Satellite‐based precipitation estimation using watershed segmentation and growing hierarchical self‐organizing map

This paper outlines the development of a multi‐satellite precipitation estimation methodology that draws on techniques from machine learning and morphology to produce high‐resolution, short‐duration rainfall estimates in an automated fashion. First, cloud systems are identified from geostationary infrared imagery using morphology based watershed segmentation algorithm. Second, a novel pattern recognition technique, growing hierarchical self‐organizing map (GHSOM), is used to classify clouds into a number of clusters with hierarchical architecture. Finally, each cloud cluster is associated with co‐registered passive microwave rainfall observations through a cumulative histogram matching approach. The network was initially trained using remotely sensed geostationary infrared satellite imagery and hourly ground‐radar data in lieu of a dense constellation of polar‐orbiting spacecraft such as the proposed global precipitation measurement (GPM) mission. Ground‐radar and gauge rainfall measurements were used to evaluate this technique for both warm (June 2004) and cold seasons (December 2004–February 2005) at various temporal (daily and monthly) and spatial (0.04° and 0.25°) scales. Significant improvements of estimation accuracy are found classifying the clouds into hierarchical sub‐layers rather than a single layer. Furthermore, 2‐year (2003–2004) satellite rainfall estimates generated by the current algorithm were compared with gauge‐corrected Stage IV radar rainfall at various time scales over continental United States. This study demonstrates the usefulness of the watershed segmentation and the GHSOM in satellite‐based rainfall estimations.     

A maximum entropy approach to satellite quantitative precipitation estimation (QPE)

This paper presents a new algorithm to generate quantitative precipitation estimates from infrared (IR) satellite imagery using passive microwave (PMW) data from Special Sensor Microwave/Imager sensor (SSM/I) satellites as ancillary information. To generate the estimates, we model the probabilistic distribution function (PDF) of the rainfall rates through the maximum entropy method (MEM), applying a cumulative histogram matching (HM) technique to the IR brightness temperatures. This results in a straightforward algorithm that can be formulated as an algebraic expression, providing a simple method to derive rainfall estimates using only IR data. The main application of the method is the direct estimation of rainfall rates and accumulated rainfall from geostationary satellites, providing appropriate temporal and spatial resolutions (up to 15 min/4 km when the Meteosat Second Generation satellite becomes available). The proposed method can be easily applied at GOES or current Meteosat satellite reception stations to generate instantaneous rainfall rates estimates with little computational cost. Here we provide examples of applications using the Global Infrared Database and Meteosat images. Our results have been compared with GOES Precipitation Index (GPI) and validated against Global Precipitation Climatology Centre (GPCC)-land rain gauge measurements, at 5°, monthly accumulations. We have obtained correlations of 0.88 for the algorithm, while the GPI yields correlations of 0.85. Preliminary comparisons with other algorithms over Australia also show how the performances of the algorithm are similar to those of more complex models. Finally, we propose some improvements and fine-tuning procedures that can be applied to the algorithm.     

Application of three satellite techniques in support of precipitation forecasts of a NWP model

In this study, the efficiency of an integrated operational system, based on three satellite infrared techniques, to support nowcasting and very short range forecasting of high precipitation rates provided by a numerical weather prediction (NWP) model is examined. Three algorithms, one for the detection and tracking of convective cloud cells and another two for rainfall estimation, are applied on satellite sensor data in order to qualitatively and quantitatively verify the precipitation forecasts provided by a NWP model. The application of the detection and tracking algorithm aims at verifying qualitatively, in real time, the precipitation forecasts by monitoring the detected convective cloud cells, at the same time intervals that the model forecasts are given. The application of the rainfall estimation techniques on satellite sensor data is needed for both quantitative and qualitative cross‐comparisons with the model outputs. The developed tool is applied in a case of intense precipitation over Greece. The results of the application are promising and show the potential for the implementation of the integrated system as a support tool for nowcasting and very short range forecasting by performing real‐time validation of NWP precipitation forecasts.     

