Examining Scale Dependent Environmental Effects on American Lobster (Homarus americanus) Spatial Distribution in a Changing Gulf of Maine

The Gulf of Maine (GOM) is a highly complex environment and previous studies have suggested the need to account for spatial nonstationarity in species distribution models (SDMs) for the American lobster (Homarus americanus). To explore impacts of spatial nonstationarity on species distribution, we compared models with the following three assumptions : (1) large-scale and stationary relationships between species distributions and environmental variables; (2) meso-scale models where estimated relationships differ between eastern and western GOM, and (3) finer-scale models where estimated relationships vary across eastern, central, and western regions of the GOM. The spatial scales used in these models were largely determined by the GOM coastal currents. Lobster data were sourced from the Maine-New Hampshire Inshore Bottom Trawl Survey from years 2000–2019. We considered spatial and environmental variables including latitude and longitude, bottom temperature, bottom salinity, distance from shore, and sediment grain size in the study. We forecasted distributions for the period 2028–2055 using each of these models under the Representative Concentration Pathway (RCP) 8.5 “business as usual” climate warming scenario. We found that the model with the third assumption (i.e., finest scale) performed best. This suggests that accounting for spatial nonstationarity in the GOM leads to improved distribution estimates. Large-scale models revealed a tendency to estimate global relationships that better represented a specific location within the study area, rather than estimating relationships appropriate across all spatial areas. Forecasted distributions revealed that the largest scale models tended to comparatively overestimate most season × sex × size group lobster abundances in western GOM, underestimate in the western portion of central GOM, and overestimate in the eastern portion of central GOM, with slightly less consistent and patchy trends amongst groups in eastern GOM. The differences between model estimates were greatest between the largest and finest scale models, suggesting that fine-scale models may be useful for capturing effects of unique dependencies that may operate at localized scales. We demonstrate how estimates of season-, sex-, and size- specific American lobster spatial distribution would vary based on the spatial scale assumption of nonstationarity in the GOM. This information may help develop appropriate local adaptation measures in a region that is susceptible to climate change.

Spatial Downscaling of Land Surface Temperature Based on a Multi-Factor Geographically Weighted Machine Learning Model

Land surface temperature (LST) is a critical parameter of surface energy fluxes and has become the focus of numerous studies. LST downscaling is an effective technique for supplementing the limitations of the coarse-resolution LST data. However, the relationship between LST and other land surface parameters tends to be nonlinear and spatially nonstationary, due to spatial heterogeneity. Nonlinearity and spatial nonstationarity have not been considered simultaneously in previous studies. To address this issue, we propose a multi-factor geographically weighted machine learning (MFGWML) algorithm. MFGWML utilizes three excellent machine learning (ML) algorithms, namely extreme gradient boosting (XGBoost), multivariate adaptive regression splines (MARS), and Bayesian ridge regression (BRR), as base learners to capture the nonlinear relationships. MFGWML uses geographically weighted regression (GWR), which allows for spatial nonstationarity, to fuse the three base learners’ predictions. This paper downscales the 30 m LST data retrieved from Landsat 8 images to 10 m LST data mainly based on Sentinel-2A images. The results show that MFGWML outperforms two classic algorithms, namely thermal image sharpening (TsHARP) and the high-resolution urban thermal sharpener (HUTS). We conclude that MFGWML combines the advantages of multiple regression, ML, and GWR, to capture the local heterogeneity and obtain reliable and robust downscaled LST data.