commonchaffinch @commonchaffinch@qoto.org
English/中文
Environmental scientist looking at global ecohydrological change using data analysis and modeling tools.
Joined May 2022
#urban_heat_islands
#heat_waves
In 2018, Lei Zhao from Princeton University and coauthors analyzed the effect of heat waves on the magnitudes of urban heat island effect.
All the findings were based on Community Earth System Models over 65 cities in North America. The heat waves were defined using the rural land unit’s temperature. The analysis were conducted based on both 2m and land surface temperatures, and the results were consistent. However, note the convective term’s contribution to land surface temperature is to the opposite of the 2m temperature.
The major conclusions are as follows:
1. During heat waves, the urban heat island magnitude is either not changed, or exacerbated.
2. The exacerbation happens to daytime temperature in the humid eastern U.S. in the present-day climate, and is caused by the evaporative effect and the increased anthropogenic waste heat from air conditioning. The evaporative effect is as such: the ample water availability in this region allows increased evaporative cooling in the rural regions during a heat wave, but the urban region, with reduced vegetation, does not have this benefit.
In the future climate, precipitation increases in this region in CESM, such that the urban land unit becomes amply watered and no longer suffers from the lack of evaporative cooling.
3. The lack of change happens to daytime temperature in the arid southeastern U.S. in the present-day climate, where the evaporative cooling contrast does not exist.
In the future climate, precipitation increases in this region in CESM create evaporative cooling contrast, and makes the urban heat island effect stronger during heat waves.
4. The exacerbation of nighttime temperature happens to all regions in the present and future climate, and is mainly caused by the increased use of air conditioning.
https://iopscience.iop.org/article/10.1088/1748-9326/aa9f73/meta
#urban_heat_islands
#physical_mechanisms
The 2014 study by Xuhui Lee from Nanjing University of Information Science and Technology, China, and coauthors investigated the drivers of urban heat island effect using 65 cities across North America.
Usually, urban heat island effect - expressed as the temperature difference between urban and rural areas, $\Delta$T - is considered to be primarily created by the reduced evaporative cooling over the urban surface.
However, Lee and coauthors showed that variations in $\Delta$T, across space, are rather created by variations in convective efficiency, which is determined by aerodynamic resistance, or surface roughness. The cities in humid regions are aerodynamically smoother than their surrounding rural areas, resulting in less efficient transfer of heat to the lower atmosphere. The cities in dry regions, on the country, are aerodynamically rougher than their surrounding rural areas, resulting in more efficient transfer of heat to the lower atmosphere. Therefore, the urban heat island effect is more important for the humid cities, and albedo management is an important means to mitigate urban heat island effect.
Background
The contributors to urban heat island effect includes (1) reduced evaporative cooling, (2) waste heat from human activities, (3) (sometimes) changed albedo, (4) daytime heat storage and nighttime heat release by buildlings, which is usually greater than by vegetation and soil, (5) energy redistribution through convection between the land surface and the atmospheric boundary layer. Among these, (1) and (5) are natural factors, and (2)(3)(4) are morphological and anthropogenic aspects of the city itself.
Method
The urban heat island effect was quantified in two ways. First, via the NASA MODIS satellite land surface temperature (LST) from 2003 to 2012 - note this is different from air temperature. Second, via a climate model (the Community Earth System Model) run at 0.23 x 0.31 degrees resolution to reflect individual cities from 1972 to 2004.
For the MODIS data, nine urban pixels (3x3) were selected from the city center, and were paired with 1-3 patches of 3x3 to 7x7 pixels in the surrounding rural land. Water pixels were excluded. [Isn't this too few?]
The climate model does not have a urban core definition, because the urban core and the rural areas are different land units, not spatially distinguished, in the same grid cell. [Now this clearly suggests potential biases in the attribution results due to model deficiencies. The study also used LST, and the findings would probably be less evident if air temperature were used]
Partition of the contributions from terms (1)-(5) were achieved via the method described in ref. 19. The partitioning equation is shown in the paper, and appears to depend on land surface model parameters. Aerodynamic resistance is calculated as a function of the thermal gradient between the temperature at the surface and at reference height, and the sensible heat flux.
Results
1. Simple regression across space between factors and the MODIS LST shows that annual mean nighttime $\Delta$T is positively correlated with the logarithm of population, but not significantly correlated with the climate (precipitation, solar radiation, temperature). On the other hand, the annual mean dayttime $\Delta$T is strongly positively correlated with precipitation, and weakly correlated with population.
Geographically, the visual effect is that the nighttime $\Delta$T does not differ much between the east and west, but the daytime $\Delta$T is clearly bigger for the eastern cities (Fig. 1).
These findings are also well-supported by past studies (refs 14,15,16,17)
[This result seems to suggest that nighttime $\Delta$T is mainly controlled by anthropogenic heat release, whereas daytime mainly by the natural evaporative and convective factors.]