Artificial neural network techniques for estimating heavy convective rainfall and recognizing cloud mergers from satellite data

This research presents an artificial neural network (ANN) technique for heavy convective rainfall estimation and cloud merger recognition from satellite data. An Artificial Neural network expert system for Satellite-derived Estimation of Rainfall (ANSER) has been developed in the NOAA/NESDIS Satellite Applications Laboratory. Using artificial neural network group techniques, the following can be achieved: automatic recognition of cloud mergers, computation of rainfall amounts that will be ten times faster, and average errors of the rainfall estimates for the total precipitation event that will be reduced to less that 10 per cent.     

Mathematical aspects of the meteorological interpretation of satellite hyperspectral infrared measurements part I: statement of the inverse problem for estimation of the cloud absorption vertical profile

Satellite high spectral resolution infrared measurements provide information for cloud vertical characterization when optically semi-transparent clouds at high altitude cover cloud layers at lower altitudes. It is an important issue because such atmospheric conditions are common and clouds are characterized by large-scale vertical development. An approximation radiative transfer model for a cloudy atmosphere is introduced. Cloud particle absorption of infrared (IR) radiation depends on the spectral frequency. So, the effective cloud parameters such as amount (absorption) and height, derived from IR spectral measurements, will be spectral functions as well. The degree of uncertainty in the determination of effective cloud parameters cannot be eliminated by increasing the number of spectral measurements. A cloud model should have extra degrees of freedom to address the spectral and spatial (vertical) variability of cloud absorption. A semi-discrete multilevel cloud model is used to describe the perturbation of the outgoing IR thermal radiation caused by cloudiness in the field of view of a satellite instrument. The model delineates cloudiness in a number of layers at fixed heights. Each layer (level) is characterized by the effective cloud absorption. An inverse problem of cloud absorption vertical profile (CAVP) estimation is described. The estimate of an effective cloud absorption profile is considered as predictor for identification of cloud presence at specific atmospheric layers. The problem is numerically examined for real satellite IR spectral measurements and solution estimates are compared with lidar measurements. Results show that the resulting estimate of CAVPs provides a realistic characterization of cloud top and cloud vertical scale.     

Augmenting satellite precipitation estimation with lightning information

We have used lightning information to augment the precipitation estimation from remotely sensed imagery using an artificial neural network cloud classification system (PERSIANN-CCS). Co-located lightning data are used to segregate cloud patches, segmented from Geostationary Operational Environmental Satellite (GOES)-12 infrared (IR) data, into either electrified patches (ECPs) or nonelectrified patches (NECPs). A set of features is extracted separately for the ECPs and NECPs. Features for the ECPs include a new feature corresponding to the number of flashes that occur within a 15 minute window around the time of the nominal scan of the satellite IR images of the cloud patches. The cloud patches are classified and clustered using a self-organizing maps (SOM) neural network. Then, brightness temperature and rain rate (T–R) relationships are derived for different clusters. Rain rates are estimated for the cloud patches based on their representative (T–R) relationship. The equitable threat scores (ETS) of the daily and hourly precipitation estimates at a range of rain rate thresholds show that incorporating lightning information can improve categorical precipitation estimation in the winter and fall seasons. In the winter, the ETS improvement is almost 15% for the daily and 12% for the hourly rainfall estimates (at thresholds below 15 mm hour^?1). During the same period, there is also a drop in the false alarm ratio (FAR) and a corresponding increase in the probability of detection (POD) at most threshold levels. During the summer and spring seasons, no categorical significant improvements have been noted, except for the BIAS scores for the hourly rainfall estimates at higher thresholds (above 5 mm hour^?1) in the summer months. A quantitative evaluation in terms of the root mean squared error (RMSE) and correlation coefficient (CORR) shows that the incorporation of lightning data does improve rainfall estimation over all seasons with the most improvement (around 11–13% CORR improvement) occurring during the winter. We speculate that during the winter, more of the ice processes are packed into a thinner stratiform layer with lower cloud tops and freezing levels. Hence, more of the ice contributes to precipitation on the ground. We also expect that information from lightning, related to the ice microphysics processes, provides surrogate information about the rain rate.     

