How well does random forest analysis model deforestation and forest fragmentation in the Brazilian Atlantic forest?

Zanella, Lisiane; Folkard, Andrew M.; Blackburn, George Alan; Carvalho, Luís Marcelo Tavares de

doi:10.1007/s10651-017-0389-8

Cited by 15 publications

(7 citation statements)

References 64 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We attributed importance of climate (ET, PR, RO, and SWE) and human water use (irrigated agriculture and population density) to the prediction of lake and wetland surface water area using randomForestSRC regression tree analysis (Ishwaran & Kogalur, ), as a nonparametric measure of variable importance (VIMP). This approach is applicable to ecological systems with typically non‐normal distributions, which most of our variables showed through time (Cutler et al, ; De'ath & Fabricius, ; Zanella, Folkard, Blackburn, & Carvalho, ). RandomForestSRC allowed for a two‐step method of randomization to de‐correlate trees, which decreased variance and bias for a stronger representative model (Zhang & Lu, ).…”

Section: Methodsmentioning

confidence: 99%

Climate and human water use diminish wetland networks supporting continental waterbird migration

Donnelly

King

Silverman

et al. 2020

Global Change Biology

View full text Add to dashboard Cite

Migrating waterbirds moving between upper and lower latitudinal breeding and wintering grounds rely on a limited network of endorheic lakes and wetlands when crossing arid continental interiors. Recent drying of global endorheic water stores raises concerns over deteriorating migratory pathways, yet few studies have considered these effects at the scale of continental flyways. Here, we investigate the resiliency of waterbird migration networks across western North America by reconstructing long‐term patterns (1984–2018) of terminal lake and wetland surface water area in 26 endorheic watersheds. Findings were partitioned regionally by snowmelt‐ and monsoon‐driven hydrologies and combined with climate and human water‐use data to determine their importance in predicting surface water trends. Nonlinear patterns of lake and wetland drying were apparent along latitudinal flyway gradients. Pervasive surface water declines were prevalent in northern snowmelt watersheds (lakes −27%, wetlands −47%) while largely stable in monsoonal watersheds to the south (lakes −13%, wetlands +8%). Monsoonal watersheds represented a smaller proportion of total lake and wetland area, but their distribution and frequency of change within highly arid regions of the continental flyway increased their value to migratory waterbirds. Irrigated agriculture and increasing evaporative demands were the most important drivers of surface water declines. Underlying agricultural and wetland relationships however were more complex. Approximately 7% of irrigated lands linked to flood irrigation and water storage practices supported 61% of all wetland inundation in snowmelt watersheds. In monsoonal watersheds, small earthen dams, meant to capture surface runoff for livestock watering, were a major component of wetland resources (67%) that supported networks of isolated wetlands surrounding endorheic lakes. Ecological trends and human impacts identified herein underscore the importance of assessing flyway‐scale change as our model depictions likely reflect new and emerging bottlenecks to continental migration.

show abstract

Section: Methodsmentioning

confidence: 99%

Climate and human water use diminish wetland networks supporting continental waterbird migration

Donnelly

King

Silverman

et al. 2020

Global Change Biology

View full text Add to dashboard Cite

show abstract

“…latitude, longitude). Therefore, many studies have unsurprisingly identified coordinates as one of the, if not the, most important predictor, such as for tree species distribution (Attorre et al, 2011), monthly precipitation (Jing et al, 2016), deforestation (Zanella et al, 2017), phytoplankton abundance (Roubeix et al, 2016) and explaining the spatial variability of soil organic carbon . According to our results, spatial variable selection would have very likely removed geolocation variables from these studies models.…”

Section: Relevance Of Spatial Variable Selectionmentioning

confidence: 99%

Importance of spatial predictor variable selection in machine learning applications – Moving from data reproduction to spatial prediction

