Wheat, as one of the three major food crops in the world, plays a crucial role in the global food supply system. It is very necessary to accurately and timely predict the winter wheat yield for national food security. Spectral information obtained from remote sensing satellites can be expected to better reflect the growth status of crops, and then predict the yield of large-scale crops. Machine learning has also been widely applied in crop classification and yield prediction in recent years, due to its strong capabilities in data mining and analysis. However, the input factors of machine learning are mostly traditional variables, such as soil moisture, meteorology, and land use. It is a high demand to introduce new variables as the input factors for machine learning, in order to improve the accuracy of crop yield estimation. Solar-induced chlorophyll fluorescence (SIF, a byproduct of vegetation photosynthesis) has gradually become a new variable in crop yield prediction, which is sensitive to water and heat stress during crop growth. The data from different time windows can be used to determine the optimal time window and the high accuracy of yield prediction of winter wheat. Meanwhile, the limited data can be used to provide reliable predictions before the winter wheat harvest. In this study, an ideal region was taken to predict the wheat yield per unit area in Henan Province in China. The yield of winter wheat was then predicted using multi-source remote sensing data and ensemble learning. The growth season of winter wheat was divided into 28 time windows. SIF feature variables were introduced to normalize the vegetation index, water index, leaf area index, monthly average temperature, and total monthly precision. Multiple remote sensing indicators and meteorological data were collected from 2003 to 2018, in order to train the yield of winter wheat from 2019 to 2021. Stacking-based ensemble learning was used to fit eight machine learning models, and then combine their predictions. The prediction errors were then reduced to improve the accuracy and generalization of the model. The results show that: (1) SIF features improved the prediction of winter wheat yield; (2) The optimal time window of machine learning was in the period of December to the following May; (3) Compared with the rest, the Stacking-based ensemble learning performed the best, where the coefficient of determination was 0.816, the root mean squared error was 580.36 kg/hm2, and mean absolute error was 476.01 kg/hm2; (4) The spatial distribution of the predicted yield was similar to the actual one, with a trend of low in the west and high in the east. This finding can also provide new ideas for predicting crop yield.
- Article type
- Year
- Co-author
Accurate identification of winter wheat irrigation events and precise estimation of irrigation frequency are crucial for the rational use of water resources and the accurate estimation of irrigation water consumption. Site observation records can provide accurate irrigation information, but such data are difficult to obtain, and the number of observation sites providing irrigation information is limited. As a result, accurately identifying regional-scale irrigation events and frequency remained a challenge. This study proposed a method for identifying winter wheat irrigation events and estimating irrigation frequency. First, the spatial distribution of winter wheat was obtained. The study used winter wheat distribution data at three different spatial scales (field scale and two pixel scales), where the pixel-scale data had spatial resolutions of 500 m and 30 m, using publicly available data. The field-scale winter wheat distribution was extracted in this study, with the process as follows: acquiring the best cloud-free Sentinel-2 images from 2018 to 2020 for the study area, extracting spectral features, constructing sample sets for each year's image, and using the nearest neighbor classifier based on GEOBIA(geographic object-based image analysis) to extract winter wheat field distribution. Next, grids of 500 m × 500 m or 2 km × 2 km were generated in the study area, and the average values of the Sentinel-1 SAR VV (Vertical-Vertical) single-polarization backscatter coefficient for winter wheat (each field or pixel) and the grid were calculated. By analyzing and comparing the standard deviation and correlation coefficient of winter wheat (each field or pixel) and the grid within a certain time range, combined with daily precipitation data, irrigation events were identified. Finally, the irrigation event identification results at the field and pixel scales from 2018 to 2020 were validated using observed irrigation time and frequency data, and irrigation frequency was calculated based on the identified irrigation events. This provided the spatial distribution of winter wheat irrigation frequency at the field and two pixel scales in the study area. The study applied the proposed method to identify winter wheat irrigation events in Yucheng, Shandong Province, at three different spatial scales (field scale, 500 m, and 30 m pixel scales), and used observed irrigation data to validate the results. Irrigation frequency was then estimated at the spatial scale with the highest irrigation event identification accuracy, and the spatial distribution of winter wheat irrigation frequency in Yucheng from 2018 to 2020 was obtained. The results showed that the recall rates for identifying irrigation events at the field scale, 500 m, and 30 m pixel scales were 85.71%, 78.57%, and 57.14%, respectively, with F-scores of 70.59%, 66.67%, and 50.00%. The accuracy of field scale applications is better than that of applications at 500 m and 30 m pixel scales. The "Strict" overall accuracy for identifying the winter wheat irrigation frequency distribution at the field scale was 69.75%, while the "Loose" overall accuracy was 90.24%. The irrigation frequency for winter wheat in Yucheng from 2018 to 2020 concentrated between 1 and 3 times. This study provides a reliable method for identifying irrigation events and estimating irrigation frequency at regional scales.
Understanding the distribution of irrigated farmland is of great significance for rational utilization of water resources, timely adjustment of agricultural production policies and protection of food security. Existing studies based on machine learning to identify irrigated farmland mostly use binarized sample labeling (that is, only labeling irrigated and non-irrigated), which may lead to the omission of irrigation samples, resulting in low identification accuracy of irrigated farmland. To avoid this problem, a scheme of assigning irrigation fractions to training samples was proposed in this paper. Firstly, three preliminary irrigation maps were generated using normalized vegetation index, enhanced vegetation index and greenness vegetation index combined with statistical irrigation area data. Then, combined with three different preliminary irrigation maps, the scheme proposed in this study assigned different irrigation scores to each pixel and obtained training samples. Then, two machine learning methods, random forest (RF) and convolutional neural network (CNN), were used to predict the pixel-by-pixel irrigation fraction of the study area, respectively. Identification of 250 m resolution irrigated farmland in Shandong Province from 2018 to 2022 was conducted. Finally, the spatial distribution of irrigated farmland in Shandong Province from 2018 to 2022 was obtained based on verification samples and statistical data, and the spatial and temporal distribution characteristics of irrigated farmland were analyzed. The results showed as follows: 1) Compared with the two binarization methods for obtaining training samples, the county R2 of RF and CNN for identifying irrigated farmland was as high as 0.95 and 0.93 when the proposed scheme was used to obtain training samples and identify irrigated farmland; The accuracy of RF identification of irrigated farmland in Shandong Province was better than that of CNN, and the performance was relatively stable in different years. 2) From 2018 to 2022, the spatial distribution of irrigated farmland in Shandong Province was relatively consistent, mainly distributed in the northwest and south of Shandong Province, while the distribution was relatively sparse in Jiaodong Peninsula and central Shandong Province. The statistical irrigated area in Shandong Province showed a slight growth trend in the past five years, and the identified irrigated farmland area in 2022 also showed a slight growth compared with 2018. The scheme of assigning irrigation scores to training samples proposed in this study can effectively improve the accuracy of irrigated farmland identification in Shandong Province, and is a reliable and effective method to identify irrigated farmland at regional scale.
京公网安备11010802044758号