Ecological Modelling 109 (1998) 77 – 98
Sensitivity of a biogeography model to soil properties D. Bachelet a,*, M. Brugnach b, R.P. Neilson c a
Department of Bioresource Engineering, Oregon State Uni6ersity, 97330 Cor6allis, Oregon, USA b Department of Forest Science, Oregon State Uni6ersity, 97330 Cor6allis, Oregon, USA c U.S. Department of Agriculture Forest Ser6ice, 3200 SW Jefferson, 97330 Cor6allis, Oregon, USA Received 21 March 1997; accepted 16 December 1997
Abstract This paper presents the changes in vegetation distribution and hydrological balance resulting from a change in soils input data to the biogeography model MAPSS (Neilson, 1995). The model was run for the conterminous United States using three different sets of soil characteristics: (1) all soils were assumed to be sandy loam; (2) soils characteristics came from the Food and Agriculture Organization (FAO) soils map of the world (FAO, 1974 – 1979) and (3) soil characteristics came from the Natural Resource Conservation Service (NRCS) National Soil Geographic (NATSGO) dataset. Resulting changes in vegetation distribution appear small on a country-wide basis, but large changes in simulated runoff in savannas, shrublands and deserts reflect the importance of using the best available soils dataset. In the state of Oregon, a 16% relative decrease in forest areal extent is accompanied by an 18% relative increase in shrubland when switching from FAO to NATSGO datasets. Conversely, forest cover increases by 24% while shrubland extent decreases by 14% when all Oregon soils are assumed to be sandy loam. MAPSS vegetation distribution projections were compared to Ku¨chler’s potential vegetation map (Ku¨chler, 1964). The generalization of all US soils to sandy loam soils decreases the similarity between MAPSS predictions and Ku¨chler’s map and is clearly inappropriate. If the similarity between MAPSS projections and Ku¨chler’s map does not clearly improve by using NATSGO rather than FAO soils data, NATSGO soil representation is more reliable and thus we recommend using it in the future. © 1998 Elsevier Science B.V. All rights reserved. Keywords: Simulation; Soil texture; Soil depth; Rock fragment; Vegetation distribution; MAPSS
1. Introduction
* Corresponding author. Forestry Sciences Laboratory, 3200 SW Jefferson Way, Corvallis OR 97330, USA. Tel: +1 541 7587754; fax: +1 541 7507329; e-mail:
[email protected]
While numerous studies have illustrated the influence of nutrient content on vegetation distribution (e.g. Goldberg, 1982; Austin et al., 1990), few have demonstrated the influence of soil physical properties. Soil physical characteristics influence water availability by determining the
0304-3800/98/$19.00 © 1998 Elsevier Science B.V. All rights reserved. PII S 0 3 0 4 - 3 8 0 0 ( 9 8 ) 0 0 0 1 0 - 6
78
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
capacity of the soil to retain water. Brush et al. (1980) found that determining soil water availability was essential to understanding the distribution of mature forest types across Maryland. Aber et al. (1982) documented, in the prairie peninsula region in Wisconsin, a gradient from silt clay loams to sandy loams along which mean leaf height, canopy layering and species composition changed continuously. Soil physical characteristics also determine how water is distributed in the soil profile and how easily plants can access it. Perry (1994) mentioned the existence of ponderosa pine on old fissured lava flows that allow deep roots to access stored water. Clearly it is important to include soils in vegetation distribution models. Kern (1995) analyzed the USDA national soil data base (NATSGO) and the United Nations soil map of the World (FAO, 1974 – 1979) and recommended using NATSGO data for their accuracy over the conterminous United States of America. Questions were then raised about simulation model sensitivity to soil characterization. MAPSS (Neilson and Marks, 1994; Neilson, 1995) is a biogeography model that requires monthly precipitation, mean monthly temperature, water vapor, wind, elevation, and soil characteristics (depth, sand, clay and rock fragment content) to predict potential vegetation distribution. It includes a water submodel that calculates plant available water and a rule-based submodel that determines the climatic zone, the lifeform and the plant type as a function of temperature thresholds and water availability. The maximum potential leaf area index (LAI) a site can support is calculated iteratively allowing grasses and trees to compete for and use up all of the site available soil water while shading by trees limits grass growth. It also includes a fire submodel that maintains transition zones such as the prairie peninsula. MAPSS was run for the conterminous USA at a 10-km resolution using three sets of soil physical characteristics (1) one corresponding to a sandy loam soil over the entire map as a first approximation to simplify soil representation (Neilson, 1995); and two other datasets derived and discussed by Kern (1995): (2) from the USDA national soil data base (NATSGO) and (3) the
United Nations soil map of the World (FAO). Changes in vegetation types, runoff, actual evapotranspiration and evaporation were analyzed to evaluate the sensitivity of the model to soil properties.
2. Materials and methods
2.1. NATSGO The area boundaries of the National Soil Geographic (NATSGO) data base are those of the Major Land Resource Area (MLRA) and Land Resource Region (LRR). The MLRAs are a system of land resource units, each with characteristic patterns of soils, climate, water resources, and physiographic features (Soil Conservation Service, 1981). The map was digitized at a scale of 1:7 500 000. The base map used is a 1970 Census Bureau state and county digital data base, Albers Equal Area projection. Map unit composition was determined by sampling done as part of the 1982 National Resources Inventory (NRI) (SCS, 1987). NRI is an inventory of land cover and use, soil erosion, prime farmland, wetlands, and other natural resource characteristics on non-Federal rural land in the United States (Reybold and TeSelle, 1989). These inventories are conducted every 5 years by the USDA’s Natural Resources Conservation Service (NRCS). The 1982 NRI is the most extensive inventory yet conducted, covering over 800 000 sample sites on non-Federal land. Nonfederal land was assumed to be representative of the Federal land (which was not sampled) within the same MLRA, each sample point having an expansion factor indicating how much non-Federal land it represents in the MLRA (Kern, 1995). NATSGO is linked to a Soil Interpretation record (SIR) attribute data base through the NRI data base. The data required by the vegetation distribution model are sand and clay contents, rock fragments and depth to bedrock. If sand and clay values were missing, values were assigned based on the centroid values of the textural classes listed first for those horizons (Kern, 1995). Geographic digital data layers were produced by Kern (1995) using expansion factors to derive area-
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
weighted means for the various soil properties from the 1982 National Resources Inventory using the sample points that were not reported as urban land, water and miscellaneous areas.