Modelling rainfall intensity from NOAA AVHRR data for operational flood forecasting in Malaysia

Many empirical studies in numerical weather prediction have been carried out that establish the relationship between top‐of‐the‐cloud brightness temperature and rainfall particularly in tropical and equatorial regions of the world. Malaysia is a tropical country that lies along the path of the north‐east and south‐west monsoon rainfall, which sometimes causes extensive flood disasters. Observations have generally shown that heavy cumulonimbus cloud formation and thunderstorms precede the usual heavy monsoon rains that cause flood disasters in the region. In this study, a model has been developed to process National Oceanic & Atmospheric Administration Advanced Very High Resolution Radiometer (AVHRR) satellite data for rainfall intensity in an attempt to improve quantitative precipitation forecasting (QPF) as input to operational hydro‐meteorological flood early warning. The thermal bands in the multispectral AVHRR data were processed for brightness temperature. Data were further processed to determine cloud height and classification performed to delineate clouds in three broad classes of low, middle, and high. A rainfall intensity of 3–12 mm h^?1 was assigned to the 1‐D cloud model to determine the maximum rain rate as a function of maximum cloud height and minimum cloud model temperature at a threshold level of 235 K. The result of establishing the rainfall intensity based on top of the cloud brightness temperature was very promising. It also showed a good areal coverage that delineated areas likely to receive intense rainfall on a regional scale. With a spatial resolution of 1.1 km, data are course but provide a good coverage for an average river catchment/basin. This raises the opportunity of simulating rainfall runoff for the river catchment through the coupling of a suitable hydro‐dynamic model and GIS to provide early warning prior to the actual rainfall event.     

大范围云导风的直接可视化方法

针对传统大范围云导风生成及显示过程中存在较大计算冗余的问题,提出一种快速风场可视化算法,可使用卫星云图序列直接生成便于人眼观察的大范围云导风态势展示视频。该算法首先使用稀疏像素块匹配分析得到相邻帧之间的运动信息,再使用匹配结果形成图像形变网格,最后使用该网格对噪声纹理图像进行形变和颜色混合叠加操作,从而输出风场运动轨迹的动态视频。由于无须复杂的光流计算和额外流场可视化后处理等步骤,该算法在程序实现时更加便捷,对计算硬件依赖性较小。实验结果显示,这种直接可视化算法可以生成易于辨识的、时间一致性强的云导风动态可视化视频,能应对红外、可见光和水汽等各种波段的卫星图像数据,对云层无规则消散、生成等干扰具有较高鲁棒性。     

Motion estimation on meteorological infrared data using a total brightness invariance hypothesis

This work investigates a novel approach for cloud motion wind (CMW) estimation of Meteosat infrared images. It is motivated by the fact that variational techniques, such as those employed for computing the optical flow, are successfully applied to many computer vision applications but fail in this particular applicative context, mainly because optical flow techniques are adapted to rigid objects on visible data. The objective of this work is not to propose a full operational process for CMW estimation, but rather to improve optical flow techniques by applying constraints adapted to the specificity of meteorological infrared imagery.     

Objectively determined 10-day African rainfall estimates created for famine early warning systems

A method of estimation of accumulated precipitation which incorporates numerical model analyses, satellite and surface data has been developed for the African continent. An estimate for accumulated convective cloud precipitation is computed from cold cloud top temperature duration with a bias removal made from the use of rain-gauge data. Orographic precipitation from relatively warm cloud sources is estimated using a combination of surface and satellite data, orography, and numerical model analyses of relative humidity and wind. The results of a comparison of these precipitation estimates with independent rainfall data show this method produces skilful analyses of estimated accumulated precipitation for the Sahel region of Africa.     

