关键词: 农业干旱, 温度植被干旱指数, Theil-Sen+Mann-Kendall, ICEEMDAN-ARIMA, 干旱预测

Abstract:

[Objective] Drought, one of the most frequent natural disasters globally, is characterized by its extensive impact area, prolonged duration, and significant harm. With the intensification of global warming and human activities, the economic consequences of drought have increased sharply, resulting in tens of billions of dollars in economic losses. Large scale irrigation areas, as important pillars of China's agricultural economy, often have their benefits severely restricted by drought disasters. Therefore, quickly and accurately grasping the regional drought situation is of great significance. It can not only effectively improve the utilization efficiency of water resources and reduce agricultural production losses but also promote the sustainable development of regional agriculture. [Methods] The Santun river irrigation area in Xinjiang, an arid - zone irrigation area, was taken as an example. Based on Landsat TM/ETM+/OLI_TIRS series data, the temperature vegetation drought index (TVDI) and the vegetation temperature condition index (VTCI) were calculated. Using in situ the soil water content of the 0 – 10 cm soil layer in the study area measured by the Smart Soil Moisture Monitor, an applicability analysis of the drought monitoring effects of TVDI and VTCI was carried out to select the remote - sensing monitoring index suitable for drought research in the study area. Based on the selected drought monitoring index, methods such as linear trend analysis and Theil - Sen + Mann - Kendall trend test were used to explore the temporal and spatial distribution characteristics and change trends of drought in the study area from 2005 to 2022. Meanwhile, with the help of machine learning algorithms, an ICEEMDAN - ARIMA combined model was constructed to predict the drought situation in the study area in spring, summer, and autumn of 2023. The prediction performance of the ICEEMDAN - ARIMA combined model was evaluated using root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). [Results and Discussions] The research results show that there were varying degrees of linear correlations between the two drought indices, TVDI and VTCI, inverted from remote - sensing data, and the soil water content of the 0－10 cm surface soil layer in the Santun river irrigation area of Xinjiang. The coefficient of determination between TVDI and the measured soil water content was greater than 0.51 in all periods, with an overall fitting coefficient of 0.57, and the slopes of the fitting equations were all negative, indicating a significant negative correlation. In contrast, the highest coefficient of determination of VTCI was only 0.33, and its overall monitoring effect was significantly weaker than that of TVDI. In terms of temporal and spatial distribution, the drought situation in the study area showed a slow-increasing trend from 2005 to 2022. The growth rate of TVDI was 0.01/10 a, and it had strong spatial heterogeneity, specifically manifested as the spatial distribution characteristic that the southern and southwestern regions of the irrigation area were drier than the northern and northeastern regions. The results of the drought trend analysis indicated that from 2005 to 2022, the distribution of Sen change rate data in the study area conforms to the normal distribution (P < 0.01), and the Sen slopes of more than 72.83% of the regions were greater than zero. At the same time, according to the classification criteria of the Sen + Mann - Kendall trend test, six types of drought change trends were divided. The area proportions of the extremely significant mitigation, significant mitigation, slight mitigation, extremely significant drying, significant drying, and slight drying categories were 0.73%, 1.78%, 24.31%, 5.33%, 9.43%, and 58.42%, respectively. The area proportions of the slight drying and slight mitigation categories were the largest, accounting for a total of 82.73% of the total area of the study area. The ICEEMDAN - ARIMA combined model constructed with the help of machine learning algorithms achieved good results in predicting the drought situation in the study area in 2023. The average value of R² reached 0.962, demonstrating high robustness and good prediction performance. [Conclusions] The research results systematically characterizes the characteristics of agricultural drought changes in the Santun river irrigation area of Xinjiang over a long - time series, and reveal that the ICEEMDAN - ARIMA combined model has good prediction accuracy in agricultural drought prediction research. This study can provide important references for the construction of drought early warning and forecasting systems, water resource management, and the sustainable development of agriculture in arid-zone irrigation areas.

Key words: agricultural drought, TVDI, Theil-Sen+Mann-Kendall, ICEEMDAN-ARIMA, drought prediction