2. The attribution plot, Fig. 2, shows that
(1) nighttime $\Delta$T is dominated by the evaporative term and the building storage term.
(2) for dayttime $\Delta$T, the evaporative term always contributed positively. The convective term contributed most positively in the southeastern U.S., moderately positively in the northern temperate zones, and negatively in the southeastern U.S. Interestingly, the radiation term (albedo) is negative in the southeastern U.S., suggesting that the cities increased the albedo of the surface and reflected more sunlight, but positive elsewhere.
3. Fig. 3 further shows the spatial correlations between the individual terms and precipitation. It is clear that the positive correlation between daytime $\Delta$T and precipitation is contributed mainly by the convection term. In fact, the covariance analysis shows that the convection term explains 200% of the covariance, whereas the other terms together explains -100%.
In rural lands in the humid areas, the aerodynamic roughness is greater, meaning lower aerodynamic resistance (39 s m-1) than the urban counterpart (62 s m-1). Lower aerodynamic resistance means greater convective efficiency.
In past studies, the urban cool island effect in dry regions was explained by the evaporative cooling of urban greenspace (refs 15, 16, 17). However, in the current results, a lot of the cooled pixels are vegetation free.
4. Fig. 4 differs from Figs. 1-3 by showing the temporal sensitivity of dayttime and nighttime $\Delta$T to precipitation. However, there is conflict between the MODIS data and the climate model with regard to whether the dayttime $\Delta$T in the majority of the cities is positively or negatively correlated with precipitation. The conflict may be due to the short length of the MODIS data, or the bias of the climate model. Nonetheless, the precipitation-induced cooling effect becomes more evident as annual mean precipitation increases, i.e. towards the east. This is clearly induced by evaporative cooling.
写条件语句的时候，要修改的变量一定不能出现在条件语句里。
参加了一个大项目的panel review，90%的内容都听不懂，怀疑自己一辈子也不可能达到管理那个级别项目的水平。且不提跨领域的海洋生物地球化学循环和气候系统的tipping point，只是和陆面相关的机器学习、野火、植物光合机理和对水文&碳循环的控制、模型权重这些，都是只知道关键词而无法跟上逻辑。我能学会这些吗，在学界的pay/time balance情况下，我有必要学会这些吗？
#ecosystem_resilience
#hydroclimate_extremes
#ecological_extremes
In 2014, J. Zscheischler and co-authors conducted a continental-scale analysis on extreme events in gross primary productivity (GPP). The datasets used included a machine learning based construction, a semi-empirical, and two land surface models (OCN and LPJmL).
They found a few important phenomena:
(1) The 50 largest positive and negative GPP extremes accounted for most of the variations in continental GPP variation.
* That is, the extreme events, though limited in number, are very important for interannual variability in GPP.
(2) The spatial extents of the GPP extremes played a larger role on the impact of the event, than the duration or maximal GPP.
(3) Water scarcity was the most important cause of negative GPP extremes. Heat waves played a secondary role. In Europe, South America, and Oceania, fire was a third important factor.
* That is, GPP extremes happened most often when there is drought, followed by heat waves, and finally, in some continents, fires.
* It's interesting that the heat wave seemed to account for the GPP extremes best in Russia. Is this because the vegetation there are adapted to cold conditions?
#urban_vegetation
#urban_rural_differences
#vegetation_cover
(The vegetation cover was quantified as Enhanced Vegetation Index (EVI)).
In 2018, Shuqing Zhao analyzed the direct and indirect effects of urbanization on vegetation growth. The direct effect of urbanization is through land cover change. The indirect effect is through the urban environment. On average, the indirect vegetation growth enhancement in the U.S. offset 29.2%, 29.5%, and 31.0% of the growth reductions caused by impervious area replacement.
Background
There has been a debate on whether the urban environment enhances or suppresses vegetation growth. High air temperature, low soil water content are stress factors on vegetation. But fertilization, irrigation, introduction of non-native species, urban heat island, climate change, and atmospheric chemistry change like ozone and carbon dioxide were found to enhance vegetation growth by later studies. The authors previous study [Shuqing Zhao et al. PNAS 133 (22) 6313-6138 https://doi.org/10.1073/pnas.1602312113] already demonstrated that vegetation growth enhancement existed in 32 major cities in China. In this study, they would like to verify the same effect existed in USA cities.
Some useful information
The Northeast (NE) and Southeast (SE) U.S. are mainly covered by forests. The East North Central (ENC) and Central (C) are covered by a large cropland area. The West North Central (WNC) are the South are heavily covered by crops, pasture, and grassland. The Southwest (SW) is dominated by shrubs. The West (W) and Northwest (NW) are mostly covered by forests and shrubs.