2.2. FAO The Food and Agriculture Organization (FAO) of the United Nations Educational and Scientific Cultural Organization (UNESCO) Soil Map of the World was published between 1971 and 1979 at the original scale of 1:5 000 000 (FAO, 1974 – 1979). It is the most detailed inventory at the global scale and is based on a compilation of national soil maps with some additional field data collected by FAO personnel. The digital soil map of the world (version 3.0) was released in May 1994 in the Geographic projection (Latitude – Longitude). The legend of the original Soil Map of the World (FAO, 1974 – 1979) includes an estimated 4930 different map units which consist of soils units or associations. There are 106 soil units grouped in 26 major soil groupings. Geographic digital data layers were produced by Kern (1995) for the United States. Example pedons or ‘typical profiles’ (26 for the United States) with laboratory data accompanying the map were used by Kern (1995) to make soil texture estimates. Soil depth was based on the definitions in the explanatory texts and was assumed to be 203 cm unless a phase was indicated on the map. Stoney and petri phases were assumed to indicate 35 and 40% rock content, respectively. The methodology used by FAO for each map unit was developed in the Agro-ecological Zones Project (FAO, 1978). Soil unit properties were combined with soil composition data to calculate area-weighted averages for each polygon. Kern (1995) notes that ‘‘due to the many data sources, no quantitative statement can be made about data quality expressed as precision or accuracy’’.
2.3. Hydrology and soil characteristics in MAPSS MAPSS calculates the leaf area index (LAI) of the different lifeforms it includes (tree, shrub and grasses) assuming that the potential vegetation
79
that is projected for the site corresponds to a mature ‘climax’ system. An iterative process calculates the maximum LAI corresponding to a lifeform or a mixture of lifeforms that can use up all available water during the driest period of the year at the site. LAI thresholds are then used to determine particular vegetation types. To understand how soil characteristics may affect MAPSS projections of vegetation distribution, a brief presentation of the main equations that determine soil water availability to plants follows. Sources of water in the model are rainfall and snow. A fraction of the rainfall is intercepted and evaporated by the canopy (eq. 1.2 in Neilson, 1995) while the rest becomes throughfall. Snowmelt (eq. 1.4 in Neilson, 1995) is also added to throughfall. Throughfall is then partitioned into surface runoff and infiltration depending on soil water content (eq. 1.5 and 1.6 in Neilson, 1995). The model includes three soil layers, two of which include roots. Layers 1 and 2 are 1.5 m deep. Trees and shrubs can access water from both layers 1 and 2 while grasses only access water in the top soil layer. Infiltration is partitioned into saturated and unsaturated percolation (eq. 1.7 and 1.8 in Neilson, 1995) using analogues of Darcy’s Law. For water to infiltrate from one layer to the next, soil water content in the lower layer must be less than its saturated water holding capacity. Saturated water holding capacity (swhc) varies with soil texture for each soil layer (equation 7 in Saxton et al., 1986) as follows: swhc= eff – thickness× WS(layer)
(1)
where swhc is saturated water holding capacity; WS is soil water content of the fine earth fraction at saturation (m3/m3); layer is soil layer; eff – thickness is soil effective thickness (m). eff – thickness= depth× (1− rock – frag)
(2)
where eff – thickness is soil effective thickness (m); depth is soil depth (m); rock – frag is rock fragment content. WS(layer)= h+j ×sand+ k× log(clay)
(3)
where WS is soil water content at saturation (m3/m3); layer is soil layer; sand is percent sand;
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
80
clay is % clay; h, j and k are coefficients defined in Saxton et al. (1986) as 0.332, −7.251 × 10 − 4 and 0.1276, respectively. To estimate saturated and unsaturated percolation, field capacity and wilting point are calculated. They are defined as follows according to equations 5 and 6 in Saxton et al. (1986):
where FC is field capacity; eff – thickness is soil effective thickness (Eq. (2)); FP is field potential which is a fixed variable, identical for all soil layers (− 0.033 MPa); parameters aa and bb are defined below.
where Qu is unsaturated percolation; W is current soil water content; WP is soil water content at wilting point (Eq. (5)); WS is soil water content at saturation (Eq. (3)); K1u and K2u are constants calibrated separately for each soil layer and thus implicitly carry soil thickness information in their value. After infiltration but before percolation, transpiration by both woody and grass vegetation occurs. Indirectly, transpiration is also dependent on soil physical characteristics. For each lifeform, actual transpiration increases as a function of LAI and canopy conductance (eq. 1.9. in Neilson, 1995).
WP = eff – thickness×(MP/aa)1/bb
AT= PET× (1− exp(− Ka× Cc))
FC= eff – thickness× (FP/aa)1/bb
(4)
(5)
where WP is wilting point; eff – thickness is soil effective thickness (Eq. (2)); MP is matrix potential, a fixed variable identical for all soil layers ( − 1.5 MPa); parameters aa and bb are defined below. aa = 100× exp(a +b ×clay +c ×sand 2 +d × clay×sand 2)
(6)
where a, b, c, d are defined in Saxton et al. (1986) as −4.396, −0.0715, −4.88 ×10 − 4 and − 4.285×10 − 5 respectively.
