Null models reveal preferential sampling, spatial autocorrelation and overfitting in habitat suitability modelling

Abstract

Nowadays, species are driven to extinction at a high rate. To reduce this rate it is important to delineate suitable habitats for these species in such a way that these areas can be suggested as conservation areas. The use of habitat suitability models (HSMs) can be of great importance for the delineation of such areas. In this study Maxent, a presence-only modelling technique, is used to develop HSMs for 223 nematode species of the Southern Bight of the North Sea. However, it is essential that these models are beyond discussion and they should be checked for potential errors. In this study we focused on two categories (1) errors which can be attributed to the database such as preferential sampling and spatial autocorrelation and (2) errors induced by the modelling technique such as overfitting. In order to quantify these adverse effects thousands of nulls models were created. The effect of preferential sampling (i.e. some areas where visited more frequently than others) was investigated by comparing null models sampling the actual sampling stations with null models sampling the entire mapping area (Raes and ter Steege, 2007). Overfitting is exposed by a fivefold cross-validation and the influence of spatial autocorrelation is assessed by separating test and training sets in space. Our results clearly show that all these effects are present: preferential sampling has a strong effect on the selection of non-random species models. Cross-validation seems to have less influence on the model selection and spatial autocorrelation is also strongly present. It is clear from this study that predefined thresholds are not readily applicable to all datasets and additional tests are needed in model selection.

A maximum entropy method to extract urban land by combining MODIS reflectance, MODIS NDVI, and DMSP-OLS data

“…We can see from the response curves that urban areas are likely to occur in certain spectral conditions in NDVI and bands 1-7 of MOD09A1, and the probability of being positive increases with the DN value of DMSP-OLS data. In addition, an AUC value of 0.884 suggests that the model is well-performed (Phillips, Anderson, and Schapire 2006;Merckx et al 2011). Since the response curves and ROC curves of different scenes are quite similar, the results of the proposed method are reliable.…”

“…The logistic output format was chosen because it can achieve better performance compared with cumulative and raw output formats (Phillips and Dudík 2008). Moreover, we chose the default option 'auto features', which automatically determines the most suitable complexity based on the sample size of presence records (Merckx et al 2011;Syfert, Smith, and Coomes 2013). The random test percentage was set as 25% (Young, Carter, and Evangelista 2011).…”

Identifying and managing risk factors for salt-affected soils: a case study in a semi-arid region in China

“…However, traditional statistical techniques such as linear regression model (Zhang et al 2010b), multiple gray relation model (Rao and Yadava 2009), and system dynamic model (Ali Kerem and Yaman 2001) are limited when they are used to analyze spatial data due to spatial autocorrelations in geographic variables (Overmars et al 2003;Merckx et al 2011;Naimi et al 2011). Spatial autoregressive (SAR) model (Anselin and Griffith 1988) is a powerful tool for spatial analysis (Kissling and Carl 2008; see also Aguiar et al 2007;Kissling and Carl 2008), and it can be used to examine the relationship between soil salinity and its risk factors (Akramkhanov et al 2011).…”

Modelling the habitat use and distribution of the threatened Javan slow loris Nycticebus javanicus

“…org/ web/ 2012 040 21 4 3802/ https:// www1 .nga.mil/ Products Services/ Geo desy and Geophysics/ World Geodetic System/ Pages/ default .aspx) were se lected. To avoid model overfitting (Dormann et al 2007, Merckx et al 2011), we corrected occurrence data for spatial autocorrelation by selecting 1 random locality in a 5 km radius, resulting in n = 13 occurrence points (n = 56, i.e. 52 sightings from the present study and 4 additional localities from independent surveys).…”