data. Comput stat Data Anal 53(8):3057–3069
Wulder MA, White JC, Nelson RF, Næsset E, Ørka HO, Coops NC, Hilker T, Bater CW, Gobakken T (2012) Lidar sampling for large-area forest characterization: a review. Remote Sens Environ 121:196–209
Zhang H (2004) Inconsistent estimation and asymptotically equal interpolations in model-based geostatistics. J Am Stat Assoc 99(465):250–261
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflicts of interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Data Availability
Data and code used during this study have been provided as an electronic supplementary material.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Gibbs Samplers
Gibbs Samplers
Here we give exact details on the Gibbs sampler for the typical and mixture model. First presenting our notation, let \({{\varvec{s}}}= [s_1 \cdots s_n]^T\) be the vector of n observation locations, \({{\varvec{y}}}({{\varvec{s}}})\) be the vector of observed log-RH98 and \({{\varvec{X}}}({{\varvec{s}}})\) be the \(n\times p\) matrix of optical intensities.
1.1 Typical Spatial Model
The Gibbs sampler for the typical spatial model iterates as follows:
-
1.
Sample spatial effects \({{\varvec{w}}}|{{\varvec{y}}}({{\varvec{s}}}), {{\varvec{X}}}({{\varvec{s}}}), \mu , {\varvec{\beta }},\theta , \tau ^2\), which is multivariate normal with mean and variance
$$\begin{aligned} \textrm{E}[{{\varvec{w}}}| \ldots ]&=({{\varvec{Q}}}(\theta ) + \frac{1}{\tau ^2}{{\varvec{A}}}({{\varvec{s}}})^T{{\varvec{A}}}({{\varvec{s}}}))^{-1}{{\varvec{A}}}({{\varvec{s}}})^T\frac{1}{\tau ^2}({{\varvec{y}}}({{\varvec{s}}}) - \mu {\varvec{1}} - {{\varvec{X}}}({{\varvec{s}}}){\varvec{\beta }}), \end{aligned}$$(22)$$\begin{aligned} \textrm{Var}[{{\varvec{w}}}| \ldots ]&= ({{\varvec{Q}}}(\theta ) + \frac{1}{\tau ^2}{{\varvec{A}}}({{\varvec{s}}})^T{{\varvec{A}}}({{\varvec{s}}}))^{-1}. \end{aligned}$$(23) -
2.
Sample jointly mean parameters \(\mu ,{\varvec{\beta }}| {{\varvec{y}}}({{\varvec{s}}}), {{\varvec{X}}}({{\varvec{s}}}), \theta , \tau ^2\) marginalizing over \({{\varvec{w}}}\), which is multivariate normal with mean and variance
$$\begin{aligned} \textrm{E}\begin{bmatrix} \mu \\ {\varvec{\beta }}\end{bmatrix}&= \left( {\tilde{{{\varvec{X}}}}}({{\varvec{s}}})^T{\varvec{\Sigma }}({{\varvec{s}}};\theta )^{-1}{\tilde{{{\varvec{X}}}}}({{\varvec{s}}})\right) ^{-1}{\tilde{{{\varvec{X}}}}}({{\varvec{s}}})^T{\varvec{\Sigma }}({{\varvec{s}}};\theta )^{-1}{{\varvec{y}}}({{\varvec{s}}}), \end{aligned}$$(24)$$\begin{aligned} \textrm{Var}\begin{bmatrix} \mu \\ {\varvec{\beta }}\end{bmatrix}&= \left( {\tilde{{{\varvec{X}}}}}({{\varvec{s}}})^T{\varvec{\Sigma }}^{-1}{\tilde{{{\varvec{X}}}}}({{\varvec{s}}})\right) ^{-1}, \end{aligned}$$(25)where \({\varvec{\Sigma }}({{\varvec{s}}};\theta ) = {{\varvec{A}}}({{\varvec{s}}}){{\varvec{Q}}}(\theta )^{-1}{{\varvec{A}}}({{\varvec{s}}})^T + \tau ^2{{\varvec{I}}}\) and \({\tilde{{{\varvec{X}}}}}({{\varvec{s}}}) = [\textbf{1},~{{\varvec{X}}}({{\varvec{s}}})]\). Using the Woodbury matrix identity,
$$\begin{aligned} {\varvec{\Sigma }}^{-1} = \frac{1}{\tau ^2}{{\varvec{I}}}- \frac{1}{\tau ^4}{{\varvec{A}}}({{\varvec{s}}})^T\left( {{\varvec{Q}}}(\theta ) + \frac{1}{\tau ^2}{{\varvec{A}}}({{\varvec{s}}})^T{{\varvec{A}}}({{\varvec{s}}})\right) ^{-1}{{\varvec{A}}}({{\varvec{s}}}). \end{aligned}$$(26) -
3.
