Search

目的/意义人工智能（Artificial Intelligence，AI）技术已在学术和工程应用领域掀起了研究高潮，在地球物理参数和农业气象遥感参数反演方面也表现出了强大的应用潜力。目前大部分AI技术在地学和农学的应用还是“黑箱”，没有物理意义或缺乏可解释性及通用性。为了促进AI在地学和农学的应用和培养交叉学科的人才，本研究提出基于AI耦合物理和统计方法的地球物理参数反演范式理论。方法首先基于物理能量平衡方程进行物理逻辑推理，从理论上构造反演方程组，然后基于物理推导构建泛化的统计方法。通过物理模型模拟获得物理方法的代表性解以及利用多源数据获得统计方法代表性的解作为深度学习的训练和测试数据库，最后利用深度学习进行优化求解。［结果和讨论］判定形成具有通用性和物理可解释的范式条件包括：（1）输入与输出变量（参数）之间必须存在因果关系；（2）输入和输出变量（参数）之间理论上可以构建闭合的方程组（未知数个数少于或等于方程组个数），也就是说输出参数可以被输入参数唯一确定。如果输入参数（变量）和输出参数（变量）之间存在很强的因果关系，则可以直接使用深度学习进行反演。如果输入参数和输出参数之间存在弱相关性，则需要添加先验知识来提高输出参数的反演精度。此外，本研究以农业气象遥感中的关键参数地表温度、发射率、近地表空气温度和大气水汽含量联合反演作为案例对理论进行了证明，分析结果表明本理论是可行的，并且可以辅助优化设计卫星传感器波段组合。结论本理论和判定条件的提出在地球物理参数反演史上具有里程碑意义。

The validation of remotely sensed evapotranspiration (ET) products is important for the development of ET estimation models and the accuracy of the scientific application of the products. In this study, different ET products such as HiTLL, MOD16A2, ETMonitor, and SoGAE were compared using multi-source remote sensing data and ground-based data to evaluate their applicability in the Heihe River Basin (HRB) during 2010–2019. The results of the comparison with the site observations show that ETMonitor provides a more stable and reliable estimation of ET than the other three products. The ET exhibited significant variations over the decade, characterized by a general increase in rates across the HRB. These changes were markedly influenced by variations in land use and topographical features. Specifically, the analysis showed that farmland and forested areas had higher ET rates due to greater vegetation cover and moisture availability, while grasslands and water bodies demonstrated lower ET rates, reflecting their respective land cover characteristics. This study further explored the influence of various factors on ET, including land use changes, NDVI, temperature, and precipitation. It was found that changes in land use, such as increases in agricultural areas or reforestation efforts, directly influenced ET rates. Moreover, meteorological conditions such as temperature and precipitation patterns also played crucial roles, with warmer temperatures and higher precipitation correlating with increased ET. This study highlights the significant impact of land use and climatic factors on spatiotemporal variations in ET within the HRB, underscoring its importance for optimizing water resource management and land use planning in arid regions. [ABSTRACT FROM AUTHOR]

Accurate estimation of surface evapotranspiration (ET) in the Heihe River Basin using remote sensing data is crucial for understanding water dynamics in arid regions. In this paper, by coupling physical constraints and machine learning for hybrid modeling, we develop a hybrid model based on surface conductance optimization. A hybrid modeling algorithm, two physical process-based ET algorithms (Penman–Monteith-based and Priestley–Taylor-based ET algorithms), and three pure machine learning algorithms (Random Forest, Extreme Gradient Boosting, and K Nearest Neighbors) are comparatively analyzed for estimating the ET. The results showed that, in general, the machine learning model optimized by parameters was able to better predict the surface conductance of the hybrid model. Driver analyses showed that radiation, normalized difference vegetation index (NDVI), and air temperature had high correlations with ET. The hybrid model had a better prediction performance for ET than the other five models, and it improved the R2 of the two physical process-based algorithms to 0.9, reduced the root mean square error (RMSE) to 0.5 mm/day, reduced the BIAS to 0.2 mm/day, and improved the Kling–Gupta efficiency (KGE) to 0.9. The hybrid model outperformed the others across different time scales, displaying lower BIAS, RMSE, and higher KGE. Spatially, its ET patterns aligned with regional vegetation changes, with superior accuracy in annual ET estimation compared to the other models. Comparison with other ET products shows that the estimation results based on the hybrid model have better performance. This approach not only improves the accuracy of ET estimation but also improves the understanding of the physical mechanism of ET estimation by pure machine learning models. This study can provide important support for understanding ET and hydrological processes under different climatic and biotic vegetation in other arid and semi-arid regions. [ABSTRACT FROM AUTHOR]

The energy and water vapor exchange between the land surface and atmospheric boundary layer plays a critical role in regional climate simulations. This paper implemented a hybrid data assimilation and machine learning framework (DA-ML method) into the Weather Research and Forecasting (WRF) model to optimize surface soil and vegetation conditions. The hybrid method can integrate remotely sensed leaf area index (LAI), multi-source soil moisture (SM) observations, and land surface models (LSMs) to accurately describe regional climate and land–atmosphere interactions. The performance of the hybrid method on the regional climate was evaluated in the Heihe River basin (HRB), the second-largest endorheic river basin in Northwest China. The results show that the estimated sensible (H) and latent heat (LE) fluxes from the WRF (DA-ML) model agree well with the large aperture scintillometer (LAS) observations. Compared to the WRF (open loop – OL), the WRF (DA-ML) model improved the estimation of evapotranspiration (ET) and generated a spatial distribution consistent with the ML-based watershed ET (ETMap). The proposed WRF (DA-ML) method effectively reduces air warming and drying biases in simulations, particularly in the oasis region. The estimated air temperature and specific humidity from WRF (DA-ML) agree well with the observations. In addition, this method can simulate more realistic oasis–desert boundaries, including wetting and cooling effects and wind shield effects within the oasis. The oasis–desert interactions can transfer water vapor to the surrounding desert in the lower atmosphere. In contrast, the dry and hot air over the desert is transferred to the oasis from the upper atmosphere. The results show that the integration of LAI and SM will induce water vapor intensification and promote precipitation in the upstream of the HRB, particularly on windward slopes. In general, the proposed WRF (DA-ML) model can improve climate modeling by implementing detailed land characterization information in basins with complex underlying surfaces. [ABSTRACT FROM AUTHOR]

Generalized multi-adaptive regression splines (MARS) and genetic expression programming (GEP)-based equations were developed to estimate Reference Evapotranspiration (ETo) in coastal regions. Following existing regression-based ETo retrieval equations, five climatic data configurations were used to train, validate, and test the MARS and GEP models (hereafter called MARS1–MARS5 and GEP1–GEP5). The performances of the MARS and GEP models with each of the five input configurations were assessed. The generalized MARS1–MARS5 and GEP1–GEP5 models could estimate ETo accurately in regions other than their training region. In addition, MARS1 performed better than MARS2–MARS5. Similarly, GEP1 outperformed GEP2–GEP5, implying that input configuration 1 contains the most important information about ETo. The results also show that MARS can estimate ETo more accurately than GEP. The findings indicate that MARS1–MARS5 and GEP1–GEP5 improved ETo values compared with the corresponding traditional equations. Finally, sensitivity analyses were conducted to evaluate the impact of each input variable on ETo. [ABSTRACT FROM AUTHOR]