Resolution dependent errors in remote sensing of cultivated areas

Remote sensing has become a common and effective method for estimating the areal coverage of land cover classes. One class of particular interest is agriculture as area estimates of cultivated lands are important for purposes such as estimating yields or irrigation needs. The synoptic coverage of satellite imagery and the relative ease of automated analysis have led to widespread mapping of agriculture using remote sensing. The accuracy of area estimates derived from these maps is known to be related to the accuracy of the maps. However, even in the situation where the map is very accurate, errors in area estimates may occur. These errors result from the behavior of the distribution of subpixel proportions of cultivated areas, and how that behavior changes as a result of sensor spatial resolution and class definitions. The sensitivity of estimates of cultivated areas to sensor spatial resolution and to the choice of threshold used to define cultivated land is explored in six agriculturally distinct locations around the world. Using a beta model for the distribution of subpixel proportions that is parameterized using variograms, it is possible to model the distribution of subpixel proportions for any spatial resolution. When the spatial resolution is small with respect to the spatial structure of the landscape (as measured by the variogram range) use of any class definition threshold produces an estimate very close to the true area coverage. On the other hand, as the resolution becomes coarse in relation to the variogram range, the subpixel proportions are no longer concentrated at the extremes of the distribution and the difference between the estimated and the true area has greater sensitivity to the selected threshold used to define classes. Thus, for the cases examined here, both the resolution and the class definition threshold have a strong influence on area estimates. The spatial resolutions where errors can be large depend on landscape spatial structure, which can be quantified using variograms. The net effect is that for the same spatial resolution, some places will exhibit much larger errors in area estimates than others. For the site in the Anhui province of China, where agricultural fields are very small (0.07 ha on the average), area estimates are highly sensitive to class definition thresholds even at the relatively fine resolution of 45 m. Conversely, in California (USA) spatial resolutions as coarse as 500 m can be used to reliably estimate cultivated areas. Results also suggest that the proportion of the total area that is cultivated significantly influences the accuracy of area estimates. When the area proportion is low, the class definition threshold must also be low to achieve an accurate area estimate. Conversely, in areas dominated by agriculture, a very stringent class definition of cultivated lands is required for accurate area estimates. While explored in the context of estimation of cultivated areas, the findings presented here are generic to the problem of area estimation using remote sensing.     

A bispectral method for the automatic determination of parameters for use in imaging satellite cloud retrievals

Two major problems in deriving cloud amounts and physical properties from satellite imagery are the selection of suitable surface-type discriminators, which may vary as a function of time and place, and the extraction of cloudiness on the subpixel scale. We present a method of retrieving suitable parameters for such discriminations on the local scale, based on the information contained in a bispectral (visible and 11 μm infrared) histogram. The application of these parameters to the retrieval of cloud information on both the pixel and subpixel scales is demonstrated.     

Reduction of the effect of clouds on satellite thermal imagery

A simple method for reducing the effect of cloud on satellite thermal imagery, while preserving full temperature and spatial resolution, is presented. The method makes use of the correlation between apparent thermal and visible radiances in an area contaminated by thin or patchy (subpixel) cloud. The method was developed for ocean images where the cloud is present over an area of low uniform albedo in the visible spectral region, but advantages are also demonstrated in processing Arctic images containing some sea ice cover.     

Validation of high‐resolution satellite rainfall products over complex terrain

High‐resolution satellite rainfall products, at daily accumulation and 0.25° spatial resolution, are evaluated using station networks located over two different parts of Africa. The first site is located over Ethiopia with a very complex terrain. The second site, located over Zimbabwe, has a less rugged topography. The evaluated satellite rainfall products are the NOAA‐CPC African rainfall estimation algorithm (RFE), TRMM‐3B42, the CPC morphing technique (CMORPH), PERSIANN, and the Naval Research Laboratory's blended product. These products perform reasonably well over both regions in detecting the occurrence of rainfall. However, performances are poor in estimating the amount of rainfall in each pixel. The correlation coefficients are low and random errors high. The performance was better over Zimbabwe as compared with Ethiopia. Comparing the different products, CMORPH and TRMM‐3B42 showed a better performance over Ethiopia, while RFE, CMORPH, and TRMM‐3B42 preformed relatively better over Zimbabwe.     

