Using nighttime DMSP/OLS images of Citylights to Estimate District-level Population Distribution in Developing Countries.
3. Results of the Linear Models Estimation
3.1. Result of the linear model applied to whole of China
At first, the authors hafe estimated a linear model for whole China. Fig. 3-1 shows the correlation value of the estimated models with various thresholding rates. Form fig.3-1, it can be said that thresholding rate should be between 5 and 36% to make better estimation.
Fig.3-1 Linear model applied to whole of China
However, the maximum correlation value obtained is 0.49 and shows that the model is not practical when applying it to the whole of China to estimate the parameters.
3.2 Results of the linear model estimation applied to teach province
(1) Model fitness
we now apply the linear model to each one of the provinces of China. Fig.3-2 shows the correlation values of the model estimation versus the census population of each county. Every value is calculated with the best thresholding rate applicable for that province.
Fig.3-2 Maximum Correlation Value of each province (Linear Model)
Values vary with provinces. To find the reason or these difference , the authors have grouped all the provinces into -(i) 3 large cities ( Beijing , Tianjin, Shanghai) and (ii) other provinces . furthermore, the second group was sub-divided into (ii-a) flat provinces , and (ii-b) Mountainous provinces.
This classification cab be summarized as follows:
(i) 3 large cities
(ii) other provinces
(ii-a) Flat Provinces
(ii-b) Mountainous provinces
After grouping the counties into classes as described above , it becomes clear that the linear model can be applied to the population distributin of large cities.
In turn , fig.3-2 also implies that accuracy of the linear model is better when it is applied to flat provinces and les applicable to mountainous provinces.
Fig. 3-3 to 3-5 shoes the detailed relationship between estimated and census population of counties of a large city ( Beijing ), a flat province ( shandong ), and a mountainous province ( Ningxia ) . the preceding argument cab; be also seen in these figures .
Fig.3-3 Linear Model fitness on Beijing
Fig.3-4 Linear Model fitness on Shandong
Fig.3-5 Linear Model fitness on Ningxia
Fig.3-6 Correlation Values with the threshold (Linear Model)
(2) Effects of Thresholding rate
Fig. 3-6 shows the change of fitness in the linear model with various thresholding rates for several provinces . for most provinces, the thresholding rate has no strong relations with the fitness of the linear model.
4. Result of the Exponential Model Estimation
4.1 Result of the exponential models applied to whole of China
Fig . 4.1 is the result of the estimation from the exp9onential model considering whole China as one district . From Fig .4-1, the best thresholding rate is around 89% to Assume population distribution
Fig.4-1 Exponential model applied to whole of China
However, like the case of linear model, the correlation value is too low to estimate detailed population distributin applying this model. It is also necessary to establish different models for each province.