(10)
where AT is actual transpiration; PET is potential evapotranspiration (turbulent transfer model by Marks, 1990); Ka is a constant translating actual LAI into canopy effective LAI as a function of light, wind and humidity in the canopy; Cc is canopy conductance. Canopy conductance is represented as the product of stomatal conductance and canopy LAI (eq. 1.9 in Neilson, 1995). Cc= (LAI/LAIm)× (Cs/Cm)
(11)
where e, f, g are defined in Saxton et al. (1986) as − 3.14, −2.22×10 − 3 and − 3.484 ×10 − 5. Saturated and unsaturated percolation are then calculated as follows:
where Cc is canopy conductance; LAIm is maximum LAI; Cs is stomatal conductance; Cm is maximum stomatal conductance. Stomatal conductance decreases with decreasing soil water potential and increasing potential evapotranspiration (PET) (eq. 1.10 in Neilson, 1995).
Qs= (W− FC)×K1s
Cs= ((a 2 × PSI 2 − 4b× PSI 2 + 4c)1/2 + a× PSI)
bb =e+f× clay 2 + g × sand 2 +g ×clay ×sand 2 (7)
×((W −FC)/(WS −FC))K2s
(8)
where Qs is saturated percolation; W is current soil water content; FC is soil water content at field capacity (Eq. (4)); WS is soil water content at saturation (Eq. (3)); K1s and K2s are constants calibrated separately for each soil layer and thus implicitly carry soil thickness information in their value. Qu =(W−WP)× K1u ×((W −WP)/(WS − WP))K2u
(9)
/2
(12)
where Cs is stomatal conductance; PSI is soil water potential; a, b and c are defined below. a=0 if PET is below a set threshold or a= (PET−threshold)× as if PET is above the set threshold
(13)
where as is a parameter controlling the sensitivity of stomatal conductance to PET above the set threshold.
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
b= (Cm/WP)2
(14)
c = Cm 2
(15)
where Cm is maximum stomatal conductance; WP is wilting point (Eq. (5)). Equation 6 from Saxton et al. (1986) is used to calculate soil water potential: PSI = aa×PSW bb
(16)
where PSI is soil water potential; PSW is percent soil water; aa and bb are constants defined in Eqs. (6) and (7). In conclusion, soil characteristics are used to define soil water potential which is used to calculate transpiration which affects the water balance and thus the projected vegetation type. Soil characteristics also determine field capacity and wilting point that are used to calculate percolation. So in the model, soil physical characteristics are important to determine both biological and physical flow of water.
2.4. Vegetation types in MAPSS MAPSS defines vegetation types based on climate and LAI thresholds. The model first determines the climatic zone for each pixel. The boreal zone is defined as the region where mean monthly temperature is below − 16°C, which is equivalent to a region where absolute minimum daily temperatures of −40°C occur (Prentice et al., 1992); −40°C corresponds to the supercooled freezing point of water and the northern limit of most temperate deciduous trees. The tropical zone is defined as a zone where freezing (minimum daily temperature below 0°C) never occurs. The subtropical zone is defined by mean monthly temperatures greater than 2.25°C which means that there cannot be any hard frost (maximum daily temper-
81
Table 1 Definition of MAPSS vegetation classes — Energy limited classes
Taiga-tundra Tundra
Boreal zone
Other zones
367.5BGDDB665 GDDB367.5
307.5BGDDB582.5 GDDB307.5
GDD, growing degree days (base 0°C).
ature below 0°C) in that zone. The temperate zone is defined as that between the boreal and subtropical zones. Taiga-tundra and tundra are energy limited vegetation types and are defined independently of LAI by thresholds in growing degree days (GDD). In the boreal zone, taiga-tundra is delimited by two GDD thresholds of 665 and 367.5 below which tundra itself exists. In the temperate zone, those thresholds are 582.5 and 307.5 (Table 1). These threshold values have been specifically calibrated to the climatic dataset used to run the model. The rest of the vegetation classification in MAPSS is based on the presence/absence and LAI values of three types of lifeforms: trees, shrubs and grasses. The woody components, trees or shrubs, are assumed to be dominant and mutually exclusive. Closed canopy above an LAI of 6 assumes no grass component while in open canopies, grasses are allowed to compete for water with trees or shrubs. Closed canopy forests are defined when tree LAI is greater than 3.75. Closed canopy shrublands have shrub LAIs greater than 0.7 in tropical and subtropical zones, 0.175 in the two other zones. Open canopy or tree savannas have LAIs below 3.75 but greater than 1.5 except in the tropical zone where the minimum value equals 0.65 (Table 2).
Table 2 Definition of MAPSS vegetation classes—Forests and savannas
Forest Savanna
Tree LAI (boreal/temperate/subtropical zones)
Tree LAI (tropical zone)
Grass LAI
LAI\3.75 1.5BLAIB3.75
LAI\3.75 0.65BLAIB3.75
0 Variable
Savanna includes forest savannas and tree savannas as defined in MAPSS (Neilson, 1995).
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
82
Table 3 Definition of MAPSS vegetation classes—Shrublands, grasslands and deserts Tropical and subtropical zones
Shrubland Grassland Desert
Boreal and temperate zones
All zones
Grass LAI
LAD
Shrub LAI
Total LAI
Shrub LAI
Grass LAI
\1.5 \1.5 \1.5 \1.5
B3 B3 B3 B1.2
\0.7 B0.7 B0.7 B0.7
\0.5 B0.5 B0.5 B0.5
\0.175 B0.175 B0.175 B0.175
Variable \0.1 B0.1 \0.1
LAD or leaf area duration is defined as the area under the growing season grass LAI curve and is used in the model as an index of productivity.