Sample the noise variance \(\tau ^2|{{\varvec{y}}}({{\varvec{s}}}), {{\varvec{X}}}({{\varvec{s}}}), \mu , {\varvec{\beta }},{{\varvec{w}}}\), which is inverse-gamma with shape n/2 and rate
$$\begin{aligned} \frac{1}{2}\left( {{\varvec{y}}}({{\varvec{s}}}) - \mu {\varvec{1}} - {{\varvec{X}}}({{\varvec{s}}}){\varvec{\beta }}- {{\varvec{A}}}({{\varvec{s}}}){{\varvec{w}}}\right) ^T\left( {{\varvec{y}}}({{\varvec{s}}}) - \mu {\varvec{1}} - {{\varvec{X}}}({{\varvec{s}}}){\varvec{\beta }}- {{\varvec{A}}}({{\varvec{s}}}){{\varvec{w}}}\right) . \end{aligned}$$(27) -
4.
Sample jointly the Matérn parameters \(\theta = \{\sigma ^2, \phi \}\) using a Metropolis–Hastings step on target density
$$\begin{aligned} p(\theta |{{\varvec{w}}}) \propto |{{\varvec{Q}}}(\theta )|^{1/2}\exp \left( -\frac{1}{2}{{\varvec{w}}}^T{{\varvec{Q}}}(\theta ){{\varvec{w}}}\right) . \end{aligned}$$(28)
1.2 Spatial Mixture Model
Let \({{\varvec{s}}}_j = \{s_i~:~z(s_i) = j\}\) for \(j\in \{0,1\}\) be the sub-vector of observations corresponding to either class. The Gibbs sampler for the spatial mixture model iterates as follows.
-
1.
Sample classifications \({{\varvec{z}}}(s_i)|\ldots ;~i\in \{1,\ldots ,n\}\), which are Bernoulli distributed with conditional probabilities
$$\begin{aligned} \pi ^*(s_i) = \frac{\pi (s_i)f\left( y(s_i)| z(s_i) =1\right) }{\pi (s_i)f\left( y(s_i)| z(s_i) =1\right) + (1-\pi (s_i))f\left( y(s_i)| z(s_i) =0\right) }, \end{aligned}$$(29)where \(f\left( \cdot |z(s_i) = j\right) \) is a normal density with mean \(\mu _j + {{\varvec{x}}}(s_i)^T{\varvec{\beta }}_j + {{\varvec{A}}}(s_i){{\varvec{w}}}_j\) and variance \(\tau _j^2\) for \(j\in \{0,1\}\), and \(\pi (s_i)\) is Eq. (4) evaluated at current parameter and effect values.
-
2.
Sample spatial effects \({{\varvec{w}}}_j|{{\varvec{y}}}({{\varvec{s}}}_j), {{\varvec{X}}}({{\varvec{s}}}_j), \mu _j, {\varvec{\beta }}_j,\theta _j, \tau ^2_j\) for \(j\in \{0,1\}\), which are multivariate normal with mean and variance
$$\begin{aligned} \textrm{E}[{{\varvec{w}}}_j| \ldots ]&=({{\varvec{Q}}}(\theta _j) + \frac{1}{\tau _j^2}{{\varvec{A}}}({{\varvec{s}}}_j)^T{{\varvec{A}}}({{\varvec{s}}}_j))^{-1}{{\varvec{A}}}({{\varvec{s}}}_j)^T\frac{1}{\tau _j^2}({{\varvec{y}}}({{\varvec{s}}}_j) - \mu _j{\varvec{1}} - {{\varvec{X}}}({{\varvec{s}}}_j){\varvec{\beta }}_j), \end{aligned}$$(30)$$\begin{aligned} \textrm{Var}[{{\varvec{w}}}_j| \ldots ]&= ({{\varvec{Q}}}(\theta _j) + \frac{1}{\tau _j^2}{{\varvec{A}}}({{\varvec{s}}}_j)^T{{\varvec{A}}}({{\varvec{s}}}_j))^{-1}. \end{aligned}$$(31) -
3.
