Abstract

The fire danger rating system implemented in New Zealand is the Canadian Fire Weather Index (FWI) System developed 40 years ago for Canadian temperate forests. Issues have been raised in relation to this system when applied in other regions with different climate and vegetation environments. For the first time, two methods were proposed for improving the Drought Code (DC) component of the FWI System for New Zealand. The first method (PotE) employs a potential evaporation (PE) scheme that considers wind speed, surface air stability, and water vapor mixing ratio gradient. The second method (soilM) uses soil moisture. For the latter, when soil moisture is derived from observations, the calculated DC represents the actual drought status of the soil. DC and FWI have been calculated with the original and the two new DC methods at 28 climate stations in New Zealand for a pair of 2-yr periods. The Joint U.K. Land Environment Simulator (JULES) was run to provide the PE and soil moisture for the two methods. The original DC method underestimated the drought status in New Zealand, especially in summer, leading to underestimation of FWI. The PotE method significantly overestimated the drought status in summer. The errors in the calculated drought status and FWI were largely reduced by using the soilM method with simulated soil moisture from JULES. In this paper, the reasons for this reduction in error are investigated by testing the sensitivity of DC to surface evaporation and to soil parameters. Potential benefit is found from using the proposed soilM method for monitoring drought status and for FWI calculations.

1. Introduction

New Zealand is a group of islands in the midlatitudes of the southwest Pacific Ocean. The two main islands are the North Island and the South Island. The central North Island is dominated by the Volcanic Plateau, an active volcanic area. The South Island has a dominant southwest-to-northeast mountain ridge, the Southern Alps. Of all the land area of New Zealand, about 31% is covered by native and plantation forests, about 31% by grassland, and about 28% by scrubland (http://www.lcdb.scinfo.org.nz). From 1991 to 2007, an average of 3033 wildfires burned 5866 ha each year, of which 54% was grasslands, 40% scrublands, and 6% forest (Anderson et al. 2008).

In New Zealand, fire danger is assessed using the Canadian Fire Weather Index (FWI) System, which is a subsystem of the Canadian Forest Fire Danger Rating System (Anderson 2005; Alexander 2008). Developed in Canada during the 1960s and 1970s, the FWI System uses daily rainfall accumulation, air temperature, relative humidity, and wind velocity measured at noon local standard time to estimate the moisture content of fuels from three levels of the forest floor (Van Wagner 1987; Wotton 2009). The fuel moisture codes, in particular, provide fire managers with useful information on the relative dryness of vegetation fuels and associated fire control difficulty [e.g., ease of ignition and involvement of different fuel layers in combustion, potential fuel consumption, depth of burning, and extinguishment (mop up) requirements]. The Fine Fuel Moisture Code (FFMC) represents the moisture content of litter and dead fine fuels on the forest floor. The Duff Moisture Code (DMC) represents the moisture content of loosely compacted organic material (~7.6 cm deep; Van Wagner 1970). The Drought Code (DC) represents the moisture content of deep compact organic soil of moderate depth (~18 cm; Van Wagner 1987).

These three fuel moisture codes are then used to calculate the initial spread index (ISI) and the buildup index (BUI). The ISI estimates the combined effects of wind speed and the FFMC on potential fire spread by using a simple exponential function of wind speed so that it doubles the FWI for increments of wind speed of about every 20 km h−1. The BUI, representing the availability of the deeper or larger-sized fuel for combustion, is a combination of the DMC and the DC. The ISI and the BUI are then used to calculate the value of the final FWI, which represents the potential peak daily intensity of a spreading fire.

In the FWI System, DC values are calculated using the ratio of current soil water content (a balance of the accumulated soil water from precipitation and the evaporation of the soil water) to saturated water content (8 in. or 200 mm; 1 in. = 25.4 mm) for the organic soil. The evaporation is assumed to be in a simple linear relationship with the moisture content and potential evaporation (PE). The PE is simply calculated using only air temperature and day-length factor. It does not consider the wind speed, air stability, and the vertical gradient of air humidity. In addition, PE is affected by the vegetation types and also by the soil texture.

Although based on simple parameterizations, initial studies (e.g., Muraro and Lawson 1970) showed that the DC index obtained by using this method tracked moisture variations in deep compact organic soil layers reasonably well. Aside from the daily operations for fire management (e.g. Girardin and Wotton 2009) the DC and FWI have recently been used to study climate change impacts on past, present, and future wildfire risks in Canada (Amiro et al. 2004; Girardin et al. 2004; Bergeron et al. 2004; Girardin and Mudelsee 2008). Although initially developed from and applied to the temperate conifer forests of Canada, the FWI System has been used in other regions of the world, such as the Mediterranean and the Iberian Peninsula in Europe (Viegas et al. 1999; Camia et al. 2006; Dimitrakopoulos et al. 2011; Bedia et al. 2012), tropical Indonesia and Malaysia (de Groot et al. 2007), parts of the United States and Mexico (Alexander and Cole 1995; Brenner et al. 1998; Wilmore 2001; Lee et al. 2002), Australia (Dowdy et al. 2010), and New Zealand (Anderson 2005; Alexander 2008). For these regions, the forest environment is likely to be different from that of the boreal temperate forest on which the evaporation method was based. In addition, in some countries (e.g., Indonesia and New Zealand), the DC component has also been used for other vegetation types such as grasslands and scrublands (e.g., Fogarty et al. 1998; de Groot 2007). For these environments with their different soil and vegetation types, the calculated DC using the original methods may not accurately represent the real situation. In fact, comments regarding the applicability of adopting the FWI System in environments differing from that in which the system was developed have been made by several authors (e.g., Fogarty et al. 1998; Wilmore 2001; Taylor and Alexander 2006; Anderson 2009; Anderson and Anderson 2010). However, to our knowledge, reformulation and improvement of the original parameterization methods for the fuel moisture codes within the FWI System have not been found in any publications when applied in other countries or vegetation types, despite some studies suggesting the need for this (e.g., Wotton et al. 2005; Johnson et al. 2013). One of the objectives of this study is therefore to investigate the reliability of the original DC method of the FWI System when used in New Zealand, especially for grasslands and scrublands, the predominant land cover types close to the 32 climate stations (Fig. 1) used in this study. Another objective is to evaluate two new DC methods for improving the calculation of DC, and to investigate the sensitivity of these methods to surface evaporation and soil parameters representative of New Zealand.

Fig. 1.

Locations of the 32 climate stations (denoted from 1 to 32) in New Zealand with soil moisture observations and meteorological observations used to run JULES.

