A recent study (Shan et al., 2024, https://doi.org/10.1016/j.rse.2024.114167) showed that using a Deep-Neural-Network (DNN)-based observation operator in land data assimilation (DA) does not guarantee improvements in surface soil moisture (WG2). Here, we conduct a synthetic experiment to explain this outcome by testing whether DNN-based operators reproduce physically consistent Jacobians (sensitivities) required by DA. The ISBA-A-gs land surface model is perturbed to generate “synthetic true” WG2 and leaf area index (LAI), and a Water Cloud Model (WCM) is used to generate synthetic backscatter. Two DNNs are trained taking simulated states from the open loop run as input and the synthetic backscatter as output. The first is trained on WG2 and LAI while the second uses LAI and soil moisture from multiple layers, that is including redundant inputs. The synthetic observations are then assimilated using the two DNNs and the WCM as observation operators. Results suggest that the assimilation using a DNN-based observation operator improves the estimates of WG2. However, DNN Jacobians are no longer physically plausible when the model simulations used as training inputs contain errors relative to the “true” data, or when redundant input variables are included. This confirms that a strong predictive performance does not automatically reflect accurate representation of underlying physical sensitivities which prove problematic in subsequent DA. This study has important implications for deep-learning-based DA and Earth system models (ESM): DA or ESM trained on reanalysis or simulations may learn inaccurate Jacobians if input contains errors and redundant variables. This potential limitation could remain hidden when evaluation focuses primarily on predictive performance.

ASCAT normalized backscatter (σ₄₀^o) and slope (σ^′) contain valuable information about soil moisture and vegetation. While σ₄₀^o has been assimilated to constrain soil moisture, sometimes together with Leaf Area Index (LAI), this study is the first to assimilate σ^′ directly to constrain vegetation states. Here, we assimilate σ₄₀^o and slope σ^′ into the ISBA-A-gs LSM using the Simplified Extended Kalman Filter (SEKF) using a Deep Neural Network (DNN) as the observation operator. The performances of the data assimilation (DA) and open loop (OL) are evaluated against in-situ soil moisture observations from the International Soil Moisture Network (ISMN), and LAI observations from the Copernicus Global Land Service (CGLS). Given an accurate and physically plausible observation operator, along with well-defined model and observation errors, the data assimilation system should yield improved estimates of the model states. However, results show that the DA performance is neutral compared to the OL in terms of the median unbiased root mean square error (ubRMSE) and Pearson correlation coefficient (ρ) across all validation sites. In addition, an analysis of the residuals and innovations confirms that DA had limited or no impact. This poor performance is perplexing. Furthermore, given the growing interest in the use of machine-learning-based observation operators, it is essential to understand the role that the use of the DNN may be playing in this poor performance. While representativeness errors and error specification play some part, it is demonstrated that the key factor constraining the efficacy of the SEKF is the correct estimation of the Jacobians that control the degree to which the observations update the states in the SEKF. It is argued that the DNN relating model states to satellite observations must have physically-plausible and robust Jacobians for the DNN to be effective in a data assimilation framework.

Current news datasets merely focus on text features on the news and rarely leverage the feature of images, excluding numerous essential features for news classification. In this paper, we propose a new dataset, N24News, which is generated from New York Times with 24 categories and contains both text and image information in each news. We use a multitask multimodal method and the experimental results show multimodal news classification performs better than text-only news classification. Depending on the length of the text, the classification accuracy can be increased by up to 8.11%. Our research reveals the relationship between the performance of a multimodal classifier and its sub-classifiers, and also the possible improvements when applying multimodal in news classification. N24News is shown to have great potential to prompt the multimodal news studies.