Estudios de predicción

The panel model 1 regression results from Table 5.1 demonstrate that most of the climate variables have a significant impact on land values (except September temperature and Rainfall).

The estimated coefficients of most of the linear and quadratic terms are statistically significant.

As expected, the climate parameters across the prairies change over the seasons. Since the squared terms for temperature of different seasons have different signs, a mixture of hill shaped and U shaped responses has been implied. Also, the parameter estimates for precipitation variables such as TPTEMP, SNOW, RHJUL and RAIN all have positive squared term implying U shaped response function.

The panel model 2 regression results reveal that climate variables based on growing degree days for different seasons are not significant and does not show any significance even at

the 10% level except for May growing degree days squared and June growing degree days squared. Also, frost free days (FFD) is significant in none of the models. Rainfall (RAIN) and the evapo-transpiration proxy (TPTEMP) climate variables in this model are at the same significance level and similar in value with respect to the panel model 1. Therefore, the two model results are consistent with the understanding of the importance of precipitation in agricultural production within the prairie landscape.

5.2.1.1 Marginal Climate Impacts

Since it is difficult to interpret the linear (constant slopes) and squared coefficients (nonlinear slopes which are a function of CLIMATE variables) in raw form, Marginal Climate Impact (MCI)¹¹ for each climate variables has been calculated. Recalling equation 4.2 from section 4.3.1, if land values are expressed as a quadratic function of climate variables then the partial derivative of land value (LVAL) with respect to climate would be:

2 2 3

LV A L

CLIMA TE CLIMATE∂ β β

= +

∂ (4.3) next, taking the mean from both sides:

2 3

( LV A L ) 2 * ( )

E E CLIMA TE

CLIMA TE^∂ =β + β

∂ (5.1) which is the MCI for any climate variable. Evaluating the marginal effects of all climate variables at their mean provides the MCI for each climate variable (Table 5.2). In fact, MCI is the amount of change in land value when one unit change occurs in any climate variable. In this case, MCIs represent the change in CAD/ha of farmland value per ˚C or mm/year, evaluated at the mean annual climate for farmland in Canadian Prairies. Equation (5.1) can be calculated based on the numbers from the estimation results. Therefore, it can be tested as a restriction for

11 It is also called marginal influence, marginal value, marginal effects of climate (Mendelsohn and Reinsborough, 2007) or Ricardian climate sensitivities (Polsky, 2004).

panel model 1¹². Now, to investigate the significant level of estimated MCIs, it is necessary to run an F-test¹³ (Gujarati, 2006). All the F- statistics of the climate variables in the model are highly significant at the 1 percent level (Table 5.2).

5.2 Marginal Climate Impacts

Variable β₂ β3 _{SD MCI} F-statistic

January Temperature 15.26 -0.46 -3.80 28.14*** 54.91 April Temperature 22.04 3.05 8.43 47.38*** 31.59

July Temperature -31.70 -5.40 -14.20 -218.96*** 237.65 September Temperature 15.50 5.77 14.23 139.42*** 96.03

Rainfall 0.57 0.03 3.14 18.98*** 36.47

Snow fall -1.80 0.01 0.38 -0.10 0.07 July Relative Humidity 9.15 -0.35 -3.55 9.15*** 6.66

Evapo-transpiration Proxy 0.04 3.69×10^-7 0.00 0.04*** 5341.92

*** denotes significant at 1% level.

The estimated MCIs for the climatic variables are consistent with expectations and have intuitive signs as well. All variables, except Snowfall, are highly significant. The marginal effects of January and September temperature on land value are significant indicating that a marginal increase in temperature for these months is beneficial for prairie agriculture. In contrast, the MCI for July temperature is negative and significant; indicating that higher July temperatures will tend to decrease agricultural land value. The reason for this relationship is that the greater than the normal warming condition along with more water evaporation (due to higher