Fig. 1.

Locations of the 32 climate stations (denoted from 1 to 32) in New Zealand with soil moisture observations and meteorological observations used to run JULES.

In this paper, the land surface models (LSMs), DC methods, experiments, and data used are described in section 2. Calculated DC and FWI values from the original and the proposed new DC methods are analyzed in section 3. The sensitivities of the original DC method and one of the suggested new DC methods (PotE) to surface evaporation are tested in section 4. Effects of soil parameters on DC for the suggested soil moisture DC method are analyzed in section 5. Results are summarized in section 6.

2. Description of DC methods, models, data, and experiments

a. Description of DC of the FWI System

The moisture status described by the DC component of the FWI System is defined to be a function of the saturated soil water content and the current soil water content (Turner 1972):

 
formula

It is assumed that the capacity of the soil layer when fully saturated is 8 in. (or 800 in units of 0.01 in., implying uniform soil), so that where q is the current soil water content in units of 0.01 in.

 
formula

When daily precipitation (mm) is measured at noon local standard time, the effective precipitation is

 
formula

or

 
formula

and is the previous day’s soil water content in units of 0.01 in. The evaporation of the soil E is expressed as

 
formula

Combining Eq. (3) with Eq. (6),

 
formula

or approximately

 
formula

In Turner (1972), α = 1.0 and the day-length factor, which changes with latitude and seasons, and T is the air temperature at noon (°C). Different values of α will be used in section 4 for sensitivity tests.

For this standard method of calculating the DC, the saturated soil water content for the organic soils of Canadian temperate forests is set as a fixed value of 800 (in units of 0.01 in., or 200 mm). For a different soil with a different saturated soil water capacity, the same DC value may not represent the same soil drought status or the same soil water content as for the standard organic soil. In other words, the absolute DC values calculated by this method are not comparable between different soil types.

b. Proposed new methods

Two methods to improve the DC (drought status) estimation were proposed. The first of these, PotE (Table 1), replaces the PE in Eq. (5) of the original DC method with PE as simulated by the Joint U.K. Land Environment Simulator (JULES) based on Eq. (9) as described in the following subsection, which considers wind speed, air stability, and the vertical gradient of air humidity. In addition, since JULES simulations were run over grasslands and the interception of precipitation by grass is smaller than that for forest, in the PotE method the condition R > 2.8 for effective rain [Eq. (3)] was changed to R > 1.0 for grass, based on experience and also to make sure the effective rain is not negative. This is only a rudimentary adjustment. The second method, soilM, following Eq. (1), has two forms (sim_soilM or obs_soilM; Table 1):

 
formula
 
formula

where is the saturated soil moisture volumetric content for a specific soil, and is different for different soil types, and sim_SM and obs_SM are the soil moisture simulated by JULES and the observed soil moisture volumetric content, respectively.

Table 1.

Description of the DC methods and experiments.

Description of the DC methods and experiments.
Description of the DC methods and experiments.

Because different soil types have different saturation points, one advantage of the use of soilM over the original DC method [Eq. (5)] and the first proposed DC method (PotE) is that the effect of different soil types is now considered. Another advantage is that when the soil moisture is based on observations, the calculated DC [Eq. (7)] represents the actual drought status.

c. Short description of JULES hydrology

This study is based to a large degree on the recent work done by Yang et al. (2013, referred to hereinafter as YY13) for simulated soil moisture and PE for a pair of 2-yr time periods made using JULES (Best et al. 2011). In JULES, the surface evaporation E, including bare-soil surface evaporation and vegetation transpiration, is expressed as

 
formula

where ψ is a function of aerodynamic resistance and stomatal or surface moisture resistance. The aerodynamic resistance is a function of surface roughness, wind speed, and atmospheric stability through a Richardson number formulation. For a wet surface, the surface evaporation E is equal to PE:

 
formula

where is aerodynamic resistance, ρ is the air density, is the saturation humidity at the surface, and is the humidity at a specific height above the surface.

The amount of water that reaches the soil surface beneath vegetation depends on the type of surface and the interception of precipitation by the vegetation canopy. The water reaching the soil surface is then split between infiltration into the soil and surface runoff. In YY13, four soil layers in JULES were configured as 0.1, 0.25, 0.65, and 2.0 m, giving a total soil depth of 3 m. Soil water contents are updated using a finite-difference form of the Richards equation. The total soil moisture content within a soil layer is incremented by diffusive water flux flowing in from the layer above and the diffusive flux flowing out to the layer below, and the evapotranspiration extracted by plant roots in the layer. The vertical water fluxes follow Darcy’s law [see Cox et al. (1999), Essery et al. (2003), and Best et al. (2011) for more detailed descriptions of JULES].

d. Data and sensitivity tests to soil parameters

JULES, version 2.1.2, was used to simulate PE for the PotE method and soil moisture for the sim_soilM method. In this study, the soil moisture simulated by JULES in the 0.0–0.35-m soil layer was used for sim_soilM to calculate DC. Refer to YY13 for more information about how the soil moisture was simulated in New Zealand.

In New Zealand, soil moisture volumetric content observations for the 0–40-cm soil layer are made on an hourly basis at 55 climate stations using Aquaflex soil-moisture-measuring equipment. This equipment consists of a 3-m “tape” laid across the appropriate soil depth, with one end of the tape deeper than the other. The measuring equipment uses fundamental differences in the dielectric constant of free water and soil materials in mineral soils, and the way that microwave pulses propagated down transmission lines interact with soil and water. These soil moisture observations were used for obs_soilM. Of the 55 available stations, only 32 climate stations (Fig. 1) have hourly surface meteorological observations as required by an LSM to simulate soil moisture. For 28 of the 32 stations, YY13 showed that the one-dimensional setting of JULES (lateral soil water flow was not considered) can adequately simulate the soil moisture with a yearly mean absolute error of ~5% or smaller in volumetric moisture content. In this study, all the experiments and sensitivity tests were conducted at only these 28 climate stations. The remaining four stations (stations 29–32) have very large errors in simulated soil moisture and were not used for the calculation of DC and FWI because of potential lateral soil water flow (YY13). The typical soil texture at each of these 28 stations is mainly one of six types: stony loam, sandy loam, fine sandy loam, stony silt loam, clay loam, or silt loam.

