Title:Estimating mesoscale linkage between land-surface conditions and marine productions in Japan: a study using a sparse high-dimensional model

Abstract:There have been several scientific studies concerning the interconnectedness between land-surface conditions (e.g., vegetation, land use, and socioeconomic activities) and marine ecosystems (e.g., biodiversity, primary production, and seafood production). This idea of the land-sea connectivity sounded reasonable for many scientists because there is an obvious connection by rivers, however, quantitative estimation of this relationship has been thought to be difficult due to the size of the target areas, the numbers of possible variables, and the amount of noises in this complex system. In this study, we applied a sparse high-dimensional modeling to overcome these difficulties and found several significant land-sea linkages. In this modeling, the key is the penalization in the number of independent variables, and thus a limited number of significant independent variables are chosen in an objective and quantitative manner. We selected 448 independent variables (geological, biological, and social) and 68 dependent variables (marine products) from a governmental database. Then we summarized the data according to the political boundaries of 47 prefectures. The sparse high-dimensional model we constructed successfully highlighted several significant variables to estimate the amount of marine products by prefecture. For example, we found that coastal, especially sessile marine products such as seaweed and shellfish had more explanatory variables than open-water marine products such as tuna and sailfishes. In addition, salmon production had a strong connection to the mesoscale (prefecture-level) land-surface conditions possibly due to their interconnected life cycle between freshwater rivers and the sea. We believe that the sparse modeling is an effective statistical tool to explain relationships in a complex system such as land-sea connections.