Remote sensing data can improve predictions of species richness by stacked species distribution models: a case study for Mexican pines

Abstract

Aim Remote sensing data have been used in a growing number of studies to directly predict species richness or to improve the performance of species distribution models (SDMs), but their suitability for stacked species distribution models (S-SDMs) remains unclear. In this case study, we evaluated the potential and limitations of remotely sensed data in S-SDMs and addressed the commonly observed overestimation of species richness by S-SDMs.Location Mexico.Methods Phenological and statistical metrics were derived from remotely sensed time series data (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)) of the Terra-MODIS enhanced vegetation index and land surface temperature products. In a series of climatic and remote sensing-based SDMs, the distribution ranges of 40 species of the genus Pinus (Pinaceae) were modelled based on presence-only herbarium and field data using the maximum entropy algorithm and summed to estimate species richness. Three different species-specific thresholds were applied to convert continuous model predictions into binary maps. Modelled species richness was compared to independent data from the Mexican National Forest Inventory. ResultsThe inclusion of remote sensing data led to significantly better predictions of species richness in comparison to the climate-based models for the summed suitabilities and all thresholds considered. Both climatic and remote sensing-based models allowed us to identify the areas with the highest pine species richness based on presence-only data. Remote sensing-based models compare closely with climate-derived patterns, but provide better spatial resolution and more detailed information on local habitat availability. Main conclusionsThe results of this case study provide general guidance for the potential and limitations of using remote sensing data in S-SDMs. Our results confirmed that remote sensing data may not only have the capability for improving individual SDMs, but also can be a potential tool for reducing the overestimation of species richness by S-SDMs. This approach opens up new possibilities for species richness predictions in areas where biological survey data are scarce and where no species richness inventory data exist.

Predicting Spatial Distribution of Key Honeybee Pests in Kenya Using Remotely Sensed and Bioclimatic Variables: Key Honeybee Pests Distribution Models

“…The fine spatial resolution of remotely sensed variables reduced the errors that emanate from the generalization in bioclimatic variables over given spatial extents. Moreover, the model products (prediction maps) were more realistic in predicting habitat suitability in model 3 and they had reduced overestimation [20,57].…”

“…Due to increased replication, random seed was used to select different random test/run partition for each run. We used the '10 percentile training presence threshold' which predicts the 10% most extreme presence observation as absent [20] to eliminate outliers from the model.…”

“…This is because remotely sensed variables are continuous observations with high spatial and temporal resolution compared to bioclimatic variables that are modelled or interpolated [14]. Well processed remotely sensed data also help in reducing over-fitting (over predicting) of the EN models as such improving the accuracy and value of species and pest distribution estimates [20].…”

“…A fine and realistic geographic distribution of honeybee pest species is attained using remotely sensed variables such as vegetation 'greenness' dynamics [20]. This is because remotely sensed variables are continuous observations with high spatial and temporal resolution compared to bioclimatic variables that are modelled or interpolated [14].…”

“…State of the art remote sensing techniques can be used to derive phenological variables (such as start and end of the season) from remotely sensed data that offer fine scaled and specific information for prediction of the spread of honeybee pests. A model that links remotely sensed and bioclimatic variables provides an operational system to monitor long term changes related to the spread of pests [20]. It also offers a tool that can model futuristic scenarios of the honeybee pest distribution since projected climate and phenological responses can be modelled.…”