Drought is one of the most damaging environmental hazards and a naturally occurring phenomenon in Central Asia that is accompanied by crucial consequences for the agriculture sector. This research aimed at understanding the nature and extent of drought over the cropland regions of Central Asia with the help of spatiotemporal information from the region. We assessed drought occurrence using the vegetation health index (VHI). An algorithm was developed to reduce the noise of heterogeneous land surfaces by adjusting the vegetation index and brightness temperature. The vegetation condition index (VCI) and temperature condition index (TCI) were calculated using Moderate Resolution Imaging Spectroradiometer (MODIS) products for the growing season (April–September) from 2000 to 2015. The intense drought years were identified and a drought map (drought probability occurrence) was generated. The findings of this research indicated regional heterogeneity in the cropland areas having experienced droughts, observed through spatiotemporal variations. Some of the rain-fed and irrigated croplands of Kazakhstan demonstrated a higher vulnerability to annual drought occurrences and climate change impacts, while other cropland regions were found to be more resistant to such changes. The development of policy tools is required to support informed decision-making and planning processes to adapt to the occurrence of droughts. This could be achieved by the timely assessment, monitoring, and evaluation of the spatiotemporal distribution trends and variabilities of drought occurrences in this region. The results from this study focus on the spatiotemporal variations in drought to reveal the bigger picture in order to better understand the regional capacity for sustainable land management and agricultural activities within a changing environment.

Landscape restoration initiatives are mainly focusing on forest regeneration and agroforestry, especially in the Global South. However, due to a lack of monitoring tools, the carbon balance of restoration efforts remains poorly quantified. Here, we use satellite images from 2008 and 2019 to calculate carbon stocks for individual trees in Rwanda, a country which has been actively engaged in restoration activities over the past decade. We show that smallholder farmers on average planted about 3 trees per farm during 2008–2019, contributing about 50.4 million new trees at the national scale. The overall C sink of the new farmland trees was 0.13 Megagrams of Carbon per hectare per year, which is 6 times lower than gains observed from restoration of degraded forests (0.76 Megagrams of Carbon per hectare per year). If national greenhouse gas emissions remain at the level of 2019, agroforestry (~61% of national area coverage) and continued restoration of degraded natural forests (~0.5% of national area coverage) have the potential to offset about 80% of the national emissions before 2050. Our work monitors and quantifies progress and impact of landscape restoration projects and outlines a pathway to engage smallholder farmers with a limited number of on-farm trees into the expanding carbon market.

Monitoring of improper soil erosion empowered by water is constantly adding more risk to the natural resource mitigation scenarios, especially in developing countries. The demographical pattern and the rate of growth, in addition to the impairments of the rainfall pattern, are consequently disposed to adverse environmental disturbances. The current research goal is to evaluate soil erosion triggered by water in the coastal area of Kenya on the district level, and also in protected areas. The Revised Universal Soil Loss Equation (RUSLE) model was exercised to estimate the soil loss in the designated study area. RUSLE input parameters were functionally realized in terms of rainfall and runoff erosivity factor (R), soil erodibility factor (K), slope length and gradient factor (LS), land cover management factor (C) and slope factor (P). The realization of RUSLE input parameters was carried out using different dataset sources, including meteorological data, soil/geology maps, the Digital Elevation Model (DEM) and processing of satellite imagery. Out of 26 districts in coastal area, eight districts were projected to have mean annual soil loss rates of >10 t·ha⁻¹·y⁻¹: Kololenli (19.709 t·ha⁻¹·y⁻¹), Kubo (14.36 t·ha⁻¹·y⁻¹), Matuga (19.32 t·ha⁻¹·y⁻¹), Changamwe (26.7 t·ha⁻¹·y⁻¹), Kisauni (16.23 t·ha⁻¹·y⁻¹), Likoni (27.9 t·ha⁻¹·y⁻¹), Mwatate (15.9 t·ha⁻¹·y⁻¹) and Wundanyi (26.51 t·ha⁻¹·y⁻¹). Out of 34 protected areas at the coastal areas, only four were projected to have high soil loss estimation rates >10 t·ha⁻¹·y⁻¹: Taita Hills (11.12 t·ha⁻¹·y⁻¹), Gonja (18.52 t·ha⁻¹·y⁻¹), Mailuganji (13.75.74 t·ha⁻¹·y⁻¹), and Shimba Hills (15.06 t·ha⁻¹·y⁻¹). In order to mitigate soil erosion in Kenya’s coastal areas, it is crucial to regulate the anthropogenic disturbances embedded mainly in deforestation of the timberlands, in addition to the natural deforestation process caused by the wildfires.