Data
377 metropolitan statistical areas in CONUS
EVI 250 m product (MOD13Q1) from MODIS, averaged over the growing season (frost-free days) in the years 2001, 2006, and 2011
Impervious area was originally 30 m and interpolated to 250 m
The pixels that were water body, or had elevations > 50 m above the highest elevation of urban core (impervious percentage > 50%) were excluded to remove the effect of water body and elevation
Methods
(1) Definitions of the relative direct effect, the relative indirect effect, and the growth offset
The conceptual decomposition of the indirect and direct effects of impervious area is:
V_{obs} = (1 + \omega)(1 - \beta)V_v + \beta V_{nv}
, where V_{obs} is the observed vegetation index, V_{v} is the background vegetation index without urbanization, V_{nv} is the vegetation index of the pixel when it is completely impervious, \beta is the percent impervious area, and \omega is the effect of urbanization on vegetation growth. Note that the pixels with no impervious area may still have enhanced vegetation growth in an urban area, but the number of these pixels is relatively small.
V_{nv} was obtained as the mean EVI of the fully urbanized pixels over all the MSAs, and found equal to 0.0064. V_v was obtained by polynomial regression between V_{obs} and \beta (V_{obs} = V_v + a_1 * \beta + a2_ * \beta^2 + a3 * \beta^3).
The “background” vegetation growth, without vegetation impacts, can then be interpolated from V_v and V_{nv} as:
V_{zi} = (1 - \beta)V_v + \beta V_{nv}
, with V_{nv} being equal to 0.0064 in this study. So the relative direct urbanization effect on vegetation growth is:
\omega_d = (V_{zi} - V_v) / V_v * 100%
, and the relative indirect urbanization effect on vegetation growth is
\omega_i = (V_{obs} - V_{zi}) / V_{zi} * 100%
Finally, “growth offset” is defined as the ratio of the absolute indirect effect to the absolute direct effect, in order to quantify how the indirect effect compensates for or worsen the direct effect:
\tau = (V_{obs} - V_{zi}) / (V_v - V_{zi}) * 100%
(2) Definition of urban intensity
The urban intensity of a pixel is defined as the percentage of developed imperviousness surfaces in the pixel.
It seems they derived V_v separately for each city and each year, based on Fig. 2b and Section 3.2. However, it is not clear if they controlled for the land cover effect by separately deriving for each land cover type.
Results
Fig.2c shows quite clearly that in the vast majority of pixels, the V_{obs} is higher than the expected value from impervious area (V_{zi}) - i.e. “urban pixels are often greener than expected given the amount of paved surface they contain”.
The estimated \omega_i values appear to increase slightly with \beta, whereas the \tau values decrease slightly. That is, as the impervious areas become higher and the vegetation becomes sparser, the compensated growth becomes larger relative to the actual growth, but becomes smaller relative to the amount of missed growth due to the direct effect.
In terms regional differences, unsurprisingly, the unperturbed vegetation (V_v) is the highest for the forested NE and the cropped ENC and C, and the lowest for the arid SW and W.
In all the individual regions, the growth offset (\tau) decreases with higher impervious area (\beta). The SW has the highest \tau, whereas the C, NE, and SE have the lowest. This may be because the western U.S. is more heavily irrigated?
In the eastern regions (ENC, S, SE, NE, C), the relative indirect effect (\omega_i) increases with impervious area (\beta), but in the western regions (NW, WNC, W, SW), the relative indirect effect is the highest at medium impervious levels.
Discussion
Their results demonstrated the indirect effect, which was rarely done by past regional scale studies. Also, their results were consistent with most ground observations that showed urbanization to enhance vegetation growth. A few studies (e.g. in temperate zone cities in Europe) showed negative impacts of urbanization on vegetation growth.
The urban to rural gradient include factors like terrain, soils, species, air pollutants, temperature, CO2 enrichment, N deposition, ozone, and traffic volume. Therefore, the potential causes of influences are myriad and complicated.
The unique hump-shaped curve in the west may be because there are too little effort spent on vegetation planting on city fringes, and the amount of impervious area is too high to allow vegetation in the urban center. As a result, the medium dense areas, intentional planting and urban management effectively increase the vegetation index. On the other hand, man-made changes in species and management in humid cities would not cause any obvious increase in vegetation index, since the background is already quite high.
#urban_vegetation
#urban_rural_differences
#phenology
These two studies in 2021 and 2019 show that, in urban areas, the start-of-season usually occur earlier, and end-of-season latter, overall enhancing the length of the growing season.
There are, however, exceptions to this rule. The responses of start and end of season depend on the background climate of the urban area (latitude, spring day- and nighttime temperatures). The northern cities have greater urban-rural distinction than the southern cities.