Sample jointly mean parameters \(\mu _j,{\varvec{\beta }}_j| {{\varvec{y}}}({{\varvec{s}}}_j), {{\varvec{X}}}({{\varvec{s}}}_j), \theta _j, \tau _j^2\) marginalizing over \({{\varvec{w}}}_j\) for \(j\in \{0,1\}\), which are multivariate normal with mean and variance
$$\begin{aligned} \textrm{E}\begin{bmatrix} \mu _j \\ {\varvec{\beta }}_j \end{bmatrix}&= \left( {\tilde{{{\varvec{X}}}}}({{\varvec{s}}}_j)^T{\varvec{\Sigma }}({{\varvec{s}}}_j;\theta _j)^{-1}{\tilde{{{\varvec{X}}}}}({{\varvec{s}}}_j)\right) ^{-1}{\tilde{{{\varvec{X}}}}}({{\varvec{s}}}_j)^T{\varvec{\Sigma }}({{\varvec{s}}}_j;\theta _j)^{-1}{{\varvec{y}}}({{\varvec{s}}}_j), \end{aligned}$$(32)$$\begin{aligned} \textrm{Var}\begin{bmatrix} \mu _j \\ {\varvec{\beta }}_j \end{bmatrix}&= \left( {\tilde{{{\varvec{X}}}}}({{\varvec{s}}}_j)^T{\varvec{\Sigma }}({{\varvec{s}}}_j;\theta _j)^{-1}{\tilde{{{\varvec{X}}}}}({{\varvec{s}}}_j)\right) ^{-1}, \end{aligned}$$(33)where \({\varvec{\Sigma }}({{\varvec{s}}}_j;\theta _j) = {{\varvec{A}}}({{\varvec{s}}}_j){{\varvec{Q}}}(\theta _j)^{-1}{{\varvec{A}}}({{\varvec{s}}}_j)^T + \tau _j^2{{\varvec{I}}}\) and \({\tilde{{{\varvec{X}}}}}({{\varvec{s}}}_j) = [\textbf{1},~{{\varvec{X}}}({{\varvec{s}}}_j)]\). Using the Woodbury matrix identity,
$$\begin{aligned} {\varvec{\Sigma }}({{\varvec{s}}}_j;\theta _j)^{-1} = \frac{1}{\tau _j^2}{{\varvec{I}}}- \frac{1}{\tau _j^4}{{\varvec{A}}}({{\varvec{s}}}_j)^T\left( {{\varvec{Q}}}(\theta _j) + \frac{1}{\tau _j^2}{{\varvec{A}}}({{\varvec{s}}}_j)^T{{\varvec{A}}}({{\varvec{s}}}_j)\right) ^{-1}{{\varvec{A}}}({{\varvec{s}}}_j). \end{aligned}$$(34) -
4.
Sample the noise variance \(\tau _j^2|{{\varvec{y}}}({{\varvec{s}}}_j), {{\varvec{X}}}({{\varvec{s}}}_j), \mu _j, {\varvec{\beta }}_j,{{\varvec{w}}}_j\) for \(j\in \{0,1\}\), which are inverse-gamma with shape \(n_j/2\) (where \(n_j\) is the length of \({{\varvec{s}}}_j\)) and rate
$$\begin{aligned} \frac{1}{2}\left( {{\varvec{y}}}({{\varvec{s}}}_j) - \mu _j{\varvec{1}} - {{\varvec{X}}}({{\varvec{s}}}_j){\varvec{\beta }}_j - {{\varvec{A}}}({{\varvec{s}}}_j){{\varvec{w}}}_j\right) ^T\left( {{\varvec{y}}}({{\varvec{s}}}_j) - \mu _j{\varvec{1}} - {{\varvec{X}}}({{\varvec{s}}}_j){\varvec{\beta }}_j - {{\varvec{A}}}({{\varvec{s}}}_j){{\varvec{w}}}_j\right) .\nonumber \\ \end{aligned}$$(35) -
5.