To investigate the sensitivity of DC to errors in soil parameters, in addition to sim_soilM that used the best soil parameters from field survey, two sensitivity tests were also conducted: NEW SOIL and OLD SOIL (Table 1). The soil parameter data for JULES coupled with the Met Office Unified Model (MetUM; Webster et al. 2003) have recently been updated (YY13). NEW SOIL refers to the experiment where the soil parameters in JULES were from the updated soil data of MetUM, while OLD SOIL refers to the experiment with the soil data from MetUM before the update.

In the FWI System, the initial values of FFMC, DMC, and DC are required to commence calculations. This means a spinup period is required to minimize the effects of errors in the initial conditions. The spinup of DC can take more than 6 months (not shown), and that of FWI may need up to 6 months (not shown). In YY13, to investigate the effect of spinup on the simulated soil moisture by JULES, two independent experiments were conducted using different but overlapping seasons. One is from 1 August 2008 to 31 July 2010. The other is from 1 February 2009 to 31 January 2011. The first year of each was used to spin up JULES. The simulation output during the second year after spinup was used separately for analysis. In this study, the simulated soil moisture and PE from YY13 were used to calculate DC [Eqs. (6) and (4)]. Accordingly, a full year of calculation (from 1 August 2008 to 31 July 2009 or 1 February 2009 to 31 January 2010) was used to spin up the FWI System codes and indices. The second year of each was used for analysis. Because the simulated soil moistures during the overlapping period (1 February–31 July 2010) were not the same for both experiments because of the effect of the spinup starting in different seasons (YY13), in this study the daily outputs of DC and FWI for the two separate periods, 1 August 2009–31 July 2010 and 1 February 2010–31 January 2011, were also analyzed separately.

3. DC and FWI from three DC methods

The proposed obs_soilM method uses observed soil moisture, and the value of DC represents the actual soil drought status. Results for this method for two representative climate stations in New Zealand, Paraparaumu (station 28) and Winchmore (station 19), are shown in Fig. 2. The DC (obs_soilM; solid lines) was much higher (drier soil) during the warm seasons (from November to April, the same hereinafter) than during the cold seasons (from May to October, the same hereinafter), because of strong evaporation and relatively less rainfall during the warm seasons. The DC was around 200 or lower (relatively moist soil) for most of the cold seasons at the two stations, but around 600 for most of the warm seasons. For the yearly mean, the DC ranges from ~150 to ~600 across the 28 stations (Figs. 3 and 4) with a low (high) yearly mean DC corresponding to a high (low) annual rainfall amount (not shown).

Fig. 2.

The calculated DC values for Old DC, PotE, sim_soilM, and obs_soilM at (left) Paraparaumu (station 28; Fig. 1) and (right) Winchmore (station 19) for (top) 1 Feb 2010–31 Jan 2011 and (bottom) 1 Aug 2009–31 Jul 2010.

Fig. 2.

The calculated DC values for Old DC, PotE, sim_soilM, and obs_soilM at (left) Paraparaumu (station 28; Fig. 1) and (right) Winchmore (station 19) for (top) 1 Feb 2010–31 Jan 2011 and (bottom) 1 Aug 2009–31 Jul 2010.

Fig. 3.

(a) Yearly mean DC, (b) MD of DC, and (c) MAD during 1 Feb 2010–31 Jan 2011 at 28 climate stations for PotE, sim_soilM, and Old DC.

Fig. 3.

(a) Yearly mean DC, (b) MD of DC, and (c) MAD during 1 Feb 2010–31 Jan 2011 at 28 climate stations for PotE, sim_soilM, and Old DC.

Fig. 4.

As in Fig. 3, but for 1 Aug 2009–31 Jul 2010.

Fig. 4.

As in Fig. 3, but for 1 Aug 2009–31 Jul 2010.

For the original DC method (Old DC, Fig. 2, dashed lines), the value of DC was much lower than that from obs_soilM for most times of the year at the two stations. The drought status was underestimated by Old DC. At times during warm seasons, this underestimation could be more than 300 points. For most of the time during the cold seasons, values of the DC were 0 or very close to 0 (a saturated soil). This is not the real situation in many areas of New Zealand (YY13). Like Old DC, the DC calculated from PotE was also close to 0 during some winter periods (Fig. 2). The near-saturated soil during winters for both PotE and Old DC can be found at most of the 28 stations. Both Old DC and PotE used Eq. (5), except the latter used PE from JULES. This implies that the parameterization of Eq. (5) is not appropriate at some times during cold seasons over grasslands or scrublands in New Zealand. In fact, without using this parameterization, sim_soilM was closer to obs_soilM than Old DC and PotE during the cold seasons.

YY13 showed an overall satisfactory soil moisture simulation by JULES at the 28 stations over New Zealand. This implies that PE was simulated reliably by JULES. However, during the warm seasons, the DC values calculated using the PotE method were much higher than the observations, implying that the evaporation [or the drying process, (α/2)PE] in Eq. (5) was overestimated for α = 1. In the following section, different values of α will be tested for the PotE method.

Figures 3a and 4a show the yearly mean DC values at the 28 climate stations (Fig. 1, stations 1–28). For all stations, the DC values calculated using the Old DC method were smaller than those calculated using either of the PotE or sim_soilM methods. Overall, the DC values calculated using PotE were higher than those from the sim_soilM method. However, large variations in DC were found among stations. Overall, the DC values at stations 1, 2, 4, and 17 were much smaller than at other stations. This is mainly a result of the inverse relationship between DC values and rainfall amount. During prevailing westerly airflow, station 1 (Stratford) is on the windward side of the North Island and stations 2 (Reefton), 4 (Greymouth), and 17 (Franz Josef) are on the windward side of the South Island (Fig. 1a) and, thus, receive much more annual rainfall than other stations (not shown).

Taking the DC calculated from obs_soilM as the “true” values, the mean difference (MD) of DC (calculated DC − true DC) and the mean absolute difference [MAD, absolute (calculated as DC− true DC)] were calculated at each station (Tables 2 and 3). For the 28 stations, Old DC had negative MD (moist bias) at 25 stations, while sim_soilM had small positive MD (dry bias) at 23 stations and PotE had large positive MD (>50) at 21 stations (Figs. 3b and 4b). The overall dry bias of PotE and moist bias of Old DC were significant, as shown by the average values of MD of DC at the 28 stations, especially during the warm seasons for PotE.

Table 2.

The MAD of DC and the MD (obs_soilM − PotE or sim_soilM or OLD SOIL or NEW SOIL) averaged across 28 stations for 1 Feb 2010–31 Jan 2011, with the warm season (numbers inside parentheses) and cold season (numbers inside brackets) shown separately.