Ongoing climate change poses a major threat to the soil resources of many African countries that mainly rely on an agricultural economy. While arid and semi-arid lands (ASALs) take up most of Kenya’s land mass, approximately 64% of its total croplands lie within mountainous areas with high rainfall, hence, areas highly vulnerable to water erosion. Flooding of the Great Lakes and increasing desertification of the ASALs are illustrative cases of the implications of recent precipitation dynamics in Kenya. This study applied the Revised Universal Soil Loss Equation (RUSLE) to estimate future soil erosion rates at the national level based on four Coupled Model Intercomparison Project v5 (CMIP5) models under two Representative Concentration Pathway (RCP) scenarios. Results showed the current soil loss rate to be at 4.76 t ha⁻¹ yr⁻¹ and projected an increase in average rainfall erosivity under the two scenarios, except for RCP-2.6 (2030s) and (2080s) for the MIROC-5 model. Future projections revealed an incremental change in rainfall erosivity from the baseline climate by a cumulative average of 39.9% and 61.1% for all scenarios by the 2030s and 2080s, respectively, while soil loss is likely to increase concomitantly by 29% and 60%, respectively. The CCCMA_CANESM2 model under the RCP 8.5 (2080s) scenario projected the highest erosion rate of 15 t ha⁻¹ yr⁻¹ over Kenya, which is a maximum increase of above 200%, with the Rift Valley region recording an increase of up to 100% from 7.05 to 14.66 t ha⁻¹ yr⁻¹. As a first countrywide future soil erosion study, this assessment provides a useful reference for preventing water erosion and improving ecosystem service security.

Remote sensing scene classification is a fundamental responsibility of earth observation, aiming at identifying information granular for land cover classification. The multi-granular land use for multi-source remotely sensed image categories is now a principal task in remote sensing data augmentation and data selection. Understanding image representations are meaningful for the scene classification task. Training deep learning model-based approaches has to do with scene classification and brings about fantastic achievement. At the same time, these high-level approaches are computationally expensive and time–consuming. This paper introduces multi-granularity Neural Network Encoding architecture base on InceptionV3, InceptionReseNetV2, VGG16, and DenseNet201 architecture into remote sensing scene classification. To improve performance and to solve intra-class variation for multi-class scene issues remote sensing dataset. By using pre-trained CNN, activation function and ensemble learning have been adopting. InceptionV3 and VGG16 are used to extract features. InceptionResNetV2 use for fine-tuning, which consists of unfreezing the entre model and retraining the new data with a lower learning rate. The proposed fine-tune whole pre-trained model produces better results of test set up to 97.84 % than features extracted by InceptionResNetV2. Also, we use DCNNs ensemble average and weighted average to achieve better outcomes for 97.36% and 99.10%. In our experiments, we attempted to fine-tune the deep convolutional neural networks (DCNNs) training method for remote sensing scene classification on two public datasets UCM, SIRI-WHU, and one dataset I collected through the google earth engine from East Africa Community Countries (EACC) within nine classes within total 2112 labeled images. The results indicate that our proposed fine-tuning of the pre-trained model with few epochs and less computational time increases accuracy.