Shrublands are assumed to have shrub LAIs below that of tree LAIs which means 0.65 in the tropical zone and 1.5 in the three other zones. Shrub LAI however are above 0.7 in tropical and subtropical zones and above 0.175 in boreal and temperate zones. LAI values are calculated to maximize the water use by the vegetation. Trees and shrubs are given different roughness lengths so that shrub LAI can be greater than tree LAI but still only use the same amount of water which is why the lower limit of forest LAI is less than shrubland LAI in tropical zones (Table 3). If shrub LAI drops below a specified level (0.175 in boreal and temperate zones, 0.7 in tropical and subtropical zones) and grass LAI drops below 0.1, the site is considered a desert. Leaf area duration (LAD) is defined as the area under the growing season grass LAI curve and is used in the model as an index of productivity. If grass LAI remains above 0.1 but productivity (LAD) is below 1.2, the site is still classified as desert (Table 3). If shrub LAI falls below 0.7 in tropical and subtropical zones or 0.175 in boreal and temperate zones, and if grass LAI remains above 0.1 with LAD above 1.2, the site is considered a grassland (Table 3). MAPSS vegetation classes are further refined by using specific LAI thresholds for each landcover type. To simplify the results, we have used broad vegetation categories: forests, savannas, shrublands, grasslands, deserts, tundra, as defined above (Tables 1–3), rather than MAPSS actual vegetation types.
3. Results
3.1. Conterminous USA 3.1.1. NATSGO 6s FAO soils data Major differences between FAO and NATSGO data sets are obvious when one compares the maps of rock fragment and depth-to-bedrock (Figs. 1 and 2); 89% of the United States has been assigned a rock fragment content of 0% in the FAO dataset versus 53% in the NATSGO dataset. Moreover, NATSGO rock fragment content exists for two soil horizons (0–50 and 50–150 cm depth) rather than the whole profile like in the FAO dataset (Fig. 1). The mapped depth-tobedrock varies in both cases from 0 (43% of the country) to 1.5 m. In the case of the FAO data 42% of the country has a depth-to-bedrock between 1.45 and 1.50 m while, with the NATSGO data, only 7% of the country has such deep soils (Fig. 2). Soil parameters per se were not defined in model runs using sandy-loam soils for the entire United States. Key hydrologic variables such as soil water holding capacity for each soil layer, field capacity (60%), matrix potential (− 1.5 bar) and field potential ( − 0.033 bar) were defined in the site-specific input file. Table 4 illustrates the differences in texture between NATSGO and FAO data sets for the projected vegetation categories. In deserts, soil clay content is at least 10% greater in both surface and deeper soil layers at the expense of the sand content in the NATSGO data set. Similarly in shrublands, clay content is 6 and 12% greater in
Fig. 1. Rock fragment content of mineral soils in the conterminous United States; top, FAO; bottom, NATSGO data for the first and second soil layer.
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98 83
Fig. 2. Depth to bedrock for mineral soils in the conterminous United States; top, FAO; bottom, NATSGO data.
84 D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
85
Table 4 Soil texture data in the top two layers of soil (maximum thickness 0.5 and 1.0 m, respectively) averaged for simplified vegetation types of the conterminous United States of America Vegetation types
Sand1 (%)
Sand2 (%)
Clay1 (%)
Clay2 (%)
Depth (mm)
Rock fragments (%)
A. FAO Forest Savanna Shrub Grass Desert Tundra
44.84 41.36 46.24 39.34 51.47 48.43
39.25 36.62 46.27 34.79 57.44 47.28
26.01 28.17 21.37 27.05 18.84 23.30
32.70 31.83 21.06 28.13 15.84 25.95
1453 1417 1270 1393 1359 1392
3.32 4.35 5.83 1.52 2.55 12.49
B. NATSGOa Forest Savanna Shrub Grass Desert Tundra
36.18 31.42 34.07 30.52 39.82 32.27
34.71 29.72 31.61 29.05 37.31 30.77
22.07 26.20 27.40 27.94 28.99 27.83
27.64 31.69 33.33 31.55 31.95 34.04
1454 1367 1192 1310 1312 1203
13.46/15.53 16.85/18.69 17.72/20.15 14.77/15.44 19.58/20.63 19.04/23.26
a
NATSGO data include rock fragment content for both upper soil layers while FAO only includes one rock fragment content for the whole profile.
the surface and deeper soil layers, respectively, at the expense of sand in the NATSGO dataset.
3.1.1.1. Lifeforms and LAI. The absolute land area of shrubs increases (+ 1%) when the NATSGO data set was used and decreases (− 0.56%) when all soils are assumed to be sandy loam (Table 5, Fig. 3). These changes are small at the scale of the continent but one needs to consider that the state of Maine represents only 1.07% of the total area of the conterminous United States. To compare results, we calculated relative changes with respect to the model output using FAO soils data. Relative change as referred to in the rest of the paper is calculated as: ((other −FAO)/FAO), ‘other’ referring to the type of soil database used, other than FAO. The relative change in shrubland area increases (+ 8%) when NATSGO data are used and decreases (− 4%) when all soils are assumed to be sandy loam (Table 5). Forest relative area decreases (− 3%) when NATSGO data are used and increases ( + 3%) when all soils are assumed to be sandy loams. The most obvious changes when switching from FAO to NATSGO soils data (Fig. 4, top graph) are shifts from forest to savanna along an axis