The MAD of DC and the MD (obs_soilM − PotE or sim_soilM or OLD SOIL or NEW SOIL) averaged across 28 stations for 1 Feb 2010–31 Jan 2011, with the warm season (numbers inside parentheses) and cold season (numbers inside brackets) shown separately.
The MAD of DC and the MD (obs_soilM − PotE or sim_soilM or OLD SOIL or NEW SOIL) averaged across 28 stations for 1 Feb 2010–31 Jan 2011, with the warm season (numbers inside parentheses) and cold season (numbers inside brackets) shown separately.
Table 3.

As in Table 2, but for 1 Aug 2009–31 Jul 2010.

As in Table 2, but for 1 Aug 2009–31 Jul 2010.
As in Table 2, but for 1 Aug 2009–31 Jul 2010.

At most of the stations (Figs. 3c and 4c), the MAD of DC for Old DC was much larger than that for sim_soilM. The MAD of DC for PotE was the largest (Tables 2 and 3). For Old DC, large moist biases were found for both warm and cold seasons, indicating a moist bias is the main reason for the underestimation of DC throughout the year within the original DC method. For PotE, the dry bias is the main reason for the large MAD of DC in warm seasons (Tables 2 and 3). It is surprising that the PotE method, with its more reliable PE estimation from JULES, had larger overall errors than Old DC, especially during the warm seasons. Smaller errors were found when using the sim_soilM method. The reasons for this will be investigated in the following section.

DC is a key part of the FWI System. It is necessary to see the differences in the calculated FWI for different DC methods. Figure 5 shows the differences in calculated FWI between using obs_soilM and Old DC (or PotE, or sim_soilM). Since the DC from obs_soilM represents the actual soil drought status, FWI values calculated from obs_soilM were taken as the “true” values. Large differences between the Old DC method and obs_soilM were found, differences of up to six points at Winchmore and up to three points at Paraparaumu. Use of the original DC method led to underestimation of FWI for some periods, especially during warm seasons. The difference between FWIs calculated using obs_soilM and sim_soilM (solid lines) was usually 0.5 points or less throughout the year at the same two stations. In comparison, the difference between FWI values calculated using PotE and obs_soilM was relatively large (~1.5 points or lower). Overall, FWIs calculated using the Old DC method had the largest departure from the true values throughout the year at these two stations.

Fig. 5.

Difference in calculated FWI between using the obs_soilM and using the Old DC methods (or the PotE or sim_soilM methods) at (left) Paraparaumu and (right) Winchmore: (top) 1 Feb 2010–31 Jan 2011 and (bottom) 1 Aug 2009–31 Jul 2010.

Fig. 5.

Difference in calculated FWI between using the obs_soilM and using the Old DC methods (or the PotE or sim_soilM methods) at (left) Paraparaumu and (right) Winchmore: (top) 1 Feb 2010–31 Jan 2011 and (bottom) 1 Aug 2009–31 Jul 2010.

Figure 6a shows the yearly mean FWI values at 28 climate stations for the three DC methods for 1 February 2010–31 January 2011. Overall, FWI values calculated from the original DC method (Old DC) were smaller than those from PotE and sim_soilM. However, in contrast, differences in FWI between stations when calculated using the same DC method were generally much larger than differences in FWI at a given station using different DC methods. This indicates that FWI is very sensitive to the local climate of New Zealand. Stations 1, 2, 4, and 17 again had the yearly mean FWI values that were much smaller than other stations for the same reason described earlier (higher rainfall). At station 12 (Akitio), the annual mean wind speed (35 km h−1) was almost double that of the other stations. As described in the introduction, FWI usually doubles for increments in wind speed of about 20 km h−1. Thus, the high mean wind speed at station 12 led to the highest yearly mean FWI of the 28 stations. Similar results can also be found for the 1 August 2009–31 July 2010 period (not shown).

Fig. 6.

(a) Yearly mean FWI and (b) yearly MAD of FWI at 28 climate stations for PotE, sim_soilM, and Old DC.

Fig. 6.

(a) Yearly mean FWI and (b) yearly MAD of FWI at 28 climate stations for PotE, sim_soilM, and Old DC.

Figure 6b shows the yearly MAD in FWI values for each DC method by using the FWIs calculated from obs_soilM as the true values for the 1 February 2010–31 January 2011 period. Relatively large MAD values of FWI (0.3–1.5 points) were found at the 28 stations for the Old DC. In contrast, the MADs of FWI for sim_soilM were much smaller. The MADs of FWI using PotE were larger than those using sim_soilM, but smaller overall than those using the Old DC method, even though the overall MADs of DC for the PotE method were larger than those of the Old DC method as described earlier. Similar results can also be found for the 1 August 2009–31 July 2010 period (not shown). It is obvious that the errors in FWI when calculated with the original DC method can be much reduced by using the sim_soilM method for DC over the grasslands and scrublands of New Zealand.

4. Sensitivity to surface evaporation

In this section, we recalculated DC values by changing the value for α in Eq. (5) for both the Old DC and PotE methods. We tried to find an optimum α (if it exists) for all 28 stations for each method so that the MAD and the MD values were as small as possible. For the Old DC method, the value of α was increased to increase surface evaporation so that the overall moist bias of this method could be reduced. For these calculations, and in the PotE method, the condition for effective rain [Eq. (3)] was changed from R > 2.8 mm for forest to R > 1 mm for grasses, because the interception of rainfall by grasses is lower than by trees. This actually increases the amount of rainfall into the soil. For the PotE method, the value of α was decreased so that this overall dry bias could be reduced.

For PotE, when α was changed from 1.0 to 0.6, the MADs and MDs of the DC values averaged at the 28 stations gradually decreased (Tables 4 and 5) from ~230 to ~145 and from ~150 to ~0.5, respectively. Further decreasing of α from 0.6 to 0.5, the MADs of DC were almost the same (~145 for 1 February 2010–31 January 2011 period; ~150 for 1 August 2009–31 July 2010 period), but the MD of DC changed to large negative values (~−44, moist bias). It seems that α = ~0.6 is very close to the value that gives the best overall performance using the PotE method at the 28 stations. However, the MAD of DC using the PotE method with α = 0.6 was still much larger than that using the sim_soilM method (cf. Tables 4 and 5 with Tables 2 and 3). From α = 1.0 to 0.6, both MADs and MDs of DC generally decreased at 18 of the 28 stations (Figs. 7 and 8). The reduction of the MADs and MDs of DC mainly occurred during the warm seasons by decreasing the surface evaporation (Fig. 9). However, during the cold seasons, the near-saturated DC values (moist bias) changed little.