Meyer

Reudenbach

Wöllauer

et al. 2019

Ecological Modelling

257

158

View full text Add to dashboard Cite

Machine learning algorithms find frequent application in spatial prediction of biotic and abiotic environmental variables. However, the characteristics of spatial data, especially spatial autocorrelation, are widely ignored. We hypothesize that this is problematic and results in models that can reproduce training data but are unable to make spatial predictions beyond the locations of the training samples. We assume that not only spatial validation strategies but also spatial variable selection is essential for reliable spatial predictions.We introduce two case studies that use remote sensing to predict land cover and the leaf area index for the "Marburg Open Forest", an open research and education site of Marburg University, Germany. We use the machine learning algorithm Random Forests to train models using non-spatial and spatial cross-validation strategies to understand how spatial variable selection affects the predictions.Our findings confirm that spatial cross-validation is essential in preventing overoptimistic model performance. We further show that highly autocorrelated predictors (such as geolocation variables, e.g. latitude, longitude) can lead to considerable overfitting and result in models that can reproduce the training data but fail in making spatial predictions. The problem becomes apparent in the visual assessment of the spatial predictions that show clear artefacts that can be traced back to a misinterpretation of the spatially autocorrelated predictors by the algorithm. Spatial variable selection could automatically detect and remove such variables that lead to overfitting, resulting in reliable spatial prediction patterns and improved statistical spatial model performance.We conclude that in addition to spatial validation, a spatial variable selection must be considered in spatial predictions of ecological data to produce reliable predictions.

show abstract

“…Statistical methods are commonly used to analyze the relationship between wetland loss and explanatory variables, such as partial least squares regression, partial correlation analysis, linear correlation, logistic regression, gray correlation, and multiple stepwise regression ( van Asselen et al 2013;Meng et al 2017). Meanwhile, machine learning (e.g., random forest method) methods have gradually been applied in land use change studies (Zanella et al 2017).…”

Section: Introductionmentioning

confidence: 99%

The spatiotemporal changes of marshland and the driving forces in the Sanjiang Plain, Northeast China from 1980 to 2016

Liu

et al. 2020

Ecol Process

View full text Add to dashboard Cite

Background Wetland loss is a global concern due to its enormous ecosystem services. Marshland, a typical natural wetland, which is concentrated in the Sanjiang Plain, has undergone dramatic loss in the last several decades. The spatiotemporal changes in marshland were studied based on Landsat images of the Sanjiang Plain from 1980 to 2016 with the land use maps in 1980, 1995, 2000, 2005, 2010, and 2016 using land use dynamic degree and landscape indices. The driving forces of marshland loss, including biophysical factors, socio-economic factors, and land management, were analyzed with boosted regression trees (BRTs) methods. Results The area of marshland loss was 7372 km2, which accounted for 65.7% of the area of marshland in 1980; however, the paddy field area was expanded by 22,313 km2. The lost marshland was mainly converted to dry farmland (47.5%) and paddy field (47.2%) during 1980–2016. Both the landscape pattern of the study area and marshland became increasingly fragmented. The relatively important factors responsible for marshland loss were biophysical factors, socio-economic factors, and land management, which accounted for 65.2%, 25.5%, and 8.4%, respectively. The most important driving forces with high “relative influence” were “distance to river,” which accounted for 20.6% of the total variance explained. The “relative influence” of potential crop yield and ditch density reached 20.2% and 8.4%, respectively. Conclusion Significant land use changes have occurred in the Sanjiang Plain over the past 37 years, with characteristics of rapid paddy field expansion and drastic marshland loss. Meanwhile, marshland fragmentation continued to increase. Reclamation was the main reason for the large-scale marshland loss. Biophysical factors influenced the decisions regarding the locations of marshland loss. Moreover, climate factors (i.e., average annual precipitation and average annual temperature) also played an important role in marshland loss. These results can provide helpful knowledge for understanding the patterns and reasons for marshland loss and protecting and managing strategies for wetlands restoration.

show abstract

How well does random forest analysis model deforestation and forest fragmentation in the Brazilian Atlantic forest?

Cited by 15 publications

References 64 publications

Climate and human water use diminish wetland networks supporting continental waterbird migration

Climate and human water use diminish wetland networks supporting continental waterbird migration

Importance of spatial predictor variable selection in machine learning applications – Moving from data reproduction to spatial prediction

The spatiotemporal changes of marshland and the driving forces in the Sanjiang Plain, Northeast China from 1980 to 2016

Contact Info

Product

Resources

About