CN103020369A - High-resolution forest fire forecasting method - Google Patents

Abstract

Translated fromChinese

一种高分辨率森林火灾预报方法，属于计算机数字图像处理技术领域。以1956-2006年气象数据作为基础，通过ANUSPLIN插值技术，生成高分辨率的林内气象要素空间预报数据库。利用1956-1990年气象数据对森林火险气候指数系统进行初步参数化，利用1991-2006年气象数据和火灾记录数据对系统预测结果进行校准，并进一步调整和优化参数，建立适合浙江省的火险气候指数模型；最后将未来气象数据导入本地化后的模型，并结合浙江省森林火险灾害发生强度和空间分布格局及地域特点，对浙江省未来森林火险进行预测，通过传媒进行预报，最终建立适合浙江省的高分辨率的森林火险预警预报系统。

A high-resolution forest fire forecasting method belongs to the technical field of computer digital image processing. Based on the meteorological data from 1956 to 2006, a high-resolution spatial forecast database of meteorological elements in the forest was generated through ANUSPLIN interpolation technology. Using meteorological data from 1956 to 1990 to preliminarily parameterize the forest fire risk climate index system, use meteorological data and fire record data from 1991 to 2006 to calibrate the prediction results of the system, and further adjust and optimize parameters to establish a fire risk climate suitable for Zhejiang Province Index model; finally import the future meteorological data into the localized model, and combine the occurrence intensity, spatial distribution pattern and regional characteristics of forest fire disasters in Zhejiang Province to predict the future forest fire risk in Zhejiang Province, and make forecasts through the media, and finally establish a model suitable for Zhejiang Province. The province's high-resolution forest fire warning and forecasting system.

Description

High resolving power Three Essential Factors of Forest Fire method

[technical field]

The invention belongs to the computer digital image processing technology field, is the Canadian risk of forest fire climatic index of a kind of foundation (FWI) system, treats the method that geodetic territory such as Forests of Zhejiang Province fire carry out early-warning and predicting.

[background technology]

Canada risk of forest fire climatic index (FWI) system is one of the most perfect, most widely used system of development on the our times, also is unique system that can adapt to any yardstick from the part to the whole world.It is take the moisture of three kinds of forest fuel and wind on the impact of fire behavior as the basis.

At present, module or the research idea of this system adopted in some other country, formed the fire size class system of oneself, such as Alaska and Florida and some European countries of New Zealand, Fiji, Mexico, the U.S..1999-2003, Canadian Forest Service and country in Southeast Asia have finished Southeast Asia risk of forest fire systematic study project jointly, and the FWI technology is used in the fire size class system of this area.The state of Michigan of Croatia, Chile and the U.S. is also assessed the application of this system.The fire Forecasting Methodology of this system is than domestic existing correlation technique all science and advanced person.Therefore, preferably set up local forest fire danger class system based on the FWI technology in China.

The implementation of this technology is sex-limited with very large sectional center, take Canada with Zhejiang Province as example compares, comprise here:

1. the weather conditions of latitude formation are different: Canada belongs to the continental climate of Northern Europe form, and Zhejiang Province belongs to typical subtropics monsoon climate.

2. the type of vegetation with distribute different: Canadian northern territory is the tundra of passing through from east to west-tundra soil band and coniferous forest-ashing soil zone, the eastern region from north south by coniferous forest-podzolic soil to temperate mixed forest-grey brown earth transition, the west area is from east and the west former-chestnut soil that is sylvosteppe-luvic phaiozems, prairie-chernozem, short grass successively, and high Mountain area vegetation is needle leaf forest-podzolic soil, Mountain Meadow, mountain region tundra-tundra soil, permanent ice nival belt from bottom to top.Zhejiang the whole province scope all belongs to the Subtropical Evergreen Broad-leaf Forest zone---territory, evergreen broadleaf forest subprovince, east (moistening)---, and Mid-subtropical Evergreen Broadleaved Forests area, evergreen broadleaf forest are the base band forests in Zhejiang Province; Say that from vegetation evergreen broadleaf forest is the zonal vegetation in Zhejiang Province.

3. main fire source is different: Canadian forest fire is mainly natural fire; The forest fire type in Zhejiang Province all is artificial fire source basically, and the main cause that causes the forest fires generation is that productivity is used fiery (open-air smoking) with fire (burn the grass on waste land and make charcoal), sacrifice (visit a grave and burn paper as sacrificial offerings) and life.