Table 4.

The MAD of DC and MD (obs_soilM − PotE) averaged across 28 stations during 1 Feb 2010–31 Jan 2011 for PotE with different α values.

The MAD of DC and MD (obs_soilM − PotE) averaged across 28 stations during 1 Feb 2010–31 Jan 2011 for PotE with different α values.
The MAD of DC and MD (obs_soilM − PotE) averaged across 28 stations during 1 Feb 2010–31 Jan 2011 for PotE with different α values.
Table 5.

As in Table 4, but for 1 Aug 2009–31 Jul 2010.

As in Table 4, but for 1 Aug 2009–31 Jul 2010.
As in Table 4, but for 1 Aug 2009–31 Jul 2010.
Fig. 7.

(a) MAD and (b) MD of DC at 28 climate stations for PotE with α = 0.6 and 1.0 for 1 Feb 2010–31 Jan 2011.

Fig. 7.

(a) MAD and (b) MD of DC at 28 climate stations for PotE with α = 0.6 and 1.0 for 1 Feb 2010–31 Jan 2011.

Fig. 8.

As in Fig. 7, but for 1 Aug 2009–31 Jul 2010.

Fig. 8.

As in Fig. 7, but for 1 Aug 2009–31 Jul 2010.

Fig. 9.

Calculated DC values from obs_soilM and PotE with α = 0.6 and 1.0 (left) Paraparaumu and (right) Winchmore for (top) 1 Feb 2010–31 Jul 2011 and (bottom) 1 Aug 2009–31 Jul 2010.

Fig. 9.

Calculated DC values from obs_soilM and PotE with α = 0.6 and 1.0 (left) Paraparaumu and (right) Winchmore for (top) 1 Feb 2010–31 Jul 2011 and (bottom) 1 Aug 2009–31 Jul 2010.

However, if each station is considered separately, we found the MADs and MDs of DC (Figs. 7 and 8) are minimized for different values of α. For example, at Winchmore (station 19), the value of α with the smallest MAD of DC is ~0.6, but it is ~0.7 at Paraparamu (station 28) (Fig. 10). At Winchmore with α = 0.6 (Fig. 10a), the calculated DC is higher than the true values (obs_soilM) for some periods and lower than the true values for other periods during the warm seasons, indicating the optimum value for α also varies with time.

Fig. 10.

Calculated DC values from obs_soilM and PotE with α = 0.6 and 0.7 at (top) Winchmore and (bottom) Paraparaumu.

Fig. 10.

Calculated DC values from obs_soilM and PotE with α = 0.6 and 0.7 at (top) Winchmore and (bottom) Paraparaumu.

For the Old DC method, after changing the conditions for effective rain [Eq. (3)] for grasslands, slightly more precipitation entered the soil, which increased the soil water, leading to lower DC values. As a result, the MAD of DC is about 10 points higher, and the negative MD (moist bias) of DC is ~50 points higher (moister) for α = 1.0 (cf. Tables 2 and 3 with Tables 6 and 7 for Old DC).

Table 6.

The average values of MAD and MD of DC across 28 stations during 1 Feb 2010–31 Jan 2011 for Old DC with different α values.

The average values of MAD and MD of DC across 28 stations during 1 Feb 2010–31 Jan 2011 for Old DC with different α values.
The average values of MAD and MD of DC across 28 stations during 1 Feb 2010–31 Jan 2011 for Old DC with different α values.
Table 7.

As in Table 6, but for 1 Aug 2009–31 Jul 2010.

As in Table 6, but for 1 Aug 2009–31 Jul 2010.
As in Table 6, but for 1 Aug 2009–31 Jul 2010.

When α was increased from 1.0 to 1.4, the negative values of MD of DC for this Old DC method decreased dramatically from ~−123 to ~−40 (Tables 6 and 7). The MAD of DC also decreased, but by a relatively smaller amount (from ~140 to ~120). Increasing α further from 1.4 to 1.6 dramatically decreased the MD of DC to −9. However, the MAD of DC increased from ~120 to 127. Further increasing α to 1.8 increased both the MADs and MDs of DC, especially the latter. It seems that the value of α that gives the overall best performance of this method for DC calculation at the 28 stations is around 1.6 for the Old DC method. For α = 1.6, about half of the 28 stations slightly decreased the MADs of DC, while the other half slightly increased their MADs (Figs. 11 and 12), leading to little change in MAD for the average of the 28 stations compared with α = 1.0 (Tables 6 and 7). For the MDs of DC, about 20 of the 28 stations decreased their negative values (Figs. 11 and 12). These results indicate that the moist bias of the Old DC method can be decreased by increasing surface evaporation, but the errors in DC change little. The reduction in moist bias occurred mainly during warm seasons, while during the cold seasons the increase of surface evaporation for α = 1.6 barely changed the moisture status of the near-saturated soil (Fig. 13).

Fig. 11.

As in Fig.7, but for Old DC with α = 1.6 and 1.0.

Fig. 11.

As in Fig.7, but for Old DC with α = 1.6 and 1.0.

Fig. 12.

As in Fig. 11, but for 1 Aug 2009–31 Jul 2010.

Fig. 12.

As in Fig. 11, but for 1 Aug 2009–31 Jul 2010.

Fig. 13.

As in Fig. 9, but from obs_soilM and Old DC with α = 1.6 and 1.0.

Fig. 13.

As in Fig. 9, but from obs_soilM and Old DC with α = 1.6 and 1.0.

As for PotE, the optimum α for the Old DC method also varies with station and time of year. Thus, there is no single value of α for either the PotE or Old DC method that gives optimum performance at each of the 28 stations at all times of the year. This implies there are significant effects of soil texture differences, and meteorological factors such as wind speed and near-surface atmospheric stability, on the surface evaporation at the 28 stations. In other words, instead of being a constant parameter, α should be a function that varies with these effects. In addition, for both the PotE and Old DC methods, the apportioning of rainfall into runoff and increases in soil water is a very simple linear function of water amount [Eq. (3)]. In fact, this partitioning of precipitation is affected by canopy, soil texture, and soil water content. All of these processes are well treated in JULES and, as a result, the performance of the sim_soilM method is much better than that of either the Old DC or PotE methods (Tables 2, 3, 6, and 7).

