Search

Wildfire occurrence and spread are affected by atmospheric and land-cover conditions, and therefore meteorological and land-cover parameters can be used in area burned prediction. We apply three forecast methods, a generalized linear model, regression trees, and neural networks (Levenberg–Marquardt backpropagation) to produce monthly wildfire predictions 1 year in advance. The models are trained using the Global Fire Emissions Database version 4 with small fires (GFEDv4s). Continuous 1-year monthly fire predictions from 2011 to 2015 are evaluated with GFEDs data for 10 major fire regions around the globe. The predictions by the neural network method are superior. The 1-year moving predictions have good prediction skills over these regions, especially over the tropics and the southern hemisphere. The temporal refined index of agreement (IOA) between predictions and GFEDv4s regional burned areas are 0.82, 0.82, 0.8, 0.75, and 0.56 for northern and southern Africa, South America, equatorial Asia and Australia, respectively. The spatial refined IOA for 5-year averaged monthly burned area range from 0.69 in low-fire months to 0.86 in high-fire months over South America, 0.3–0.93 over northern Africa, 0.69–0.93 over southern Africa, 0.47–0.85 over equatorial Asia, and 0.53–0.8 over Australia. For fire regions in the northern temperate and boreal regions, the temporal and spatial IOA between predictions and GFEDv4s data in fire seasons are 0.7–0.79 and 0.24–0.83, respectively. The predictions in high-fire months are better than low-fire months. This study illustrates the feasibility of global fire activity outlook forecasts using a neural network model and the method can be applied to quickly assess the potential effects of climate change on wildfires.

Land cover mapping for large regions often employs satellite images of medium to coarse spatial resolution, which complicates mapping of discrete classes. Class memberships, which estimate the proportion of each class for every pixel, have been suggested as an alternative. This paper compares different strategies of training data allocation for discrete and continuous land cover mapping using classification and regression tree algorithms. In addition to measures of discrete and continuous map accuracy the correct estimation of the area is another important criteria. A subset of the 30 m national land cover dataset of 2006 (NLCD2006) of the United States was used as reference set to classify NADIR BRDF-adjusted surface reflectance time series of MODIS at 900 m spatial resolution. Results show that sampling of heterogeneous pixels and sample allocation according to the expected area of each class is best for classification trees. Regression trees for continuous land cover mapping should be trained with random allocation, and predictions should be normalized with a linear scaling function to correctly estimate the total area. From the tested algorithms random forest classification yields lower errors than boosted trees of C5.0, and Cubist shows higher accuracies than random forest regression.

Global change affects vegetation cover and processes through multiple pathways. Long time series of surface land surface properties derived from satellite remote sensing give unique abilities to observe these changes, particularly in areas with complex topography and limited research infrastructure. Here, we focus on Nepal, a biodiversity hotspot where vegetation productivity is limited by moisture availability (dominated by a summer monsoon) at lower elevations and by temperature at high elevations. We analyze the normalized difference vegetation index (NDVI) from 1981 to 2015 semimonthly, at an 8 km spatial resolution. We use a random forest (RF) of regression trees to generate a statistical model of the NDVI as a function of elevation, land use, CO2 level, temperature, and precipitation. We find that the NDVI increased over the studied period, particularly at low and middle elevations and during the fall (post-monsoon). We infer from the fitted RF model that the NDVI linear trend is primarily due to CO2 level (or another environmental parameter that is changing quasi-linearly), and not primarily due to temperature or precipitation trends. On the other hand, interannual fluctuation in the NDVI is more correlated with temperature and precipitation. The RF accurately fits the available data and shows promise for estimating trends and testing hypotheses about their causes. [ABSTRACT FROM AUTHOR]

The estimation of ground sampling distance (GSD) from a remote sensing image enables measurement of the size of an object as well as more accurate segmentation in the image. In this paper, we propose a regression tree convolutional neural network (CNN) for estimating the value of GSD from an input image. The proposed regression tree CNN consists of a feature extraction CNN and a binomial tree layer. The proposed network first extracts features from an input image. Based on the extracted features, it predicts the GSD value that is represented by the floating-point number with the exponent and its mantissa. They are computed by coarse scale classification and finer scale regression, respectively, resulting in improved results. Experimental results with a Google Earth aerial image dataset and a mixed dataset consisting of eight remote sensing image public datasets with different GSDs show that the proposed network reduces the GSD prediction error rate by 25% compared to a baseline network that directly estimates the GSD.