The corresponding authors to these studies were Shuqing Zhao from Peking University, and Xinchang Zhang from Guangzhou University & Qinchuang Xin from Chinese Academy of Sciences and Sun-Yat Sen University.
https://onlinelibrary.wiley.com/doi/full/10.1111/gcb.15602?
https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2019JG005262
#urban_vegetation
#urban_trees
#species_distribution_models
2019 study by Hugh Burley from Macquarie University, Australia and co-authors
Background information
(1) Although global governments are placing great emphasis on using urban trees to improve the adaptability and sustainability of cities under cilmate change, there are little consideration on whether the selected urban tree species will be resilient to climate changes that may happen during their lifespan. Climate change has been demonstrated to shift species taxa for many natural plants, crops, and invasive species, but few such attention has been paid to urban trees.
(2) Several previous studies reported shifts in the growth of urban trees, which were likely due to shifting climate [Lanza & Stone 2016; Nitschke et al. 2017; Pretzsch et al. 2017]. Analysis across climate gradients showed that trees that were adapted to cool environments in the natural setting tended to be planted in warmer cities, and vice versa [Kendal et al. 2018].
(3) The urban heat island may exacerbate heat stress on trees, but irrigation may be available to urban trees [Jenerette et al. 2016; Vogt et al. 2017].
(4) Climate suitability models, i.e. species distribution models, can describe the environmental tolerances of the species, and can be used to identify un-occupied suitable areas, or project climate impacts on species.
Study objective
(1) Assess the climatically suitable habitat for 176 native Australia urban tree species during the baseline (1960-1990), short term (2030) and long term (2070) periods
- which tree species are more likely to experience increases and decreases in the habitat sizes
- which urban areas are likely to see more/fewer tree species that can use it as a suitable habitat
- hypothesis: urban areas in cooler regions will gain in species, and warmer regions lose species; previous findings in the natural setting revealed this to be true [O'Donnell et al. 2012]
Data and methods
(1) The native tree species kept by Australia nurseries were obtained from online resources. The existence of these tree species were further verified by contacting local government authorities, resulting in data for 44 local councils spanning 49 "significant urban areas" (SUAs). The final results were 248 species.
(2) The spatial occurrences of the species were obtained from GBIF and the Atlas of Living Australia (www.ala.org.au, rgbif package, ALA4R package). Due to the relative lack of data for the arid and tropical regions of Australia, the analysis was limited to the temperate Köppen zones in Australia.
(3) The climatic variables for the species distribution modeling were obtained from WorldClim and included 8 bioclimatic variables: 1) annual mean temperature, 2) temperature seasonality, 3) maximum temperature of the warmest month, 4) minimum temperature of the coldest month, 5) annual precipitation, 6) precipitation seasonality, 7) precipitation of the wettest month, 8) precipitation of the driest month.
(4) The species distribution model was Maxent. The performance was assessed using the Area Under the Receiver Operating Characteristic (AUC) and the Truee Skill Statistic (TSS) through five-fold cross-validation. The Multivariate Environmental Similarity Surfaces (MESS) was further used to indicate whether the models were applied on novel states of individual variables, which can cause the model to over-project suitability. Note MESS cannot indicate whether novel combinations of variables occurred. The MESS results further excluded 72 species, due to the insufficient ability of the above 8 variables to characterize their niche. Therefore, 176 species were analyzed in the end.
Results
(1) The climatically suitable habitat will decline from 16,152 km2 in the baseline to 13,043 km2 in 2030, and 12,300 in 2070. Among the 176 species, 18% will lose >50% of their habitat by 2030, 34% will lose >50% by 2070, and 11 species will increase their habitat by >50%. Generally, suitable habitat will shift polewards, so most of the species gained habitat in the south, and lost in the north.
(2) The hotter places (higher mean annual temperature & mean annual maximum temperature) tended to gain fewer species, and lose more species.
(3) At present, each SUA contained suitable climate for 10-139 of the 176 species, and on average 74 species, but by 2070, the average suitable species declined to 63. 21 of the 82 SUAs, mainly in the cooler regions, increased in the number of suitable species between the baseline and 2070.
Discussion
(1) The poleward shift in climatically suitable habitat in urban areas (SUAs) parallels the change in natural areas.
(2) Species and urban areas in cooler regions would fare better than those in warmer regions, which also supports the findings of previous studies [Jenerette et al. 2016; Kendal et al. 2018].
(3) Some caveats are that: the study area is limited to the temperate regions of Australia, and the tropical & subtropical & arid area trees may respond differently; only macroclimatic predictors were considered, microclimate, extreme weather, and edaphic conditions need to be considered by future studies; management factors like irrigation may increase the habitat range of some species despite warming.
(4) Heat waves and drought can affect the growth of urban tree species [Nitschke et al. 2017]. Climate change also affects the growth rate [Pretzsch et al. 2017; Jia et al. 2018].