5. Sensitivity to soil parameters

Section 3 showed that the sim_soilM DC method using simulated soil moisture is the best for determining DC and FWI values. However, given reliable observations of the surface meteorological factors needed by an LSM, the simulated soil moisture is sensitive to soil parameters (YY13). Uncertainties and errors in the soil parameters (such as saturated soil moisture, wilting soil moisture, etc.) can lead to large errors in simulated soil moisture. Errors in both saturated soil moisture and simulated soil moisture can lead to errors in the DC values calculated using Eq. (6). In this section, in addition to sim_soilM, we also conducted experiments referred to as OLD SOIL and NEW SOIL (Table 1) to investigate the sensitivity of DC calculations to soil parameters. As described earlier, sim_soilM employed the soil moisture as simulated by JULES, using more reliable soil parameters to derive DC values (YY13). In contrast, both OLD SOIL and NEW SOIL used the soil moisture simulated by JULES based on different soil parameters, as described in section 2d. Any differences in DC values calculated using Eq. (6) obtained from these three experiments at each station are therefore solely due to differences in the soil parameters.

Among the three experiments (Fig. 14), only very small differences in DC were found during the cold seasons. However, relatively large differences (up to 300 points) can be found for some periods during the warm seasons (Fig. 14). The highest DC values were calculated for sim_soilM, and the lowest were for NEW SOIL, with OLD SOIL in the middle. In fact, for both NEW SOIL and OLD SOIL, larger MADs of DC mainly occurred during warm seasons at most of the stations (Tables 2 and 3). This indicates that the sim_soilM DC method is sensitive to soil parameters mainly in warm seasons. DC values calculated using NEW SOIL were generally smaller than those from sim_soilM or OLD SOIL during the warm seasons (Tables 2 and 3). This is due to the soil parameters in NEW SOIL allowing the soil to store more soil moisture (i.e., become wetter).

Fig. 14.

Calculated DC for sim_soilM, NEW SOIL, and OLD SOIL at (left) Paraparaumu and (right) Winchmore from (top) 1 Feb 2010 to 31 Jan 2011 and (bottom) 1 Aug 2009 to 31 Jul 2010.

Fig. 14.

Calculated DC for sim_soilM, NEW SOIL, and OLD SOIL at (left) Paraparaumu and (right) Winchmore from (top) 1 Feb 2010 to 31 Jan 2011 and (bottom) 1 Aug 2009 to 31 Jul 2010.

The yearly mean DC values from both NEW SOIL and OLD SOIL at most stations were generally smaller than from sim_soilM (Figs. 15a and 16a). At most of the 28 stations, the difference in the yearly mean DC calculated using the three different soil types at a station was much smaller than that calculated using the three DC methods for the same soil type at the same station (cf. Figs. 15a and 16a with Figs. 3a and 4a). This indicates that the DC calculation is more sensitive to the DC methods than to soil types in this study. The variation in DC among the 28 stations could be as large as ~400 points. This is larger than the maximum difference (~220 points) between the three soil experiments, indicating that the sim_soilM DC method is more sensitive to local climate than to soil types.

Fig. 15.

(a) Mean DC, (b) MAD of DC, and (c) MD of DC at 28 climate stations for the NEW SOIL, OLD Soil, and sim_soilM for 1 Feb 2010–31 Jan 2011.

Fig. 15.

(a) Mean DC, (b) MAD of DC, and (c) MD of DC at 28 climate stations for the NEW SOIL, OLD Soil, and sim_soilM for 1 Feb 2010–31 Jan 2011.

Fig. 16.

As in Fig. 15, but for 1 Aug 2009–31 Jul 2010.

Fig. 16.

As in Fig. 15, but for 1 Aug 2009–31 Jul 2010.

Figures 15b and 16b show the yearly MAD of DC for the three experiments. Even though the soil parameters for OLD SOIL had larger errors than those for sim_soilM, the overall MAD and MD results of DC for OLD SOIL were very close to those for sim_soilM. It is interesting to note that OLD SOIL was better than the Old DC and PotE methods in terms of yearly mean MAD and MD (Tables 2 and 3), while NEW SOIL was better than PotE and almost the same as Old DC. As described in YY13, for the average of the yearly MAD of simulated soil moisture at the 28 stations, NEW SOIL and OLD SOIL were significantly larger than sim_soilM. This indicates that the sensitivity of DC to soil parameters is less than that of soil moisture to soil parameters. The effect of errors in the soil parameters on simulated soil moisture was reduced in the calculation of DC by using the sim_soilM DC method, indicating the potential skill of the proposed sim_soilM method in determining drought status and for FWI calculation and forecasting.

Please note that for the three soil experiments, we used exactly the same surface meteorological observations to run JULES to simulate the soil moisture. For a forecast system with the atmospheric model coupled with an LSM, errors in soil parameters can affect the simulated soil moisture and the surface thermal forcing, and thus affect the simulated surface meteorological conditions in New Zealand (Yang et al. 2011). Errors in the latter would in turn feed back onto the simulated soil moisture. Thus, if using sim_soilM for DC and FWI prediction, the sensitivity of DC to soil parameters obtained by using simulated surface meteorological conditions may be slightly different from that obtained by using observations.

6. Summary

The Canadian FWI System developed in the early 1970s for the boreal temperate forest environment with deep organic soil has been used in other countries with almost no modification, and not only used for forest but also for other vegetation types such as grasslands. This raises the issue of whether components of the FWI System need to be modified for different vegetation and climate regions.

In this study, we focused on the DC component of the FWI System, a daily index of water stored in the soil. We proposed two alternative new methods for improving the DC calculation. The first of these, PotE, employs PE calculated from an LSM (JULES) that includes the effects of wind speed, surface air stability, and the vertical gradient of surface air humidity on evaporation. The second method, soilM (including use of either observed (obs_soilM) or simulated (sim_soilM) soil moisture), uses the ratio of saturated soil moisture content to the actual soil moisture content.

Method soilM has three advantages:

  1. The effect of different soil types is considered by using saturated soil moisture content, which changes for different soils.

  2. When the soil moisture is based on observations, the calculated DC represents the actual drought status and can be used for verification.

  3. The absolute values of DC are comparable regarding drought status for different areas with different soil types.

DC and FWI values were calculated using the original DC method (Old DC), and the proposed new PotE, obs_soilM and sim_soilM methods at 28 climate stations over grasslands of New Zealand for two separate 2-yr periods. The spinup period for initializing the DC and FWI values and removing the effects of the starting values may take up to half a year.

