As a fundamental and important task in remote sensing, remote sensing image scene understanding (RSISU) has attracted tremendous research interest in recent years. RSISU includes the following sub-tasks: remote sensing image scene classification, remote sensing image scene retrieval, and scene-driven remote sensing image object detection. Although these sub-tasks have different goals, they share some communal hints. Hence, this paper tries to discuss them as a whole. Similar to other domains (e.g., speech recognition and natural image recognition), deep learning has also become the state-of-the-art technique in RSISU. To facilitate the sustainable progress of RSISU, this paper presents a comprehensive review of deep-learning-based RSISU methods, and points out some future research directions and potential applications of RSISU.

Summary: The spring phenology has advanced significantly over recent decades with climate change, impacting large‐scale biogeochemical cycles, climate feedback, and other essential ecosystem processes. Although numerous prognostic models have been developed for spring phenology, regional analyses of the optimality (OPT) strategy model that incorporate environmental variables beyond temperature and photoperiod remain lacking. We investigated the roles of solar radiation (SR) and three water stress factors (precipitation (P), soil moisture, and vapor pressure deficit (VPD)) on spring phenology from 1982 to 2015 using the OPT model with Global Inventory Modeling and Mapping Studies NDVI3g dataset and environmental data from TerraClimate, CRU_TS, and Global Land Data Assimilation System across the Northern Hemisphere (> 30°N). Our results show that SR and water stress factors significantly impacted intrasite decadal spring phenology variability, with water stress factors dominant in grassland ecosystems while SR dominated in the rest of the ecosystem types. Enhanced models incorporating SR (OPT‐S) and VPD (OPT‐VPD) outperformed the original OPT model, likely due to improved representation of the adaptive strategy of spring phenology to optimize photosynthetic carbon gain while minimizing frost risk. Our research enhances the understanding of the key environmental drivers influencing decadal spring phenology variation in the Northern Hemisphere and contributes to more accurate forecasts of ecological responses to global environmental change.

The changing of vegetation is a sensitive signature of global warming, and satellite photon-counting laser altimeters provide an effective way to monitor the changing of vegetation. Based on the background noise difference between vegetation-covered areas and bare lands, we proposed a classification method to distinguish vegetation-covered areas from the raw photons measured by photon-counting laser altimeters in relatively flat areas. First, a theoretical noise model was established considering the influence of the sunlight incident direction and reflection characteristics of different surfaces. Second, the thresholds from the proposed theoretical model were calculated and tested to classify the along-track land-cover types for the Ice, Cloud, and Elevation Satellite-2 (ICESat-2) photon-counting laser altimeter. Then, the study areas near Seattle and Romania in summer were selected and the classification method was verified to achieve an overall accuracy of over 77% (the strong beam) and over 76% (the weak beam) for both thresholds and areas. Our method utilized the noise photons with vegetation canopy reflection information, which are enormous in quantity and easy to extract compared to the signal photons. More importantly, this method reduces the requirements of the optical images (that are used as prior knowledge). The results show that using the noise photons of the weak beam may be more potential for the classification of vegetation and land than using the signal photons of the weak beam. We extended the research on the mechanism and application of ICESat-2 in forestry.

As an essential parameter in forestry, crown base height (CBH) faces many tasks. The methods are still developing for estimating it. Unmanned aerial vehicles (UAV)light detection and ranging (LiDAR) supplies new, massive and high-density data for estimating CBH. Many methods hadbeen generated to compute CBH indirectly using regression-based ways or directly using geometric/statistical LiDAR-based ways. However, there werefew methods to deal with the problem of understory, trunkand noise points caused by high-density UAV data. A robust method was firstlyproposed in this studyto directly estimate CBH from LiDAR data, which containedtwo significant skills: 1)understory vegetation removal for each tree using a polynomial curve,and 2) computing CBH by kernel densification ofthe elevation frequency histogram of LiDAR data. It could tolerate the understoryand trunk points better through kernel convolution. The method proposed in this study and a previous simple model were applied in a crabapple plot in Huailai Remote Sensing Comprehensive Experimental Station, Hebei Province, China and verified byfield-measured data. Itwasinspiringthat our method is slightly better, the mean CBH of LiDAR-derived trees was only 1.60 cm higher than that of field-measured trees.The mean absolute error (MAE) of CBH was 4.91 cm, R2 was 0.73, RMSE was 8.29 cm and bias was 2.68%for these trees. Generally, this method showedstrong usability for high-density UAV LiDAR data and high precision for measuringCBH of low trees.

The precision of remote sensing methodologies in estimating live aboveground biomass (AGB) exhibits significant variability across ecosystems. Dryland systems, owing to their critical contributions to global carbon cycling, necessitate highly resolved spatiotemporal AGB assessments to elucidate their role in climate feedback mechanisms. However, the structural characteristics of arid and semiarid vegetation introduce substantial methodological constraints in biomass quantification via remote sensing platforms. This paper put forward a new method to retrieve AGB using simulated ICESat-2 LiDAR data. By using simulated photon-counting LiDAR data of spaceborne ICESat-2 and using multiple machine learning algorithms (random forest, extreme gradient boosting tree, support vector machine) to train simulated photon-counting LiDAR data, we created a novel model to retrieve AGB based on simulated ICESat2 data. The results showed that the RF algorithm was the best (R² = 0.61, RMSE = 17.45, RE = 49.24%) in AGB inversion. The result maps could provide timely guidance and suggestions for ecological protection and economic construction in the study area, and also provide reliable basis for large-scale vegetation AGB remote sensing monitoring, carbon cycle research.