from eastern Texas to Michigan, also in Virginia and Alabama, and in Oregon. In the western part of the country, there is a significant switch from savanna to shrubland in Washington, Oregon, northern Nevada and California. When soils are assumed to be sandy loam across the country (Fig. 4 bottom graph), there is a large switch from savanna to forest in Florida and in Oregon, northern California and Idaho. In California, Nevada and Arizona, a fraction of the desert vegetation type switches to shrubland. In Texas, the switch from grassland to savanna is more pronounced than when NATSGO data were used but is not accompanied by a switch from savanna to grassland nor one from grassland to desert like with NATSGO. Tree LAI in forests decreases (− 6%) when NATSGO soils data are used and increases (+ 2%) when all soils are assumed to be sandy loam (Table 5). Grass LAI greatly increases in forests (+ 18%) when NATSGO soils data are used and (+ 27%) when all soils are assumed to be sandy loam. In grasslands and deserts shrub LAI increases (+ 20 and + 15%, respectively) when NATSGO soils data are used and decreases (− 13 and − 17%, respectively) when all soils are assumed to be sandy loam. In savannas, grass LAI
B0.1 0.23 0.41 1.8 0.2 0.00
Grass LAI
29.14 −7.10 4.64 0.65 2.25
16.18 −8.92 −4.63 −0.51 −1.80
10 84 108 301 152 0.00
AT Grass (mm)
2.90 1.37 0.05 −3.05 −10.31 −0.37
−2.10 1.57 −1.28 1.19 15.18 −0.42
557 374 143 3 37 140
AT Tree (mm)
0.42 −2.08 0.11 1.45 −9.18
0.38 2.88 2.31 1.36 5.95 0.00
176 138 51 80 19 51
Evap. (mm)
−4.24 −8.56 −38.41 0.02 −52.79 0.19
(−4.61) (−5.50) (−35.61) (−0.55) (−0.55) (−49.81)
(2.90) (14.27) (25.23) (3.56) (−4.56) (0.18)
(33) (24) (15) (28) (9) (25) 3.27 19.99 28.47 4.91 −3.45 0.18
381 193 53 153 22 298
Runoff (mm) (% ppt)
3.86:14.26 6.91:17.81 5.16:28.92 2.21:10.78 0.03:15.67 13.39:27.70
−6.36:−14.61 −6.57:−19.77 −10.84:−26.28 −10.12:−21.24 −20.03:−25.15 −5.14:−31.87
154:280 150:272 152:248 157:289 160:277 141:251
Avail. water (mm)
AT is actual transpiration (Eq. (9) in the text). Evap. is intercepted precipitation (function of LAI) that is evaporated by the canopy. Runoff is the sum of surface runoff and base flow with, in parentheses, the fraction of annual precipitation that runoff represents. Avail. water is soil water available in the surface soil layer and in the second or intermediate soil layer (surface:intermediate). Fractional change is calculated as: (not FAO−FAO)/FAO. a All values in %.
26.79 −6.88 3.30 2.64 −0.75
0.68 B0.1 B0.1 1.8
Shrub LAI
C. SANDY LOAM: Fractional change with respect to FAOa Forest 2.82 2.00 Savanna 0.63 0.08 Shrub −4.19 1.51 Grass −0.54 −12.50 Desert −7.21 −17.18 Tundra −1.05 0.00
8.5 2.4
Tree LAI
17.86 −8.29 −6.81 −0.30 −5.58
36.51 15.94 13.37 27.53 5.69 0.95
% Land area
B. NATSGO: Fractional change with respect to FAOa Forest −3.42 −5.61 Savanna 2.76 1.57 Shrub 7.55 5.52 Grass −0.80 20.00 Desert 0.35 14.59 Tundra 0.00 0.03
A. FAO Forest Savanna Shrub Grass Desert Tundra
Vegetation types
Table 5 Average LAIs and hydrological characteristics for simplified vegetation types of the conterminous United States of America using FAO soils data (A). Percent change from results obtained with FAO are presented for (B) a model run with NATSGO soils data and (C) a model run with ubiquitous sandy loam soils
86 D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
Fig. 3. Distribution of simplified vegetation types in the conterminous United States; top left, Ku¨chler; top right, FAO; bottom left, NATSGO; bottom right, ubiquitous sandy loam soil.
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98 87
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
Fig. 4.
88
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
decreases both with NATSGO soils data and generalized sandy loam soils.
3.1.1.2. Hydrology. Interception/evaporation increases when NATSGO data are used rather than FAO data (Table 5). Concurrently, runoff predicted using NATSGO data is always greater than when using FAO data, except in deserts, and corresponds to a greater relative fraction of total rainfall (Table 5). The largest relative changes when using NATSGO data are recorded for runoff in savannas and shrublands where significant increases occur (+ 20 and +28%, respectively) (Fig. 5, top graph). This is accompanied by a decrease in grass transpiration in savannas (− 9%) and in shrublands (− 5%) and a small increase in tree transpiration in savannas ( + 1.6%). We deduce from these numbers that soil surface permeability was reduced (we concurrently observe that grass LAI decrease in both savannas and shrublands) and surface runoff was increased, decreasing the amount of available water to the plants. Indeed, we calculated a 10% decrease in available water in the top soil layer (accessible to both trees/shrubs and grasses) in both savannas and shrublands (Table 5). Decreases in available water in the second soil layer (only accessible to trees/shrubs) vary from −20% in savannas to − 26% in shrublands. The largest relative changes when all soils are assumed to be sandy loams are recorded for runoff in shrublands (−38%) and deserts (−53%) (Fig. 5, bottom graph). In deserts, both shrub transpiration (− 10%) and interception/evaporation (−9%) decrease while we concurrently observe a decrease in shrub LAI ( − 17%). In shrublands, grass LAI and grass
89
transpiration increase while shrub LAIs decrease. In general, runoff on sandy loam soils are a smaller fraction of total rainfall than runoff on FAO soils and a relatively smaller fraction of total rainfall except in grasslands and tundra
Fig. 5. Fractional changes in runoff and actual evapotranspiration with respect to the FAO soils data: top graph include results with NATSGO soils data, bottom graph includes results obtained with generalized sandy loam soils. Fractional change was calculated as: ((Soil −FAO)/FAO).