Referenced studies
Jenerette, G.D., Clarke, L.W., Avolio, M.L., Pataki, D.E., Gillespie, T.W., Pincetl, S., Nowak, D.J., Hutyra, L.R., McHale, M., McFadden, J.P., Alonzo, M., 2016. Climate tolerances and trait choices shape continental patterns of urban tree biodiversity. Glob. Ecol. Biogeogr. 25, 1367–1376.
Jia, W., Zhao, S., Liu, S., 2018. Vegetation growth enhancement in urban environments of the conterminous United States. Glob. Chang. Biol. 24, 4084–4094.
Kendal, D., Dobbs, C., Gallagher, R.V., Beaumont, L.J., Baumann, J., Williams, N.S.G., Livesley, S.J., 2018. A global comparison of the climatic niches of urban and native tree populations. Glob. Ecol. Biogeogr. 27, 629–637.
Nitschke, C.R., Nichols, S., Allen, K., Dobbs, C., Livesley, S.J., Baker, P.J., Lynch, Y., 2017. The influence of climate and drought on urban tree growth in Southeast Australia and the implications for future growth under climate change. Landsc. Urban Plan. 167, 275–287.
Pretzsch, H., Biber, P., Uhl, E., Dahlhausen, J., Schütze, G., Perkins, D., Rötzer, T., Caldentey, J., Koike, T., Con, T.V., Chavanne, A., Toit, B.D., Foster, K., Lefer, B., 2017. Climate change accelerates growth of urban trees in metropolises worldwide/631/158/858/704/158/ 2165 article. Sci. Rep. 7, 15403.
https://www.sciencedirect.com/science/article/pii/S0048969719323289?via%3Dihub
#wildfires
#emergent_constraints
#machine_learning
Yan Yu from Peking University and co-authors in 2022.
Background
(1) It is necessary to create constrained future projections of fire emissions
Current earth system models only include incomplete and highly parameterized driving processes of fire, and cannot accurately characterize the human-vegetation-fire-climate feedback.
(2) The conventional emergent constraint approach does not satisfy the need of wildfire constraint, because
- Only 13 of the currently available CMIP6 models provide fire carbon emissions in both historical and future simulations, and they lack diversity in structures and parameters
- The linear relationship estimated by traditional emergent constraint cannot deal with the complex, nonlinear relationship between meteorological, ecological, and socioeconomic factors and fire
- The traditional emergent constraint may be suitable for large-scale averaged quantities, but not for the detailed spatial distributions of wildfire
- The emergent constraint approach, however, is still more suitable than bias-correction or model weighting.
Method
(1) The authors developed a machine learning based emergent constraint method to project future global fire emissions and socioeconomic risks
(2) The machine learning models were trained on the Earth system model-simulated relationships between the historical climate, ecosystem, and socioeconomic variables, and future fire carbon emissions under SSP585. The observed variables were then substituted in. The predictors included fuel abundance (leaf area index, temperature, precipitation), fuel moisture (soil moisture, relative humidity, precipitation, temperature), fire spread conditions (wind, orography), ignition sources (flash rate, land use, population).
Results
(1) The machine learning constraint reduces the biases in fire carbon emission bias. Before and after the emergent constraint, the grid-level biases always ranged from -0.32 to 0.16 kg m-2 yr-1, but the grids that have large errors were much, much more before the constraint than after.
(2) The constrained results showed weaker trends in fire carbon emissions compared to the un-constrained results, but the socioeconomic exposure became greater, in terms of population, gross domestic production, and agricultural area.
Jiafu Mao from the Oak Ridge National Laboratory and co-authors in 2021 examined the interannual variability in global burned area using satellite-derived wildfire products and the outputs of the E3SM Land Model v1. The investigation included multiple aspects of the interannual variability, i.e.
(1) the spatial and temporal structures during 1997-2018
(2) contributions from individual biomes
(3) covariations with climate factors
(4) comparison between the ELM and several remote sensing wildfire products.
The data sources of the observations were
(1) The Global Fire Emission Database (GFED): 1997-2016, 0.25 degrees, monthly burned area. This product is derived from several satellite sources (TRMM, ATSR, MODIS).
(2) The FireCCI5.1 burned area: 2001-present, 250m, MODIS image.
Main findings
(1) The highest observed interannual variability is in the boreal area and semi-arid regions, and the lowest in tropical and subtropical regions - especially the African subtropical savannah systems.
(2) The ELM simulations underestimated the interannual variability's magnitudes in the boreal area, and overestimated the variability's magnitudes over Africa. The highest interannual variability in ELM-simulated burned area is still in the boreal forests, but the lowest is in the temperate grassland and shrubland.