4. dividing of long-term and the short-term of having called time in advance: Canadian Three Essential Factors of Forest Fire is take short-time forecast as main, and our forecasting procedure of exploitation is as main take Long-term forecasting.

5. precautionary approach is different: Canadian Three Essential Factors of Forest Fire mainly stresses the forecast of fire origination point, the forecasting procedure of our exploitation is by point and face, set up the Spatial Distribution Pattern forecast that the Forests of Zhejiang Province fire occurs, realize the forecast to the risk of forest fire of zones of different.

Therefore, should set up a kind of targetedly, have novelty, adapt to local high resolving power Forest Fire Alarm forecast system.

[summary of the invention]

The technical issues that need to address of the present invention are to set up a kind of high resolving power Three Essential Factors of Forest Fire method of suitable Zhejiang Province.

For solving the problems of the technologies described above, technical scheme of the present invention is to carry out as follows:

(1) collection of the historical weather data data of region to be forecast and forest fire statistics data: now be example take Zhejiang Province as region to be forecast, collect Zhejiang Province and periphery weather data every day of totally 31 meteorological stations, time span is 1956-2006; Meteorological element comprises: 1. daily maximum temperature, day minimum relative humidity, every day 24h quantity of precipitation and per day wind speed, the 2. average annual highest temperature, average annual relative humidity, annual precipitation and average annual wind speed; Collect the forest fire record data of the 1991-2006 of Zhejiang Province, comprise fire time of origin, place, burnt area, vegetation pattern and fire source;

(2) collection of following weather data data and integration: be that the climatic model of the inter-governmental climate change council generates following Climate Scenarios data from IPCC, adopt the weather data of two kinds of lower four kinds of sights of pattern; Be respectively IIC3AA and two kinds of sights of HC3GG under the HadCM3 pattern, two kinds of sights of the SresA2 under the CGCM3 pattern and SresB1; Four kinds of context data time spans are 2007～2099;

(3) carry out the high resolution space interpolation of meteorological element: utilizing thin dish Smoothing spline method of interpolation is the ANUSPLIN interpolation, and the space interpolation that carries out to each meteorological element obtains high-resolution meteorological element space distribution; The method is used the normal thin dish and local thin dish splines interpolation theory carries out calculating and the optimization of space curved surface to discrete data point, and introduces the line style covariant factor, thereby consider a plurality of factors of influence multivariate data is analyzed interpolation calculation;

(4) foundation of localized fire climate exponential model: according to treating geodetic 1956-1990 weather data, calculate six indexs of risk of forest fire weather: tiny wetness of fuel code, soil ulmin humidity codes, arid code, index of bunching, fire climate index, and then the Canadian risk of forest fire climatic index system to sharing, be FWI, carry out preliminary parametrization, utilize 1991-2006 weather data and fire record data that the system prediction result is calibrated, whole and the Optimal Parameters of the step of going forward side by side is set up the internal heat that is fit to Zhejiang Province and is waited exponential model;

(5) early-warning and predicting of Future Forest fire: utilize following weather data, the internal heat after the operation localization is waited exponential model, and is combined with medium, and the Future Forest fire is carried out early-warning and predicting.

[description of drawings]

Fig. 1 is fire forecast system schematic flow sheet.

Fig. 2 is the structural representation of FWI.

Fig. 3 is the FWI of Zhejiang Province spatial distribution map.

[embodiment]

Also the invention will be further described with reference to the accompanying drawings below in conjunction with embodiment:

Idiographic flow is seen accompanying drawing 1.

Fire climate index system (FWI) comprises six parts.See accompanying drawing 2.The FWI system only requires four kinds of data of every day, i.e. 24 hours gross precipitation of dry-bulb temperature, relative humidity, ten meters high wind speed of field and local meam time measurement at noon.First three index of FWI system is the combustible index, represents respectively the humidity of different layers in the Forest Litter, comprises tiny wetness of fuel code, surface fuel humidity codes and arid code.The value of the humidity codes on the same day is calculated by the weather data value of measuring the same day and the humidity codes of the previous day.The core that each humidity codes is calculated is a simple moisture index exchange model.The rate of drying of dissimilar forest fuel is not identical, and along with Changes in weather every day, combustible humidity also changes.Rear three indexs of system are the fire behavior codes, are generated by three humidity codes and wind speed, have represented respectively the combustible rate of propagation in the forest, Effective fuel quantity and live wire intensity, i.e. initial rate of propagation, index of bunching and weather conditions conducive to wildfires index.

