Towards a Reproducible LULC Hierarchical Class Legend for Use in the Southwest of Pará State, Brazil: A Comparison with Remote Sensing Data-Driven Hierarchies

Reis, Mariane Souza; Escada, Maria Isabel Sobral; Dutra, Luciano Vieira; Sant’Anna, Sidnei J. S.; Vogt, Nathan

doi:10.3390/land7020065

Cited by 11 publications

(7 citation statements)

References 63 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Spatial representativeness should be considered as a limiting factor for legend design [53], and to the extent possible, researchers should attempt to use categories that are supported by both the spatial resolution of the model data and the field sampling protocols to be used; we recommend that researchers consult the extensive literature on legend design [25,[144][145][146][147][211][212][213].…”

Section: Legend Designmentioning

confidence: 99%

Accounting for Training Data Error in Machine Learning Applied to Earth Observations

et al. 2020

View full text Add to dashboard Cite

Remote sensing, or Earth Observation (EO), is increasingly used to understand Earth system dynamics and create continuous and categorical maps of biophysical properties and land cover, especially based on recent advances in machine learning (ML). ML models typically require large, spatially explicit training datasets to make accurate predictions. Training data (TD) are typically generated by digitizing polygons on high spatial-resolution imagery, by collecting in situ data, or by using pre-existing datasets. TD are often assumed to accurately represent the truth, but in practice almost always have error, stemming from (1) sample design, and (2) sample collection errors. The latter is particularly relevant for image-interpreted TD, an increasingly commonly used method due to its practicality and the increasing training sample size requirements of modern ML algorithms. TD errors can cause substantial errors in the maps created using ML algorithms, which may impact map use and interpretation. Despite these potential errors and their real-world consequences for map-based decisions, TD error is often not accounted for or reported in EO research. Here we review the current practices for collecting and handling TD. We identify the sources of TD error, and illustrate their impacts using several case studies representing different EO applications (infrastructure mapping, global surface flux estimates, and agricultural monitoring), and provide guidelines for minimizing and accounting for TD errors. To harmonize terminology, we distinguish TD from three other classes of data that should be used to create and assess ML models: training reference data, used to assess the quality of TD during data generation; validation data, used to iteratively improve models; and map reference data, used only for final accuracy assessment. We focus primarily on TD, but our advice is generally applicable to all four classes, and we ground our review in established best practices for map accuracy assessment literature. EO researchers should start by determining the tolerable levels of map error and appropriate error metrics. Next, TD error should be minimized during sample design by choosing a representative spatio-temporal collection strategy, by using spatially and temporally relevant imagery and ancillary data sources during TD creation, and by selecting a set of legend definitions supported by the data. Furthermore, TD error can be minimized during the collection of individual samples by using consensus-based collection strategies, by directly comparing interpreted training observations against expert-generated training reference data to derive TD error metrics, and by providing image interpreters with thorough application-specific training. We strongly advise that TD error is incorporated in model outputs, either directly in bias and variance estimates or, at a minimum, by documenting the sources and implications of error. TD should be fully documented and made available via an open TD repository, allowing others to replicate and assess its use. To guide researchers in this process, we propose three tiers of TD error accounting standards. Finally, we advise researchers to clearly communicate the magnitude and impacts of TD error on map outputs, with specific consideration given to the likely map audience.

show abstract

Section: Legend Designmentioning

confidence: 99%

Accounting for Training Data Error in Machine Learning Applied to Earth Observations

et al. 2020

View full text Add to dashboard Cite

show abstract

“…Additionally, an important feature in the region is the Tapajós National Forest, a forest reserve created by the Act number 73,684 of February 19, 1974, with approximately 600,000 ha. The study area has been selected due to the available information about land cover classes over the years, collected in field works realized by the research team between August and September of years 2009, 2010, and 2013, as well as those published in varied studies [22]- [24].…”

