Remote sensing can transform the speed, scale, and cost of biodiversity and forestry surveys. Data acquisition currently outpaces the ability to identify individual organisms in high resolution imagery. We outline an approach for identifying tree-crowns in RGB imagery while using a semi-supervised deep learning detection network. Individual crown delineation has been a long-standing challenge in remote sensing and available algorithms produce mixed results. We show that deep learning models can leverage existing Light Detection and Ranging (LIDAR)-based unsupervised delineation to generate trees that are used for training an initial RGB crown detection model. Despite limitations in the original unsupervised detection approach, this noisy training data may contain information from which the neural network can learn initial tree features. We then refine the initial model using a small number of higher-quality hand-annotated RGB images. We validate our proposed approach while using an open-canopy site in the National Ecological Observation Network. Our results show that a model using 434,551 self-generated trees with the addition of 2848 hand-annotated trees yields accurate predictions in natural landscapes. Using an intersection-over-union threshold of 0.5, the full model had an average tree crown recall of 0.69, with a precision of 0.61 for the visually-annotated data. The model had an average tree detection rate of 0.82 for the field collected stems. The addition of a small number of hand-annotated trees improved the performance over the initial self-supervised model. This semi-supervised deep learning approach demonstrates that remote sensing can overcome a lack of labeled training data by generating noisy data for initial training using unsupervised methods and retraining the resulting models with high quality labeled data.

Anthropogenic land cover changes (LCC) affect regional and global climate through biophysical variations of the surface energy budget mediated by albedo, evapotranspiration, and roughness. This change in surface energy budget may exacerbate or counteract biogeochemical greenhouse gas effects of LCC, with a large body of emerging assessments being produced, sometimes apparently contradictory. We reviewed the existing scientific literature with the objective to provide an overview of the state-of-the-knowledge of the biophysical LCC climate effects, in support of the assessment of mitigation/adaptation land policies. Out of the published studies that were analyzed, 28 papers fulfilled the eligibility criteria, providing surface air temperature and/or precipitation change with respect to LCC regionally and/or globally. We provide a synthesis of the signal, magnitude and uncertainty of temperature and precipitation changes in response to LCC biophysical effects by climate region (boreal/temperate/tropical) and by key land cover transitions. Model results indicate that a modification of biophysical processes at the land surface has a strong regional climate effect, and non-negligible global impact on temperature. Simulation experiments of large-scale (i.e. complete) regional deforestation lead to a mean reduction in precipitation in all regions, while air surface temperature increases in the tropics and decreases in boreal regions. The net global climate effects of regional deforestation are less certain. There is an overall consensus in the model experiments that the average global biophysical climate response to complete global deforestation is atmospheric cooling and drying. Observed estimates of temperature change following deforestation indicate a smaller effect than model-based regional estimates in boreal regions, comparable results in the tropics, and contrasting results in temperate regions. Regional/local biophysical effects following LCC are important for local climate, water cycle, ecosystems, their productivity and biodiversity, and thus important to consider in the formulation of adaptation policy. However before considering the inclusion of biophysical climate effects of LCC under the UNFCCC, science has to provide robust tools and methods for estimation of both country and global level effects.

Abstract. This study evaluates the performances of the new version (v.5.1) of 3D-CMCC Forest Ecosystem Model (FEM) in simulating gross primary productivity (GPP), against eddy covariance GPP data for 10 FLUXNET forest sites across Europe. A new carbon allocation module, coupled with new both phenological and autotrophic respiration schemes, was implemented in this new daily version. Model ability in reproducing timing and magnitude of daily and monthly GPP fluctuations is validated at intra-annual and inter-annual scale, including extreme anomalous seasons. With the purpose to test the 3D-CMCC FEM applicability over Europe without a site-related calibration, the model has been deliberately parametrized with a single set of species-specific parametrizations for each forest ecosystem. The model consistently reproduces both in timing and in magnitude daily and monthly GPP variability across all sites, with the exception of the two Mediterranean sites. We find that 3D-CMCC FEM tends to better simulate the timing of inter-annual anomalies than their magnitude within measurements' uncertainty. In six of eight sites where data are available, the model well reproduces the 2003 summer drought event. Finally, for three sites we evaluate whether a more accurate representation of forest structural characteristics (i.e. cohorts, forest layers) and species composition can improve model results. In two of the three sites results reveal that model slightly increases its performances although, statistically speaking, not in a relevant way.