Spaceborne remote sensing provides valuable information about the state of the Earth‐atmosphere system and its components in an area‐wide and continuous manner. Over the past 50 years a range of satellite platforms carrying many different sensors has been constructed to monitor atmospheric parameters used in meteorological and climatological studies, and the information retrieved from satellite‐based sensors has greatly enhanced our understanding of the processes and dynamics within the Earth‐atmosphere system. The present paper gives an overview of existing satellites and sensors, together with the developed algorithms to retrieve meteorological and climatological parameters. Furthermore, it gives an outlook on new systems planned for the near future. Copyright © 2011 Royal Meteorological Society

Information about clouds is important for observing and predicting weather and climate as well as for generating and distributing solar power. Most existing approaches extract cloud information from satellite data by classifying individual pixels instead of using closely integrated spatial information, ignoring the fact that clouds are highly dynamic, spatially continuous entities. This paper proposes a novel cloud classification method based on deep learning. Relying on a Convolutional Neural Network (CNN) architecture for image segmentation, the presented Cloud Segmentation CNN (CS-CNN), classifies all pixels of a scene simultaneously rather than individually. We show that CS-CNN can successfully process multispectral satellite data to classify continuous phenomena such as highly dynamic clouds. The proposed approach produces excellent results on Meteosat Second Generation (MSG) satellite data in terms of quality, robustness, and runtime compared to other machine learning methods such as random forests. In particular, comparing CS-CNN with the CLAAS-2 cloud mask derived from MSG data shows high accuracy (0.94) and Heidke Skill Score (0.90) values. In contrast to a random forest, CS-CNN produces robust results and is insensitive to challenges created by coast lines and bright (sand) surface areas. Using GPU acceleration, CS-CNN requires only 25 ms of computation time for classification of images of Europe with $508\times 508$ pixels.

In the megadiverse tropical mountain forest in the Andes of southern Ecuador, a global biodiversity hotspot, the use of fire to clear land for cattle ranching is leading to the invasion of an aggressive weed, the bracken fern, which is threatening diversity and the provisioning of ecosystem services. To find sustainable land use options adapted to the local situation, a profound knowledge of the long-term spatiotemporal patterns of land cover change and its drivers is necessary, but hitherto lacking. The complex topography and the high cloud frequency make the use of remote sensing in this area a challenge. To deal with these conditions, we pursued specific pre-processing steps before classifying five Landsat scenes from 1975 to 2001. Then, we quantified land cover changes and habitat fragmentation, and we investigated landscape changes in relation to key spatial elements (altitude, slope, and distance from roads). Good classification results were obtained with overall accuracies ranging from 94.5% to 98.5% and Kappa statistics between 0.75 and 0.98. Forest was strongly fragmented due to the rapid expansion of the arable frontier and the even more rapid invasion by bracken. Unexpectedly, more bracken-infested areas were converted to pastures than vice versa, a practice that could alleviate pressure on forests if promoted. Road proximity was the most important spatial element determining forest loss, while for bracken the altitudinal range conditioned the degree of invasion in deforested areas. The annual deforestation rate changed notably between periods: ~1.5% from 1975 to 1987, ~0.8% from 1987 to 2000, and finally a very high rate of ~7.5% between 2000 and 2001. We explained these inconstant rates through some specific interrelated local and national political and socioeconomic drivers, namely land use policies, credit and tenure incentives, demography, and in particular, a severe national economic and bank crisis.

The distinction made by satellite data between ground fog and low stratus is still an open problem. A proper detection scheme would need to make a determination between low stratus thickness and top height. Based on this information, stratus base height can be computed and compared with terrain height at a specific picture element. In the current paper, a procedure for making the distinction between ground fog and low-level stratus is proposed based on Moderate Resolution Imaging Spectroradiometer (MODIS, flying on board the NASA Terra and Aqua satellites) daytime data for Germany. Stratus thickness is alternatively derived from either empirical relationships or a newly developed retrieval scheme (lookup table approach), which relies on multiband albedo and radiative transfer calculations. A trispectral visible–near-infrared (VIS–NIR) approach has been proven to give the best results for the calculation of geometrical thickness. The comparison of horizontal visibility data from synoptic observing (SYNOP) stations of the German Weather Service and the results of the ground fog detection schemes reveals that the lookup table approach shows the best performance for both a valley fog situation and an extended layer of low stratus with complex local visibility structures. Even if the results are very encouraging [probability of detection (POD) = 0.76], relatively high percentage errors and false alarm ratios still occur. Uncertainties in the retrieval scheme are mostly due to possible collocation errors and known problems caused by comparing point and pixel data (time lag between satellite overpass and ground observation, etc.). A careful inspection of the pixels that mainly contribute to the false alarm ratio reveals problems with thin cirrus layers and the fog-edge position of the SYNOP stations. Validation results can be improved by removing these suspicious pixels (e.g., percentage error decreases from 28% to 22%).

The Random Forest (RF) algorithm, a decision-tree-based technique, has become a promising approach for applications addressing runoff forecasting in remote areas. This machine learning approach can overcome the limitations of scarce spatio-temporal data and physical parameters needed for process-based hydrological models. However, the influence of RF hyperparameters is still uncertain and needs to be explored. Therefore, the aim of this study is to analyze the sensitivity of RF runoff forecasting models of varying lead time to the hyperparameters of the algorithm. For this, models were trained by using (a) default and (b) extensive hyperparameter combinations through a grid-search approach that allow reaching the optimal set. Model performances were assessed based on the R², %Bias, and RMSE metrics. We found that: (i) The most influencing hyperparameter is the number of trees in the forest, however the combination of the depth of the tree and the number of features hyperparameters produced the highest variability-instability on the models. (ii) Hyperparameter optimization significantly improved model performance for higher lead times (12- and 24-h). For instance, the performance of the 12-h forecasting model under default RF hyperparameters improved to R² = 0.41 after optimization (gain of 0.17). However, for short lead times (4-h) there was no significant model improvement (0.69 < R² < 0.70). (iii) There is a range of values for each hyperparameter in which the performance of the model is not significantly affected but remains close to the optimal. Thus, a compromise between hyperparameter interactions (i.e., their values) can produce similar high model performances. Model improvements after optimization can be explained from a hydrological point of view, the generalization ability for lead times larger than the concentration time of the catchment tend to rely more on hyperparameterization than in what they can learn from the input data. This insight can help in the development of operational early warning systems.