Building multi-scale EGLSN system and simulating cultivated land productivity

Evaluation of cultivated land productivity is the basis of fertilization. Traditional evaluation of cultivated land productivity is based on the Delphi method and fuzzy mathematics theory to establish analytic hierarchy process (AHP) framework using integrated index and split it into given grades at county level. Traditional methods need to consider factors selection depending on regional variation and requires spatially explicit information on cultivated land distribution and soil property, which is not all available. The traditional methods also cannot figure out the scales of cultivated land productivity and the inner relations between the factors. In order to overcome these deficiencies, this paper used GPS sampling points data of soil survey and statistical yearbook data to build a multi-scale EGLSN system, which was in turn used to simulate cultivated land productivity in Shanxi Province on the basis of 'process-scale-hierarchy' theory of landscape ecology. The multi-scale EGLSN system included five scales, from macro to micro which were eco-climate, geomorphological landscape, land use and measurement, soil condition and nutrient management scales. The eco-climate scale was set up as the first scale due to high correlation between crop yield and climate factors. The other scales were selected by analyzing soil survey data with principal component analysis (PCA) method. The control area of each scale was calculated in ArcGIS. The crop yield of eco-climate scale was used as the basic standard of cultivated land productivity, which was from yearbook. The models of productivity on other four scales were integrated based on the top scale. The paper run the models in different scales in three different parts of Shanxi Province ― Xinfu District (XD), Yuci District (YC) and Xiangfen County (XF), from north to south, to evaluate the cultivated land productivity. Standard root mean square error (NRMSE) of multi-scale ENLSN evaluation results was calculated with simulated data and GPS sampling points data. NRMSEs of different scales for XD were 31% (geomorphological landscape scale), 26% (land use and measure scale), 14% (soil condition scale) and 3% (nutrient management scale). NRMSEs for YC were 32% (geomorphological landscape scale), 27% (land use and measure scale), 15% (soil condition scale) and 4% (nutrient management scale). Then NRMSEs for XF were 35% (geomorphological landscape scale), 28% (land use and measure scale), 13% (soil condition scale) and 7% (nutrient management scale). Concurrently, on geomorphological landscape scale, the precisions of total productivity in each region relative to eco-climate total productivity were 83% (XD), 80% (YC) and 82% (XF); the ones on land use and measurement scale were 93% (XD), 90% (YC) and 91% (XF); the ones on soil condition scale were 95% (XD), 95% (YC) and 91% (XF); while the ones on nutrient management scale were 96% (XD), 95% (YC) and 93% (XF). The results therefore showed that it was feasible to apply the EGLSN method in simulating cultivated land productivity. The EGLSN models described in this paper extended present understanding of spatial distribution of cultivated land productivity in Shanxi Province with increased accuracy. The multi-scale EGLSN method was preferable over the traditional method because it not only simulated cultivated land productivity at multiple scales, but also provided reference for simulations of cultivated land productivity on larger scales and regional formula fertilization.