Validating three infrared‐based rainfall retrieval algorithms for intense convective activity over Greece

Three different rainfall estimation techniques based on infrared data from Meteosat, the Arkin technique (ARKT), the Negri‐Adler‐Wetzel technique (NAWT) and the Convective‐Stratiform technique (CST), were applied to four convective systems over Greece to test their performance in case of flood episodes of varying intensities and to examine the possibility of their optimization for this particular geographical region. The comparison between satellite‐derived estimates and the corresponding 12‐hourly accumulated precipitation data from ground stations proved that all three techniques have the common characteristic of overestimating the precipitation in Greece. In general, the CST method was found to best represent the rainfall pattern observed in the rain gauge network, when the comparison is made on a single station basis. On the contrary, the overall performances of ARKT and NAWT methods were not satisfactory. Moreover, a sensitivity analysis of the ARKT and NAWT method to their different parameters indicated that the main parameter for improving their overall performance is the threshold temperature whereas the adjustment of the assigned rain rates has little influence on results. In general, the optimized NAWT technique may provide a very good representation of convective rainfall in the particular geoclimatic conditions of Greece, especially if the estimated values are averaged over suitable space‐time intervals.     

Mesoscale grid rainfall estimation from AVHRR and GMS data

Areal rainfall averages derived from rain-gauge observations suffer from limitations not only due to sampling but also because gauges are usually distributed with a spatial bias towards populated areas and against areas with high elevation and slope. For a large river basin, however, heavy rainfall in the mountain upstream can result in severe flooding downstream. In this study, cloud-indexing and cloud model-based techniques were applied to Advanced Very High Resolution Radiometer (AVHRR) and Geostationary Meteorological Satellite (GMS) imager data based on the cloud-top brightness temperature (T _B) and processed for estimating mesoscale grid rainfall. This study aims to improve and refine rainfall estimation in Malaysian monsoons based on cloud model techniques for operational pre-flood forecasting using readily available near-real-time satellite data such as the National Oceanic and Atmospheric Administration (NOAA)-AVHRR and GMS imager. Rain rates between 3 and 12 mm h^?1 were assigned to cloud pixels of hourly coverage AVHRR or GMS data over the Langat Basin area for the duration of the monsoon rainfall event of 27 September to 8 October 2000 in Malaysia. The observed rainfall and quantitative precipitation forecast (QPF) showed an R ² value of 0.9028, while the observed rainfall run-off (RR; recorded) and its simulated data had an R ² value of 0.9263 and the QPF run-off and its simulated data had an R ² value of 0.815. The rainfall estimate was used to simulate the flood event of the catchment. The estimated rainfall over the catchment showed similar flood area coverage to the observed flood event.     

Development of the Microwave calibrated Infrared Split‐window Technique (MIST) for rainfall estimation

An infrared rainfall estimation technique that includes information from a split window is developed. The split window refers to the difference in the brightness between the far infrared (IR) channels situated at around 10 µm and 12 µm, which has been used to estimate atmospheric water vapour and for rain area detection. The technique, called the Microwave calibrated Infrared Split‐window Technique (MIST), can be considered an extension of the Adjusted GOES Precipitation Index (AGPI). IR rain rates are first estimated from an IR‐rain rate relation derived from matching the monthly histograms of combined microwave rain estimates (3B40RT) produced by the Tropical Rainfall Measuring Mission (TRMM) Multi‐satellite Precipitation Analysis (TMPA) and the infrared data observed from a geostationary satellite. The novelty is the inclusion of the split‐window information to eliminate non‐rainy pixels as a second step. The technique has been applied to Geostationary Meteorological Satellite (GMS) and Geostationary Operational Environmental Satellite (GOES) data and tested for a dry and a wet period. The results show that the MIST has comparable biases and better rain event detection skill than the TMPA, although the TMPA is constrained by the gauge analysis by design while the MIST has no direct gauge input.     