1. tiny wetness of fuel code

In the FWI system, tiny wetness of fuel code (FFMC) has represented that dry weight is 0.25kg/m in the forest litter², thickness is the dry branches and fallen leaves of 1.2cm and the water percentage of other the tiny fuel that has solidified.Tiny fuel major part is comprised of dead and spicule that wither and fall, leaf, lichens, liver moss and other little loose fragment.FFMC is the combustibility and a flammable relative simple index of the tiny fuel of representative.FFMC is subjected to the impact of temperature, precipitation, relative humidity and wind speed, and the value of FFMC changes along with the variation of water-in-fuel rate, and minimum value is 0 (the water-in-fuel rate is 100%), and maximal value is 101 (water percentage of combustible is 0).The larger fire size class that shows of the value of FFMC is higher.The core of FFMC is the exponential model of a simple exchange of moisture, that is:

$m_O=147.2\×\frac{101-\mathrm{FFMC}}{59.5+\mathrm{FFMC}}---\left(1.1\right)$

M wherein₀The water percentage that represents the slow fire fuel of the previous day.

2. soil ulmin humidity codes

Soil ulmin humidity codes (DMC) represents the water percentage of the surface fuel on forest humus upper strata, and namely forest litter the superiors thickness is about 7cm, dry weight is 5kg/m²The water percentage of organic substance.DMC can show the fuel consumption of medium lower floor mulch cover mulch-covering and medium-sized xyloid material.The minimum value of DMC is 0 (the saturated aqueous rate that represents surface fuel is 100%) in the FWI system, and maximal value does not have the upper bound (but in all tests, maximal value seldom surpasses 150).The DMC model also is the exponential model of a simple exchange of moisture, that is:

$M_O=20+\ln \left(\frac{\mathrm{DMC}-244.73}{-43.43}\right)---\left(1.2\right)$

M wherein_OThe water percentage that represents the surface fuel of the previous day.

3. arid yard

Arid code (DC) is one and calculates prolonged drought to the simple index number of the impact of forest fuel.Arid code has represented that dry weight is 25kg/m in the forest litter², thickness is the deep layer combustible of 18cm and the water percentage of the thick residual body of dead-wood.For weighing the impact of seasonal drought on forest fuel and deep layer lower floor mulch cover mulch-covering and large-scale section wood, the arid code is a useful index.The minimum value of DC is 0, and maximal value does not have the upper bound, but seldom surpasses 1000.The core of DC model also is a simple exponential model.That is:

$Q_o=400\×e\frac{-\mathrm{DC}}{400}---\left(1.3\right)$

Q wherein_oThe humidity equivalent of expression arid the previous day code.

4. index of bunching

Index of bunching (BUI) is calculated by DMC and DC, has represented the humidity level of combustible.Although the value of BUI is the weighted mean value of DMC and CD, the computing formula of BUI is very complicated.In the FWI system, BUI be one without the unit index, relatively represented the amount of the potential burning of forest fuel.

5. initial Sprawl Indices

Initial Sprawl Indices (ISI) is calculated by FFMC and wind speed, has represented the potential grade of fire spread.In different Forest Types, ISI is the good indicator of expression fire spread grade.

6. fire climate index

Fire climate index (FWI) is last index in the FWI system, is calculated by ISI and BUI.FWI is that the internal heat in an area is waited the result that condition combines with the combustible water percentage.By dividing different FWI numerical ranges, just can fire size class be described to people.But more index is predicted fire in the now fire administrative authority dependence FWI system, is not single FWI index.

Step 1 is collected Zhejiang Province and the periphery weather data data of totally 17 meteorological stations, and Forests of Zhejiang Province fire record data data.Wherein, weather data derives from observation data and the meteorological element Sharing Center of National Meteorological Bureau of meteorological station.Meteorological element comprises: 1. daily maximum temperature, day minimum relative humidity, every day 24h quantity of precipitation and per day wind speed.2. the average annual highest temperature, average annual relative humidity, average annual precipitation and average annual wind speed; The weather data of collecting is 1956 the earliest, 2006 is that distance was collected data immediate 1 year then, time span is 1956-2006, forest fire is data from Zhejiang Forestry Room forestry monitoring center, comprise fire time of origin, place, point of origin geographic coordinate, burnt area, Forest Types, reason of fire and the information of putting out a fire to save life and property, wherein, Zhejiang Province carries out record since 1991 to fire, 2006 is that distance was collected data immediate 1 year then, and time span is 1991-2006.