The original DC method underestimated the drought status (by up to 300 points) with a moist bias, especially in summer. This led to the underestimation of FWI values. For PotE, DC values were much overestimated with a dry bias, especially in summer. For the Old DC method, PE was simply calculated using air temperature and day-length factor, as described earlier. For both Old DC and PotE, the calculation of surface evaporation uses a simple linear function of PE and soil water content. Meteorological factors such as wind speed and surface air stability, and differences in canopy and soil texture, were not considered. Sensitivity tests for surface evaporation showed that at the 28 stations, failure to consider these factors in the calculations of surface evaporation was a key factor contributing to quite large MAD values of DC for the Old DC and PotE methods. In addition, for the Old DC and PotE methods, the partitioning of water reaching the surface between runoff and increases to soil water is a very simple linear function of water amount whereas, in reality, this split of water is affected by the canopy, soil texture, and soil water content. These processes are well treated in JULES, and as a result the errors in the calculated drought status and FWI values in the Old DC and PotE methods were largely reduced by use of simulated soil moisture as in the sim_soilM method.

Although soil moisture observations are rare, simulated soil moisture can be easily obtained by running an LSM using surface observations or analyses. However, the soil moisture simulated by an LSM is very sensitive to soil types. Errors in the soil parameters can cause quite large errors in the simulated soil moisture. Our sensitivity tests for different soil parameters showed that the DC calculated using sim_soilM was less sensitive to soil parameters than that of soil moisture itself to soil parameters. The effect of soil parameter errors on simulated soil moisture was reduced in the calculation of DC using the sim_soilM DC method, indicating the potential skill of the proposed sim_soilM method in monitoring drought status and for FWI System calculations and forecasts. However, for more accurate calculations of DC using a sim_soilM method, more reliable soil parameters for a LSM are required, because the differences in the DC values for the three soil tests compared in the study were significant (up to 300 points during some warm seasons).

Comparison of the FWI values calculated using all the DC methods with fire occurrence was not possible because of a lack of suitable data. There were no records of wildfires around any of the 28 station during the time periods of the data used in this study. In addition, for all available wildfire cases in New Zealand, surface radiation data, needed to run JULES for PotE and sim_soilM, were not available. This comparison work will be done in the future if suitable data become available.

In this study, all experiments for the DC and FWI calculations were based on observations from the 28 climate stations. Their land surface environments are typically grasslands and scrublands, and the land mainly consists of mineral soil. Thus, the results from this study should mainly be applied to areas that are covered by these two vegetation types and where the soil is predominantly mineral. Because of a lack of meteorological observations and soil parameters to run JULES in New Zealand forests, similar experiments have not been performed for forests with deeper organic layers more akin to the reference FWI System fuel type at the present time. However, for forests where the land is mainly dominated by mineral soil, the results from this study may also serve as a reference.

Acknowledgments

This research was carried out under Research Collaboration SC0128 with the Met Office and funded by NIWA under its Hazards Research Programme (2012/13 SCI), and partially supported by the New Zealand Fire Service Commission’s Contestable Research Fund project Improving Forecasts of Fire Danger with New Coupled Weather and Land Models.