Fig. 4. Change from one simplified vegetation type to another as predicted by MAPSS using FAO versus NATSGO (top) or sandy loam (bottom) soils data. F–S = Forest to savanna when shifting from FAO to NATSGO or sandy loam soils data; F – G = Forest to grassland when shifting from FAO to sandy loam; S–F = Savanna to forest when shifting from FAO to NATSGO or sandy loam soils data; S – s = Savanna to shrubland when shifting from FAO to NATSGO or sandy loam soils data; S – G = Savanna to grassland when shifting from FAO to NATSGO or sandy loam soils data; s – S = Shrubland to savanna when shifting from FAO to NATSGO or sandy loam soils data; s–G =Shrubland to grassland when shifting from FAO to NATSGO or sandy loam soils data; s – D = Shrubland to desert when shifting from FAO to NATSGO or sandy loam soils data; G – F = Grassland to forest when shifting from FAO to NATSGO or sandy loam soils data; G– S= Grassland to savanna when shifting from FAO to NATSGO or sandy loam soils data; G–s =Grassland to shrubland when shifting from FAO to NATSGO or sandy loam soils data; G–D= Grassland to desert when shifting from FAO to NATSGO or sandy loam soils data; D – s =Desert to shrubland when shifting from FAO to NATSGO or sandy loam soils data; D – G =Desert to grassland when shifting from FAO to NATSGO or sandy loam soils data.
Fig. 6. Rock fragment content of mineral soils in the state of Oregon; top, FAO; bottom, NATSGO data for the first and second soil layer.
90 D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
Fig. 7. Depth to bedrock for mineral soils in the state of Oregon; top, FAO; bottom, NATSGO data.
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98 91
92
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
Fig. 8. Change from one simplified vegetation type to another as predicted by MAPSS using FAO versus NATSGO (top) or sandy loam (bottom) soils data. See legend for Fig. 4.
(Table 5). Total soil available water increases in all cases when soils are assumed to be sandy loams.
equal to 1.5 m in the FAO soils database while it varies from 1 to 1.5 m in the NATSGO dataset (Fig. 7).
3.2. Oregon
3.2.1.1. Lifeforms and LAI. The two major changes in vegetation distribution when shifting from FAO to NATSGO soils datasets are: (1) from forest to savannas in the coast range and, to a lesser extent, in the Cascades and north-eastern Oregon; and (2) from savannas to shrublands in central and southern Oregon–northern California (Fig. 8, top graph). When all soils are assumed to be sandy loam, major changes with regard to FAO projections are (1) from savannas to forests in the Willamette valley, the southern Cascades,
3.2.1. NATSGO 6s FAO soils data One of the major differences between FAO and NATSGO soils data is that the percentage of rock fragment is 0 for about half of the state of Oregon in the FAO soils data: mostly in southeastern Oregon, the Willamette valley and the southern coastal areas (Fig. 6). In contrast, NATSGO data indicate no significant areas without coarse fragments. Secondly, soil depth is homogenous and
Fig. 9. Distribution of simplified vegetation types in the state of Oregon; top left, Ku¨chler; top right, FAO; bottom left, NATSGO; bottom right, ubiquitous sandy loam soil.
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98 93
94
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
and northeastern Oregon; (2) from shrublands to savannas in eastern Oregon and northern California (Fig. 8, bottom graph). NATSGO soils data correspond to a smaller relative extent of forests ( −16%) and savannas ( − 4%) but a greater area of shrublands (+ 18%) and grasslands (+17%) with respect to FAO soils data (Fig. 9, Table 6). Grass LAIs in forests and savannas are much greater when NATSGO soils data are used ( +117 and +96%, respectively) while tree LAIS are lower ( − 4% and −1%) (Table 6). When all soils are assumed to be sandy loam, the extent of forests increases ( + 24%) at the expense of shrublands and grasslands (− 14 and −83%, respectively) (Fig. 9). Tree LAIs in forests are slightly greater with sandy loam soils ( + 5%) and grass LAIs are much larger in forests, savannas, and grasslands (+58%, + 44% and + 46% respectively) with respect to FAO soils data (Table 6).
3.2.1.2. Hydrology. Runoff increases (+5%) while tree transpiration ( −7%) and tree LAI ( −4%) decrease when shifting from FAO to NATSGO soils data (Table 6, Fig. 10) in forests. This means that there is less water available to the trees ( − 6% in top soil layer and −33% in the second soil layer). When all soils are assumed to be sandy loam, runoff in forests decreases (−10%) while tree transpiration (+ 8%) and tree LAI ( + 5%) increase which is probably due to the increased soil water holding capacity (Table 6, Fig. 10). Available soil water increases by 19% in the top soil layer (available to both grasses and trees/ shrubs) and by 26% in the second soil layer (only available to trees). In savannas, the increase in runoff (+50%) with NATSGO soils data is accompanied with an increase in interception/evaporation (+9%) and a decrease in tree transpiration ( − 7%), and a large concurrent increase in grass LAI. Available soil water decreases by 5% in the top soil layer and 37% in the second soil layer. When all soils are assumed to be sandy loam, runoff and interception/evaporation decrease in savannas (−34 and − 9%, respectively) which increases soil water availability both in the top and the second soil layer (+ 14 and + 18%, respectively) while grass and tree transpiration in-
crease ( + 23 and + 3%, respectively) (Table 6, Fig. 10). In shrublands, runoff increases (+ 45%) while soil water availability decreases (− 37% in the second soil layer) with NATSGO data but runoff decreases (−54%) while soil water availability increases with sandy loam soils. The decrease in runoff in grasslands (− 64%) with sandy loam soils corresponds to an increase in interception/evaporation (+ 41%), an increase in grass transpiration (+ 20%) (concurrent with an increase in grass LAI) and an increase in soil available water (+ 48% in the top soil layer and + 79% in the second soil layer) (Table 6, Fig. 10). At the scale of the country, the magnitude of the changes in runoff is similar to that of the changes in the fraction of rainfall runoff represents. In
Fig. 10. Fractional changes in runoff and actual evapotranspiration with respect to the FAO soils data in the state of Oregon: top graph include results with NATSGO soils data, bottom graph includes results obtained with generalized sandy loam soils. Fractional change was calculated as: ((Soil − FAO)/FAO).
