Landsat-Scale Regional Forest Canopy Height Mapping Using ICESat-2 Along-Track Heights: Case Study of Eastern Texas

Malambo, Lonesome; Popescu, Sorin C.; Li, Meng

doi:10.3390/rs15010001

Cited by 11 publications

(13 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Examples of papers where ML structures are applied include the following: Neural networks have been utilized in studies of vegetation canopy height, using ICESat-2 and Landsat data [99,100]. In a case study of a forest in Texas, [100] investigate the potential of using a Deep Neural Net (DNN) or a Random Forest (RF) model for above-ground biomass assessment based on ICESat-2 and Landsat data, finding similar performance values for the DNN and the RF.…”

Section: Recent Applications Of Nns In Geosciencesmentioning

confidence: 99%

Combining "Deep Learning" and Physically Constrained Neural Networks to Derive Complex Glaciological Change Processes from Modern High-Resolution Satellite Imagery: Application of the GEOCLASS-image System to Create VarioCNN for Glacier Surges

Herzfeld,

Hessburg,

Trantow

et al. 2024

Preprint

View full text Add to dashboard Cite

The objectives of this paper are to investigate the tradeoffs between a physically constrained neural network and a deep, convolutional neural network and to design a combined ML approach ("VarioCNN"). Our solution is provided in the framework of a cyberinfrastructure that includes a newly designed ML software, GEOCLASS-image, modern high-resolution satellite image data sets (Maxar WorldView data) and instructions/descriptions that may facilitate solving similar spatial classification problems. Combining the advantages of the physically-driven connectionist-geostatistical classification method with those of an efficient CNN, VarioCNN provides a means for rapid and efficient extraction of complex geophysical information from submeter resolution satellite imagery. A retraining loop overcomes the difficulties of creating a labeled training data set. Computational analyses and developments are centered on a specific, but generalizable, geophysical problem: The classification of crevasse types that form during the surge of a glacier system. A surge is a glacial catastrophe, an acceleration of a glacier to typically 100-200 times its normal velocity. GEOCLASS-image is applied to study the current (2016-2024) surge in the Negribreen Glacier System, Svalbard. The geophysical result is a description of the structural evolution and expansion of the surge, based on crevasse types that capture ice deformation in six simplified classes.

show abstract

Section: Recent Applications Of Nns In Geosciencesmentioning

confidence: 99%

Combining "Deep Learning" and Physically Constrained Neural Networks to Derive Complex Glaciological Change Processes from Modern High-Resolution Satellite Imagery: Application of the GEOCLASS-image System to Create VarioCNN for Glacier Surges

Herzfeld,

Hessburg,

Trantow

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…Examples of papers where ML structures are applied include the following: Neural networks have been utilized in studies of vegetation canopy height, using ICESat-2 and Landsat data (Malambo & Popescu, 2023;Narine et al, 2019). In a case study of a forest in Texas, Narine et al (2019) investigate the potential of using a Deep Neural Net (DNN) or a Random Forest (RF) model for above-ground biomass assessment based on ICESat-2 and Landsat data, finding similar performance values for the DNN and the RF.…”

Section: Recent Applications Of Nns In Geosciencesmentioning

confidence: 99%

Combining “Deep Learning” and Physically Constrained Neural Networks to Derive Complex Glaciological Change Processes from Modern High-Resolution Satellite Imagery: Application of the GEOCLASS-image System to Create VarioCNN for Glacier Surges