This paper presents the methodology and results of two ecological-based net ecosystem production (NEP) regression tree models capable of up scaling measurements made at various flux tower sites throughout the U.S. Great Plains. Separate grassland and cropland NEP regression tree models were trained using various remote sensing data and other biogeophysical data, along with 15 flux towers contributing to the grassland model and 15 flux towers for the cropland model. The models yielded weekly mean daily grassland and cropland NEP maps of the U.S. Great Plains at 250 m resolution for 2000-2008. The grassland and cropland NEP maps were spatially summarized and statistically compared. The results of this study indicate that grassland and cropland ecosystems generally performed as weak net carbon (C) sinks, absorbing more C from the atmosphere than they released from 2000 to 2008. Grasslands demonstrated higher carbon sink potential (139 g C.m-2.year-1) than non-irrigated croplands. A closer look into the weekly time series reveals the C fluctuation through time and space for each land cover type. [ABSTRACT FROM AUTHOR]

Thermal infrared (TIR) imagery is normally acquired at coarser pixel resolution than that of shortwave sensors on the same satellite platform and often the TIR resolution is not suitable for monitoring crop conditions of individual fields or the impacts of land cover changes that are at significantly finer spatial scales. Consequently, thermal sharpening techniques have been developed to sharpen TIR imagery to shortwave band pixel resolutions, which are often fine enough for field-scale applications. A classic thermal sharpening technique, TsHARP, uses a relationship between land surface temperature (LST) and Normalized Difference Vegetation Index (NDVI) developed empirically at the TIR pixel resolution and applied at the NDVI pixel resolution. However, recent studies show that unique relationships between temperature and NDVI may only exist for a limited class of landscapes, with mostly green vegetation and homogeneous air and soil conditions. To extend application of thermal sharpening to more complex conditions, a new data mining sharpener (DMS) technique is developed. The DMS approach builds regression trees between TIR band brightness temperatures and shortwave spectral reflectances based on intrinsic sample characteristics. A comparison of sharpening techniques applied over a rainfed agricultural area in central Iowa, an irrigated agricultural region in the Texas High Plains, and a heterogeneous naturally vegetated landscape in Alaska indicates that the DMS outperformed TsHARP in all cases. The artificial box-like patterns in LST generated by the TsHARP approach are greatly reduced using the DMS scheme, especially for areas containing irrigated crops, water bodies, thin clouds or terrain. While the DMS technique can provide fine resolution TIR imagery, there are limits to the sharpening ratios that can be reasonably implemented. Consequently, sharpening techniques cannot replace actual thermal band imagery at fine resolutions or missions that provide high quality thermal band imagery at high temporal and spatial resolution critical for many agricultural, land use and water resource management applications.

Global change affects vegetation cover and processes through multiple pathways. Long time series of surface land surface properties derived from satellite remote sensing give unique abilities to observe these changes, particularly in areas with complex topography and limited research infrastructure. Here, we focus on Nepal, a biodiversity hotspot where vegetation productivity is limited by moisture availability (dominated by a summer monsoon) at lower elevations and by temperature at high elevations. We analyze the normalized difference vegetation index (NDVI) from 1981 to 2015 semimonthly, at an 8 km spatial resolution. We use a random forest (RF) of regression trees to generate a statistical model of the NDVI as a function of elevation, land use, CO 2 level, temperature, and precipitation. We find that the NDVI increased over the studied period, particularly at low and middle elevations and during the fall (post-monsoon). We infer from the fitted RF model that the NDVI linear trend is primarily due to CO 2 level (or another environmental parameter that is changing quasi-linearly), and not primarily due to temperature or precipitation trends. On the other hand, interannual fluctuation in the NDVI is more correlated with temperature and precipitation. The RF accurately fits the available data and shows promise for estimating trends and testing hypotheses about their causes.

