Temporal transferability of wildlife habitat models: implications for habitat monitoring

Abstract

Aim Temporal transferability is an important issue when habitat models are used beyond the time frame corresponding to model development, but has not received enough attention, particularly in the context of habitat monitoring. While the combination of remote sensing technology and habitat modelling provides a useful tool for habitat monitoring, the effect of incorporating remotely sensed data on model transferability is unclear. Therefore, our objectives were to assess how different satellite-derived variables affect temporal transferability of habitat models and their usefulness for habitat monitoring.Location Wolong Nature Reserve, Sichuan Province, China.Methods We modelled giant panda habitat with the maximum entropy algorithm using panda presence data collected in two time periods and four different sets of predictor variables representing land surface phenology. Each predictor variable set contained either a time series of smoothed wide dynamic range vegetation index (WDRVI) or 11 phenology metrics, both derived from single-year or multi-year (i.e. 3-year) remotely sensed imagery acquired by the Moderate Resolution Imaging Spectroradiometer (MODIS). We evaluated the ability of models obtained with these four variable sets to predict giant panda habitat within and across time periods by using threshold-independent and threshold-dependent evaluation methods and five indices of temporal transferability. ResultsOur results showed that models developed with the four variable sets were all useful for characterizing and monitoring giant panda habitat. However, the models developed using multi-year data exhibited significantly higher temporal transferability than those developed using single-year data. In addition, models developed with phenology metrics, especially when using multi-year data, exhibited significantly higher temporal transferability than those developed with the time series.Main conclusions The integration of land surface phenology, captured by high temporal resolution remotely sensed imagery, with habitat modelling constitutes a suitable tool for characterizing wildlife habitat and monitoring its temporal dynamics. Using multi-year phenology metrics reduces model complexity, multicollinearity among predictor variables and variability caused by interannual climatic fluctuations, thereby increasing the temporal transferability of models. This study provides useful guidance for habitat monitoring through the integration of remote sensing technology and habitat modelling, which may be useful for the conservation of the giant panda and many other species.

Impact of livestock on giant pandas and their habitat

“…Giant panda habitat suitability models often rely on binary classifications of forest versus non-forest as a primary measure to delineate areas suitable for panda inhabitance (Liu et al 1999Wang et al 2010). Our findings suggest that it is also important to include other human disturbances in addition to timber harvesting He et al 2009;Linderman et al 2006;Tuanmu et al 2011), as current forested areas in otherwise suitable giant panda habitat may be subjected to threats such as livestock grazing that may only be detected using field surveys.…”

Mapping Rubber Plantations and Natural Forests in Xishuangbanna (Southwest China) Using Multi-Spectral Phenological Metrics from MODIS Time Series

“…These more simple metrics facilitate the interpretation of phenological differences and their importance for land cover classifications. These more generalized metrics might also increase the transferability of the results to other study areas [49]. Third, metrics that have been identified as relevant for differentiating between natural forests and rubber plantations can be monitored over time more easily than tracking changes in multi-dimensional annual time series.…”

“…This suggests that the phenological metrics were overall sufficient to capture the spectral-temporal differences between the different land cover classes, but using the full time series generally achieved higher accuracies. Nonetheless, using phenological metrics has several advantages: The feature space is substantially reduced (i.e., 46 vs. 18 features for the EVI-SWIR model), which reduces redundancies and multi-collinearity in the explanatory variable set [49]. Second, phenological differences in land cover can be broken down into single key metrics describing a specific land cover.…”

Outstanding Challenges in the Transferability of Ecological Models

“…Similarly, projections of Asian tiger mosquito (Aedes albopictus) distribution models onto all continents helped identify areas at greatest risk of invasion, with important implications for human health [81]. Temporal transfers have largely been applied to forecast species' responses to climate warming, retrospectively describe pristine population states, characterize evolutionary patterns of speciation, quantify the repercussions of land use changes, or estimate future ecosystem dynamics (e.g., [68,72,82]). Despite being difficult to quantify, the societal and economic gains from transferring models can be substantial, and are most readily illustrated by the mitigation of costs associated with invasive species [83].…”