Yunfeng Lyu

Accurate prediction of maize yields is crucial for effective crop management. In this paper, we propose a novel deep learning framework (CNNAtBiGRU) for estimating maize yield, which is applied to typical black soil areas in Northeast China. This framework integrates a one-dimensional convolutional neural network (1D-CNN), bidirectional gated recurrent units (BiGRU), and an attention mechanism to effectively characterize and weight key segments of input data. In the predictions for the most recent year, the model demonstrated high accuracy (R² = 0.896, RMSE = 908.33 kg/ha) and exhibited strong robustness in both earlier years and during extreme climatic events. Unlike traditional yield estimation methods that primarily rely on remote sensing vegetation indices, phenological data, meteorological data, and soil characteristics, this study innovatively incorporates anthropogenic factors, such as Degree of Cultivation Mechanization (DCM), reflecting the rapid advancement of agricultural modernization. The relative importance analysis of input variables revealed that Enhanced Vegetation Index (EVI), Sun-Induced Chlorophyll Fluorescence (SIF), and DCM were the most influential factors in yield prediction. Furthermore, our framework enables maize yield prediction 1–2 months in advance by leveraging historical patterns of environmental and agricultural variables, providing valuable lead time for decision-making. This predictive capability does not rely on forecasting future weather conditions but rather captures yield-relevant signals embedded in early-season data.

Climate change and human activities affect the biomass of different algal and the succession of dominant species. In the past, phytoplankton phyla inversion has been focused on oceanic and continental shelf waters, while phytoplankton phyla inversion in inland lakes and reservoirs is still in the initial and exploratory stage, and the research results are relatively few. Especially for mid-to-high latitude lakes, the research is even more blank. Therefore, this study proposes a machine learning method based on OLCI/Sentinel-3 satellite imagery to retrieve algal biomass abundance. Remote sensing models were developed to estimate the biomass abundance of three major algal groups: Cyanophyta, Chlorophyta, and Bacillariophyta. This study compared and evaluated 6 commonly used machine learning models, including extreme gradient boosting (XGBoost), support vector regression (SVR), backpropagation neural network (BP), gradient boosting decision tree (GBDT), random forest (RF), and categorical boosting (CatBoost). The results indicated that XGBoost exhibited the highest accuracy (R² = 0.92, RMSE = 1.78%, MAPE = 9.96%) in estimating Cyanophyta’s biomass abundance. The RF model demonstrated the highest accuracy for estimating Chlorophyta’s biomass abundance (R² = 0.72, RMSE = 6.57%, MAPE = 50.8%), while the GBDT model exhibited the highest accuracy for estimating Bacillariophyta’s biomass abundance (R² = 0.9, RMSE = 4.66%, MAPE = 47.87%). The models were subsequently applied to all cloud-free OLCI images from Hulun Lake during the ice-free periods from 2016 to 2023, producing spatiotemporal distribution maps of the different phytoplankton biomass abundance. Cyanophyta dominated the biomass abundance (44.62 ± 3.47%), followed by Bacillariophyta (36.35 ± 2.68%), and Chlorophyta had the lowest proportion (10.42 ± 1.08%). Together, these three algae groups constituted 91.4 ± 1.55% of all phytoplankton in Hulun Lake. Significant annual variations in the biomass abundance of Cyanophyta and Bacillariophyta were observed, whereas those of Chlorophyta remained stable. Additionally, this study examined the effects of climatic factors and water quality parameters on the biomass abundance of algae. The findings suggest that temperature, wind speed, and atmospheric pressure are critical factors influencing the biomass abundance of the different algae groups. This study not only fills the gaps in the related field, but also provides a new method for monitoring algae, as well as a strong support for realizing the goals of sustainable management of water resources and ecological protection.

Chromophoric dissolved organic matter (CDOM) is a significant contributor to the biogeochemical cycle and energy dynamics within aquatic ecosystems. Hence, the implementation of a systematic and comprehensive monitoring and governance framework for the CDOM in inland waters holds significant importance. This study conducted the retrieval of CDOM in Lake Khanka. Specifically, we use the GBDT (R² = 0.84) algorithm which performed best in retrieving CDOM levels and an empirical relationship based on the situ data between CDOM and dissolved organic carbon (DOC) to indicate the distribution of DOC indirectly. The performance of the CDOM-DOC retrieval scheme was reasonably good, achieving an R² value of 0.69. The empirical algorithms were utilized for the analysis of Sentinel-3 datasets from the period 2016 to 2020 in Lake Khanka. The potential factors that contributed to the sources of DOM were also analyzed with the humification index (HIX). The significant relationship between CDOM and DOC (HIX and chemical oxygen demand (COD)) indicated the potential remote sensing application of water quality monitoring for water management. An analysis of our findings suggests that the water quality of the Great Khanka is superior to that of the Small Khanka. Moreover, the distribution of diverse organic matter exhibits a pattern where concentrations are generally higher along the shoreline compared to the center of the lake. Efficient measures should be promptly implemented to safeguard the water resources in international boundary lakes such as Lake Khanka and comprehensive monitoring systems including DOM distribution, DOM sources, and water quality management would be essential for water resource protection and government management.

Algal blooms frequently occur in numerous lakes in China, risking human health and the environment. In contrast, aquatic vegetation contributes to water purification. Due to the similar spectral characteristics shared by algal and aquatic vegetation, both are hardly distinguishable in remote sensing imaging, especially in turbid water bodies. To address this challenge, this study constructed a method to effectively extract algal blooms and aquatic vegetation from the turbid water bodies using Sentinel 2 images with high spatial resolution. Our results showed that the accuracy of the extraction of vegetation information could reach 96.1%. Since this method combined the vegetation extraction results from multiple indices, it effectively tackled the mis-extraction when only the Floating Algae Index (FAI) or the Normalized Difference Vegetation Index (NDVI) is used in water with high turbidity. By combining the image time series information with the natural phenological characteristics of the aquatic vegetation and algal blooms, an improved Vegetation Presence Frequency (VPF) was developed. It effectively distinguished algal blooms and aquatic vegetation without actual measurement data. Based on the above method and process, the information of algal blooms and aquatic vegetation was sufficiently distinguished in five typical lakes in China (Lake Hulun, Lake Hongze, Lake Chaohu, Lake Taihu, and Lake Dianchi), and the spatial distribution was reasonably mapped. The overall identification accuracy of aquatic vegetation and algal blooms using the improved VPF ranged 71.8–84.3%. The spatial transferability test of the method in the independent lakes with the various optical properties indicated the prospects of its application in other turbid water bodies. This study should provide strong methodological and theoretical support for future monitoring of algal blooms in turbid water bodies with vigorous aquatic vegetation, especially in the absence of actual measurement data. This should have practical relevance for water environment management and governance departments.