This paper presents the methodology and results of two ecological-based net ecosystem production (NEP) regression tree models capable of up scaling measurements made at various flux tower sites throughout the U.S. Great Plains. Separate grassland and cropland NEP regression tree models were trained using various remote sensing data and other biogeophysical data, along with 15 flux towers contributing to the grassland model and 15 flux towers for the cropland model. The models yielded weekly mean daily grassland and cropland NEP maps of the U.S. Great Plains at 250 m resolution for 2000–2008. The grassland and cropland NEP maps were spatially summarized and statistically compared. The results of this study indicate that grassland and cropland ecosystems generally performed as weak net carbon (C) sinks, absorbing more C from the atmosphere than they released from 2000 to 2008. Grasslands demonstrated higher carbon sink potential (139 g C·m−2·year−1) than non-irrigated croplands. A closer look into the weekly time series reveals the C fluctuation through time and space for each land cover type.

Land cover mapping for large regions often employs satellite images of medium to coarse spatial resolution, which complicates mapping of discrete classes. Class memberships, which estimate the proportion of each class for every pixel, have been suggested as an alternative. This paper compares different strategies of training data allocation for discrete and continuous land cover mapping using classification and regression tree algorithms. In addition to measures of discrete and continuous map accuracy the correct estimation of the area is another important criteria. A subset of the 30 m national land cover dataset of 2006 (NLCD2006) of the United States was used as reference set to classify NADIR BRDF-adjusted surface reflectance time series of MODIS at 900 m spatial resolution. Results show that sampling of heterogeneous pixels and sample allocation according to the expected area of each class is best for classification trees. Regression trees for continuous land cover mapping should be trained with random allocation, and predictions should be normalized with a linear scaling function to correctly estimate the total area. From the tested algorithms random forest classification yields lower errors than boosted trees of C5.0, and Cubist shows higher accuracies than random forest regression. [ABSTRACT FROM AUTHOR]

Wildfire occurrence and spread are affected by atmospheric and land-cover conditions, and therefore meteorological and land-cover parameters can be used in area burned prediction. We apply three forecast methods, a generalized linear model, regression trees, and neural networks (Levenberg–Marquardt backpropagation) to produce monthly wildfire predictions 1 year in advance. The models are trained using the Global Fire Emissions Database version 4 with small fires (GFEDv4s). Continuous 1-year monthly fire predictions from 2011 to 2015 are evaluated with GFEDs data for 10 major fire regions around the globe. The predictions by the neural network method are superior. The 1-year moving predictions have good prediction skills over these regions, especially over the tropics and the southern hemisphere. The temporal refined index of agreement (IOA) between predictions and GFEDv4s regional burned areas are 0.82, 0.82, 0.8, 0.75, and 0.56 for northern and southern Africa, South America, equatorial Asia and Australia, respectively. The spatial refined IOA for 5-year averaged monthly burned area range from 0.69 in low-fire months to 0.86 in high-fire months over South America, 0.3–0.93 over northern Africa, 0.69–0.93 over southern Africa, 0.47–0.85 over equatorial Asia, and 0.53–0.8 over Australia. For fire regions in the northern temperate and boreal regions, the temporal and spatial IOA between predictions and GFEDv4s data in fire seasons are 0.7–0.79 and 0.24–0.83, respectively. The predictions in high-fire months are better than low-fire months. This study illustrates the feasibility of global fire activity outlook forecasts using a neural network model and the method can be applied to quickly assess the potential effects of climate change on wildfires. [ABSTRACT FROM AUTHOR]

The estimation of ground sampling distance (GSD) from a remote sensing image enables measurement of the size of an object as well as more accurate segmentation in the image. In this paper, we propose a regression tree convolutional neural network (CNN) for estimating the value of GSD from an input image. The proposed regression tree CNN consists of a feature extraction CNN and a binomial tree layer. The proposed network first extracts features from an input image. Based on the extracted features, it predicts the GSD value that is represented by the floating-point number with the exponent and its mantissa. They are computed by coarse scale classification and finer scale regression, respectively, resulting in improved results. Experimental results with a Google Earth aerial image dataset and a mixed dataset consisting of eight remote sensing image public datasets with different GSDs show that the proposed network reduces the GSD prediction error rate by 25% compared to a baseline network that directly estimates the GSD. [ABSTRACT FROM AUTHOR]