(3) The GFED and ELM simulations both showed that the burned area interannual variability were positively correlated with temperature and shortwave radiation, but negative with precipitation. The only disagreement was in North Australia, where GFED had negative correlations with temperature, but ELM had positive.
https://www.sciencedirect.com/science/article/pii/S1674927821000903
#wildfires
#machine_learning
#seasonal_forecasts
Yu Yan from Peking University and co-authors in 2020.
The objectives of this study are to (1) deepen the understanding of the environmental drivers of African fire, (2) improve the capacity to accurately predict seasonal fire activity in Africa. These contribute to better fire management in these vulnerable ecoregions.
1. Background
The Africa subtropical savannah ecosystems emit ~50% of the global fire-related carbon emissions, and occupy ~70% of the global burned area. The interannual variability in atmospheric CO2 concentration is mediated by African fires. African wildfires also affect human health by emitting atmospheric pollutants.
Previous studies indicate that variations in the extent of burned area in Africa are attributable to vegetation composition and distribution, and air & soil controls on fuel drying. Human activities like cropland expansion and population growth also cause decline in African burned area since 1998. However, these factors cannot predict wildfire activities at the seasonal timescale. Previous seasonal climate forecasts resulted in insignificant correlations between the observed and predicted burned areas across most of Africa.
Since ocean sea-surface temperature (SST) and the land surface have longer memory than the atmosphere, the seasonal predictability of African wildfire may be improved if we focus on SST and terrestrial controls, instead of air controls. But past studies mainly focused on ENSO, ignoring the tropical Atlantic SSTs or the interactions between different modes of SST variability.
2. Method
In order to disentangle the vegetation impacts on fire from the fire impacts on vegetation, and the oceanic and land surfaces influences, the Stepwise Generalied Equilibrium Feedback Assessment (SGEFA) method was used. The SGEFA can disentangle the individual linear influences, but machine learning is necessary for building seasonal forecasting models.
The final results show that SST, leaf area index (LAI), and soil moisture are the most important predictors of African fire, and the machine learning model based on these predictors can effectively predict African fire activity 1 month ahead.
Key findings
I. SGEFA results
(1) Northern Africa's fire is sensitive to tropical Atlantic Ocean SST during the dry season (Nov-Mar), North Atlantic Ocean SST during the boreal winter (fire-active season), and tropical Indian Ocean SST during the wet season (Apr-Sep).
(2) Southern Africa's fire is sensitive to South Atlantic Ocean SSTs during the fire-active season (May-Nov/boreal summer).
(3) Soil moisture exerts important controls on the northern Africa fire during the wet season (Apr-Jan), and slightly smaller control on the southern Africa fire than the South Atlantic SSTs during the fire-active season (May-Nov).
(4) LAI is not an important control in Northern Africa. In Southern Africa, the impact of LAI was akin to that of the North Atlantic, and smaller than soil moisture or the South Atlantic.
(5) If one focuses on specific oceanic modes (ENSO, Atlantic Nino), the response to both of these modes are the highest during the fire-active season (boreal winter in Northern Africa, and boreal summer in Southern Africa). The fire-incuding SST anomalies are generally conducitve to warm and/or dry conditions.
(6) Soil moisture generally supresses fire through low-level cooling and elevated precipitation. LAI enhances fire in Northern Africa and the grasslands in Southern Africa, but supresses fire through surface cooling and reduced wind speed in most of Southern Africa.
II. Seasonal predictability
Fig. 3 shows the predictability of African wildfires by leading time. In both Northern and Southern Africa, the predictability at 1-month lead is significantly enhanced if one uses oceanic + terrestrial predictors, instead of atmospheric + socioeconomic predictors. Using all predictors resulted in only very slight improvements compared to using oceanic + terrestrial predictors.
Also, using season-specific predicting models resulted in much better performances than annual models.
3. Implications of the work
The work will aid the development of fire models, which currently can capture the spatial distribution of observed global fire, but not the seasonal or interannual variations.
The work also has some limitation caused by observational data availability (e.g. lightening, agricultural practice).
The results are good, outperforming past prediction models.
4. Data sources
The data sources are in their Supplemental Table 1. The fire carbon emissions and burned area fractions are from the Global Fire Emissions Database (GFED), which covers 1997-2016 at 0.25 degrees resolution. This is an observation-based analysis.
Landsat data analysis https://emapr.github.io/LT-GEE/introduction.html
#factual_information
#land_use_and_land_cover_change
In 2021, Karina Winkler from Wageningen University & Research (WUR) and co-authors presented maps of global land use changes between 1960 and 2019.
Fig. 1 shows the total net and gross changes for all the land use categories.
Fig. 2 shows the changes in three individual land use types: forests, crops, and pasture/rangeland.