REFERENCES

REFERENCES
Alexander
,
M. E.
,
2008
: Proposed revision of fire danger class criteria for forest and rural areas in New Zealand. 2nd ed. National Rural Fire Authority Rep., 62 pp. [Available online at http://www.scionresearch.com/__data/assets/pdf_file/0012/4620/Fire-Danger-Classification-Criteria-reprint_final-corr.pdf.]
Alexander
,
M. E.
, and
F. V.
Cole
,
1995
: Predicting and interpreting fire intensities in Alaskan black spruce forests using the Canadian system of fire danger rating. Managing Forests to Meet People’s Needs: Proceedings of the 1994 Society of American Foresters/Canadian Institute of Foresters Convention, SAF Publ. 95-02, 185–192.
Amiro
,
B. D.
,
K. A.
Logan
,
B. M.
Wotton
,
M. D.
Flannigan
,
J. B.
Todd
,
B. J.
Stocks
, and
D. L.
Martell
,
2004
:
Fire Weather Index System components of large fires in the Canadian boreal forest
.
Int. J. Wildland Fire
,
13
,
391
400
, doi:.
Anderson
,
S.
,
2005
: Forest and rural fire danger rating in New Zealand. Forestry Handbook, M Colley, Ed., New Zealand Institute of Forestry, 241–244.
Anderson
,
S.
,
2009
:
Future options for fire behaviour modelling and fire danger rating in New Zealand
.
Proc. Royal Soc. Queensl.
,
115
,
119
127
.
Anderson
,
S.
, and
W. R.
Anderson
,
2010
:
Ignition and fire spread thresholds in gorse (Ulex europaeus)
.
Int. J. Wildland Fire
,
19
,
589
598
, doi:.
Anderson
,
S.
,
J. J.
Doherty
, and
H. G.
Pearce
,
2008
:
Wildfires in New Zealand from 1991 to 2007
.
N. Z. J. Forestry
,
53
,
19
22
.
Bedia
,
J.
,
S.
Herrera
,
J. M.
Gutiérrez
,
G.
Zavala
,
I. R.
Urbieta
, and
J. M.
Moreno
,
2012
:
Sensitivity of fire weather index to different reanalysis products in the Iberian Peninsula
.
Nat. Hazards Earth Syst. Sci.
,
12
,
699
708
, doi:.
Bergeron
,
Y.
,
S.
Gauthier
,
M.
Flannigan
, and
V.
Kafka
,
2004
:
Fire regimes at the transition between mixed wood and coniferous boreal forest in northwestern Quebec
.
Ecology
,
85
,
1916
1932
, doi:.
Best
,
M. J.
, and Coauthors
,
2011
:
The Joint UK Land Environment Simulator (JULES), model description—Part 1: Energy and water fluxes
.
Geosci. Model Dev.
,
4
,
677
699
, doi:.
Brenner
,
J.
,
L. G.
Arvanitis
,
D. P.
Brackett
,
B. S.
Lee
,
R. J.
Carr
, and
R. M.
Suddaby
,
1998
: Integrating GIS, mesoscale fire weather prediction, smoke plume dispersion modeling, and the Internet for enhanced open burning authorizations and wildfire response in Florida. Proc. Int. Conf. on Geospatial Information in Agriculture and Forestry, Vol. I, Lake Buena Vista, FL, ERIM International,
140
147
.
Camia
,
A.
,
P.
Barbosa
,
G.
Amatulli
, and
J.
San-Miguel-Ayanz
,
2006
: Fire danger rating in the European Forest Fire Information System (EFFIS): Current developments. For. Ecol. Manage.,
234
(Suppl.), S20, doi:.
Cox
,
P. M.
,
R. A.
Betts
,
C. B.
Bunton
,
R. L. H.
Essery
,
P. R.
Rowntree
, and
J.
Smith
,
1999
:
The impact of new land surface physics on the GCM simulation of climate and climate sensitivity
.
Climate Dyn.
,
15
,
183
203
, doi:.
de Groot
,
W.
,
R.
Field
,
A.
Brady
,
O.
Roswintiarti
, and
M.
Mohamad
,
2007
:
Development of the Indonesian and Malaysian fire danger rating systems
.
Mitigation Adapt. Strategies Global Change
,
12
,
165
180
, doi:.
Dimitrakopoulos
,
A.
,
A.
Bemmerzouk
, and
I.
Mitsopoulos
,
2011
:
Evaluation of the Canadian Fire Weather Index System in an eastern Mediterranean environment
.
Meteor. Appl.
,
18
,
83
93
, doi:.
Dowdy
,
A. J.
,
G. A.
Mills
,
K.
Finkele
, and
W.
de Groot
,
2010
:
Index sensitivity analysis applied to the Canadian forest fire weather index and the McArthur forest fire danger index
.
Meteor. Appl.
,
17
,
298
312
, doi:.
Essery
,
R. L. H.
,
M. J.
Best
,
R. A.
Betts
,
P. M.
Cox
, and
C. M.
Taylor
,
2003
:
Explicit representation of subgrid heterogeneity in a GCM land surface scheme
.
J. Hydrometeor.
,
4
,
530
543
, doi:.
Fogarty
,
L. G.
,
H. G.
Pearce
,
W. R.
Catchpole
, and
M. E.
Alexander
,
1998
: Adoption vs. adaptation: Lessons from applying the Canadian Forest Fire Danger Rating System in New Zealand. Proc. III Int. Conf. on Forest Fire Research/14th Conf. on Fire and Forest Meteorology, Vol. I, Luso, Portugal, University of Coimbra, 1011–1028.
Girardin
,
M. P.
, and
M.
Mudelsee
,
2008
:
Past and future changes in Canadian boreal wildfire activity
.
Ecol. Appl.
,
18
,
391
406
, doi:.
Girardin
,
M. P.
, and
B. M.
Wotton
,
2009
:
Summer moisture and wildfire risks across Canada
.
J. Appl. Meteor. Climatol.
,
48
,
517
533
, doi:.
Girardin
,
M. P.
,
J.
Tardif
,
M. D.
Flannigan
,
B. M.
Wotton
, and
Y.
Bergeron
,
2004
:
Trends and periodicities in the Canadian drought code and their relationships with atmospheric circulation for the southern Canadian boreal forest
.
Can. J. For. Res.
,
34
,
103
119
, doi:.
Johnson
,
E. A.
,
D. M.
Keith
, and
Y. E.
Martin
,
2013
:
Comparing measured duff moisture with a water budget model and duff and drought codes of the Canadian fire weather index
.
For. Sci.
,
59
,
78
92
, doi:.
Lee
,
B. S.
,
M. E.
Alexander
,
B. C.
Hawkes
,
T. J.
Lynham
,
B. J.
Stocks
, and
P.
Englefield
,
2002
:
Information systems in support of wildland fire management decision making in Canada
.
Comput. Electron. Agric.
,
37
,
185
198
, doi:.
Muraro
,
S. J.
, and
B. D.
Lawson
,
1970
: Prediction of duff moisture distribution for prescribed burning. Canadian Forestry Service Dept. of Fisheries and Forestry Information Rep. BC-X-46, 13 pp.
Taylor
,
S. W.
, and
M. E.
Alexander
,
2006
:
Science, technology, and human factors in fire danger rating: The Canadian experience
.
Int. J. Wildland Fire
,
15
,
121
135
, doi:.
Turner
,
J. A.
,
1972
: The drought code component of the Canadian forest fire behaviour system. Canadian Forest Service Publ. 1316, 14 pp.
Van Wagner
,
C. E.
,
1970
: An index to estimate the current moisture of the forest floor. Canadian Forestry Service Dept. of Fisheries and Forestry Publ. 1288, 23 pp.
Van Wagner
,
C. E.
,
1987
: Development and structure of the Canadian forest fire weather index. Canadian Forestry Service Forestry Tech. Rep. 35, 37 pp. [Available online at http://cfs.nrcan.gc.ca/pubwarehouse/pdfs/19927.pdf.]
Viegas
,
D.
,
G.
Bovio
,
A.
Ferreira
,
A.
Nosenzo
, and
B.
Sol
,
1999
:
Comparative study of various methods of fire danger evaluation in southern Europe
.
Int. J. Wildland Fire
,
9
,
235
246
, doi:.
Webster
,
S.
,
A. R.
Brown
,
C. P.
Jones
, and
D. R.
Cameron
,
2003
:
Improvements to the representation of orography in the Met Office Unified Model
.
Quart. J. Roy. Meteor. Soc.
,
129
,
1989
2010
, doi:.
Wilmore
,
B.
,
2001
: Duff moisture dynamics in black spruce–feather moss stands and their relation to the Canadian Forest Fire Danger Rating System. M.S. thesis, School of Natural Resources and Extension, University of Alaska Fairbanks, 84 pp. [Available online at http://depts.washington.edu/nwfire/publication/Wilmore_2001.pdf.]
Wotton
,
B. M.
,
2009
:
Interpreting and using outputs from the Canadian Forest Fire Danger Rating System in research applications
.
Environ. Ecol. Stat.
,
16
,
107
131
, doi:.
Wotton
,
B. M.
,
B. J.
Stocks
, and
D. L.
Martell
,
2005
:
An index for tracking sheltered forest floor moisture within the Canadian Forest Fire Weather Index System
.
Int. J. Wildland Fire
,
14
,
169
182
, doi:.
Yang
,
Y.
,
M.
Uddstrom
,
M.
Revell
,
P.
Andrews
,
H.
Oliver
,
R.
Turner
, and
T.
Carey-Smith
,
2011
:
Numerical simulations of effects of soil moisture and modification by mountains over New Zealand in summer
.
Mon. Wea. Rev.
,
139
,
494
510
, doi:.
Yang
,
Y.
,
M.
Uddstrom
,
M.
Revell
, and
S.
Moore
,
2013
: Soil moisture simulation by JULES in New Zealand: Verification and sensitivity tests. Meteor. Appl.,21, 888–897, doi:.

Footnotes

*

Current affiliation: Rural Fire Research Group, Scion, Christchurch, New Zealand.