Clouds represent the biggest source of uncertainty in climate response to CO$_2$ forcing. Of special concern are the stratocumulus decks found in the subtropics. Modeling their dynamics is challenging at scale, and so their behavior must be approximated in global models. Satellite observations provide an opportunity to learn about cloud physics, but represent a serious data challenge.
Using a self-supervised cloud classification system based on an auto-encoder applied to MODIS multi-spectral imagry, we can identify cloud types based on texture, which may diferentiate physical processes. When employed on 20 years of MODIS observations and co-located we reanalysis weather data. We find that low cloud texture in the subtropics is well-explained by the temperature structure of the lower tropopshere, specifically the estimated inversion strenght and the near surface temperature. In each temperature regime, you are most likely to find a certain class of clouds (boundaries are less clear).
See more details: https://github.com/jamesafranke/AICCA/blob/master/ms/low_marine_clouds_met_PNAS.pdf
Recent advances deep learning architctures have motivage a resurgence in in data-driven weather forecasting. However, all recent transformer-based forecast models are trained on reanalysis model data (usually ERA5, which does not contain all the revelant cloud phyisics to begin with. We developed a new machine-learning ready cloud dataset based on the 5 main geostationary satelites we call CloudBench (following WeatherBench) in order to facilitate data-driven cloud forecasting.
Using these observations, we employ a decoder-only transformer (GPT-like) to learn the relationship between two historical timesteps and one future timestep. Forecasts can in principle be rolled out indefinitely. Research is preliminary.
Stratospheric aerosol injection may impact low clouds by changing the radiative cloud top cooling or vertical temperature structure of the atmosphere. I am investigating this potential through a combination of large eddy simulations and satelite observations. Prelimary analysis in global scale models shows that low clouds repond relativly more to SAI forcing than to CO2.
Cattle farming is a major source of global food production and livelihoods that is being impacted by climate change. However, despite numerous studies reporting local-scale heat impacts, quantifying the global risk of heat stress to cattle from climate change remains challenging. We conducted a global synthesis of documented heat stress for cattle using 164 records to identify temperature-humidity conditions associated with decreased production and increased mortality, then projected how future greenhouse gas emissions and land-use decisions will limit or exacerbate heat stress, and mapped this globally. The median threshold for the onset of negative impacts on cattle was a temperature-humidity index of 68.8 (95% C.I.: 67.3–70.7). Currently, almost 80% of cattle globally are exposed to conditions exceeding this threshold for at least 30 days a year. For global warming above 4°C, heat stress of over 180 days per year emerges in temperate regions, and year-round heat stress expands across all tropical regions by 2100. Limiting global warming to 2°C, limits expansion of 180 days of heat stress to sub-tropical regions. In all scenarios, severity of heat stress increases most in tropical regions, reducing global milk yields. Future land-use decisions are an important driver of risk. Under a low environmental protection scenario (SSP3-RCP7.0), the greatest expansion of cattle farming is projected for tropical regions (especially Amazon, Congo Basin, and India), where heat stress is projected to increase the most. This would expose over 500 million more cattle in these regions to severe heat risk by 2090 compared to 2010. A less resource-intensive and higher environmental protection scenario (SSP1-RCP2.6) reduces heat risk for cattle by at least 50% in Asia, 63% in South America, and 84% in Africa. These results highlight how societal choices that expand cattle production in tropical forest regions are unsustainable, both worsening climate change and exposing hundreds of millions more cattle to large increases in severe, year-round heat stress.
See more details: https://iopscience.iop.org/article/10.1088/1748-9326/aceb79
Modern food production is spatially concentrated in global “breadbaskets.” A major unresolved question is whether these peak production regions will shift poleward as the climate warms, allowing some recovery of potential climate-related losses. While agricultural impacts studies to date have focused on currently cultivated land, the Global Gridded Crop Model Intercomparison Project (GGCMI) Phase 2 experiment allows us to assess changes in both yields and the location of peak productivity regions under warming. We examine crop responses under projected end of century warming using seven process-based models simulating five major crops (maize, rice, soybeans, and spring and winter wheat) with a variety of adaptation strategies. We find that in no-adaptation cases, when planting date and cultivar choices are held fixed, regions of peak production remain stationary and yield losses can be severe, since growing seasons contract strongly with warming. When adaptations in management practices are allowed (cultivars that retain growing season length under warming and modified planting dates), peak productivity zones shift poleward and yield losses are largely recovered. While most growing-zone shifts are ultimately limited by geography, breadbaskets studied here move poleward over 600 km on average by end of the century under RCP 8.5. These results suggest that agricultural impacts assessments can be strongly biased if restricted in spatial area or in the scope of adaptive behavior considered. Accurate evaluation of food security under climate change requires global modeling and careful treatment of adaptation strategies.
See more details: https://onlinelibrary.wiley.com/doi/abs/10.1111/gcb.15868