27.55 38.73 32.83 0.12 0.77
% Land area
7.37 2.28
Tree LAI
2.28
1.00
Shrub LAI
a
B0.01 0.01 0.31 0.96
Grass LAI
58.33 44.44 4.69 45.83 −26.89 22.68 4.72 20.19
25.10 59.64 −6.08 −1.06
1 4 45 151
AT Grass (mm)
−0.29
7.67 2.72 −2.91
−0.35
−7.47 −6.69 −2.89
142
434 308 170
AT Tree (mm)
−1.66 −8.83 −9.57 40.52 0.00
0.43 9.29 6.13 −0.87 0.00
183 101 48 31 75
Evap. (mm)
−10.15 −34.17 −54.09 −64.01 0.16
(−4.60) (−24.22) (−47.57) (−25.18) (0.16)
(1.91) (27.88) (35.69) (0.01) (0.12)
(50) (37) (19) (86) (21) 5.36 50.17 44.87 −1.36 0.12
870 249 61 1131 350
Runoff (mm) (% ppt)
19.35:25.52 14.20:18.16 4.66:10.26 48.15:78.97 28.29:34.95
−6.29:−32.75 −4.64:−36.56 −9.33:−37.95 26.85:−8.09 0.83:−28.87
134:255 140:271 153:290 108:179 125:237
Avail. water (mm)
AT is actual transpiration (Eq. (9) in the text). Evap. is intercepted precipitation (function of LAI) that is evaporated by the canopy. Runoff is the sum of surface runoff and base flow with, in parentheses, the fraction of annual precipitation that runoff represents. Avail. water is soil water available in the surface soil layer and in the second soil layer (surface:intermediate). Fractional change is calculated as: (not FAO−FAO)/FAO. a All values in %.
C. Sandy loam: Fractional change with respect to FAO Forest 23.77 4.97 Savanna −4.75 −3.36 Shrub −14.01 −3.57 Grass −83.33 Tundra −1.30 0.00
a
B. NATSGO: Fractional change with respect to FAO soils data Forest −15.75 −4.20 116.67 Savanna −3.74 −1.12 96.30 Shrub 17.58 3.11 −1.57 Grass 16.67 5.91 Tundra 0.00 0.00
A. FAO Forest Savanna Shrub Grass Tundra
Vegetation types
Table 6 Average LAIs and hydrological characteristics for simplified vegetation types in the state of Oregon using FAO soils data (A). Percent change from results obtained with FAO are presented for (B) a model run with NATSGO soils data and (C) a model run with ubiquitous sandy loam soils
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98 95
96
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
Oregon, because changes in runoff are accompanied with changes in grass transpiration, they correspond to smaller changes in the fraction of total rainfall runoff represents (Table 6).
4. Discussion MAPSS simulates potential vegetation assuming that leaf area will be optimized to utilize the available soil water. When plants have ample access to water, they grow and increase their LAI values. The larger the LAI, the more water is intercepted by the canopy. The more water is intercepted, the more it evaporates from the leaf surfaces thus reducing the amount of water that can infiltrate into the soil. This feedback in the model then forces plants to reduce their LAI to optimize the amount of water that enters the soil. Within any precipitation regime, the amount of water available to plants depends on the amount of water that can infiltrate and be held in the soil. Soil texture, coarse fragment content and depth determine water availability in the soil profile. The distribution of the water in the soil profile, which depends on the characteristics of the various soil layers, will affect the competition for water between trees and grasses. MAPSS assumes trees can access water in both the top and middle layers while grasses have access only to the surface layer. It is thus extremely important to represent soil characteristics accurately to simulate the various water fluxes through the soil, plant and atmosphere and predict the type of vegetation present at any site. Differences between NATSGO and FAO soils data are significant (Table 4). Resulting changes in potential vegetation distribution may appear small on a country-wide basis (Fig. 4) but they reveal areas where either model assumptions are weak or where potential vegetation is ill-defined. In these cases, using the most accurate soil representation is essential. For example, differences in vegetation types line up along an axis from eastern Texas to Michigan. This area corresponds to the transition zone between the eastern deciduous and southeastern forests and the tall grass prairie. In the prairie peninsula region across Illinois,
annual precipitation is the same as in neighboring forests but the bulk of it falls during summer rather than being distributed throughout the year like in the forests. MAPSS associates a savanna type with the region and maintains it by assuming that periodic summer drought and fires restrict trees. There is a significant shift from forested areas to savannas when using NATSGO rather than FAO soils data. This means that there is a reduction in available water (shallower soil depth, coarser texture) such that closed forest cannot be supported any more. It also means that, in this case, NATSGO data give a better estimate of available water than FAO data. In Oregon, shallower soils (with the NATSGO soils database) in the coast range and the Western Cascades mean less available water for forests and thus a switch to savanna type vegetation which is not observed in reality. In this case, FAO data give a better estimate than NATSGO of soil water availability with the current calibration. In southern Oregon and northern California, the reduction in soil depth when using NATSGO rather than FAO soils data triggers a switch from savannas to shrublands with a more open canopy and greater grass component. So in general, using NATSGO soil characteristics reduces the available water in the soil thus triggering a switch (or coming close to do so) to a vegetation type adapted to a drier environment than what FAO soils data would lead us to believe. When soils are assumed to be sandy loams, the trend is reversed. In most cases, there is a shift to a vegetation type adapted to wetter conditions. In Florida, northern Idaho and in western Oregon (Willamette valley), there is a shift from savannas to forests. In central California, there is a shift from desert to shrubland. In central Texas, there is a shift from grasslands to savannas. On the north–south axis from Texas to Illinois, where soils are really deep and silty, there are shifts to vegetation types adapted to drier conditions (northwestern Texas, Indiana and Ohio) but they are not as extensive as with NATSGO soils data. So, we can conclude that assuming sandy loam soils would mean assuming a greater water availability than when using FAO soils data overall, with exceptions in areas where soils are finer than sandy loam.