Rainfall sensitivity analyses for the HSB sounder: an Amazon case study

This work examines the sensitivity of the different channels of the HSB (Humidity Sensor for Brazil), on board the AQUA satellite, for the purpose of retrieving surface rainfall over land. The analysis is carried out in two steps: (a) a theoretical study performed using two radiative transfer models, RTTOV and the so‐called Eddington method; and (b) the determination of the correlation between coincident measurements of HSB brightness temperatures and radar rainfall estimates during the DRY‐TO‐WET/AMC/LBA field campaign held in the Amazon region during September and October 2002. Theoretical results indicate the sensitivity of the HSB to water vapour content and cloud liquid water in the precipitation estimation. Theoretical and experimental analyses show that the channels 150 and 183±7 GHz are more adapted to estimate precipitation than the 183±1 and 183±3 GHz channels. The simulation analyses clearly show a hierarchy in physical effects that determine the brightness temperature of these channels. The rain and ice scattering dominate over the absorption of liquid water, and the liquid water absorption effect dominates over the absorption of water vapour. The results show that the 150 and 183±7 channels are more sensitive to the variation of liquid water and ice than the 183±1 and 183±3 channels. For the precipitation estimation using these channels, it was found that it is best adapted to the low precipitation rate situations, since the brightness temperature is rapidly saturated in the presence of high intense precipitation. A case study to estimate precipitation using the radar data has shown that it is possible to adjust a curve that relates the precipitation rate to the brightness temperature of the 150 GHz channel with a good level of accuracy for low precipitation rates.     

Study of the potential of alternative crops by integration of multisource data using a neuro‐fuzzy technique

This work proposes a neuro‐fuzzy method for suggesting alternative crop production over a region using integrated data obtained from land‐survey maps as well as satellite imagery. The methodology proposed here uses an artificial neural network (multilayer perceptron, MLP) to predict alternative crop production. For each pixel, the MLP takes vector input comprising elevation, rainfall and goodness values of different existing crops. The first two components of the aforementioned input, that is, elevation and rainfall, are determined from contour information of land‐survey maps. The other components, such as goodness values of different existing crops, are based on the productivity estimates of soil determined by fuzzyfication and expert opinion (on soil) along with production quality by the Normalized Difference Vegetation Index (NDVI) obtained from satellite imagery. The methodology attempts to ensure that the suggested crop will also be a high productivity crop for that region.     

Accuracy assessment of MODIS,NOAA and IRS data in snow cover mapping under Himalayan conditions

Snow cover information is an essential parameter for a wide variety of scientific studies and management applications, especially in snowmelt runoff modelling. Until now NOAA and IRS data were widely and effectively used for snow‐covered area (SCA) estimation in several Himalayan basins. The suit of snow cover products produced from MODIS data had not previously been used in SCA estimation and snowmelt runoff modelling in any Himalayan basin. The present study was conducted with the aim of assessing the accuracy of MODIS, NOAA and IRS data in snow cover mapping under Himalayan conditions. The total SCA was estimated using these three datasets for 15 dates spread over 4 years. The results were compared with ground‐based estimation of snow cover. A good agreement was observed between satellite‐based estimation and ground‐based estimation. The influence of aspect in SCA estimation was analysed for the three satellite datasets and it was observed that MODIS produced better results. Snow mapping accuracy with respect to elevation was tested and it was observed that at higher elevation MODIS sensed more snow and proved better at mapping snow under mountain shadow conditions. At lower elevation, IRS proved better in mapping patchy snow cover due to higher spatial resolution. The temporal resolution of MODIS and NOAA data is better than IRS data, which means that the chances of getting cloud‐free scenes is higher. In addition, MODIS has an automated snow‐mapping algorithm, which reduces the time and errors incorporated during processing satellite data manually. Considering all these factors, it was concluded that MODIS data could be effectively used for SCA estimation under Himalayan conditions, which is a vital parameter for snowmelt runoff estimation.