Herzfeld,

Hessburg,

Trantow

et al. 2024

Preprint

View full text Add to dashboard Cite

The objectives of this paper are to investigate the tradeoffs between a physically constrained neural network and a deep, convolutional neural network and to design a combined ML approach (“VarioCNN”). Our solution is provided in the framework of a cyberinfrastructure that includes a newly designed ML software, GEOCLASS-image, modern high-resolution satellite image datasets (Maxar WorldView data) and instructions/descriptions that may facilitate solving similar spatial classification problems. Combining the advantages of the physically-driven connectionist-geostatistical classification method with those of an efficient CNN, VarioCNN, provides a means for rapid and efficient extraction of complex geophysical information from submeter resolution satellite imagery. A retraining loop overcomes the difficulties of creating a labeled training data set. Computational analyses and developments are centered on a specific, but generalizable, geophysical problem: The classification of crevasse types that form during the surge of a glacier system. A surge is a glacial catastrophe, an acceleration of a glacier to typically 100-200 times its normal velocity, which for a marine-terminating glacier leads to sudden and substantial mass transfer from the cryosphere to the oceans, contributing significantly to sea-level-rise. The sudden and rapid acceleration characteristic of a surge results in formation of crevasses, whose spatial characteristics provide informants on the ice-dynamic processes that occur during the surge. GEOCLASS-image is applied to study the current (2016-2024) surge in the Negribreen Glacier System, Svalbard. The geophysical result is a description of the structural evolution and expansion of the surge, based on crevasse types that capture ice deformation in 6 simplified classes.

show abstract

“…These ecosystems play a vital role in supporting a myriad of plant and animal species, contributing to the overall ecological health and biodiversity of the region [1,2]. Forest canopy height measurements play a crucial role in understanding and managing ecosystems effectively, providing valuable insights into the vertical structure of forests and enabling assessment of forest health and productivity [3,4]. In parts of Texas, such as the Edwards Plateau, woody plant encroachment is causing fundamental ecological shifts.…”

Section: Introductionmentioning

confidence: 99%

Image to Image Deep Learning for Enhanced Vegetation Height Modeling in Texas

Malambo,

Popescu

2023

Remote Sensing

Self Cite

View full text Add to dashboard Cite

Vegetation canopy height mapping is vital for forest monitoring. However, the high cost and inefficiency of manual tree measurements, coupled with the irregular and limited local-scale acquisition of airborne LiDAR data, continue to impede its widespread application. The increasing availability of high spatial resolution imagery is creating opportunities to characterize forest attributes at finer resolutions over large regions. In this study, we investigate the synergy of airborne lidar and high spatial resolution USDA-NAIP imagery for detailed canopy height mapping using an image-to-image deep learning approach. Our main inputs were 1 m NAIP image patches which served as predictor layers and corresponding 1 m canopy height models derived from airborne lidar data, which served as output layers. We adapted a U-Net model architecture for canopy height regression, training and validating the models with 10,000 256-by-256 pixel image patches. We evaluated three settings for the U-Net encoder depth and used both 1 m and 2 m datasets to assess their impact on model performance. Canopy height predictions from the fitted models were highly correlated (R2 = 0.70 to 0.89), precise (MAE = 1.37–2.21 m), and virtually unbiased (Bias = −0.20–0.07 m) with respect to validation data. The trained models also performed adequately well on the independent test data (R2 = 0.62–0.78, MAE = 3.06–4.1 m). Models with higher encoder depths (3,4) and trained with 2 m data provide better predictions than models with encoder depth 2 and trained on 1 m data. Inter-comparisons with existing canopy height products also showed our canopy height map provided better agreement with reference airborne lidar canopy height estimates. This study shows the potential of developing regional canopy height products using airborne lidar and NAIP imagery to support forest productivity and carbon modeling at spatially detailed scales. The 30 m canopy height map generated over Texas holds promise in advancing economic and sustainable forest management goals and enhancing decision-making in natural resource management across the state.

show abstract

Landsat-Scale Regional Forest Canopy Height Mapping Using ICESat-2 Along-Track Heights: Case Study of Eastern Texas

Cited by 11 publications

References 35 publications

Combining "Deep Learning" and Physically Constrained Neural Networks to Derive Complex Glaciological Change Processes from Modern High-Resolution Satellite Imagery: Application of the GEOCLASS-image System to Create VarioCNN for Glacier Surges

Combining "Deep Learning" and Physically Constrained Neural Networks to Derive Complex Glaciological Change Processes from Modern High-Resolution Satellite Imagery: Application of the GEOCLASS-image System to Create VarioCNN for Glacier Surges

Combining “Deep Learning” and Physically Constrained Neural Networks to Derive Complex Glaciological Change Processes from Modern High-Resolution Satellite Imagery: Application of the GEOCLASS-image System to Create VarioCNN for Glacier Surges

Image to Image Deep Learning for Enhanced Vegetation Height Modeling in Texas

Contact Info

Product

Resources

About