Analyze characteristics and rule thereof that 1991-2006 Forests of Zhejiang Province fire occurs, comprise that year variation, month border of forest fire changes, relation, the forest fire burning things which may cause a fire disaster of year variation, moon border variation, forest fire Density Distribution, forest fire and the vegetation of forest fire burnt area.

Step 2, the collection of following meteorological data and integration: utilize the IPCC climatic model to generate following Climate Scenarios data, adopt altogether the weather data of two kinds of lower four kinds of sights of pattern; Be respectively HC3AA and two kinds of sights of HC3GG under the HadCM3 pattern, two kinds of sights of the SresA2 under the CGCM3 pattern and SresB1; The period span is 2007～2099;

HadCM3 is the GCM pattern by the extra large gas coupling of Britain's Hadley weather center exploitation, the horizontal space resolution of pattern is 2.5 degree *, 3.75 degree, consider the impact on climate change of greenhouse gases, sulfate aerosol and ozone variation in the HC3AA sight, considered CO in the HC3GG sight₂And other trace greenhouse climate is on the impact of climate change; CGCM3 is the 3rd version of Canadian climate model and analytic centre's whole world coupled mode, the horizontal space resolution of pattern has two version T47 and T63, the former spatial resolution is 3.75 degree, the latter's spatial resolution is 2.8 degree, adopt SresA2 and two kinds of Climate Scenarios data of SresB1 in the T63 version, a very unbalanced world has been described by SresA2 sight family, principal character is: self-sufficient, keep local characteristic, the advolution of yield-power mode is unusually slow between each region, causes the population growth, the main facing area of economic development, economic growth and technique variation are discontinuous per capita, are lower than the speed of development of other sight; The world of an advolution has been described by SresB1 sight family: economic consequence is adjusted rapidly to service and information economy aspect, be accompanied by the decline of material dense degree, and the introduction of cleaning and resource high-efficiency technology, it focuses on the global solutions of economy, society and environment sustainable development, comprising the raising of fairness, but do not take extra Policies on Climate intervention;

Step 3, carry out the high resolution space interpolation of meteorological element: ANUSPLIN based on normal thin dish and local thin dish splines interpolation theory, local thin dish Smoothing spline method is the expansion to thin dish Smoothing spline prototype, except common batten independent variable, allow to introduce linear covariant submodel, such as the direct correlationship of temperature and height above sea level, precipitation and shore line, the theoretical statistical model of local thin dish Smoothing spline is:

z_i＝f(x_i)+b^Ty_i+e_i(i＝1，…，N) (1)

In the formula: Z_iTo be positioned at the dependent variable that i is ordered; x_iBe d dimension batten independent variable; F (x_i) for need estimation about x_iUnknown smooth function; B is y_iP be coefficient; y_iFor p ties up independent covariant; e_iFor have expectation value be 0 and variance be w_iδ₂The independent variable stochastic error, w wherein_iFor as the known Local Phase of weight to the coefficient of variation, δ²Being error variance, is constant in all data points, but usually unknown; By formula (1) as seen, lack second in formula, namely during covariant dimension p=0, model can be reduced to normal thin dish Smoothing spline; When lacking first independent variable, model becomes multiple linear regression model, but does not allow this situation to occur among the ANUSPLIN.

Function f (x_i) and coefficient b can be definite by minimizing of following formula, i.e. least-squares estimation is determined:

$\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left\{\frac{z_i-f\left(x_i\right){\mathrm{pb}}^T{y}_i}{w_i}\right\}}^2+{\mathrm{\ρJ}}_m\left(f\right)=\min ---\left(2\right)$

In the formula: J_m(f) be function f (x_i) the roughness measure function, m is called the batten number of times in ANUSPLIN, also be the roughness number of times; ρ is positive smoothing parameter, between the roughness of data fidelity and curved surface, play equilibrium activity, when ρ close to zero the time, fitting function is a kind of precise interpolation method, when ρ when infinite, function is close to being the least square polynomial expression, order is determined by coarse number of times m, and the smoothing parameter value usually by Generalized Cross Validation (Generalized Cross Validation, GCV) minimizing to determine;

Select daily maximum temperature, day minimum relative humidity, per day wind speed and the daily precipitation amount of every day.Then carry out the ANUSPLIN interpolation, generate meteorological element Space Forecast database in the woods.

