The Effects of Point or Polygon Based Training Data on RandomForest Classification Accuracy of Wetlands

Corcoran, Jennifer; Knight, Joseph F.; Pelletier, Keith C.; Rampi, Lian P.; Wang, Yan

doi:10.3390/rs70404002

Cited by 32 publications

(17 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, perennial canopy cover from coniferous forests can obscure small wetlands and transitional zones from optical sensors, particularly moderate resolution Landsat data (Corcoran et al. ; Gallant ). These issues are further compounded by the relative scarcity of wetlands and riparian areas on the landscape, which makes it difficult to distinguish small wetlands nested within larger land cover classes (Wright and Gallant ; Millard and Richardson ).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

An integrative modeling approach to mapping wetlands and riparian areas in a heterogeneous Rocky Mountain watershed

Chignell

Luizza

Young

et al. 2017

Remote Sens Ecol Conserv

View full text Add to dashboard Cite

Accurate maps of wetlands and riparian areas are critical for targeting conservation and monitoring efforts. However, detailed inventories in mountain regions are largely non-existent, as conventional mapping approaches are hindered by high costs, remoteness, and landscape variability. Contemporary modeling techniques can circumvent many of these issues, but are often difficult to interpret and tend to rely on specialized datasets that prevent their wider application. In this study, we used machine learning, Landsat 8 imagery and geomorphometric indices to map the distribution of wetlands and riparian areas in the Cache la Poudre River watershed, Colorado, USA. We used a presence-background approach to develop and compare predictions from three popular algorithms: boosted regression trees, MaxEnt and random forests. In addition, we developed the models within three elevation-based life zones to account for altitudinal changes in ecohydrology and land use. Our results showed strong predictive performance, with top-performing models achieving area under the curve values as high as 0.98 and correctly classifying up to 95% of test data. Model performance varied by elevation zone, and no algorithm consistently outperformed the others. The boosted regression trees approach was uniquely able to differentiate wetlands from irrigated agriculture and residential areas in lower elevations. Multi-seasonal greenness and wetness indices were highly influential predictors in all models, underscoring the importance of capturing local phenological characteristics and hydrological regimes. Dissection and roughness terrain metrics were key predictors for identifying valley bottom meadows and emergent wetlands in high-elevation forests. We demonstrate how integrating ecological interpretation into the modeling workflow can inform conventional accuracy statistics and help bridge field-based and remote sensing perspectives. We also show how continuous model outputs can facilitate this process by depicting nuances of the wetland-upland continuum. Our approach requires only public data that are widely available, and can be easily adapted to other heterogeneous mountain settings.

show abstract

Section: Introductionmentioning

confidence: 99%

“…This is particularly important in forested regions, as remotely sensed elevation data can describe the vertical structure of terrain obscured by canopy (Corcoran et al. ). LiDAR DEMs have been shown to be especially effective for capturing these features (Millard and Richardson ; Huang et al.…”

Section: Introductionmentioning

confidence: 99%

An integrative modeling approach to mapping wetlands and riparian areas in a heterogeneous Rocky Mountain watershed

Chignell

Luizza

Young

et al. 2017

Remote Sens Ecol Conserv

View full text Add to dashboard Cite

show abstract

“…A LiDAR dataset is available for the entire bog and surrounding area. A number of recent studies have demonstrated that LiDAR derivatives provide superior classification accuracies compared with use of optical and radar imagery (e.g., [25][26][27][28][29]) for both general land cover mapping and specifically in peatland mapping. Optical imagery captures spectral reflectance from the different vegetation species visible from above.…”

Section: Study Area and Datamentioning

confidence: 99%

On the Importance of Training Data Sample Selection in Random Forest Image Classification: A Case Study in Peatland Ecosystem Mapping

2015

View full text Add to dashboard Cite

Random Forest (RF) is a widely used algorithm for classification of remotely sensed data. Through a case study in peatland classification using LiDAR derivatives, we present an analysis of the effects of input data characteristics on RF classifications (including RF out-ofbag error, independent classification accuracy and class proportion error). Training data selection and specific input variables (i.e., image channels) have a large impact on the overall accuracy of the image classification. High-dimension datasets should be reduced so that only uncorrelated important variables are used in classifications. Despite the fact that RF is an ensemble approach, independent error assessments should be used to evaluate RF results, and iterative classifications are recommended to assess the stability of predicted classes. Results are also shown to be highly sensitive to the size of the training data set. In addition to being as large as possible, the training data sets used in RF classification should also be (a) randomly distributed or created in a manner that allows for the class proportions of the training data to be representative of actual class proportions in the landscape; and (b) should have minimal spatial autocorrelation to improve classification results and to mitigate inflated estimates of RF out-of-bag classification accuracy.

show abstract

“…Corcoran and colleagues [86] evaluated the benefits of using points, point-centered windows, and image objects to train the Random Forests decision-tree classifier, then considered the benefits contributed by different variables derived from color-infrared orthoimagery and lidar data to generate classifications for two wetland landscapes in Minnesota, USA. Millard and Richardson [87] used lidar-derived variables to evaluate a suite of effects of different characteristics of training data selection on the Random Forests classifier.…”

Section: Purpose Of This Special Issuementioning

confidence: 99%

The Challenges of Remote Monitoring of Wetlands

2015

View full text Add to dashboard Cite

Wetlands are highly productive and support a wide variety of ecosystem goods and services. Various forms of global change impose compelling needs for timely and reliable information on the status of wetlands worldwide, but several characteristics of wetlands make them challenging to monitor remotely: they lack a single, unifying land-cover feature; they tend to be highly dynamic and their energy signatures are constantly changing; and steep environmental gradients in and around wetlands produce narrow ecotones that often are below the resolving capacity of remote sensors. These challenges and needs set the context for a special issue focused on wetland remote sensing. Contributed papers responded to one of three overarching questions aimed at improving remote, large-area monitoring of wetlands: (1) What approaches and data products are being developed specifically to support regional to global long-term monitoring of wetland landscapes? (2) What are the promising new technologies and sensor/multisensor approaches for more accurate and consistent detection of wetlands? (3) Are there studies that demonstrate how remote long-term monitoring of wetland landscapes can reveal changes that correspond with changes in land cover and land use and/or changes in climate?

show abstract

The Effects of Point or Polygon Based Training Data on RandomForest Classification Accuracy of Wetlands

Cited by 32 publications

References 42 publications

An integrative modeling approach to mapping wetlands and riparian areas in a heterogeneous Rocky Mountain watershed

An integrative modeling approach to mapping wetlands and riparian areas in a heterogeneous Rocky Mountain watershed

On the Importance of Training Data Sample Selection in Random Forest Image Classification: A Case Study in Peatland Ecosystem Mapping

The Challenges of Remote Monitoring of Wetlands

Contact Info

Product

Resources

About