#terrestrial_water_storage
#GRACE
In 2019, Bin He from Beijing Normal University and co-authors analyzed the response time of terrestrial water storage (TWS) to precipitation (P) over 168 global river basins.
The study period was 2003-2014. The trend metric was Mann-Kendall. The trends were all calculated at the basin level for the 168 global basins.
Key findings
(1) In low- and mid-latitude basins, TWS is correlated with P with a shorter lag (1-2 months) than in high-latitude basins (6-9 months).
(2) Three of the individual components of TWS - surface, ground, soil - have the same correlation lags pattern with P as TWS. The other two components - canopy and snow - have different correlation lags, i.e. 0 months and 3-8 months.
(3) Groundwater and soil moisture contributions are generally the largest. In high-latitude basins, snow contribution is greater, which can explain the longer lag in high latitude basins. Interestingly, snow contribution is also large in northern India. Soil moisture contribution is not negligible in very wet regions (e.g. Mackenzie in Canada, Amazon region, Yangtze, Yellow River, Yenisei & Lena in Russia) and a few arid basins (Niger and Nile).
https://journals.ametsoc.org/view/journals/hydr/20/9/jhm-d-18-0253_1.xml
#terrestrial_water_storage
#GRACE
#global_hydrological_models
#land_surface_models
2018 study designed by Bridget R. Scanlon from University of Texas, Austin and Zizhan Zhang from Wuhan University, China.
This is a basin-level analysis for the globe of TWS trends during 2002-2014.
The concept of terrestrial/land terrestrial water storage (TWS), which does not include glaciers.
TWS = SnowWS + CanopyWS + SoilWS + SurfaceWS + GroundWS (WS = water storage)
Caveats of different data sources:
(1) Land surface models have greater emphasis on fluxes, whereas global hydrological models have more emphasis on water storage and human water use.
(2) Many land surface models only simulate the snow and soil moisture storage components, while most global hydrological models simulate all except glaciers.
(3) GRACE data estimated by mascons can have leakage, i.e., the decreasing TWS trends in melting glaciers can cause decreasing trends in the adjacent pixels
The assessments were conducted at large-basin levels (<100 across the globe). The units of basin-level TWS were km^3/year, but larger trends in km^3/year generally corresponded to larger trends in mm/year.
Major findings:
(1) All the models generally under-estimate the magnitudes of trends, which may be increasing or decreasing trends. The agreements based on regression analysis were also pretty poor.
(2) The more (less) irrigated a basin is, the more negative (positive) is the TWS trends in GRACE. The non-irrigated basins are mainly in the humid regions. Nonetheless, human intervention is not the greatest driver of global trends in water storage, because the land surface models, which do not simulate human interventions, produced more negative global trends than
(3) Larger (smaller) basins have smaller (larger) GRACE measurement and leakage uncertainties. Examples of leakage existed in the Yukon Basin (from the Alaskan glaciers), the Ganges (from the Asian High Mountain Glaciers), and the Salado basins in South America (glacker & Chile 2010 Maule earthquake).
(4) Land TWS affects global mean sea level rise by taking water from or putting water into the sea. GRACE suggests the global land is increasing in TWS, but the land surface and hydrological models all suggests negative global land average TWS trends.
(5) Sources of discrepancy between GRACE and the model outputs are discussed and analyzed in detail.
#data_products
#terrestrial_water_storage
Name: GRACE-REC
Spatial coverage: global
Spatial resolution: 0.5 degrees
Temporal coverage: MSWEP-based reconstruction 1979-2016, ERA5-based 1979-present, GSWP3-based 1901-2014
Temporal resolution: Monthly (100 ensemble members for each combination of the three meteorological datasets and two GRACE satellite mascons), daily (ensemble mean only)
Gap-free: Yes
Year of publication: 2019
Algorithm: (1) linear water storage model, driven by temperature and precipitation, calibrated against de-seasonalized, de-trendedd GRACE satellite data (2002-2017), (2) spatial autoregressive model was used to generate an ensemble of spatially autocorrelated residuals, to facilitate uncertainty propagation from grid-level to regional or global averages. This method to propagate uncertainty is noteworthy.
Pros: outperforms hydrological and land surface models when evaluated against de-seasonalized, de-trended GRACE terrestrial water data, sea-level budget, basin-scale water balance, and streamflow.
Cons: The trends in GRACE-REC is purely driven by precipitation changes, therefore missing a ton of relevant factors (evapotranspiration change, dams, human water withdrawal, ice melt...). The seasonality is constant and not particularly reliable, either.
#remote_sensing
#vegetation_index
#ndvi
#evi
#review
2022. Dalei Hao from Pacific Northwest National Lab, and coauthors.
Satellite-measured spectral reflectance is co-determined by leaf reflectance, background soil reflectance, canopy structure, and sensor geometry.
