Land use–land cover (LULC) is an important feature for ecological environment research, land resource management and evaluation. Although global high-resolution LULC data sets are booming, their regional performances were still evaluated in limited regions. To demonstrate the local applicability of global LULC data products, six emerging LULC data products were evaluated and compared in Guangxi, China. The six products used are European Space Agency GlobCover (ESAGC), ESRI Land Use–Land Cover (ESRI–LULC), Finer Resolution Observation and Monitoring of Global Land Cover (FROM–GLC), the China Land Cover Dataset (CLCD), the Global Land Cover product with Fine Classification System at 30 m (GLC_FCS30) and GlobeLand30 (GLC30). Reference data were obtained from the local government statistical yearbook and high-resolution remote sensing images on Google Earth. The results showed that CLCD, ESRI–LULC and GLC30 were found to agree well with the forest reference data, with the highest correlation coefficient of 0.999. For the cropland areas, GLC30, CLCD and ESAGC agreed well with the reference data, and the highest correlation coefficient was 0.957. Combined with the comparison with the high-resolution images obtained by Google Earth, we finally concluded that ESAGC, CLCD and GLC30 can best represent the LULCs in Guangxi. Furthermore, the spatial consistency analysis showed that three or more products identified the same LULC type as high as 96.98% of the area. We suggest that majority voting might be applied to global LULC products to provide fused products with better performances on a regional or local scale to avoid the error caused by a single data product.

As Arctic sea ice rapidly declines, accurate daily sea ice concentration (SIC) forecasts become crucial for Arctic research and operations. Emerging deep learning (DL) forecast models have the advantage of consuming fewer computational resources compared with numerical forecast systems. However, DL forecast models face challenges of lacking sea ice domain knowledge, low forecast accuracy in the marginal ice zone (MIZ), and poor explainability. This study proposes an explainable DL model, SIFNet, for daily Arctic SIC forecasts, which considers the domain knowledge of teleconnections and lagged correlations by integrating the convolutional block attention module (CBAM) and temporal feature convolution module (TFCM). To improve SIFNet’s accuracy in the MIZ, an MIZ-weighted loss function is used for fine-tuning, leading to a notable 1.36% reduction in the mean absolute error (MAE) of MIZ forecasts (MIZ MAE). In the January 2020–December 2023 test scenario, forecasting SIC for the future 49 days using the past 21 days’ data, SIFNet demonstrates forecast accuracies with the MAE of 4.69%, binary accuracy (BACC) of 95.16%, structural similarity index (SSIM) of 95.13%, and MIZ MAE of 18.22%. Compared with the other two DL models and nine numerical forecast systems, SIFNet’s forecasts achieve higher accuracies. Using layerwise relevance propagation (LRP) technique, the LRP- $\vert z\vert $ rule is constructed to improve SIFNet’s explainability. During Arctic minimum sea ice extent (SIE) periods, SIFNet forecasts confirm upward and downward surface solar radiation as predictors for minimum SIE. Similarly, over the past 30 years, precipitation phase is considered a potentially significant factor influencing SIC forecasts.

Sustainable development has become a scientific and decision-making consensus in countries and regions around the world. The current research on sustainable development mainly focuses on urban areas with a high level of economic development and intensive land use. Small-scale research, especially for underdeveloped areas, is required to reveal the spatial patterns and differences within administrative units. This study focuses on 1241 towns in Guangxi to assess the sustainable development process and variations at the township scale from 2014 to 2020 by using the entropy method, the coupled coordination degree method, and cluster analysis. The results show that the average sustainable development goal composite index (SDGCI) of towns in Guangxi is around 0.12, and their overall sustainable development level is low. The SDGCI of towns in the central region shows an increasing trend, while that in the northern and southern regions shows a decreasing trend. Then, the SDGs are divided into three systems, namely people, planet, and prosperity. The prosperity system contributes the most to the overall SDGCI, and the low values of the people and planet systems are the reason for the low level of sustainable development in Guangxi. The coupling degree of the SDGCI among the three systems is at a high level, and the level of coupling coordination degree is good, which indicates high consistency and close linkage of the towns in Guangxi when pursuing the SDGs. The results of hierarchical cluster analysis show that towns in Guangxi can be divided into three categories to adapt to different features based on SDGCI values. The assessment of SDG process in towns in Guangxi could guide local governments to refine their development policy, formulate and adjust development strategies in a targeted manner, and promote balanced and sustainable development under townships.

Understanding historical patterns of changes in drought is essential for drought adaptation and mitigation.While the negative impacts of drought over east Africa have attracted increasing attention,a comprehensive and long-term spatiotemporal assessment of drought is still lacking. Here, we provideda comprehensive spatiotemporal drought pattern analysis during the period of 1964–2015 overthe GHA. The Standardised Precipitation-Evapotranspiration Index (SPEI) at various timescales (1 month(SPEI-01), 3 month (SPEI-03), 6 month (SPEI-06), and 12 month (SPEI-12)) was used to investigate droughtpatterns on a monthly, seasonal, and interannual basis. The results showed that despite regional differences,an overall increasing tendency of the drought was observed across the GHA over the past 52 yr, with trends ofchange of -0.0017 yr-1, -0.0036 yr-1, -0.0031 yr-1, and -0.0023 yr-1 for SPEI-01, SPEI-03, SPEI-06, andSPEI-12, respectively. Droughts were more frequent, persistent, and intense in Sudan and Tanzania, whilemore severe droughts were found in Somalia, Ethiopia, and Kenya. Droughts occurred frequently beforethe 1990 s, and then became intermittent with large-scale impacts occurred during 1973–1974, 1984–1985, and 2010–2011. A turning point was also detected in 1989, with the SPEI showing a statistically significantdownward trend during 1964–1989 and a non-statistically significant downward trend from 1990to 2015. Seasonally, droughts exhibited an increasing trend in winter, spring, and summer, but a decreasingtrend in autumn. The research findings have significant implications for drought adaptation and mitigationstrategies through identifying the hotspot regions over east Africa at various timescales. Area-specificefforts are required to alleviate environmental and societal vulnerabilities to drought events.

A better detection of landslide occurrence is critical for disaster prevention and mitigation, and a standing pursuit owing to increasing and widespread impact of slope failures on human activities and natural environment in a changing world. However, the detection of rainfall-induced landslide is limited in some areas by data scarcity and method applicability. In this study, we proposed distributed rainfall thresholds within homogeneous slope units, by considering the interaction of landslide-influencing geo-environmental conditions and landslide-triggering rainfall variables. Homogeneous slope units are extracted based on detailed terrain analysis. Various landforms are identified and used to obtain slope units with homogeneous slope traits. The concept behind the distributed rainfall threshold models is that rainfall threshold for landslide occurrence varies with geo-environmental conditions such as slope gradient. Thus, a link can be established between landslide-influencing geo-environmental conditions and landslide-triggering rainfall variables. We used elevation, slope, plan and profile curvature, mean annual precipitation and temperature, soil texture and land cover as independent variables. Rainfall duration and cumulated rainfall of landslide-triggering rainfall events are automatically calculated and used, the former as one of independent variables, and the latter as the dependent variable. A support vector regression (SVR) and a multiple linear regression (MLR) method are used. The error and correlation coefficient measurement indicate a better performance of SVR method. Compared with grid units, the model scores high accuracy for slope units. The models are implemented at a regional scale (Guangdong, China). The SVR model in slope units ran with error of 0.16 mm and correlation coefficient of 0.93.