12 The restriction to test is β₂+2β₃Α = Βwhere A, and B are numerical amount.

13 The F-test to test the numerical amount of restriction (5.1) in the model can be estimated by the following way.

Taking Standard Deviation (SD) from equation 5.1gives:

( ^∂ ) 2= β3 ( )

∂ LV A L

SD SD CLIMA TE

CLIMA TE

which is presented as SD in column 4 of Table 5.4. Now, F-statistics of the joint significance is:

2

{ / 2β3 ( )}

= ×

F MCI SD CLIMATE

which is presented in the last Column of Table 5.4.

air temperature, which takes available water out of reach of plants) can cause heat stress on crops and reduces the crop productivity. This discussion about change in productivity and yields of different crop needs to be used with caution since there are different perspectives on the effects of climate change on crop yield. Tubiello et al (2007) show that among some agronomical studies on the yield effects of climate change, high temperature during the critical flowering period of a crop may lower positive CO2 effects on yield by reducing grain number, size, and quality. Also, increased temperatures during the growing period may also reduce CO2 effects indirectly, by increasing water demand. This is justifying the negative MCI for July on the prairies.

It is also important to identify that the results cannot be interpreted explicitly as land value reflecting change in yield and crop productivity. There are other regional differences that might affect agricultural land values, especially for non- agricultural based CSDs. Irrigation, livestock, and urban development are some of those regional factors that directly and indirectly might affect land value. In fact, depending on dominant activity within each CSDs (agricultural or non-agricultural base), regional factors may have significant impact on the land value. For example, agricultural land values will be affected by the metropolitan spillovers such as competition over land for a range of non-agricultural uses.

The MCI results indicate that with a temperature increase of 1˚C in April, farmland value will increase, on average, by 47 CAD per hectare, while the same increase in temperature in July will decrease land value, on average, by 219 CAD per hectare. Amongst all temperature variables, September’s temperature has the most influence on Canadian prairie agriculture (with 139 MCI) and January’s temperature has the least effect (with 25 MCI). There are no crops on the land in January, and September is harvesting time for most of the crops on the prairies.

Moreover, warmer Septembers provide longer growing season which in turn can results in greater productivity.

Since the Prairies are Canada’s main dry land, it is expected that water deficits will have significant harmful effects on agricultural production. As increasing water scarcity is a serious problem, it is also expected that there will be a positive relationship between precipitation and farmland value in CSDs where agriculture is primary driver of land values. According to the findings of this study, the Ricardian climate sensitivities (i.e. MCI) for precipitation variables are highly significant and positive in sign. Keeping all other variables constant, a 1 mm per month increase in Rain on average results in 19 CAD per hectare increase in farmland value. Moreover, RHJUL (relative humidity in July), another water related variable, is strongly significant but appears to have less strong of an impact on agriculture. Finally, TPTEMP which is a proxy for evapo-transpiration has the least influence on the land value. In fact, the results show that 1 mm/month decrease in TPTEMP (keeping temperature constant) will cause only 4 cents per hectare decrease in farmland value. Also, based on the definition of TPTEMP¹⁴, if temperature increases (holding precipitation constant), TPTMP decreases causing land value to decrease. If precipitation increases (no change in temperature) then TPTMP will rise and thereby causing agricultural land value to increase.

Several interesting results appear from the regression analysis; first, the evapo-transpiration proxy (TPTEMP), rainfall (RAIN) and July relative humidity (RHJUL) are highly significant with positive signs which are consistent with the expectation of having a direct and positive relationship between agricultural land values and water related climate variables.

Furthermore, July temperature negatively impacts land value which can be interpreted as an increase in water deficits for plants (more evaporation than normal). Again, it is consistent with

14 TPTEP= (TPERC/TEMPAV) which is total annual mean precipitation divided by total annual mean temperature.

the claim that agriculture in the Prairies is very vulnerable to the water scarcity. In summary, as agriculture production on the Canadian prairies is highly constrained by precipitation, land use and land value strongly depend on the precipitation, at least for agricultural based CSDs.