The meteorological element spatial data that forms is put into the program that FWI that we have compiled calculates, calculate each index of fire climate: 3 wetness of fuel codes (FFMC, DMC, DC), 3 kinds of fire behavior indexes (ISI, BUI, FWI).

With the fire climate index that draws, in Compaq Visual Fortran, compile and move, make the fire climate index be compiled into the form that we need, i.e. the output file of output.

The Output file that draws is put into the FWI calculation procedure, and add Zhejiang Province's elevation map (ASC file), draw the spatial database of fire climate index.Each index is loaded among the ARCGIS, draws each fire climate index spatial distribution map on the same day, its resolution is 1 kilometer of 1 kilometer *.Specifically see accompanying drawing 3.

Step 4, the foundation of localization fire climate exponential model: according to the computational data of each index of 17 the meteorological site FWI in 1956-2006 Zhejiang Province, adopt the FWI stage division at forestry research center, the Pacific Ocean, according to " national weather grade of forest " standard risk of forest fire is divided into 5 grades, namely basic, normal, high, very high and high, wherein the fire danger low fate accounts for 45%～50%, middle fire fate accounts for 25%～30%, high fire fate accounts for 12%～15%, very high fire fate accounts for 7%～8%, and high fire fate accounts for 2%～3%.

According to Zhejiang Province's division of fire size class, the fire danger index data that day, the 1991-2006 fire was occured are added up, according to each fire size class fire frequency and actual each fire size class fate, further adjust and Optimal Parameters, the final internal heat that is fit to Zhejiang Province of setting up is waited exponential model again.

Step 5, Future Forest fire early-warning and predicting: with the fire climate exponential model after the following meteorogical phenomena database importing localization, Zhejiang Province's Future Forest fire is predicted, and forecast by medium.

Through behind the above-mentioned five steps, the dividing condition of Zhejiang Province's fire size class is seen the following form:

Table 5.1 Zhejiang Province division of fire size class

Claims (1)

1. high resolving power Three Essential Factors of Forest Fire method is characterized in that carrying out as follows:

(1) collection of the historical weather data data of region to be forecast and forest fire statistics data: now be example take Zhejiang Province as region to be forecast, collect Zhejiang Province and periphery weather data every day of totally 31 meteorological stations, time span is 1956-2006; Meteorological element comprises: 1. daily maximum temperature, day minimum relative humidity, every day 24h quantity of precipitation and per day wind speed, the 2. average annual highest temperature, average annual relative humidity, annual precipitation and average annual wind speed; Collect the forest fire record data of the 1991-2006 of Zhejiang Province, comprise fire time of origin, place, burnt area, vegetation pattern and fire source;

(2) collection of following weather data data and integration: be that the climatic model of the inter-governmental climate change council generates following Climate Scenarios data from IPCC, adopt the weather data of two kinds of lower four kinds of sights of pattern; Be respectively HC3AA and two kinds of sights of HC3GG under the HadCM3 pattern, two kinds of sights of the SresA2 under the CGCM3 pattern and SresB1; Four kinds of context data time spans are 2007-2099;

(3) carry out the high resolution space interpolation of meteorological element: utilizing thin dish Smoothing spline method of interpolation is the ANUSPLIN interpolation, and the space interpolation that carries out to each meteorological element obtains high-resolution meteorological element space distribution; The method is used the normal thin dish and local thin dish splines interpolation theory carries out calculating and the optimization of space curved surface to discrete data point, and introduces the line style covariant factor, thereby consider a plurality of factors of influence multivariate data is analyzed interpolation calculation;

(4) foundation of localized fire climate exponential model: according to treating geodetic 1956-1990 weather data, calculate six indexs of risk of forest fire weather: tiny wetness of fuel code, soil ulmin humidity codes, arid code, index of bunching, fire climate index, and then the Canadian risk of forest fire climatic index system to sharing, be FWI, carry out preliminary parametrization, utilize 1991-2006 weather data and fire record data that the system prediction result is calibrated, whole and the Optimal Parameters of the step of going forward side by side is set up the internal heat that is fit to Zhejiang Province and is waited exponential model;

(5) early-warning and predicting of Future Forest fire: utilize following weather data, the internal heat after the operation localization is waited exponential model, and is combined with medium, and the Future Forest fire is carried out early-warning and predicting.

Images