In recent years, air pollution has become an important public health concern. The high concentration of fine particulate matter with diameter less than 2.5 µm (PM_2.5) is known to be associated with lung cancer, cardiovascular disease, respiratory disease, and metabolic disease. Predicting PM_2.5 concentrations can help governments warn people at high risk, thus mitigating the complications. Although attempts have been made to predict PM_2.5 concentrations, the factors influencing PM_2.5 prediction have not been investigated. In this work, we study feature importance for PM_2.5 prediction in Tehran’s urban area, implementing random forest, extreme gradient boosting, and deep learning machine learning (ML) approaches. We use 23 features, including satellite and meteorological data, ground-measured PM_2.5, and geographical data, in the modeling. The best model performance obtained was R² = 0.81 (R = 0.9), MAE = 9.93 µg/m³, and RMSE = 13.58 µg/m³ using the XGBoost approach, incorporating elimination of unimportant features. However, all three ML methods performed similarly and R² varied from 0.63 to 0.67, when Aerosol Optical Depth (AOD) at 3 km resolution was included, and 0.77 to 0.81, when AOD at 3 km resolution was excluded. Contrary to the PM_2.5 lag data, satellite-derived AODs did not improve model performance.

Remotely sensed data are often adversely affected by many types of noise, which influences the classification result. Supervised machine-learning (ML) classifiers such as random forest (RF), support vector machine (SVM), and back-propagation neural network (BPNN) are broadly reported to improve robustness against noise. However, only a few comparative studies that may help investigate this robustness have been reported. An important contribution, going beyond previous studies, is that we perform the analyses by employing the most well-known and broadly implemented packages of the three classifiers and control their settings to represent users’ actual applications. This facilitates an understanding of the extent to which the noise types and levels in remotely sensed data impact classification accuracy using ML classifiers. By using those implementations, we classified the land cover data from a satellite image that was separately afflicted by seven-level zero-mean Gaussian, salt–pepper, and speckle noise. The modeling data and features were strictly controlled. Finally, we discussed how each noise type affects the accuracy obtained from each classifier and the robustness of the classifiers to noise in the data. This may enhance our understanding of the relationship between noises, the supervised ML classifiers, and remotely sensed data.

With the economic growth and increasing urbanization in the last three decades, the air quality over China has continuously degraded, which poses a great threat to human health. The concentration of fine particulate matter (PM_2.5) directly affects the mortality of people living in the polluted areas where air quality is poor. The Beijing-Tianjin-Hebei (BTH) region, one of the well organized urban regions in northern China, has suffered with poor air quality and atmospheric pollution due to recent growth of the industrial sector and vehicle emissions. In the present study, we used the back propagation neural network model approach to estimate the spatial distribution of PM_2.5 concentration in the BTH region for the period January 2014–December 2016, combining the satellite-derived aerosol optical depth (S-DAOD) and meteorological data. The results were validated using the ground PM_2.5 data. The general method including all PM_2.5 training data and 10-fold cross-method have been used for validation for PM_2.5 estimation (R² = 0.68, RMSE = 20.99 for general validation; R² = 0.54, RMSE = 24.13 for cross-method validation). The study provides a new approach to monitoring the distribution of PM_2.5 concentration. The results discussed in the present paper will be of great help to government agencies in developing and implementing environmental conservation policy.

Surface soil moisture (SSM) retrieval over agricultural fields using synthetic aperture radar (SAR) data is often obstructed by the vegetation effects on the backscattering during the growing season. This paper reports the retrieval of SSM from RADARSAT-2 SAR data that were acquired over wheat and soybean fields throughout the 2015 (April to October) growing season. The developed SSM retrieval algorithm includes a vegetation-effect correction. A method that can adequately represent the scattering behavior of vegetation-covered area was developed by defining the backscattering from vegetation and the underlying soil individually to remove the effect of vegetation on the total SAR backscattering. The Dubois model was employed to describe the backscattering from the underlying soil. A modified Water Cloud Model (MWCM) was used to remove the effect of backscattering that is caused by vegetation canopy. SSM was derived from an inversion scheme while using the dual co-polarizations (HH and VV) from the quad polarization RADARSAT-2 SAR data. Validation against ground measurements showed a high correlation between the measured and estimated SSM (R² = 0.71, RMSE = 4.43 vol.%, p < 0.01), which suggested an operational potential of RADARSAT-2 SAR data on SSM estimation over wheat and soybean fields during the growing season.

Spatially-detailed forest height data are useful to monitor local, regional and global carbon cycle. LiDAR remote sensing can measure three-dimensional forest features but generating spatially-contiguous forest height maps at a large scale (e.g., continental and global) is problematic because existing LiDAR instruments are still data-limited and expensive. This paper proposes a new approach based on an artificial neural network (ANN) for modeling of forest canopy heights over the China continent. Our model ingests spaceborne LiDAR metrics and multiple geospatial predictors including climatic variables (temperature and precipitation), forest type, tree cover percent and land surface reflectance. The spaceborne LiDAR instrument used in the study is the Geoscience Laser Altimeter System (GLAS), which can provide within-footprint forest canopy heights. The ANN was trained with pairs between spatially discrete LiDAR metrics and full gridded geo-predictors. This generates valid conjugations to predict heights over the China continent. The ANN modeled heights were evaluated with three different reference data. First, field measured tree heights from three experiment sites were used to validate the ANN model predictions. The observed tree heights at the site-scale agreed well with the modeled forest heights (R = 0.827, and RMSE = 4.15 m). Second, spatially discrete GLAS observations and a continuous map from the interpolation of GLAS-derived tree heights were separately used to evaluate the ANN model. We obtained R of 0.725 and RMSE of 7.86 m and R of 0.759 and RMSE of 8.85 m, respectively. Further, inter-comparisons were also performed with two existing forest height maps. Our model granted a moderate agreement with the existing satellite-based forest height maps (R = 0.738, and RMSE = 7.65 m (R² = 0.52, and RMSE = 8.99 m). Our results showed that the ANN model developed in this paper is capable of estimating forest heights over the China continent with a satisfactory accuracy. Forth coming research on our model will focus on extending the model to the estimation of woody biomass.

Desertification is of significant concern as one of the world’s most serious ecological and environmental problems. China has made great achievements in afforestation and desertification control in recent years. The climate varies greatly across northern China. Using a long-time series of remote sensing data to study the effects of desertification will further the understanding of China’s desertification control engineering and climate change mechanisms. The moist index was employed in this research to determine the climate type and delineate the potential occurrence range of desertification in China. Then, based on the Google Earth Engine platform, MODIS data were used to construct various desertification monitoring indicators and applied to four machine learning models. By comparing different combinations of indicators and machine learning models, it was concluded that the random forest model with four indicator combinations had the highest accuracy of 86.94% and a Kappa coefficient of 0.84. Therefore, the random forest model with four indicator combinations was used to monitor desertification in the study area from 2000 to 2020. According to our studies, the area of desertification decreased by more than 237,844 km² between 2000 and 2020 due to the impact of human activities and in addition to climatic factors such as the important role of precipitation. This research gives a database for the cause and control of desertification as well as a reference for national-scale desertification monitoring.