Section: A Study Areamentioning

confidence: 99%

“…Ten main land cover classes were identified in the study area [24] and organized in four different legend levels. The definition of each land cover class is presented in Table 1 and the defined legends are illustrated in Fig.…”

Section: B Land Cover Legends and Transition Matricesmentioning

confidence: 99%

Avoiding Invalid Transitions in Land Cover Trajectory Classification With a Compound Maximum a Posteriori Approach

et al. 2020

View full text Add to dashboard Cite

Classifying remote sensing time-series in land cover trajectories can provide essential information about ecosystems functioning and about the impacts of natural phenomena or human activities over the environment. The existing approaches commonly used for this task can present serious drawbacks, such as the possibility to derive invalid trajectories, use of complex data inputs, and several classification steps. To provide a simple method for land cover trajectory classification, we present a novel algorithm named Compound Maximum a Posteriori (CMAP) classifier. CMAP incorporates the knowledge of land cover dynamics and the information of multi-temporal data sets to produce only valid land cover trajectories using a global generative classification approach and simple inputs. CMAP was tested in two case studies, in which we compared land cover trajectories obtained by CMAP to those obtained using the traditional Maximum Likelihood (ML) classifier in a post-classification comparison approach. In the first case study, we classified 6 images from the same sensor, using the same land cover legend. In the second case study, we classified 3 images from different types of sensors, using different land cover legend levels. Because of its formulation, CMAP does not return invalid transitions/trajectories as classification results. The use of ML and postclassification comparison, on the other hand, resulted in invalid land cover trajectories in more than 50% of the used images in both case studies. Furthermore, the use of CMAP leads to better accuracy indexes for land cover classification of each date and reduces the classification noise. INDEX TERMS Classification algorithms, land cover trajectory mapping, multi-temporal classification, remote sensing, remote monitoring.

show abstract

“…It is a highly dynamic and heterogeneous area that presents patches of secondary vegetation, pasture, and agriculture within the forest matrix, thus being adequate for the proposed analysis. In this area, we can identify four major land cover classes (REIS et al, 2018): Forest, Pasture/Agriculture, Bare Soil, and Water.…”

Section: Study Areamentioning

confidence: 99%

Subpixel Analysis of MODIS Imagery Time Series using Transfer Learning and Relative Calibration

Moreira¹,

Reis²,

Körting³

et al. 2020

Rev. Bras. Cartogr.

View full text Add to dashboard Cite

Transfer learning reuses a pre-trained model on a new related problem, which can be useful for monitoring large areas such as the Amazon biome. A given object must have similar spectral characteristics in the data used for this type of analysis, which can be achieved using relative calibration techniques. In this article, we present a relative calibration process in multitemporal images and evaluate its impacts on a subpixel classification process. MODIS images from the Amazon region, collected between 2013 and 2017, were relatively calibrated using a 2012 image as reference and classified by transfer learning. Classifications of calibrated and uncalibrated images were compared with data from the PRODES project, focusing on forest areas. A great variation was observed in the spectral responses of the forest class, even in images of proximate dates and from the same sensor. These variations significantly impacted the land cover classifications in the subpixel, with cases of agreement between the uncalibrated data maps and PRODES of 0%. For calibrated data, the agreement values were greater than 70%. The results indicate that the method used, although quite simple, is adequate and necessary for the subpixel classification of MODIS images by transfer learning.

show abstract

Towards a Reproducible LULC Hierarchical Class Legend for Use in the Southwest of Pará State, Brazil: A Comparison with Remote Sensing Data-Driven Hierarchies

Cited by 11 publications

References 63 publications

Accounting for Training Data Error in Machine Learning Applied to Earth Observations

Accounting for Training Data Error in Machine Learning Applied to Earth Observations

Avoiding Invalid Transitions in Land Cover Trajectory Classification With a Compound Maximum a Posteriori Approach

Subpixel Analysis of MODIS Imagery Time Series using Transfer Learning and Relative Calibration

Contact Info

Product

Resources

About