Programa de Apoio aos Planos de Reestruturação e Expansão das Universidades Federais / This study analyzes simulations of Net Primary Production (NPP) from 15 different landsurface models (LSMs) and biomass pools from 6 different LSMs using meteorological conditions measured at 8 sites from Large-Scale Biosphere-Atmosphere Experiment in Amazonia (LBA) project as drivers. The models were not calibrated for the sites. The sites are divided into four biome types: Evergreen Broadleaf Forests (4 sites); Deciduous Broadleaf Forest (1 site); Savanna (1 site); Pasture/Agriculture (2 sites). The mean daily cycles, monthly and annual means of NPP were intercompared and evaluated. There were considerable differences among the NPP simulations, and some of these differences reached up to two orders of magnitude in nocturnal values. Seasonality in dry periods of the NPP could be observed in some models for all biome types. The annual mean NPP simulations from two Evergreen Broadleaf Forests (K34 and K67 sites) were compared with the observations. In general, the simulations by most models do not represent very well the observations; however, the mean value from all simulations is able to represent the observed data. In general, models that represented the Dynamic Vegetation Carbon Fluxes and Nitrogen Cycling Models (DVN) were those that better represented the observed values, suggesting that a more specific description of the vegetation dynamics capture, even without calibration, the carbon exchanges with enough accuracy. The simulated biomass is also divergent between the models, although the distribution of that biomass follows the expected patterns for each biome type. Therefore, we believe that a model calibration can improve the simulations results. / Este estudo analisa simulações de Produção Primária Líquida (NPP) de 15 diferentes modelos de superfície (LSMs) e reservatórios de biomassa de 6 diferentes LSMs usando condições meteorológicas medidas em 8 sítios do projeto Large-Scale Biosphere-Atmosphere Experiment in Amazônia (LBA) como forçantes. Os modelos não foram calibrados para os sítios. Os sítios foram divididos em quatro tipos de biomas: Florestas de Folhas Largas Sempre-Verdes (4 sítios); Florestas de Folhas Largas Decídua (1 sítio); Savana (cerrado, 1 sítio); Pasto/Agricultura (2 sítios). Os ciclos de NPP médios diários, mensais e anuais foram intercomparados e avaliados. Existem diferenças consideráveis entre as simulações de NPP, e algumas destas diferenças alcançaram até duas ordens de magnitude em valores noturnos. Pôde ser observada sazonalidade de NPP em alguns modelos para todos os tipos de bioma. O NPP médio anual simulado em duas Florestas Sempre-Verdes (sítios K34 e K67) foi comparado com as observações. Em geral, as simulações da maior parte dos modelos não representam muito bem as observações; entretanto, o valor médio de todas as simulações consegue representar os dados observados. Em geral, modelos que representam a vegetação dinâmica, fluxos de carbono e ciclo do nitrogênio (DVN) foram aqueles que melhor representaram os valores observados, sugerindo que uma descrição mais específica da dinâmica da vegetação pode capturar, mesmo sem calibração, as trocas de carbono com suficiente precisão. A biomassa simulada é também divergente entre os modelos, embora a distribuição dessa biomassa segue os padrões esperados para cada tipo de bioma. Logo, acreditamos que uma calibração do modelo pode melhorar os resultados das simulações.

Land surfaces have changed as a result of human and natural processes, such asdeforestation, urbanization, desertification and natural disasters like wildfires. Land use and landcover change impacts local and regional climates through various bio geophysical processes acrossmany time scales. More realistic representation of land surface parameters within the land surfacemodels are essential to for climate models to accurately simulate the effects of past, current andfuture land surface processes. In this study, we evaluated the sensitivity and accuracy of theWeather Research and Forecasting (WRF) model though the default MODIS land cover data andannually updated land cover data over southeast of United States. Findings of this study indicatedthat the land surface fluxes, and moisture simulations are more sensitive to the surfacecharacteristics over the southeast US. Consequently, we evaluated the WRF temperature andprecipitation simulations with more accurate observations of land surface parameters over thestudy area. We evaluate the model performance for the default and updated land cover simulationsagainst observational datasets. Results of the study showed that updating land cover resulted insubstantial variations in surface heat fluxes and moisture balances. Despite updated land use andland cover data provided more representative land surface characteristics, the WRF simulated 2- m temperature and precipitation did not improved due to use of updated land cover data. Further,we conducted machine learning experiments to post-process the Noah-MP land surface modelsimulations to determine if post processing the model outputs can improve the land surfaceparameters. The results indicate that the Noah-MP simulations using machine learning remarkablyimproved simulation accuracy and gradient boosting, and random forest model had smaller meanerror bias values and larger coefficient of determination over the majority of stations. Moreover,the findings of the current study showed that the accuracy of surface heat flux simulations byNoah-MP are influenced by land cover and vegetation type.