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
The next question is: which soil dataset gives the most reliable projection of what potential vegetation really is? We cannot compare predicted potential vegetation distribution with actual vegetation records so we used Ku¨chler (1964) potential vegetation map. We simplified the classification of vegetation types to match the simple generalized types used for MAPSS. When we compare Ku¨chler’s vegetation map and the simulations, the simulations contain MAPSS-specific idiosyncrasies that are independent of soil type. For example, there is an area in north Dakota that Ku¨chler describes as a mixture of wheatgrass, bluestem and needle grass and that MAPSS calls savanna. This area shows up in the three maps corresponding to the three soil types we tested but not in Ku¨chler’s map. Another example is the overall shape of savannas along the north – south axis between Texas and Illinois. That shape is much more similar among the three maps produced by MAPSS than between any of them and Ku¨chler’s reclassified map. In this paper, we will only discuss changes that are due to a switch in soils data. At the scale of the country (Fig. 3), assuming all soils are sandy loam reduces the extent of deserts in central California which are well defined in Ku¨chler. Savannas in the Willamette valley of Oregon (Fig. 8) also disappear. Moreover, grasslands are found in Florida where Ku¨chler only defines savannas and forests. NATSGO soils data correspond to a greater extent in the deserts of the southwest especially in western Texas. Savannas are more extensive in the central grasslands of Texas with FAO soils data than with either NATSGO data or Ku¨chler’s potential vegetation. At the scale of Oregon, NATSGO data correspond to a greater extent of savannas along the southern coast than FAO soils data. Ku¨chler’s map shows no trace of savannas along the southern coast.
5. Conclusion Since Kern (1995) analyzed both NATSGO and FAO soils data sets and recommended using NATSGO data for their accuracy, questions were raised about the sensitivity of simulation models
97
to soil characterization by the two data sets. Originally, MAPSS was calibrated using a ubiquitous sandy loam soil as a first simple and easy approximation. When FAO digital soil data layers became available, the model was recalibrated with that data. Recently Kern (1995) made available the NATSGO digital data layers and the question was raised about the advantage of including the most recent US soils data as model inputs. This analysis clearly shows that the model is sensitive to soil type and that using a generalized soil type such as sandy loam decreases the goodness of fit of the predictions when compared with Ku¨chler’s map. However, the similarity between MAPSS vegetation distribution projections and Ku¨chler’s map is not clearly improved when NATSGO rather than FAO soils data are used when different regions are analyzed. But if the vegetation map accuracy does not improve by using NATSGO data, the soil database reliability does. By calibrating MAPSS with NATSGO soils data rather than FAO like it originally was, some of the inaccuracies described here are minimized.
Acknowledgements The authors want to thank Steven M. Wondzell, Christine V. Evans, Pablo H. Rosso and Christopher Daly for reviewing the manuscript. We also want to thank Ray Drapek for providing us with MAPSS output in a digital distributed format.
References Aber, J.D., Pastor, J., Melillo, J.M., 1982. Changes in forest canopy structure along a site quality gradient in southern Wisconsin. Am. Midland Naturalist 108, 256 – 265. Austin, M.P., Nicholls, A.O., Margules, C.R., 1990. Measurement of the realized niche: environmental niches of five Eucalyptus species. Ecol. Monogr. 60, 161 – 177. Brush, G.S., Lenk, C., Smith, J., 1980. The natural forests of Maryland: an explanation of the vegetation map of Maryland. Ecol. Monogr. 50 (1), 77 – 92. Food and Agriculture Organization, 1974 – 1979. Soil Map of the World, Vol. I – X. UN Educational, Scientific and Cultural Organization, Paris.
D. Bachelet et al. / Ecological Modelling 109 (1998) 77–98
98
Food and Agriculture Organization, 1978. Report on the Agro-ecological Zones Project, Vol 1. World Soil Resource Report 48. FAO, Rome. Goldberg, D., 1982. The distribution of evergreen and deciduous trees relative to soil type: an example from the Sierra Madre, Mexico, and a general model. Ecology 63 (4), 942 – 951. Kern, J.S., 1995. Geographic patterns of soil water-holding capacity in the contiguous United States. Soil Sci. Soc. Am. J. 59, 1126 – 1133. Ku¨chler, A.W., 1964. Potential natural vegetation of the conterminous United States. Spec. Publ. 36. American Geographical Society, New York, 116 pp. Marks, D., 1990. The sensitivity of potential evapotranspiration to climate change over the continental United States. In: Gucinski, H., Marks, D., Turner, D.P. (Eds.), Biospheric Feedbacks to Climate Change: The Sensitivity of Regional Trace Gas Emissions, Evapotranspiration and Energy Balance to Vegetation Redistribution. EPA 600/390/078. US EPA, Corvallis, OR, pp. IV1–31. Neilson, R.P., 1995. A model for predicting continental-scale vegetation distribution and water balance. Ecol. Appl. 5
(2), 362 – 385. Neilson, R.P., Marks, D., 1994. A global perspective of regional vegetation and hydrologic sensitivities from climate change. J. Vegetation Sci. 5, 715 – 730. Perry, D., 1994. Forest Ecosystems. Johns Hopkins University Press, Baltimore, MD, 649 pp. Prentice, I.C., Cramer, W., Harrison, S., Leemans, R., Monserud, R.A., Solomon, A.M., 1992. A global biome model based on plant physiology and dominance, soil properties and climate. J. Biogr. 19, 117 – 134. Reybold, W.U., TeSelle, G.W., 1989. Soil geographic data bases. J. Soil Water Conserv. 41, 28 – 29. Saxton, K.E., Rawls, W.J., Romberger, J.S., Papendick, R.I., 1986. Estimating generalized soil-water characteristics from texture. Soil Sci. Soc. Am. 50, 1031 – 1036. Soil Conservation Service, 1981. Land Resources Regions and Major Land Resources Areas of the US, USDA-SCS Agric. Handbook 296. US Government Printing Office, Washington, DC. Soil Conservation Service, 1987. Basic Statistics, 1982. National Resources Inventory, Stat. Bull. 756. Iowa State University, Ames.
. .