Sample jointly \(\mu _p,{\varvec{\beta }}_p, {{\varvec{w}}}_p| {{\varvec{z}}}(s), \theta _z\) using a Laplace approximation. Let \({{\varvec{b}}}= [\mu _p,{\varvec{\beta }}_p, {{\varvec{w}}}_p]^T\) and define design matrix \({\tilde{{{\varvec{A}}}}}=[\textbf{1}, {{\varvec{X}}}({{\varvec{s}}}),{{\varvec{A}}}({{\varvec{s}}})]\) and the prior precision \({\tilde{{{\varvec{Q}}}}}(\theta _p) = \textrm{blockdiag}\{\textbf{0}_{p+1}, {{\varvec{Q}}}(\theta _p)\}\), where \(\textbf{0}_{p+1}\) is a \(p+1\times p+1\) matrix of zeros, representing the prior precision of the scalar intercept \(\mu _z\) and p regression coefficients \({\varvec{\beta }}_z\). A Newton–Raphson routine is used to find posterior mode \({\hat{{{\varvec{b}}}}}\). Letting \({{\varvec{b}}}_{(0)}\) be some initial value, we iterate
$$\begin{aligned} {{\varvec{b}}}_{(j+1)}&= {{\varvec{b}}}_{(j)} - {{\varvec{H}}}\left( f({{\varvec{b}}}_{(j)}|{{\varvec{z}}}({{\varvec{s}}}),\theta )\right) ^{-1}\frac{\partial f}{\partial {{\varvec{b}}}}({{\varvec{b}}}_{(j)}|{{\varvec{z}}}({{\varvec{s}}}),\theta ) \end{aligned}$$(36)$$\begin{aligned}&= {{\varvec{b}}}_{(j)} + \left( {\tilde{{{\varvec{Q}}}}}(\theta _p) + {\tilde{{{\varvec{A}}}}}^T{{\varvec{D}}}({{\varvec{b}}}_{(j)}){\tilde{{{\varvec{A}}}}}\right) ^{-1}\left( {\tilde{{{\varvec{A}}}}}^T\left( {{\varvec{z}}}({{\varvec{s}}}) - {{\varvec{p}}}({{\varvec{s}}}|{{\varvec{b}}}_{(j)})\right) - {\tilde{{{\varvec{Q}}}}}(\theta _p){{\varvec{b}}}_{(j)}\right) , \end{aligned}$$(37)where \({{\varvec{D}}}({{\varvec{b}}}_{(j)}) = \textrm{diag}\{\pi (s_i|{{\varvec{b}}}_{(j)})(1 - \pi (s_i|{{\varvec{b}}}_{(j)})\}\), until a convergence criterion is met. Then the Laplace approximation is
$$\begin{aligned} {{\varvec{b}}}|{{\varvec{z}}}(s),\theta _z \sim \textrm{MVN}\left( {\hat{{{\varvec{b}}}}},~~\left( {\tilde{{{\varvec{Q}}}}}(\theta _z) + {\tilde{{{\varvec{A}}}}}^T{{\varvec{D}}}({\hat{{{\varvec{b}}}}}){\tilde{{{\varvec{A}}}}}\right) ^{-1}\right) . \end{aligned}$$(38) -
6.
Sample jointly the Matérn parameters \(\theta _j = \{\sigma _j^2, \phi _j\}\) for \(j\in \{0,1,z\}\), using Metropolis-Hastings steps on the target densities
$$\begin{aligned} p(\theta _j|{{\varvec{w}}}_j) \propto |{{\varvec{Q}}}(\theta _j)|^{1/2}\exp \left( -\frac{1}{2}{{\varvec{w}}}_j^T{{\varvec{Q}}}(\theta _j){{\varvec{w}}}_j\right) . \end{aligned}$$(39)
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
May, P.B., Finley, A.O. & Dubayah, R.O. A Spatial Mixture Model for Spaceborne Lidar Observations Over Mixed Forest and Non-forest Land Types. JABES (2024). https://doi.org/10.1007/s13253-024-00600-6
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s13253-024-00600-6