[1] Tropical savanna (locally known as cerrado) composes 24% of Brazil and is characterized by high climatic variation; however, patterns of energy exchange are poorly understood, especially for mixed grasslands (locally known as campo sujo). We used eddy covariance to measure latent (L_e) and sensible (H) heat flux of a mixed grassland and linked meteorological and remote sensing data to determine the controls on these fluxes. We hypothesized that (1) seasonal variations in H and L_e would be large due to variations in precipitation; (2) ecosystem phenology, estimated using the enhanced vegetation index (EVI), would be the best predictor of seasonal variation in L_e; and (3) cerrado, transitional, and humid evergreen forests would have similar rates of average annual L_e despite large seasonal variation in cerrado L_e. Our data suggest that campo sujo exhibits large seasonal fluctuations in energy balance that are driven by rainfall and that responses to rainfall pulses are rapid and dynamic, especially during the dry season. Path analysis indicated that temporal variations in the EVI did not significantly affect L_e or G_c, but this was because all three variables (EVI, L_e, and G_c) responded similarly to temporal variations in surface water availability. Compared to other tropical ecosystems, wetter sites had higher rates of L_e during the dry season but similar rates during the wet season when water was not limiting. Over annual time periods, average rates of L_e increased significantly as average annual rainfall increased, due to dry-season water limitations in the more seasonal tropical ecosystems.

Many remote sensing‐based evapotranspiration (RSBET) algorithms have been proposed in the past decades and evaluated using flux tower data, mainly over North America and Europe. Model evaluation across South America has been done locally or using only a single algorithm at a time. Here, we provide the first evaluation of multiple RSBET models, at a daily scale, across a wide variety of biomes, climate zones, and land uses in South America. We used meteorological data from 25 flux towers to force four RSBET models: Priestley–Taylor Jet Propulsion Laboratory (PT‐JPL), Global Land Evaporation Amsterdam Model (GLEAM), Penman–Monteith Mu model (PM‐MOD), and Penman–Monteith Nagler model (PM‐VI). was predicted satisfactorily by all four models, with correlations consistently higher () for GLEAM and PT‐JPL, and PM‐MOD and PM‐VI presenting overall better responses in terms of percent bias (%). As for PM‐VI, this outcome is expected, given that the model requires calibration with local data. Model skill seems to be unrelated to land‐use but instead presented some dependency on biome and climate, with the models producing the best results for wet to moderately wet environments. Our findings show the suitability of individual models for a number of combinations of land cover types, biomes, and climates. At the same time, no model outperformed the others for all conditions, which emphasizes the need for adapting individual algorithms to take into account intrinsic characteristics of climates and ecosystems in South America.

Fire in Brazilian Pantanal represents a serious threat to biodiversity. The Brazilian National Institute of Spatial Research (INPE) has a program named Queimadas, which estimated from January 2020 to October 2020, a burned area in Pantanal of approximately 40,606 km${}^2$. This program also provides daily data of active fire (fires spots) from a methodology that uses MODIS (Aqua and Terra) sensor data as reference satellites, which presents limitations mainly when dealing with small active fires. Remote sensing researches on active fire dynamics have contributed to wildfire comprehension, despite generally applying low spatial resolution data. Convolutional Neural Networks (CNN) associated with high- and medium-resolution remote sensing data may provide a complementary strategy to small active fire detection. We propose an approach based on object detection methods to map active fire in the Pantanal. In this approach, a post-processing strategy is adopted based on Non-Max Suppression (NMS) to reduce the number of highly overlapped detections. Extensive experiments were conducted, generating 150 models, as five-folds were considered. We generate a public dataset with 775-RGB image patches from the Wide Field Imager (WFI) sensor onboard the China Brazil Earth Resources Satellite (CBERS) 4A. The patches resulted from 49 images acquired from May to August 2020 and present a spatial and temporal resolutions of 55 m and five days, respectively. The proposed approach uses a point (active fire) to generate squared bounding boxes. Our findings indicate that accurate results were achieved, even considering recent images from 2021, showing the generalization capability of our models to complement other researches and wildfire databases such as the current program Queimadas in detecting active fire in this complex environment. The approach may be extended and evaluated in other environmental conditions worldwide where active fire detection is still a required information in fire fighting and rescue initiatives.