The below are broad-band vegetation indices for measuring vegetation structure:
* NDVI is simple and requires only the NIR and Red band, but sensitive to the soil background variations and insensitive to dense vegetation.
* EVI requires the blue band as well as the NIR and Red band, but minimizes both soil and atmospheric effects and is more sensitive to dense vegetation than NDVI. However, EVI is sensitive to the BRDF (bidirectional reflectance distribution function, related to the sun-target-sensor geometry) effect.
** The blue band helps remove atmospheric effects. With the advancement in atmospheric correction algorithms, EVI2 is developed and requires only the red and NIR bands.
* PPI (Plant Phenology Index) uses red and NIR bands, and is nearly linearly correlated with green LAI. But PPI is moderately sensitive to soil brightness.
** Still, the original paper (https://www.sciencedirect.com/science/article/abs/pii/S0034425714002557?via%3Dihub) seems to suggest it is more useful for GPP than LAI.
* NIRv is not sensitive to soil background variations, but sensitive to the BRDF effect.
Narrow-band vegetation indices can measure (1) biochemical properties, (2) physiological properties (e.g. photosynthetic light-use efficiency, or environmental stress). These indices make use of measurements made at specific wavelengths. The former includes indices for chlorophyll content, carotenoid content, normalized difference water index (vegetation water condition), land surface water index (vegetation water condition), normalized difference lignin index, etc. The latter include the PRI (photochemical reflectance index) and CCI (chlorophyll/carotenoid index), etc.
Artefacts in vegetation indices arise from (1) differences between sensors, (2) satellite product versions, (3) atmospheric and directional corrections, (3) compositing algorithms, (4) application of different levels of quality assurance and quality control flags. Some notable known facts are:
(1) AVHRR and MODIS NDVI time series can show trends in opposite directions.
(2) NDVI values are substantially different between AVHRR, MODIS, and VIIRS. The level of the differences varies across land cover types
(3) Products from AVHRR and MODIS have instrumental drifts. AVHRR products, especially, are from a series of satellites. The NDVI values from AVHRR aboard NOAA-11 were higher than the values measured by other AVHRR sensors. The trends in the NIRv values are not sufficiently trustworthy to measure the global CO2 fertilization effect. The MODIS Collection 5 and Collection 6 products have inconsistent greening trends. The MODIS Terra satellite suffers from sensor degradation, more so in the blue band than in the longer-wavelength bands, resulting in artificial negative drifts in NDVI and EVI values in Collection 5. The Collection 6 products remedied the sensor degradation problem.
(4) MODIS-based NDVI exhibited increasing trend during 2001-2016, but GIMMS-based NDVI showed the opposite trend, especially after 2012. This highlights uncertainty in global greening.
(5) VIP3 and LTDR4 NDVI, and the GIMMS-3g NDVI in the more humid areas, have orbital drift effects.
(6) SPOT-VGT NDVI time series have an abrupt jump at the shift from VGT-1 to VGT-2.
(7) As a rule of thumb, ratio-based indices (e.g. NDVI) have smaller biases than straight indices (e.g. NIRv = NDVI*NIR), because of the cancellation between denominator and numerator. Therefore, it is especially important to check for instrumental drifts in the latter.
(8) BRDF effect does not influence the long-term trends or interannual variations, since the sun returns its orbit every year. But it affects phenology. Single-season analysis should be rigorously corrected for BRDF [refs 64, 122, 123, 124]. Landsat's viewing angle is relatively restricted (+/-7.5 degrees from nadir), but other sensors like AVHRR and MODIS must be corrected for this effect.
(9) GIMMS-3g datasets used relatively old compositing algorithm (maximum value compositing). The MOD13A1 and MYD13A1 used a newer one. The MCD43A4 C6 product used a even newer one and removed the view-angle effects; the angle was always set at the local solar noon zenith angle. GIMMS-3g also had only limited atmospheric correction.
Soil artefacts are worse in sparsely vegetated regions (<50% cover, or LAI ~= 1). Snow and ice lead to discrepancies in temperate regions. Topograph also create artefacts, but decreases with spatial averaging.
Finally, the pixels can be mismatched with the actual point on the ground. MODIS has accurate geolocation, but the offset can still be as much as 0.5 pixel between scenes. This is a problem for single-site study. Hyper-resolution (<3m) remote sensing products have their own problems because there can be shadows.
There is need to more extensively measure reflected radiation from vegetation to verify the remote sensing products. There is also much uncertainty in greening.
Let me start this chain and see how it goes.
`xarray` decode_times = True recognizes any variable with a time unit (e.g. "days") as a datetime variable. Clearly, this can cause problem when there are variables that are intended to measure length of time instead of date.
English/中文
Environmental scientist looking at global ecohydrological change using data analysis and modeling tools.
Joined May 2022
