Increasingly frequent extreme rainfall as a result of climate change is strongly damaging the global soil and water environment. However, few studies have focused on daily extreme sediment events (DESE) in heterogeneous karst watersheds based on long-term in -situ observations. This study quantitatively assessed the time effect of DESE on rainfall response, decoupled the impact of environmental factors on DESE by using structural equation modelling, and finally explored the modelling scheme of DESE based on the hybrid model. The results showed that DESE had the highest frequency of occurrence in May -July, with dispersed distribution in the value domain. Rainfall with a time lag of 1 day and a time accumulation of 2 or 3 days was an important contribution to DESE ( P < 0.01, R = 0.47 -0.68). Combined effects of environmental factors explained 53.6 % -64.1 % of the variation in DESE. Runoff and vegetation exerted the strongest direct and indirect effects on DESE, respectively (8 = 0.66/ -0.727). Vegetation was the dominant driver of DESE in Dabanghe and Yejihe (8 = -0.725/-0.758), while the dominant driver in Tongzhihe was climate (8 = 0.743). In the future, the risk of extreme sediments should be prevented and resolved through the comprehensive regulation of multiple paths, such as runoff and vegetation. Hybrid models significantly improved the modelling performance of machine learning models. Generalized additive model -Extreme gradient boost had the best performance, while Partial least squares regression -Extreme gradient boost was the most valuable when considering performance and input data cost. Two methods can be used as recommended solutions for DESE modelling. This study provides new and in-depth insights into DESE in

A climate transition towards warm-wet conditions in Northwest China has drawn much attention. With continuous climate change and universal glacier degradation, increasing water-related hazards and vulnerability have become one of the important problems facing the Tarim Basin. However, the impacts of the climate transition on streamflow abrupt change and extreme hydrological events were less discussed, especially in glacial basins. In the present study, the discharge datasets in four glacial basins of Tarim Basin from 1979 to 2018 were constructed using the GRU-GSWAT thorn model first. The differences in streamflow characteristics, the shift of hydrological extreme pattern, and potential changes of the controlling factors before and after the abrupt changes were investigated. The results indicated that the abrupt change point (ACP) in streamflow occurred in 2000 in the Qarqan River Basin, 2002 in the Weigan River Basin, and 1994 in the Aksu River Basin and the Yarkant River Basin. A general decrease in streamflow before the ACP has shifted to a notable upward trend in the Qarqan River Basin and the Aksu River Basin, while minor upward fluctuations were observed in other basins. Moreover, the hydrological characteristics in extreme events vary dramatically before and after the ACPs, characterized by a pronouncing shift from drought-dominant pattern to wet events dominated pattern. The driven climate factors have been altered after the ACPs with notable spatial heterogeneity, in which temperature remained as the dominant role in meltwater-dominated basins while the influence of precipitation has increased after the ACPs, whereas the sensitivity of temperature on streamflow change has been enhanced in basins dominated by precipitation such as the Qarqan River Basin. Owing to the evident warming-wetting trend and glacier compensation effect, both the inter-annual and intra-annual streamflow fluctuations can be efficiently smoothed in basins with a high glacier area ratio (GAR). These findings provide a further understanding of the abrupt change in streamflow under the exacerbated climate and glacier change in mountainous arid regions.

Fire is a crucial factor in terrestrial ecosystems playing a role in disturbance for vegetation dynamics. Process-based fire models quantify fire disturbance effects in stand-alone dynamic global vegetation models (DGVMs) and their advances have incorporated both descriptions of natural processes and anthropogenic drivers. Nevertheless, these models show limited skill in modeling fire events at the global scale, due to stochastic characteristics of fire occurrence and behavior as well as the limits in empirical parameterizations in process-based models. As an alternative, machine learning has shown the capability of providing robust diagnostics of fire regimes. Here, we develop a deep-learning-based fire model (DL-fire) to estimate daily burnt area fraction at the global scale and couple it within JSBACH4, the land surface model used in the ICON-ESM. The stand-alone DL-fire model forced with meteorological, terrestrial and socio-economic variables is able to simulate global total burnt area, showing 0.8 of monthly correlation (rm) with GFED4 during the evaluation period (2011-2015). The performance remains similar with the hybrid modeling approach JSB4-DL-fire (rm = 0.79) outperforming the currently used uncalibrated standard fire model in JSBACH4 (rm = -0.07). We further quantify the importance of each predictor by applying layer-wise relevance propagation (LRP). Overall, land properties, such as fuel amount and water content in soil layers, stand out as the major factors determining burnt fraction in DL-fire, paralleled by meteorological conditions over tropical and high latitude regions. Our study demonstrates the potential of hybrid modeling in advancing fire prediction in ESMs by integrating deep learning approaches in physics-based dynamical models. We develop a fire-vegetation model based on a hybrid approach integrating artificial intelligence (AI) techniques into physics-based models. Given the weather conditions, vegetation states, and human factors, our model estimates daily burned area fraction. The spatiotemporal variations in burned area are closely reproduced, especially over fire-prone regions, such as Africa, South America, and Australia. Our model is able to represent regional variations in the drivers of fire occurrence, showing different importance of input predictors for different regions. This approach shows the possibilities of using deep learning (DL) models to provide in-depth fire predictions in Earth system models. Deep neural networks (DNN) can accurately predict global burnt area fraction on a daily scaleIntegration of the DNN into a physics-based land model significantly improves the estimation of biomass burnt damage in vegetation dynamicsThe DNN accounts for regional fire variations by assigning varying degrees of importance to each predictor