Farmland quality assessment using deep fully convolutional neural networks.

Farmland is the cornerstone of agriculture and is important for food security and social production. Farmland assessment is essential but traditional methods are usually expensive and slow. Deep learning methods have been developed and widely applied recently in image recognition, semantic understanding, and many other application domains. In this research, we used fully convolutional networks (FCN) as the deep learning model to evaluate farmland grades. Normalized difference vegetation index (NDVI) derived from Landsat images was used as the input data, and the China National Cultivated Land Grade Database within Jiangsu Province was used to train the model on cloud computing. We also applied an image segmentation method to improve the original results from the FCN and compared the results with classical machine learning (ML) methods. Our research found that the FCN can predict farmland grades with an overall F1 score (the harmonic mean of precision and recall) of 0.719 and F1 score of 0.909, 0.590, 0.740, 0.642, and 0.023 for non-farmland, level I, II, III, and IV farmland, respectively. Combining the FCN and image segmentation method can further improve prediction accuracy with results of fewer noise pixels and more realistic edges. Compared with conventional ML, at least in farmland evaluation, FCN provides better results with higher precision, recall, and F1 score. Our research indicates that by using remote sensing NDVI data, the deep learning method can provide acceptable farmland assessment without fieldwork and can be used as a novel supplement to traditional methods. The method used in this research will save a lot of time and cost compared with traditional means.

Determination of influencing factors on historical concentration variations of PAHs in West Taihu Lake, China.

The adsorption of polycyclic aromatic hydrocarbons (PAHs) by components such as elemental carbon (EC), total organic carbon (TOC), and particles is different, and EC and PAHs are good materials for reconstructing historical human activity patterns and pollution conditions. In this study, the effects of EC (soot and char), TOC and particles of different grain size on PAHs in surface sediments were quantitatively analysed, and their historical concentrations in a sediment core from western Taihu Lake were reconstructed. The contents of soot, TOC, clay, EC and char explained 57.2%, 27.6%, 26.0%, 24.0% and 16.4%, respectively, of the PAH concentrations in surface sediments. The correlation between the soot and PAH levels was significantly higher than that between the char, TOC, and clay contents and PAH levels, and PAHs were mainly affected by the local economic development and human activity, as indicated by metrics of population, highway mileage, coal burning, and industrial output. With the development of the economy of the Taihu Lake Basin, the composition of PAHs in the sediments has changed: the proportion of low-molecular-weight PAHs decreased from 42.4% to 17.5%, and that of high-molecular-weight PAHs increased from 58.7% to 82.5%. The concentration of PAHs in pore water from Taihu Lake over the past 100 years was reconstructed and ranged from 43.1 to 961.2 µg L-1, with an average of 180.7 µg L-1. After China's reform and opening up, the concentrations of various PAHs in Taihu Lake changed from safe to chronic pollution levels. The ratios of lead (Pb) isotopes and the diagnostic ratios of PAHs showed that the main sources of PAHs in western Taihu Lake sediments were human activities such as coal and petroleum combustion.

Influence of Industrialization and Environmental Protection on Environmental Pollution: A Case Study of Taihu Lake, China.

In order to quantitatively study the effect of environmental protection in China since the twenty-first century and the environmental pollution projected for the next ten years (under the model of extensive economic development), this paper establishes a Bayesian regulation back propagation neural network (BRBPNN) to analyze the typical pollutants (i.e., cadmium (Cd) and benzopyrene (BaP)) for Taihu Lake, a typical Chinese freshwater lake. For the periods 1950⁻2003 and 1950⁻2015, the neural network model estimated the BaP concentration for the database with Nash-Sutcliffe model efficiency (NS) = 0.99 and 0.99 and root-mean-square error (RMSE) = 3.1 and 9.3 for the total database and the Cd concentration for the database with NS = 0.93 and 0.98 and RMSE = 45.4 and 65.7 for the total database, respectively. In the model of extensive economic development, the concentration of pollutants in the sediments of Taihu reached the maximum value at the end of the twentieth century and early twenty-first century, and there was an inflection point. After the early twenty-first century, the concentration of pollutants was controlled under various environmental policies and measures. In 2015, the environmental protection ratio of Cd and BaP reached 52% and 89%, respectively. Without environmental protection measures, the concentrations of Cd and BaP obtained from the neural network model is projected to reach 2015.5 µg kg-1 and 407.8 ng g-1, respectively, in 2030. Based on the results of this study, the Chinese government will need to invest more money and energy to clean up the environment.

Comparison Study on the Estimation of the Spatial Distribution of Regional Soil Metal(loid)s Pollution Based on Kriging Interpolation and BP Neural Network.

Soil pollution by metal(loid)s resulting from rapid economic development is a major concern. Accurately estimating the spatial distribution of soil metal(loid) pollution has great significance in preventing and controlling soil pollution. In this study, 126 topsoil samples were collected in Kunshan City and the geo-accumulation index was selected as a pollution index. We used Kriging interpolation and BP neural network methods to estimate the spatial distribution of arsenic (As) and cadmium (Cd) pollution in the study area. Additionally, we introduced a cross-validation method to measure the errors of the estimation results by the two interpolation methods and discussed the accuracy of the information contained in the estimation results. The conclusions are as follows: data distribution characteristics, spatial variability, and mean square errors (MSE) of the different methods showed large differences. Estimation results from BP neural network models have a higher accuracy, the MSE of As and Cd are 0.0661 and 0.1743, respectively. However, the interpolation results show significant skewed distribution, and spatial autocorrelation is strong. Using Kriging interpolation, the MSE of As and Cd are 0.0804 and 0.2983, respectively. The estimation results have poorer accuracy. Combining the two methods can improve the accuracy of the Kriging interpolation and more comprehensively represent the spatial distribution characteristics of metal(loid)s in regional soil. The study may provide a scientific basis and technical support for the regulation of soil metal(loid) pollution.

Optimization of Sample Points for Monitoring Arable Land Quality by Simulated Annealing while Considering Spatial Variations.

With China's rapid economic development, the reduction in arable land has emerged as one of the most prominent problems in the nation. The long-term dynamic monitoring of arable land quality is important for protecting arable land resources. An efficient practice is to select optimal sample points while obtaining accurate predictions. To this end, the selection of effective points from a dense set of soil sample points is an urgent problem. In this study, data were collected from Donghai County, Jiangsu Province, China. The number and layout of soil sample points are optimized by considering the spatial variations in soil properties and by using an improved simulated annealing (SA) algorithm. The conclusions are as follows: (1) Optimization results in the retention of more sample points in the moderate- and high-variation partitions of the study area; (2) The number of optimal sample points obtained with the improved SA algorithm is markedly reduced, while the accuracy of the predicted soil properties is improved by approximately 5% compared with the raw data; (3) With regard to the monitoring of arable land quality, a dense distribution of sample points is needed to monitor the granularity.